UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
/*
|
|
|
|
|
* Copyright (c) International Business Machines Corp., 2006
|
|
|
|
|
* Copyright (c) Nokia Corporation, 2006, 2007
|
|
|
|
|
*
|
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
|
* (at your option) any later version.
|
|
|
|
|
*
|
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
|
|
|
|
|
* the GNU General Public License for more details.
|
|
|
|
|
*
|
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
|
* along with this program; if not, write to the Free Software
|
|
|
|
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
|
*
|
|
|
|
|
* Author: Artem Bityutskiy (Битюцкий Артём)
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* This file includes volume table manipulation code. The volume table is an
|
|
|
|
|
* on-flash table containing volume meta-data like name, number of reserved
|
|
|
|
|
* physical eraseblocks, type, etc. The volume table is stored in the so-called
|
|
|
|
|
* "layout volume".
|
|
|
|
|
*
|
|
|
|
|
* The layout volume is an internal volume which is organized as follows. It
|
|
|
|
|
* consists of two logical eraseblocks - LEB 0 and LEB 1. Each logical
|
|
|
|
|
* eraseblock stores one volume table copy, i.e. LEB 0 and LEB 1 duplicate each
|
|
|
|
|
* other. This redundancy guarantees robustness to unclean reboots. The volume
|
|
|
|
|
* table is basically an array of volume table records. Each record contains
|
|
|
|
|
* full information about the volume and protected by a CRC checksum.
|
|
|
|
|
*
|
|
|
|
|
* The volume table is changed, it is first changed in RAM. Then LEB 0 is
|
|
|
|
|
* erased, and the updated volume table is written back to LEB 0. Then same for
|
|
|
|
|
* LEB 1. This scheme guarantees recoverability from unclean reboots.
|
|
|
|
|
*
|
|
|
|
|
* In this UBI implementation the on-flash volume table does not contain any
|
|
|
|
|
* information about how many data static volumes contain. This information may
|
|
|
|
|
* be found from the scanning data.
|
|
|
|
|
*
|
|
|
|
|
* But it would still be beneficial to store this information in the volume
|
|
|
|
|
* table. For example, suppose we have a static volume X, and all its physical
|
|
|
|
|
* eraseblocks became bad for some reasons. Suppose we are attaching the
|
|
|
|
|
* corresponding MTD device, the scanning has found no logical eraseblocks
|
|
|
|
|
* corresponding to the volume X. According to the volume table volume X does
|
|
|
|
|
* exist. So we don't know whether it is just empty or all its physical
|
|
|
|
|
* eraseblocks went bad. So we cannot alarm the user about this corruption.
|
|
|
|
|
*
|
|
|
|
|
* The volume table also stores so-called "update marker", which is used for
|
|
|
|
|
* volume updates. Before updating the volume, the update marker is set, and
|
|
|
|
|
* after the update operation is finished, the update marker is cleared. So if
|
|
|
|
|
* the update operation was interrupted (e.g. by an unclean reboot) - the
|
|
|
|
|
* update marker is still there and we know that the volume's contents is
|
|
|
|
|
* damaged.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <linux/crc32.h>
|
|
|
|
|
#include <linux/err.h>
|
|
|
|
|
#include <asm/div64.h>
|
|
|
|
|
#include "ubi.h"
|
|
|
|
|
|
|
|
|
|
#ifdef CONFIG_MTD_UBI_DEBUG_PARANOID
|
|
|
|
|
static void paranoid_vtbl_check(const struct ubi_device *ubi);
|
|
|
|
|
#else
|
|
|
|
|
#define paranoid_vtbl_check(ubi)
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
/* Empty volume table record */
|
|
|
|
|
static struct ubi_vtbl_record empty_vtbl_record;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* ubi_change_vtbl_record - change volume table record.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @idx: table index to change
|
|
|
|
|
* @vtbl_rec: new volume table record
|
|
|
|
|
*
|
|
|
|
|
* This function changes volume table record @idx. If @vtbl_rec is %NULL, empty
|
|
|
|
|
* volume table record is written. The caller does not have to calculate CRC of
|
|
|
|
|
* the record as it is done by this function. Returns zero in case of success
|
|
|
|
|
* and a negative error code in case of failure.
|
|
|
|
|
*/
|
|
|
|
|
int ubi_change_vtbl_record(struct ubi_device *ubi, int idx,
|
|
|
|
|
struct ubi_vtbl_record *vtbl_rec)
|
|
|
|
|
{
|
|
|
|
|
int i, err;
|
|
|
|
|
uint32_t crc;
|
2007-12-17 01:00:38 +07:00
|
|
|
|
struct ubi_volume *layout_vol;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
ubi_assert(idx >= 0 && idx < ubi->vtbl_slots);
|
2008-01-24 16:23:23 +07:00
|
|
|
|
layout_vol = ubi->volumes[vol_id2idx(ubi, UBI_LAYOUT_VOLUME_ID)];
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
if (!vtbl_rec)
|
|
|
|
|
vtbl_rec = &empty_vtbl_record;
|
|
|
|
|
else {
|
|
|
|
|
crc = crc32(UBI_CRC32_INIT, vtbl_rec, UBI_VTBL_RECORD_SIZE_CRC);
|
2007-05-21 21:41:46 +07:00
|
|
|
|
vtbl_rec->crc = cpu_to_be32(crc);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
memcpy(&ubi->vtbl[idx], vtbl_rec, sizeof(struct ubi_vtbl_record));
|
|
|
|
|
for (i = 0; i < UBI_LAYOUT_VOLUME_EBS; i++) {
|
2007-12-17 01:00:38 +07:00
|
|
|
|
err = ubi_eba_unmap_leb(ubi, layout_vol, i);
|
2007-12-17 17:46:48 +07:00
|
|
|
|
if (err)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return err;
|
2007-12-17 17:46:48 +07:00
|
|
|
|
|
2007-12-17 01:00:38 +07:00
|
|
|
|
err = ubi_eba_write_leb(ubi, layout_vol, i, ubi->vtbl, 0,
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi->vtbl_size, UBI_LONGTERM);
|
2007-12-17 17:46:48 +07:00
|
|
|
|
if (err)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return err;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
paranoid_vtbl_check(ubi);
|
2008-02-01 18:48:49 +07:00
|
|
|
|
return 0;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
}
|
|
|
|
|
|
2008-07-14 01:47:47 +07:00
|
|
|
|
/**
|
|
|
|
|
* ubi_vtbl_rename_volumes - rename UBI volumes in the volume table.
