2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* SBP2 driver (SCSI over IEEE1394)
|
2006-12-20 07:58:40 +07:00
|
|
|
*
|
2007-02-07 02:49:39 +07:00
|
|
|
* Copyright (C) 2005-2007 Kristian Hoegsberg <krh@bitplanet.net>
|
2006-12-20 07:58:40 +07:00
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* The basic structure of this driver is based on the old storage driver,
|
2007-02-07 02:49:39 +07:00
|
|
|
* drivers/ieee1394/sbp2.c, originally written by
|
|
|
|
* James Goodwin <jamesg@filanet.com>
|
|
|
|
* with later contributions and ongoing maintenance from
|
|
|
|
* Ben Collins <bcollins@debian.org>,
|
|
|
|
* Stefan Richter <stefanr@s5r6.in-berlin.de>
|
|
|
|
* and many others.
|
|
|
|
*/
|
|
|
|
|
2008-02-04 05:12:17 +07:00
|
|
|
#include <linux/blkdev.h>
|
2008-08-15 02:47:21 +07:00
|
|
|
#include <linux/bug.h>
|
2009-06-05 02:09:38 +07:00
|
|
|
#include <linux/completion.h>
|
2008-02-04 05:12:17 +07:00
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/device.h>
|
|
|
|
#include <linux/dma-mapping.h>
|
firewire: reorganize header files
The three header files of firewire-core, i.e.
"drivers/firewire/fw-device.h",
"drivers/firewire/fw-topology.h",
"drivers/firewire/fw-transaction.h",
are replaced by
"drivers/firewire/core.h",
"include/linux/firewire.h".
The latter includes everything which a firewire high-level driver (like
firewire-sbp2) needs besides linux/firewire-constants.h, while core.h
contains the rest which is needed by firewire-core itself and by low-
level drivers (card drivers) like firewire-ohci.
High-level drivers can now also reside outside of drivers/firewire
without having to add drivers/firewire to the header file search path in
makefiles. At least the firedtv driver will be such a driver.
I also considered to spread the contents of core.h over several files,
one for each .c file where the respective implementation resides. But
it turned out that most core .c files will end up including most of the
core .h files. Also, the combined core.h isn't unreasonably big, and it
will lose more of its contents to linux/firewire.h anyway soon when more
firewire drivers are added. (IP-over-1394, firedtv, and there are plans
for one or two more.)
Furthermore, fw-ohci.h is renamed to ohci.h. The name of core.h and
ohci.h is chosen with regard to name changes of the .c files in a
follow-up change.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2009-06-05 21:26:18 +07:00
|
|
|
#include <linux/firewire.h>
|
2009-06-05 02:09:38 +07:00
|
|
|
#include <linux/firewire-constants.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/jiffies.h>
|
2006-12-20 07:58:40 +07:00
|
|
|
#include <linux/kernel.h>
|
2009-06-05 02:09:38 +07:00
|
|
|
#include <linux/kref.h>
|
|
|
|
#include <linux/list.h>
|
2008-02-04 05:12:17 +07:00
|
|
|
#include <linux/mod_devicetable.h>
|
2006-12-20 07:58:40 +07:00
|
|
|
#include <linux/module.h>
|
2007-06-18 04:55:41 +07:00
|
|
|
#include <linux/moduleparam.h>
|
2006-12-28 05:49:23 +07:00
|
|
|
#include <linux/scatterlist.h>
|
2009-06-05 02:09:38 +07:00
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/spinlock.h>
|
2007-07-01 18:54:57 +07:00
|
|
|
#include <linux/string.h>
|
2007-08-13 22:48:25 +07:00
|
|
|
#include <linux/stringify.h>
|
2007-08-12 17:51:18 +07:00
|
|
|
#include <linux/workqueue.h>
|
2009-06-05 02:09:38 +07:00
|
|
|
|
|
|
|
#include <asm/byteorder.h>
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
#include <scsi/scsi.h>
|
|
|
|
#include <scsi/scsi_cmnd.h>
|
|
|
|
#include <scsi/scsi_device.h>
|
|
|
|
#include <scsi/scsi_host.h>
|
|
|
|
|
2007-06-18 04:55:41 +07:00
|
|
|
/*
|
|
|
|
* So far only bridges from Oxford Semiconductor are known to support
|
|
|
|
* concurrent logins. Depending on firmware, four or two concurrent logins
|
|
|
|
* are possible on OXFW911 and newer Oxsemi bridges.
