An Open-Channel SSD shall be initialized before use. To initialize, we
define an on-disk format, that keeps a small set of metadata to bring up
the media manager on top of the device.
The initial step is introduced to allow a user to format the disks for a
given media manager. During format, a system block is stored on one to
three separate luns on the device. Each lun has the system block
duplicated. During initialization, the system block can be retrieved and
the appropriate media manager can initialized.
The on-disk format currently covers (struct nvm_system_block):
- Magic value "NVMS".
- Monotonic increasing sequence number.
- The physical block erase count.
- Version of the system block format.
- Media manager type.
- Media manager superblock physical address.
The interface provides three functions to manage the system block:
int nvm_init_sysblock(struct nvm_dev *, struct nvm_sb_info *)
int nvm_get_sysblock(struct nvm *dev, struct nvm_sb_info *)
int nvm_update_sysblock(struct nvm *dev, struct nvm_sb_info *)
Each implement a part of the logic to manage the system block. The
initialization creates the first system blocks and mark them on the
device. Get retrieves the latest system block by scanning all pages in
the associated system blocks. The update sysblock writes new metadata
and allocates new block if necessary.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
NAND MLC memories have both lower and upper pages. When programming,
both of these must be written, before data can be read. However,
these lower and upper pages might not placed at even and odd flash
pages, but can be skipped. Therefore each flash memory has its lower
pages defined, which can then be used when programming and to know when
padding are necessary.
This patch implements the lower page definition in the specification,
and exposes it through a simple lookup table at dev->lptbl.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Some flash media has extended capabilities, such as programming SLC
pages on MLC/TLC flash, erase/program suspend, scramble and encryption.
MCCAP is introduced to detect support for these capabilities in the
command set.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
LightNVM targets need to know the state of the flash block when doing
flash optimizations. An example is implementing a write buffer to
respect the flash page size. Currently, block state is not accounted
for; the media manager only differentiates among free, bad and in-use
blocks.
This patch adds the logic in the generic media manager to enable
targets manage blocks into open and close separately, and it implements
such management in rrpc. It also adds a set of flags to describe the
state of the block (open, closed, free, bad).
In order to avoid taking two locks (nvm_lun and rrpc_lun) consecutively,
we introduce lockless get_/put_block primitives so that the open and
close list locks and future common logic is handled within the nvm_lun
lock.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Currently, a rrpc block only points to its nvm_lun. If a user wants to
find the associated rrpc lun, it will have to calculate the index and
look it up manually. By referencing the rrpc lun directly, this step can
be omitted, at the cost of a larger memory footprint.
This is important for upcoming patches that implement write buffering in
rrpc.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Internal logic for both core and media managers, does not have a
backing bio for issuing I/Os. Introduce nvm_submit_ppa to allow raw
I/Os to be submitted to the underlying device driver.
The function request the device, ppa, data buffer and its length and
will submit the I/O synchronously to the device. The return value may
therefore be used to detect any errors regarding the issued I/O.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Instead of passing request error into the LightNVM modules, incorporate
it into the nvm_rq.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Sometimes a user want to erase multiple PPAs at the same time. Extend
nvm_erase_ppa to take multiple ppas and number of ppas to be erased.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
There is no need to check whether dev's pages per block is
beyond rrpc support every time we init a lun, we only need
to check it once before enter the lun init loop.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The Westlake controller requires that the PPA list has sectors defined
sequentially. Currently, the PPA list is created with planes first, then
sectors. Change this to sectors first, then planes.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
This patch fix two issues in rrpc_lun_gc
1. prio_list is protected by rrpc_lun's lock not nvm_lun's, so
acquire rlun's lock instead of lun's before operate on the list.
2. we delete block from prio_list before allocating gcb, but gcb
allocation may fail, we end without putting it back to the list,
this makes the block won't get reclaimed in the future. To solve
this issue, delete block after gcb allocation.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
We delete a block from the gc list before reclaim it, so
put it back to the list on its reclaim fail, otherwise
this block will not get reclaimed and be programmable
in the future.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
We should check last io completion status before
starting another one.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
To implement sync I/O support within the LightNVM core, the end_io
functions are refactored to take an end_io function pointer instead of
testing for initialized media manager, followed by calling its end_io
function.