|
|
|
|
|
* @ubi: UBI device description object
|
2008-07-18 17:34:32 +07:00
|
|
|
|
* @rename_list: list of &struct ubi_rename_entry objects
|
2008-07-14 01:47:47 +07:00
|
|
|
|
*
|
|
|
|
|
* This function re-names multiple volumes specified in @req in the volume
|
|
|
|
|
* table. Returns zero in case of success and a negative error code in case of
|
|
|
|
|
* failure.
|
|
|
|
|
*/
|
|
|
|
|
int ubi_vtbl_rename_volumes(struct ubi_device *ubi,
|
|
|
|
|
struct list_head *rename_list)
|
|
|
|
|
{
|
|
|
|
|
int i, err;
|
|
|
|
|
struct ubi_rename_entry *re;
|
|
|
|
|
struct ubi_volume *layout_vol;
|
|
|
|
|
|
|
|
|
|
list_for_each_entry(re, rename_list, list) {
|
|
|
|
|
uint32_t crc;
|
|
|
|
|
struct ubi_volume *vol = re->desc->vol;
|
|
|
|
|
struct ubi_vtbl_record *vtbl_rec = &ubi->vtbl[vol->vol_id];
|
|
|
|
|
|
|
|
|
|
if (re->remove) {
|
|
|
|
|
memcpy(vtbl_rec, &empty_vtbl_record,
|
|
|
|
|
sizeof(struct ubi_vtbl_record));
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
vtbl_rec->name_len = cpu_to_be16(re->new_name_len);
|
|
|
|
|
memcpy(vtbl_rec->name, re->new_name, re->new_name_len);
|
|
|
|
|
memset(vtbl_rec->name + re->new_name_len, 0,
|
|
|
|
|
UBI_VOL_NAME_MAX + 1 - re->new_name_len);
|
|
|
|
|
crc = crc32(UBI_CRC32_INIT, vtbl_rec,
|
|
|
|
|
UBI_VTBL_RECORD_SIZE_CRC);
|
|
|
|
|
vtbl_rec->crc = cpu_to_be32(crc);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
layout_vol = ubi->volumes[vol_id2idx(ubi, UBI_LAYOUT_VOLUME_ID)];
|
|
|
|
|
for (i = 0; i < UBI_LAYOUT_VOLUME_EBS; i++) {
|
|
|
|
|
err = ubi_eba_unmap_leb(ubi, layout_vol, i);
|
|
|
|
|
if (err)
|
|
|
|
|
return err;
|
|
|
|
|
|
|
|
|
|
err = ubi_eba_write_leb(ubi, layout_vol, i, ubi->vtbl, 0,
|
|
|
|
|
ubi->vtbl_size, UBI_LONGTERM);
|
|
|
|
|
if (err)
|
|
|
|
|
return err;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
/**
|
2008-07-18 17:34:32 +07:00
|
|
|
|
* vtbl_check - check if volume table is not corrupted and sensible.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @vtbl: volume table
|
|
|
|
|
*
|
|
|
|
|
* This function returns zero if @vtbl is all right, %1 if CRC is incorrect,
|
|
|
|
|
* and %-EINVAL if it contains inconsistent data.
|
|
|
|
|
*/
|
|
|
|
|
static int vtbl_check(const struct ubi_device *ubi,
|
|
|
|
|
const struct ubi_vtbl_record *vtbl)
|
|
|
|
|
{
|
|
|
|
|
int i, n, reserved_pebs, alignment, data_pad, vol_type, name_len;
|
2008-05-14 20:10:33 +07:00
|
|
|
|
int upd_marker, err;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
uint32_t crc;
|
|
|
|
|
const char *name;
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < ubi->vtbl_slots; i++) {
|
|
|
|
|
cond_resched();
|
|
|
|
|
|
2007-05-21 21:41:46 +07:00
|
|
|
|
reserved_pebs = be32_to_cpu(vtbl[i].reserved_pebs);
|
|
|
|
|
alignment = be32_to_cpu(vtbl[i].alignment);
|
|
|
|
|
data_pad = be32_to_cpu(vtbl[i].data_pad);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
upd_marker = vtbl[i].upd_marker;
|
|
|
|
|
vol_type = vtbl[i].vol_type;
|
2007-05-21 21:41:46 +07:00
|
|
|
|
name_len = be16_to_cpu(vtbl[i].name_len);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
name = &vtbl[i].name[0];
|
|
|
|
|
|
|
|
|
|
crc = crc32(UBI_CRC32_INIT, &vtbl[i], UBI_VTBL_RECORD_SIZE_CRC);
|
2007-05-21 21:41:46 +07:00
|
|
|
|
if (be32_to_cpu(vtbl[i].crc) != crc) {
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi_err("bad CRC at record %u: %#08x, not %#08x",
|
2007-05-21 21:41:46 +07:00
|
|
|
|
i, crc, be32_to_cpu(vtbl[i].crc));
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi_dbg_dump_vtbl_record(&vtbl[i], i);
|
|
|
|
|
return 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (reserved_pebs == 0) {
|
|
|
|
|
if (memcmp(&vtbl[i], &empty_vtbl_record,
|
|
|
|
|
UBI_VTBL_RECORD_SIZE)) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 2;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (reserved_pebs < 0 || alignment < 0 || data_pad < 0 ||
|
|
|
|
|
name_len < 0) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 3;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (alignment > ubi->leb_size || alignment == 0) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 4;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
2008-05-22 08:32:18 +07:00
|
|
|
|
n = alignment & (ubi->min_io_size - 1);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (alignment != 1 && n) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 5;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
n = ubi->leb_size % alignment;
|
|
|
|
|
if (data_pad != n) {
|
|
|
|
|
dbg_err("bad data_pad, has to be %d", n);
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 6;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (vol_type != UBI_VID_DYNAMIC && vol_type != UBI_VID_STATIC) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 7;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (upd_marker != 0 && upd_marker != 1) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 8;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (reserved_pebs > ubi->good_peb_count) {