|
|
|
|
*
|
|
|
|
* Concurrent logins are useful together with cluster filesystems.
|
|
|
|
*/
|
2012-01-13 06:02:20 +07:00
|
|
|
static bool sbp2_param_exclusive_login = 1;
|
2007-06-18 04:55:41 +07:00
|
|
|
module_param_named(exclusive_login, sbp2_param_exclusive_login, bool, 0644);
|
|
|
|
MODULE_PARM_DESC(exclusive_login, "Exclusive login to sbp2 device "
|
|
|
|
"(default = Y, use N for concurrent initiators)");
|
|
|
|
|
2007-08-13 22:48:25 +07:00
|
|
|
/*
|
|
|
|
* Flags for firmware oddities
|
|
|
|
*
|
|
|
|
* - 128kB max transfer
|
|
|
|
* Limit transfer size. Necessary for some old bridges.
|
|
|
|
*
|
|
|
|
* - 36 byte inquiry
|
|
|
|
* When scsi_mod probes the device, let the inquiry command look like that
|
|
|
|
* from MS Windows.
|
|
|
|
*
|
|
|
|
* - skip mode page 8
|
|
|
|
* Suppress sending of mode_sense for mode page 8 if the device pretends to
|
|
|
|
* support the SCSI Primary Block commands instead of Reduced Block Commands.
|
|
|
|
*
|
|
|
|
* - fix capacity
|
|
|
|
* Tell sd_mod to correct the last sector number reported by read_capacity.
|
|
|
|
* Avoids access beyond actual disk limits on devices with an off-by-one bug.
|
|
|
|
* Don't use this with devices which don't have this bug.
|
|
|
|
*
|
2008-02-04 05:04:38 +07:00
|
|
|
* - delay inquiry
|
|
|
|
* Wait extra SBP2_INQUIRY_DELAY seconds after login before SCSI inquiry.
|
|
|
|
*
|
2008-05-11 05:35:04 +07:00
|
|
|
* - power condition
|
|
|
|
* Set the power condition field in the START STOP UNIT commands sent by
|
|
|
|
* sd_mod on suspend, resume, and shutdown (if manage_start_stop is on).
|
|
|
|
* Some disks need this to spin down or to resume properly.
|
|
|
|
*
|
2007-08-13 22:48:25 +07:00
|
|
|
* - override internal blacklist
|
|
|
|
* Instead of adding to the built-in blacklist, use only the workarounds
|
|
|
|
* specified in the module load parameter.
|
|
|
|
* Useful if a blacklist entry interfered with a non-broken device.
|
|
|
|
*/
|
|
|
|
#define SBP2_WORKAROUND_128K_MAX_TRANS 0x1
|
|
|
|
#define SBP2_WORKAROUND_INQUIRY_36 0x2
|
|
|
|
#define SBP2_WORKAROUND_MODE_SENSE_8 0x4
|
|
|
|
#define SBP2_WORKAROUND_FIX_CAPACITY 0x8
|
2008-02-04 05:04:38 +07:00
|
|
|
#define SBP2_WORKAROUND_DELAY_INQUIRY 0x10
|
|
|
|
#define SBP2_INQUIRY_DELAY 12
|
2008-05-11 05:35:04 +07:00
|
|
|
#define SBP2_WORKAROUND_POWER_CONDITION 0x20
|
2007-08-13 22:48:25 +07:00
|
|
|
#define SBP2_WORKAROUND_OVERRIDE 0x100
|
|
|
|
|
|
|
|
static int sbp2_param_workarounds;
|
|
|
|
module_param_named(workarounds, sbp2_param_workarounds, int, 0644);
|
|
|
|
MODULE_PARM_DESC(workarounds, "Work around device bugs (default = 0"
|
|
|
|
", 128kB max transfer = " __stringify(SBP2_WORKAROUND_128K_MAX_TRANS)
|
|
|
|
", 36 byte inquiry = " __stringify(SBP2_WORKAROUND_INQUIRY_36)
|
|
|
|
", skip mode page 8 = " __stringify(SBP2_WORKAROUND_MODE_SENSE_8)
|
|
|
|
", fix capacity = " __stringify(SBP2_WORKAROUND_FIX_CAPACITY)
|
2008-02-04 05:04:38 +07:00
|
|
|
", delay inquiry = " __stringify(SBP2_WORKAROUND_DELAY_INQUIRY)
|
2008-05-11 05:35:04 +07:00
|
|
|
", set power condition in start stop unit = "
|
|
|
|
__stringify(SBP2_WORKAROUND_POWER_CONDITION)
|
2007-08-13 22:48:25 +07:00
|
|
|
", override internal blacklist = " __stringify(SBP2_WORKAROUND_OVERRIDE)
|
|
|
|
", or a combination)");
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
/*
|
|
|
|
* We create one struct sbp2_logical_unit per SBP-2 Logical Unit Number Entry
|
|
|
|
* and one struct scsi_device per sbp2_logical_unit.