Sync I/O can then be implemented using a callback that signal I/O
completion. This is similar to the logic found in blk_to_execute_io().
By implementing it this way, the underlying device I/Os submission logic
is abstracted away from core, targets, and media managers.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
A device may be driven in single, double or quad plane mode. In that
case, the rqd must have either one, two, or four PPAs set for a single
PPA sent to the device. Refactor this logic into their own
functions to be shared by program/erase/read in the core.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
A device may function in single, dual or quad plane mode. The gennvm
media manager manages this with explicit helpers. They convert a single
ppa to 1, 2 or 4 separate ppas in a ppa list. To aid implementation of
recovery and system blocks, this functionality can be moved directly
into the core.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
When rrpc_write_ppalist_rq and rrpc_read_ppalist_rq succeed, we setup
rq correctly, but nvm_submit_io may afterward fail since it cannot
allocate request or nvme_nvm_command, we return error but forget to
cleanup the previous work.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The mempool allocation might fail. Make sure to return error when it
does, instead of causing a kernel panic.
Signed-off-by: Javier Gonzalez <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
When initing bad block list in gennvm_block_bb, once we move bad block
from free_list to bb_list, we should maintain both stat info
nr_free_blocks and nr_bad_blocks. So this patch fixes to add missing
operation related to nr_free_blocks.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Put bio when submission fails, since we get it
before submission. And return error when backend
device driver doesn't provide a submit_io method,
thus we can end IO properly.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
dev->nr_luns reports the total number of luns available in a device
while dev->luns_per_chnl is the number of luns per channel.
When multiple channels are available, the offset is calculated from a
channel and lun id into a linear array. As it multiplies with
the total number of luns, we go out of bound when channel id > 0 and
causes the kernel to panic when we read a protected kernel memory area.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The LightNVM module exposes a debug interface when CONFIG_NVM_DEBUG is
set. This interfaces takes a string to configure media managers and
targets. Make sure this interface is only exposed when chosen
deliberately.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
After the gennvm module has been initialized. It might be attached to
one or several devices. In that case, the module is in use. Make sure
that it can not be unloaded.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
This patch fixes two issues during media manager registration.
1. The ppa pool can be used at media manager registration. Allocate the
ppa pool before that.
2. If a media manager can't be found, this should not lead to the
device being unallocated. A media manager can be registered later, that
can manage the device. Only warn if a media manager fails
initialization.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
In the case where a request queue is passed to the low lever lightnvm
device drive integration, the device driver might pass its admin
commands through another queue. Instead pass nvm_dev, and let the
low level drive the appropriate queue.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The spin_unlock is duplicated multiple times. Jump to a single unlock
to improve the code flow.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Put the allocated blocks back to the free list
when the luns configure failed, to make these
blocks useable to others.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
rrpc_get_blk use constant 0 as the input parameter
of nvm_get_blk, this may result in getting gc block
failed unexpectedly.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
To avoid race conditions, traverse dev, media manager,
and target lists and also register, unregister entries
to/from them, should be always under the nvm_lock control.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The get_bb_tbl function takes ppa as a generic address, which is
converted to the ppa device address within the device driver. When
the update_bbtbl callback is called from get_bb_tbl, the device
specific ppa is used, instead of the generic ppa.
Make sure to pass the generic ppa.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
do device max_phys_sect boundary check first, otherwise
we will allocate dma_pools for devices whose max sectors
are beyond lightnvm support and register them.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
If copy_to_user() fails we returned error but we missed releasing
devices.
Signed-off-by: Sudip Mukherjee <sudip@vectorindia.org>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Add free block, used block, and bad block information to the show debug
interface. This information is used to debug how targets track blocks.
Also, change debug function name to make it more generic.