|
2008-10-09 02:56:24 +07:00
|
|
|
|
dbg_err("too large reserved_pebs %d, good PEBs %d",
|
|
|
|
|
reserved_pebs, ubi->good_peb_count);
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 9;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (name_len > UBI_VOL_NAME_MAX) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 10;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (name[0] == '\0') {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 11;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (name_len != strnlen(name, name_len + 1)) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 12;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Checks that all names are unique */
|
|
|
|
|
for (i = 0; i < ubi->vtbl_slots - 1; i++) {
|
|
|
|
|
for (n = i + 1; n < ubi->vtbl_slots; n++) {
|
2007-05-21 21:41:46 +07:00
|
|
|
|
int len1 = be16_to_cpu(vtbl[i].name_len);
|
|
|
|
|
int len2 = be16_to_cpu(vtbl[n].name_len);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
if (len1 > 0 && len1 == len2 &&
|
|
|
|
|
!strncmp(vtbl[i].name, vtbl[n].name, len1)) {
|
|
|
|
|
ubi_err("volumes %d and %d have the same name"
|
|
|
|
|
" \"%s\"", i, n, vtbl[i].name);
|
|
|
|
|
ubi_dbg_dump_vtbl_record(&vtbl[i], i);
|
|
|
|
|
ubi_dbg_dump_vtbl_record(&vtbl[n], n);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
bad:
|
2008-05-14 20:10:33 +07:00
|
|
|
|
ubi_err("volume table check failed: record %d, error %d", i, err);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi_dbg_dump_vtbl_record(&vtbl[i], i);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* create_vtbl - create a copy of volume table.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
* @copy: number of the volume table copy
|
|
|
|
|
* @vtbl: contents of the volume table
|
|
|
|
|
*
|
|
|
|
|
* This function returns zero in case of success and a negative error code in
|
|
|
|
|
* case of failure.
|
|
|
|
|
*/
|
2007-08-29 18:51:52 +07:00
|
|
|
|
static int create_vtbl(struct ubi_device *ubi, struct ubi_scan_info *si,
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
int copy, void *vtbl)
|
|
|
|
|
{
|
|
|
|
|
int err, tries = 0;
|
|
|
|
|
static struct ubi_vid_hdr *vid_hdr;
|
|
|
|
|
struct ubi_scan_volume *sv;
|
|
|
|
|
struct ubi_scan_leb *new_seb, *old_seb = NULL;
|
|
|
|
|
|
|
|
|
|
ubi_msg("create volume table (copy #%d)", copy + 1);
|
|
|
|
|
|
2007-08-29 01:29:32 +07:00
|
|
|
|
vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (!vid_hdr)
|
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Check if there is a logical eraseblock which would have to contain
|
|
|
|
|
* this volume table copy was found during scanning. It has to be wiped
|
|
|
|
|
* out.
|
|
|
|
|
*/
|
2008-01-24 16:23:23 +07:00
|
|
|
|
sv = ubi_scan_find_sv(si, UBI_LAYOUT_VOLUME_ID);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (sv)
|
|
|
|
|
old_seb = ubi_scan_find_seb(sv, copy);
|
|
|
|
|
|
|
|
|
|
retry:
|
|
|
|
|
new_seb = ubi_scan_get_free_peb(ubi, si);
|
|
|
|
|
if (IS_ERR(new_seb)) {
|
|
|
|
|
err = PTR_ERR(new_seb);
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
vid_hdr->vol_type = UBI_VID_DYNAMIC;
|
2008-01-24 16:23:23 +07:00
|
|
|
|
vid_hdr->vol_id = cpu_to_be32(UBI_LAYOUT_VOLUME_ID);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
vid_hdr->compat = UBI_LAYOUT_VOLUME_COMPAT;
|
|
|
|
|
vid_hdr->data_size = vid_hdr->used_ebs =
|
2007-05-21 21:41:46 +07:00
|
|
|
|
vid_hdr->data_pad = cpu_to_be32(0);
|
|
|
|
|
vid_hdr->lnum = cpu_to_be32(copy);
|
|
|
|
|
vid_hdr->sqnum = cpu_to_be64(++si->max_sqnum);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
/* The EC header is already there, write the VID header */
|
|
|
|
|
err = ubi_io_write_vid_hdr(ubi, new_seb->pnum, vid_hdr);
|
|
|
|
|
if (err)
|
|
|
|
|
goto write_error;
|
|
|
|
|
|
|
|
|
|
/* Write the layout volume contents */
|
|
|
|
|
err = ubi_io_write_data(ubi, vtbl, new_seb->pnum, 0, ubi->vtbl_size);
|
|
|
|
|
if (err)
|
|
|
|
|
goto write_error;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* And add it to the scanning information. Don't delete the old
|
|
|
|
|
* @old_seb as it will be deleted and freed in 'ubi_scan_add_used()'.
|
|
|
|
|
*/
|
|
|
|
|
err = ubi_scan_add_used(ubi, si, new_seb->pnum, new_seb->ec,
|
|
|
|
|
vid_hdr, 0);
|
|
|
|
|
kfree(new_seb);
|
|
|
|
|
ubi_free_vid_hdr(ubi, vid_hdr);
|
|
|
|
|
return err;
|
|
|
|
|
|
|
|
|
|
write_error:
|
2007-05-05 15:24:02 +07:00
|
|
|
|
if (err == -EIO && ++tries <= 5) {
|
|
|
|
|
/*
|
|
|
|
|
* Probably this physical eraseblock went bad, try to pick
|
|
|
|
|
* another one.