|
|
|
|
*/
|
|
|
|
struct sbp2_logical_unit {
|
|
|
|
struct sbp2_target *tgt;
|
|
|
|
struct list_head link;
|
2006-12-20 07:58:40 +07:00
|
|
|
struct fw_address_handler address_handler;
|
|
|
|
struct list_head orb_list;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
u64 command_block_agent_address;
|
2007-08-25 19:05:28 +07:00
|
|
|
u16 lun;
|
2006-12-20 07:58:40 +07:00
|
|
|
int login_id;
|
|
|
|
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
2007-08-25 19:05:28 +07:00
|
|
|
* The generation is updated once we've logged in or reconnected
|
|
|
|
* to the logical unit. Thus, I/O to the device will automatically
|
|
|
|
* fail and get retried if it happens in a window where the device
|
|
|
|
* is not ready, e.g. after a bus reset but before we reconnect.
|
2007-05-08 07:33:32 +07:00
|
|
|
*/
|
2006-12-20 07:58:40 +07:00
|
|
|
int generation;
|
2007-02-07 02:49:34 +07:00
|
|
|
int retries;
|
2014-03-07 22:19:57 +07:00
|
|
|
work_func_t workfn;
|
2007-02-07 02:49:34 +07:00
|
|
|
struct delayed_work work;
|
2008-02-27 05:30:02 +07:00
|
|
|
bool has_sdev;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
bool blocked;
|
2006-12-20 07:58:40 +07:00
|
|
|
};
|
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
static void sbp2_queue_work(struct sbp2_logical_unit *lu, unsigned long delay)
|
|
|
|
{
|
|
|
|
queue_delayed_work(fw_workqueue, &lu->work, delay);
|
|
|
|
}
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
/*
|
|
|
|
* We create one struct sbp2_target per IEEE 1212 Unit Directory
|
|
|
|
* and one struct Scsi_Host per sbp2_target.