Signed-off-by: Javier Gonzalez <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Maintain number of in use blocks, free blocks, and bad blocks in a per
lun basis. This allows the upper layers to get information about the
state of each lun.
Also, account for blocks reserved to the device on the free block count.
nr_free_blocks matches now the actual number of blocks on the free list
when the device is booted.
Signed-off-by: Javier Gonzalez <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
If either max_phys_sect is out of bound, the nvm_dev structure is not
freed.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The return value should be non-zero under error conditions.
Remove nvme_free(dev) to avoid free dev more than once.
Signed-off-by: Wenwei Tao <ww.tao0320@gmail.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
This prevents outstanding IOs to be sent for completion to target after
the target has been removed. The flow is now: stop new IOs > cleanup
queue > remove target.
Signed-off-by: Javier Gonzalez <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The linear and device specific address modes can be replaced with a
simple offset and bit length conversion that is generic across all
devices.
This both simplifies the specification and removes the special case for
qemu nvme, that previously relied on the linear address mapping.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Both the nvm_register and nvm_init does a kfree(dev) on error. Make sure
to only free it once.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
We register with nvm_devices when there registration can still fail.
Move the final registration at the end of the nvm_register function
to make sure we are fully registered when added to the nvm_devices list.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Only NAND flash with SLC and MLC is supported. Make sure to not try to
initialize TLC memory or other non-volatile memory types.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The specification was changed to reflect a multi-value bad block table.
Instead of bit-based bad block table, the bad block table now allows
eight bad block categories. Currently four are defined:
* Factory bad blocks
* Grown bad blocks
* Device-side reserved blocks
* Host-side reserved blocks
The factory and grown bad blocks are the regular bad blocks. The
reserved blocks are either for internal use or external use. In
particular, the device-side reserved blocks allows the host to
bootstrap from a limited number of flash blocks. Reducing the flash
blocks to scan upon super block initialization.
Support for both get bad block table and set bad block table is added.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
No functional changes in this patch, but it prepares us for returning
a more useful cookie related to the IO that was queued up.
Signed-off-by: Jens Axboe <axboe@fb.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Acked-by: Keith Busch <keith.busch@intel.com>
For cases where CONFIG_LBDAF is not set. The struct ppa_addr exceeds its
type on 32 bit architectures. ppa_addr requires a 64bit integer to hold
the generic ppa format. We therefore refactor it to u64 and
replaces the sector_t usages with u64 for physical addresses.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
This target allows an Open-Channel SSD to be exposed asas a block
device.
It implements a round-robin approach for sector allocation,
together with a greedy cost-based garbage collector.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
The implementation for Open-Channel SSDs is divided into media
management and targets. This patch implements a generic media manager
for open-channel SSDs. After a media manager has been initialized,
single or multiple targets can be instantiated with the media managed as
the backend.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
Open-channel SSDs are devices that share responsibilities with the host
in order to implement and maintain features that typical SSDs keep
strictly in firmware. These include (i) the Flash Translation Layer
(FTL), (ii) bad block management, and (iii) hardware units such as the
flash controller, the interface controller, and large amounts of flash
chips. In this way, Open-channels SSDs exposes direct access to their
physical flash storage, while keeping a subset of the internal features
of SSDs.
LightNVM is a specification that gives support to Open-channel SSDs
LightNVM allows the host to manage data placement, garbage collection,
and parallelism. Device specific responsibilities such as bad block
management, FTL extensions to support atomic IOs, or metadata
persistence are still handled by the device.
The implementation of LightNVM consists of two parts: core and
(multiple) targets. The core implements functionality shared across
targets. This is initialization, teardown and statistics. The targets
implement the interface that exposes physical flash to user-space
applications. Examples of such targets include key-value store,
object-store, as well as traditional block devices, which can be
application-specific.
Contributions in this patch from:
Javier Gonzalez <jg@lightnvm.io>
Dongsheng Yang <yangds.fnst@cn.fujitsu.com>
Jesper Madsen <jmad@itu.dk>
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