|
|
|
|
|
*/
|
|
|
|
|
list_add_tail(&new_seb->u.list, &si->corr);
|
2007-05-03 22:49:57 +07:00
|
|
|
|
goto retry;
|
2007-05-05 15:24:02 +07:00
|
|
|
|
}
|
|
|
|
|
kfree(new_seb);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
out_free:
|
|
|
|
|
ubi_free_vid_hdr(ubi, vid_hdr);
|
|
|
|
|
return err;
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* process_lvol - process the layout volume.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
* @sv: layout volume scanning information
|
|
|
|
|
*
|
|
|
|
|
* This function is responsible for reading the layout volume, ensuring it is
|
|
|
|
|
* not corrupted, and recovering from corruptions if needed. Returns volume
|
|
|
|
|
* table in case of success and a negative error code in case of failure.
|
|
|
|
|
*/
|
2007-08-29 18:51:52 +07:00
|
|
|
|
static struct ubi_vtbl_record *process_lvol(struct ubi_device *ubi,
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
struct ubi_scan_info *si,
|
|
|
|
|
struct ubi_scan_volume *sv)
|
|
|
|
|
{
|
|
|
|
|
int err;
|
|
|
|
|
struct rb_node *rb;
|
|
|
|
|
struct ubi_scan_leb *seb;
|
|
|
|
|
struct ubi_vtbl_record *leb[UBI_LAYOUT_VOLUME_EBS] = { NULL, NULL };
|
|
|
|
|
int leb_corrupted[UBI_LAYOUT_VOLUME_EBS] = {1, 1};
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* UBI goes through the following steps when it changes the layout
|
|
|
|
|
* volume:
|
|
|
|
|
* a. erase LEB 0;
|
|
|
|
|
* b. write new data to LEB 0;
|
|
|
|
|
* c. erase LEB 1;
|
|
|
|
|
* d. write new data to LEB 1.
|
|
|
|
|
*
|
|
|
|
|
* Before the change, both LEBs contain the same data.
|
|
|
|
|
*
|
|
|
|
|
* Due to unclean reboots, the contents of LEB 0 may be lost, but there
|
|
|
|
|
* should LEB 1. So it is OK if LEB 0 is corrupted while LEB 1 is not.
|
|
|
|
|
* Similarly, LEB 1 may be lost, but there should be LEB 0. And
|
|
|
|
|
* finally, unclean reboots may result in a situation when neither LEB
|
|
|
|
|
* 0 nor LEB 1 are corrupted, but they are different. In this case, LEB
|
|
|
|
|
* 0 contains more recent information.
|
|
|
|
|
*
|
|
|
|
|
* So the plan is to first check LEB 0. Then
|
|
|
|
|
* a. if LEB 0 is OK, it must be containing the most resent data; then
|
|
|
|
|
* we compare it with LEB 1, and if they are different, we copy LEB
|
|
|
|
|
* 0 to LEB 1;
|
|
|
|
|
* b. if LEB 0 is corrupted, but LEB 1 has to be OK, and we copy LEB 1
|
|
|
|
|
* to LEB 0.
|
|
|
|
|
*/
|
|
|
|
|
|
2008-07-16 21:40:22 +07:00
|
|
|
|
dbg_gen("check layout volume");
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
/* Read both LEB 0 and LEB 1 into memory */
|
|
|
|
|
ubi_rb_for_each_entry(rb, seb, &sv->root, u.rb) {
|
2007-05-06 20:12:54 +07:00
|
|
|
|
leb[seb->lnum] = vmalloc(ubi->vtbl_size);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (!leb[seb->lnum]) {
|
|
|
|
|
err = -ENOMEM;
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
2007-05-06 20:12:54 +07:00
|
|
|
|
memset(leb[seb->lnum], 0, ubi->vtbl_size);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
err = ubi_io_read_data(ubi, leb[seb->lnum], seb->pnum, 0,
|
|
|
|
|
ubi->vtbl_size);
|
|
|
|
|
if (err == UBI_IO_BITFLIPS || err == -EBADMSG)
|
2008-05-20 13:54:02 +07:00
|
|
|
|
/*
|
|
|
|
|
* Scrub the PEB later. Note, -EBADMSG indicates an
|
|
|
|
|
* uncorrectable ECC error, but we have our own CRC and
|
|
|
|
|
* the data will be checked later. If the data is OK,
|
|
|
|
|
* the PEB will be scrubbed (because we set
|
|
|
|
|
* seb->scrub). If the data is not OK, the contents of
|
|
|
|
|
* the PEB will be recovered from the second copy, and
|
|
|
|
|
* seb->scrub will be cleared in
|
|
|
|
|
* 'ubi_scan_add_used()'.
|
|
|
|
|
*/
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
seb->scrub = 1;
|
|
|
|
|
else if (err)
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
err = -EINVAL;
|
|
|
|
|
if (leb[0]) {
|
|
|
|
|
leb_corrupted[0] = vtbl_check(ubi, leb[0]);
|
|
|
|
|
if (leb_corrupted[0] < 0)
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!leb_corrupted[0]) {
|
|
|
|
|
/* LEB 0 is OK */
|
|
|
|
|
if (leb[1])
|
2008-07-18 17:19:52 +07:00
|
|
|
|
leb_corrupted[1] = memcmp(leb[0], leb[1],
|
|
|
|
|
ubi->vtbl_size);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (leb_corrupted[1]) {
|
|
|
|
|
ubi_warn("volume table copy #2 is corrupted");
|
|
|
|
|
err = create_vtbl(ubi, si, 1, leb[0]);
|
|
|
|
|
if (err)
|
|
|
|
|
goto out_free;
|
|
|
|
|
ubi_msg("volume table was restored");
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Both LEB 1 and LEB 2 are OK and consistent */
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vfree(leb[1]);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return leb[0];
|
|
|
|
|
} else {
|
|
|
|
|
/* LEB 0 is corrupted or does not exist */
|
|
|
|
|
if (leb[1]) {
|
|
|
|
|
leb_corrupted[1] = vtbl_check(ubi, leb[1]);
|
|
|
|
|
if (leb_corrupted[1] < 0)
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
|
|
|
|
if (leb_corrupted[1]) {
|
|
|
|
|
/* Both LEB 0 and LEB 1 are corrupted */
|
|
|
|
|
ubi_err("both volume tables are corrupted");
|
|
|
|
|
goto out_free;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ubi_warn("volume table copy #1 is corrupted");
|
|
|
|
|
err = create_vtbl(ubi, si, 0, leb[1]);
|
|
|
|
|
if (err)
|
|
|
|
|
goto out_free;
|
|
|
|
|
ubi_msg("volume table was restored");
|
|
|
|
|
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vfree(leb[0]);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return leb[1];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
out_free:
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vfree(leb[0]);
|
|
|
|
|
vfree(leb[1]);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return ERR_PTR(err);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* create_empty_lvol - create empty layout volume.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
*
|
|
|
|
|
* This function returns volume table contents in case of success and a
|
|
|
|
|
* negative error code in case of failure.