|
|
|
|
*/
|
|
|
|
struct sbp2_target {
|
|
|
|
struct fw_unit *unit;
|
2008-01-26 23:42:45 +07:00
|
|
|
struct list_head lu_list;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
|
|
|
u64 management_agent_address;
|
2008-03-25 02:54:28 +07:00
|
|
|
u64 guid;
|
2007-08-25 19:05:28 +07:00
|
|
|
int directory_id;
|
|
|
|
int node_id;
|
|
|
|
int address_high;
|
2008-01-26 23:42:45 +07:00
|
|
|
unsigned int workarounds;
|
2008-01-26 11:31:12 +07:00
|
|
|
unsigned int mgt_orb_timeout;
|
2009-01-25 01:41:46 +07:00
|
|
|
unsigned int max_payload;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
|
|
|
|
int dont_block; /* counter for each logical unit */
|
|
|
|
int blocked; /* ditto */
|
2007-08-25 19:05:28 +07:00
|
|
|
};
|
|
|
|
|
2011-12-21 03:34:12 +07:00
|
|
|
static struct fw_device *target_parent_device(struct sbp2_target *tgt)
|
2009-06-06 23:35:27 +07:00
|
|
|
{
|
|
|
|
return fw_parent_device(tgt->unit);
|
|
|
|
}
|
|
|
|
|
2011-12-21 03:34:12 +07:00
|
|
|
static const struct device *tgt_dev(const struct sbp2_target *tgt)
|
|
|
|
{
|
|
|
|
return &tgt->unit->device;
|
|
|
|
}
|
|
|
|
|
|
|
|
static const struct device *lu_dev(const struct sbp2_logical_unit *lu)
|
|
|
|
{
|
|
|
|
return &lu->tgt->unit->device;
|
|
|
|
}
|
|
|
|
|
2008-10-25 02:26:20 +07:00
|
|
|
/* Impossible login_id, to detect logout attempt before successful login */
|
|
|
|
#define INVALID_LOGIN_ID 0x10000
|
|
|
|
|
2009-10-08 05:39:31 +07:00
|
|
|
#define SBP2_ORB_TIMEOUT 2000U /* Timeout in ms */
|
2006-12-20 07:58:40 +07:00
|
|
|
#define SBP2_ORB_NULL 0x80000000
|
2008-03-07 13:43:31 +07:00
|
|
|
#define SBP2_RETRY_LIMIT 0xf /* 15 retries */
|
|
|
|
#define SBP2_CYCLE_LIMIT (0xc8 << 12) /* 200 125us cycles */
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2009-07-01 01:27:59 +07:00
|
|
|
/*
|
|
|
|
* There is no transport protocol limit to the CDB length, but we implement
|
|
|
|
* a fixed length only. 16 bytes is enough for disks larger than 2 TB.
|
|
|
|
*/
|
|
|
|
#define SBP2_MAX_CDB_SIZE 16
|
|
|
|
|
2008-08-15 02:47:21 +07:00
|
|
|
/*
|
2012-05-19 03:26:21 +07:00
|
|
|
* The maximum SBP-2 data buffer size is 0xffff. We quadlet-align this
|
|
|
|
* for compatibility with earlier versions of this driver.
|
2008-08-15 02:47:21 +07:00
|
|
|
*/
|
|
|
|
#define SBP2_MAX_SEG_SIZE 0xfffc
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
/* Unit directory keys */
|
2008-01-26 11:31:12 +07:00
|
|
|
#define SBP2_CSR_UNIT_CHARACTERISTICS 0x3a
|
2007-08-25 19:05:28 +07:00
|
|
|
#define SBP2_CSR_FIRMWARE_REVISION 0x3c
|
|
|
|
#define SBP2_CSR_LOGICAL_UNIT_NUMBER 0x14
|
2012-02-15 21:59:08 +07:00
|
|
|
#define SBP2_CSR_UNIT_UNIQUE_ID 0x8d
|
2007-08-25 19:05:28 +07:00
|
|
|
#define SBP2_CSR_LOGICAL_UNIT_DIRECTORY 0xd4
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
/* Management orb opcodes */
|
|
|
|
#define SBP2_LOGIN_REQUEST 0x0
|
|
|
|
#define SBP2_QUERY_LOGINS_REQUEST 0x1
|
|
|
|
#define SBP2_RECONNECT_REQUEST 0x3
|
|
|
|
#define SBP2_SET_PASSWORD_REQUEST 0x4
|
|
|
|
#define SBP2_LOGOUT_REQUEST 0x7
|
|
|
|
#define SBP2_ABORT_TASK_REQUEST 0xb
|
|
|
|
#define SBP2_ABORT_TASK_SET 0xc