|
|
|
|
|
*/
|
2007-08-29 18:51:52 +07:00
|
|
|
|
static struct ubi_vtbl_record *create_empty_lvol(struct ubi_device *ubi,
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
struct ubi_scan_info *si)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
struct ubi_vtbl_record *vtbl;
|
|
|
|
|
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vtbl = vmalloc(ubi->vtbl_size);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (!vtbl)
|
|
|
|
|
return ERR_PTR(-ENOMEM);
|
2007-05-06 20:12:54 +07:00
|
|
|
|
memset(vtbl, 0, ubi->vtbl_size);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
for (i = 0; i < ubi->vtbl_slots; i++)
|
|
|
|
|
memcpy(&vtbl[i], &empty_vtbl_record, UBI_VTBL_RECORD_SIZE);
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < UBI_LAYOUT_VOLUME_EBS; i++) {
|
|
|
|
|
int err;
|
|
|
|
|
|
|
|
|
|
err = create_vtbl(ubi, si, i, vtbl);
|
|
|
|
|
if (err) {
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vfree(vtbl);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return ERR_PTR(err);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return vtbl;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* init_volumes - initialize volume information for existing volumes.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
* @vtbl: volume table
|
|
|
|
|
*
|
|
|
|
|
* This function allocates volume description objects for existing volumes.
|
|
|
|
|
* Returns zero in case of success and a negative error code in case of
|
|
|
|
|
* failure.
|
|
|
|
|
*/
|
|
|
|
|
static int init_volumes(struct ubi_device *ubi, const struct ubi_scan_info *si,
|
|
|
|
|
const struct ubi_vtbl_record *vtbl)
|
|
|
|
|
{
|
|
|
|
|
int i, reserved_pebs = 0;
|
|
|
|
|
struct ubi_scan_volume *sv;
|
|
|
|
|
struct ubi_volume *vol;
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < ubi->vtbl_slots; i++) {
|
|
|
|
|
cond_resched();
|
|
|
|
|
|
2007-05-21 21:41:46 +07:00
|
|
|
|
if (be32_to_cpu(vtbl[i].reserved_pebs) == 0)
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
continue; /* Empty record */
|
|
|
|
|
|
|
|
|
|
vol = kzalloc(sizeof(struct ubi_volume), GFP_KERNEL);
|
|
|
|
|
if (!vol)
|
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
2007-05-21 21:41:46 +07:00
|
|
|
|
vol->reserved_pebs = be32_to_cpu(vtbl[i].reserved_pebs);
|
|
|
|
|
vol->alignment = be32_to_cpu(vtbl[i].alignment);
|
|
|
|
|
vol->data_pad = be32_to_cpu(vtbl[i].data_pad);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
vol->vol_type = vtbl[i].vol_type == UBI_VID_DYNAMIC ?
|
|
|
|
|
UBI_DYNAMIC_VOLUME : UBI_STATIC_VOLUME;
|
2007-05-21 21:41:46 +07:00
|
|
|
|
vol->name_len = be16_to_cpu(vtbl[i].name_len);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
vol->usable_leb_size = ubi->leb_size - vol->data_pad;
|
|
|
|
|
memcpy(vol->name, vtbl[i].name, vol->name_len);
|
|
|
|
|
vol->name[vol->name_len] = '\0';
|
|
|
|
|
vol->vol_id = i;
|
|
|
|
|
|
UBI: add auto-resize feature
The problem: NAND flashes have different amount of initial bad physical
eraseblocks (marked as bad by the manufacturer). For example, for 256MiB
Samsung OneNAND flash there might be from 0 to 40 bad initial eraseblocks,
which is about 2%. When UBI is used as the base system, one needs to know
the exact amount of good physical eraseblocks, because this number is
needed to create the UBI image which is put to the devices during
production. But this number is not know, which forces us to use the
minimum number of good physical eraseblocks. And UBI additionally
reserves some percentage of physical eraseblocks for bad block handling
(default is 1%), so we have 1-3% of PEBs reserved at the end, depending
on the amount of initial bad PEBs. But it is desired to always have
1% (or more, depending on the configuration).
Solution: this patch adds an "auto-resize" flag to the volume table.
The volume which has the "auto-resize" flag will automatically be re-sized
(enlarged) on the first UBI initialization. UBI clears the flag when
the volume is re-sized. Only one volume may have the "auto-resize" flag.
So, the production UBI image may have one volume with "auto-resize"
flag set, and its size is automatically adjusted on the first boot
of the device.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
2008-01-16 20:44:24 +07:00
|
|
|
|
if (vtbl[i].flags & UBI_VTBL_AUTORESIZE_FLG) {
|
|
|
|
|
/* Auto re-size flag may be set only for one volume */
|
|
|
|
|
if (ubi->autoresize_vol_id != -1) {
|
|
|
|
|
ubi_err("more then one auto-resize volume (%d "
|
|
|
|
|
"and %d)", ubi->autoresize_vol_id, i);
|
2008-03-04 01:07:52 +07:00
|
|
|
|
kfree(vol);
|
UBI: add auto-resize feature
The problem: NAND flashes have different amount of initial bad physical
eraseblocks (marked as bad by the manufacturer). For example, for 256MiB
Samsung OneNAND flash there might be from 0 to 40 bad initial eraseblocks,
which is about 2%. When UBI is used as the base system, one needs to know
the exact amount of good physical eraseblocks, because this number is
needed to create the UBI image which is put to the devices during
production. But this number is not know, which forces us to use the
minimum number of good physical eraseblocks. And UBI additionally
reserves some percentage of physical eraseblocks for bad block handling
(default is 1%), so we have 1-3% of PEBs reserved at the end, depending
on the amount of initial bad PEBs. But it is desired to always have
1% (or more, depending on the configuration).
Solution: this patch adds an "auto-resize" flag to the volume table.
The volume which has the "auto-resize" flag will automatically be re-sized
(enlarged) on the first UBI initialization. UBI clears the flag when
the volume is re-sized. Only one volume may have the "auto-resize" flag.
So, the production UBI image may have one volume with "auto-resize"
flag set, and its size is automatically adjusted on the first boot
of the device.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
2008-01-16 20:44:24 +07:00
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ubi->autoresize_vol_id = i;
|
|
|
|
|
}
|
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi_assert(!ubi->volumes[i]);
|
|
|
|
|
ubi->volumes[i] = vol;
|
|
|
|
|
ubi->vol_count += 1;
|
|
|
|
|
vol->ubi = ubi;
|
|
|
|
|
reserved_pebs += vol->reserved_pebs;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* In case of dynamic volume UBI knows nothing about how many
|
|
|
|
|
* data is stored there. So assume the whole volume is used.
|
|
|
|
|
*/
|
|
|
|
|
if (vol->vol_type == UBI_DYNAMIC_VOLUME) {
|
|
|
|
|
vol->used_ebs = vol->reserved_pebs;
|
|
|
|
|
vol->last_eb_bytes = vol->usable_leb_size;
|
2007-07-10 17:04:59 +07:00
|
|
|
|
vol->used_bytes =
|
|
|
|
|
(long long)vol->used_ebs * vol->usable_leb_size;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Static volumes only */
|
|
|
|
|
sv = ubi_scan_find_sv(si, i);
|
|
|
|
|
if (!sv) {
|
|
|
|
|
/*
|
|
|
|
|
* No eraseblocks belonging to this volume found. We
|
|
|
|
|
* don't actually know whether this static volume is
|
|
|
|
|
* completely corrupted or just contains no data. And
|
|
|
|
|
* we cannot know this as long as data size is not
|
|
|
|
|
* stored on flash. So we just assume the volume is
|
|
|
|
|
* empty. FIXME: this should be handled.
|
|
|
|
|
*/
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (sv->leb_count != sv->used_ebs) {
|
|
|
|
|
/*
|
|
|
|
|
* We found a static volume which misses several
|
|
|
|
|
* eraseblocks. Treat it as corrupted.
|
|
|
|
|
*/
|
|
|
|
|
ubi_warn("static volume %d misses %d LEBs - corrupted",
|
|
|
|
|
sv->vol_id, sv->used_ebs - sv->leb_count);
|
|
|
|
|
vol->corrupted = 1;
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
vol->used_ebs = sv->used_ebs;
|
2007-07-10 17:04:59 +07:00
|
|
|
|
vol->used_bytes =
|
|
|
|
|
(long long)(vol->used_ebs - 1) * vol->usable_leb_size;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
vol->used_bytes += sv->last_data_size;
|
|
|
|
|
vol->last_eb_bytes = sv->last_data_size;
|
|
|
|
|
}
|
|
|
|
|
|
2007-12-17 20:42:57 +07:00
|
|
|
|
/* And add the layout volume */
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
vol = kzalloc(sizeof(struct ubi_volume), GFP_KERNEL);
|
|
|
|
|
if (!vol)
|
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
|
|
vol->reserved_pebs = UBI_LAYOUT_VOLUME_EBS;
|
|
|
|
|
vol->alignment = 1;
|
|
|
|
|
vol->vol_type = UBI_DYNAMIC_VOLUME;
|
|
|
|
|
vol->name_len = sizeof(UBI_LAYOUT_VOLUME_NAME) - 1;
|
|
|
|
|
memcpy(vol->name, UBI_LAYOUT_VOLUME_NAME, vol->name_len + 1);
|
|
|
|
|
vol->usable_leb_size = ubi->leb_size;
|
|
|
|
|
vol->used_ebs = vol->reserved_pebs;
|
|
|
|
|
vol->last_eb_bytes = vol->reserved_pebs;
|
2007-07-10 17:04:59 +07:00
|
|
|
|
vol->used_bytes =
|
|
|
|
|
(long long)vol->used_ebs * (ubi->leb_size - vol->data_pad);
|
2008-01-24 16:23:23 +07:00
|
|
|
|
vol->vol_id = UBI_LAYOUT_VOLUME_ID;
|
2007-12-17 20:42:57 +07:00
|
|
|
|
vol->ref_count = 1;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
ubi_assert(!ubi->volumes[i]);
|
|
|
|
|
ubi->volumes[vol_id2idx(ubi, vol->vol_id)] = vol;
|
|
|
|
|
reserved_pebs += vol->reserved_pebs;
|
|
|
|
|
ubi->vol_count += 1;
|
|
|
|
|
vol->ubi = ubi;
|
|
|
|
|
|
|
|
|
|
if (reserved_pebs > ubi->avail_pebs)
|
|
|
|
|
ubi_err("not enough PEBs, required %d, available %d",
|
|
|
|
|
reserved_pebs, ubi->avail_pebs);
|
|
|
|
|
ubi->rsvd_pebs += reserved_pebs;
|
|
|
|
|
ubi->avail_pebs -= reserved_pebs;
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* check_sv - check volume scanning information.
|
|
|
|
|
* @vol: UBI volume description object
|
|
|
|
|
* @sv: volume scanning information
|
|
|
|
|
*
|
|
|
|
|
* This function returns zero if the volume scanning information is consistent
|
|
|
|
|
* to the data read from the volume tabla, and %-EINVAL if not.
|
|
|
|
|
*/
|
|
|
|
|
static int check_sv(const struct ubi_volume *vol,
|
|
|
|
|
const struct ubi_scan_volume *sv)
|
|
|
|
|
{
|
2008-05-14 20:10:33 +07:00
|
|
|
|
int err;
|
|
|
|
|
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (sv->highest_lnum >= vol->reserved_pebs) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 1;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
if (sv->leb_count > vol->reserved_pebs) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 2;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
if (sv->vol_type != vol->vol_type) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 3;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
if (sv->used_ebs > vol->reserved_pebs) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 4;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
if (sv->data_pad != vol->data_pad) {
|
2008-05-14 20:10:33 +07:00
|
|
|
|
err = 5;
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
goto bad;
|
|
|
|
|
}
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
bad:
|
2008-05-14 20:10:33 +07:00
|
|
|
|
ubi_err("bad scanning information, error %d", err);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
ubi_dbg_dump_sv(sv);
|
|
|
|
|
ubi_dbg_dump_vol_info(vol);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* check_scanning_info - check that scanning information.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
*
|
|
|
|
|
* Even though we protect on-flash data by CRC checksums, we still don't trust
|
|
|
|
|
* the media. This function ensures that scanning information is consistent to
|
|
|
|
|
* the information read from the volume table. Returns zero if the scanning
|
|
|
|
|
* information is OK and %-EINVAL if it is not.
|
|
|
|
|
*/
|
|
|
|
|
static int check_scanning_info(const struct ubi_device *ubi,
|
|
|
|
|
struct ubi_scan_info *si)
|
|
|
|
|
{
|
|
|
|
|
int err, i;
|
|
|
|
|
struct ubi_scan_volume *sv;
|
|
|
|
|
struct ubi_volume *vol;
|
|
|
|
|
|
|
|
|
|
if (si->vols_found > UBI_INT_VOL_COUNT + ubi->vtbl_slots) {
|
|
|
|
|
ubi_err("scanning found %d volumes, maximum is %d + %d",
|
|
|
|
|
si->vols_found, UBI_INT_VOL_COUNT, ubi->vtbl_slots);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
2008-05-22 08:32:18 +07:00
|
|
|
|
if (si->highest_vol_id >= ubi->vtbl_slots + UBI_INT_VOL_COUNT &&
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
si->highest_vol_id < UBI_INTERNAL_VOL_START) {
|
|
|
|
|
ubi_err("too large volume ID %d found by scanning",
|
|
|
|
|
si->highest_vol_id);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < ubi->vtbl_slots + UBI_INT_VOL_COUNT; i++) {
|
|
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
|
|
sv = ubi_scan_find_sv(si, i);
|
|
|
|
|
vol = ubi->volumes[i];
|
|
|
|
|
if (!vol) {
|
|
|
|
|
if (sv)
|
|
|
|
|
ubi_scan_rm_volume(si, sv);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (vol->reserved_pebs == 0) {
|
|
|
|
|
ubi_assert(i < ubi->vtbl_slots);
|
|
|
|
|
|
|
|
|
|
if (!sv)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* During scanning we found a volume which does not
|
|
|
|
|
* exist according to the information in the volume
|
|
|
|
|
* table. This must have happened due to an unclean
|
|
|
|
|
* reboot while the volume was being removed. Discard
|
|
|
|
|
* these eraseblocks.
|
|
|
|
|
*/
|
|
|
|
|
ubi_msg("finish volume %d removal", sv->vol_id);
|
|
|
|
|
ubi_scan_rm_volume(si, sv);
|
|
|
|
|
} else if (sv) {
|
|
|
|
|
err = check_sv(vol, sv);
|
|
|
|
|
if (err)
|
|
|
|
|
return err;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
2008-07-18 17:34:32 +07:00
|
|
|
|
* ubi_read_volume_table - read the volume table.
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
* @si: scanning information
|
|
|
|
|
*
|
|
|
|
|
* This function reads volume table, checks it, recover from errors if needed,
|
|
|
|
|
* or creates it if needed. Returns zero in case of success and a negative
|
|
|
|
|
* error code in case of failure.
|
|
|
|
|
*/
|
|
|
|
|
int ubi_read_volume_table(struct ubi_device *ubi, struct ubi_scan_info *si)
|
|
|
|
|
{
|
|
|
|
|
int i, err;
|
|
|
|
|
struct ubi_scan_volume *sv;
|
|
|
|
|
|
2007-05-21 21:41:46 +07:00
|
|
|
|
empty_vtbl_record.crc = cpu_to_be32(0xf116c36b);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The number of supported volumes is limited by the eraseblock size
|
|
|
|
|
* and by the UBI_MAX_VOLUMES constant.
|
|
|
|
|
*/
|
|
|
|
|
ubi->vtbl_slots = ubi->leb_size / UBI_VTBL_RECORD_SIZE;
|
|
|
|
|
if (ubi->vtbl_slots > UBI_MAX_VOLUMES)
|
|
|
|
|
ubi->vtbl_slots = UBI_MAX_VOLUMES;
|
|
|
|
|
|
|
|
|
|
ubi->vtbl_size = ubi->vtbl_slots * UBI_VTBL_RECORD_SIZE;
|
|
|
|
|
ubi->vtbl_size = ALIGN(ubi->vtbl_size, ubi->min_io_size);
|
|
|
|
|
|
2008-01-24 16:23:23 +07:00
|
|
|
|
sv = ubi_scan_find_sv(si, UBI_LAYOUT_VOLUME_ID);
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
if (!sv) {
|
|
|
|
|
/*
|
|
|
|
|
* No logical eraseblocks belonging to the layout volume were
|
|
|
|
|
* found. This could mean that the flash is just empty. In
|
|
|
|
|
* this case we create empty layout volume.
|
|
|
|
|
*
|
|
|
|
|
* But if flash is not empty this must be a corruption or the
|
|
|
|
|
* MTD device just contains garbage.
|
|
|
|
|
*/
|
|
|
|
|
if (si->is_empty) {
|
|
|
|
|
ubi->vtbl = create_empty_lvol(ubi, si);
|
|
|
|
|
if (IS_ERR(ubi->vtbl))
|
|
|
|
|
return PTR_ERR(ubi->vtbl);
|
|
|
|
|
} else {
|
|
|
|
|
ubi_err("the layout volume was not found");
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
if (sv->leb_count > UBI_LAYOUT_VOLUME_EBS) {
|
|
|
|
|
/* This must not happen with proper UBI images */
|
|
|
|
|
dbg_err("too many LEBs (%d) in layout volume",
|
|
|
|
|
sv->leb_count);
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ubi->vtbl = process_lvol(ubi, si, sv);
|
|
|
|
|
if (IS_ERR(ubi->vtbl))
|
|
|
|
|
return PTR_ERR(ubi->vtbl);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ubi->avail_pebs = ubi->good_peb_count;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The layout volume is OK, initialize the corresponding in-RAM data
|
|
|
|
|
* structures.
|
|
|
|
|
*/
|
|
|
|
|
err = init_volumes(ubi, si, ubi->vtbl);
|
|
|
|
|
if (err)
|
|
|
|
|
goto out_free;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Get sure that the scanning information is consistent to the
|
|
|
|
|
* information stored in the volume table.
|
|
|
|
|
*/
|
|
|
|
|
err = check_scanning_info(ubi, si);
|
|
|
|
|
if (err)
|
|
|
|
|
goto out_free;
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
out_free:
|
2007-05-06 20:12:54 +07:00
|
|
|
|
vfree(ubi->vtbl);
|
2008-07-18 17:19:52 +07:00
|
|
|
|
for (i = 0; i < ubi->vtbl_slots + UBI_INT_VOL_COUNT; i++) {
|
|
|
|
|
kfree(ubi->volumes[i]);
|
|
|
|
|
ubi->volumes[i] = NULL;
|
|
|
|
|
}
|
UBI: Unsorted Block Images
UBI (Latin: "where?") manages multiple logical volumes on a single
flash device, specifically supporting NAND flash devices. UBI provides
a flexible partitioning concept which still allows for wear-levelling
across the whole flash device.
In a sense, UBI may be compared to the Logical Volume Manager
(LVM). Whereas LVM maps logical sector numbers to physical HDD sector
numbers, UBI maps logical eraseblocks to physical eraseblocks.
More information may be found at
http://www.linux-mtd.infradead.org/doc/ubi.html
Partitioning/Re-partitioning
An UBI volume occupies a certain number of erase blocks. This is
limited by a configured maximum volume size, which could also be
viewed as the partition size. Each individual UBI volume's size can
be changed independently of the other UBI volumes, provided that the
sum of all volume sizes doesn't exceed a certain limit.
UBI supports dynamic volumes and static volumes. Static volumes are
read-only and their contents are protected by CRC check sums.
Bad eraseblocks handling
UBI transparently handles bad eraseblocks. When a physical
eraseblock becomes bad, it is substituted by a good physical
eraseblock, and the user does not even notice this.
Scrubbing
On a NAND flash bit flips can occur on any write operation,
sometimes also on read. If bit flips persist on the device, at first
they can still be corrected by ECC, but once they accumulate,
correction will become impossible. Thus it is best to actively scrub
the affected eraseblock, by first copying it to a free eraseblock
and then erasing the original. The UBI layer performs this type of
scrubbing under the covers, transparently to the UBI volume users.
Erase Counts
UBI maintains an erase count header per eraseblock. This frees
higher-level layers (like file systems) from doing this and allows
for centralized erase count management instead. The erase counts are
used by the wear-levelling algorithm in the UBI layer. The algorithm
itself is exchangeable.
Booting from NAND
For booting directly from NAND flash the hardware must at least be
capable of fetching and executing a small portion of the NAND
flash. Some NAND flash controllers have this kind of support. They
usually limit the window to a few kilobytes in erase block 0. This
"initial program loader" (IPL) must then contain sufficient logic to
load and execute the next boot phase.
Due to bad eraseblocks, which may be randomly scattered over the
flash device, it is problematic to store the "secondary program
loader" (SPL) statically. Also, due to bit-flips it may become
corrupted over time. UBI allows to solve this problem gracefully by
storing the SPL in a small static UBI volume.
UBI volumes vs. static partitions
UBI volumes are still very similar to static MTD partitions:
* both consist of eraseblocks (logical eraseblocks in case of UBI
volumes, and physical eraseblocks in case of static partitions;
* both support three basic operations - read, write, erase.
But UBI volumes have the following advantages over traditional
static MTD partitions:
* there are no eraseblock wear-leveling constraints in case of UBI
volumes, so the user should not care about this;
* there are no bit-flips and bad eraseblocks in case of UBI volumes.
So, UBI volumes may be considered as flash devices with relaxed
restrictions.
Where can it be found?
Documentation, kernel code and applications can be found in the MTD
gits.
What are the applications for?
The applications help to create binary flash images for two purposes: pfi
files (partial flash images) for in-system update of UBI volumes, and plain
binary images, with or without OOB data in case of NAND, for a manufacturing
step. Furthermore some tools are/and will be created that allow flash content
analysis after a system has crashed..
Who did UBI?
The original ideas, where UBI is based on, were developed by Andreas
Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others
were involved too. The implementation of the kernel layer was done by Artem
B. Bityutskiy. The user-space applications and tools were written by Oliver
Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem.
Joern Engel contributed a patch which modifies JFFS2 so that it can be run on
a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander
Schmidt made some testing work as well as core functionality improvements.
Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de>
Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 15:22:22 +07:00
|
|
|
|
return err;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#ifdef CONFIG_MTD_UBI_DEBUG_PARANOID
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* paranoid_vtbl_check - check volume table.
|
|
|
|
|
* @ubi: UBI device description object
|
|
|
|
|
*/
|
|
|
|
|
static void paranoid_vtbl_check(const struct ubi_device *ubi)
|
|
|
|
|
{
|
|
|
|
|
if (vtbl_check(ubi, ubi->vtbl)) {
|
|
|
|
|
ubi_err("paranoid check failed");
|
|
|
|
|
BUG();
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#endif /* CONFIG_MTD_UBI_DEBUG_PARANOID */
|