|
|
|
|
#define SBP2_LOGICAL_UNIT_RESET 0xe
|
|
|
|
#define SBP2_TARGET_RESET_REQUEST 0xf
|
|
|
|
|
|
|
|
/* Offsets for command block agent registers */
|
|
|
|
#define SBP2_AGENT_STATE 0x00
|
|
|
|
#define SBP2_AGENT_RESET 0x04
|
|
|
|
#define SBP2_ORB_POINTER 0x08
|
|
|
|
#define SBP2_DOORBELL 0x10
|
|
|
|
#define SBP2_UNSOLICITED_STATUS_ENABLE 0x14
|
|
|
|
|
|
|
|
/* Status write response codes */
|
|
|
|
#define SBP2_STATUS_REQUEST_COMPLETE 0x0
|
|
|
|
#define SBP2_STATUS_TRANSPORT_FAILURE 0x1
|
|
|
|
#define SBP2_STATUS_ILLEGAL_REQUEST 0x2
|
|
|
|
#define SBP2_STATUS_VENDOR_DEPENDENT 0x3
|
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
#define STATUS_GET_ORB_HIGH(v) ((v).status & 0xffff)
|
|
|
|
#define STATUS_GET_SBP_STATUS(v) (((v).status >> 16) & 0xff)
|
|
|
|
#define STATUS_GET_LEN(v) (((v).status >> 24) & 0x07)
|
|
|
|
#define STATUS_GET_DEAD(v) (((v).status >> 27) & 0x01)
|
|
|
|
#define STATUS_GET_RESPONSE(v) (((v).status >> 28) & 0x03)
|
|
|
|
#define STATUS_GET_SOURCE(v) (((v).status >> 30) & 0x03)
|
|
|
|
#define STATUS_GET_ORB_LOW(v) ((v).orb_low)
|
|
|
|
#define STATUS_GET_DATA(v) ((v).data)
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
struct sbp2_status {
|
|
|
|
u32 status;
|
|
|
|
u32 orb_low;
|
|
|
|
u8 data[24];
|
|
|
|
};
|
|
|
|
|
|
|
|
struct sbp2_pointer {
|
2008-02-29 02:51:11 +07:00
|
|
|
__be32 high;
|
|
|
|
__be32 low;
|
2006-12-20 07:58:40 +07:00
|
|
|
};
|
|
|
|
|
|
|
|
struct sbp2_orb {
|
|
|
|
struct fw_transaction t;
|
2007-08-25 05:59:58 +07:00
|
|
|
struct kref kref;
|
2006-12-20 07:58:40 +07:00
|
|
|
dma_addr_t request_bus;
|
|
|
|
int rcode;
|
2007-05-08 07:33:34 +07:00
|
|
|
void (*callback)(struct sbp2_orb * orb, struct sbp2_status * status);
|
2006-12-20 07:58:40 +07:00
|
|
|
struct list_head link;
|
|
|
|
};
|
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
#define MANAGEMENT_ORB_LUN(v) ((v))
|
|
|
|
#define MANAGEMENT_ORB_FUNCTION(v) ((v) << 16)
|
|
|
|
#define MANAGEMENT_ORB_RECONNECT(v) ((v) << 20)
|
2007-06-18 04:55:41 +07:00
|
|
|
#define MANAGEMENT_ORB_EXCLUSIVE(v) ((v) ? 1 << 28 : 0)
|
2007-05-08 07:33:35 +07:00
|
|
|
#define MANAGEMENT_ORB_REQUEST_FORMAT(v) ((v) << 29)
|
|
|
|
#define MANAGEMENT_ORB_NOTIFY ((1) << 31)
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
#define MANAGEMENT_ORB_RESPONSE_LENGTH(v) ((v))
|
|
|
|
#define MANAGEMENT_ORB_PASSWORD_LENGTH(v) ((v) << 16)
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
struct sbp2_management_orb {
|
|
|
|
struct sbp2_orb base;
|
|
|
|
struct {
|
|
|
|
struct sbp2_pointer password;
|
|
|
|
struct sbp2_pointer response;
|
2008-02-29 02:51:11 +07:00
|
|
|
__be32 misc;
|
|
|
|
__be32 length;
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_pointer status_fifo;
|
|
|
|
} request;
|
|
|
|
__be32 response[4];
|
|
|
|
dma_addr_t response_bus;
|
|
|
|
struct completion done;
|
|
|
|
struct sbp2_status status;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct sbp2_login_response {
|
2008-02-29 02:51:11 +07:00
|
|
|
__be32 misc;
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_pointer command_block_agent;
|
2008-02-29 02:51:11 +07:00
|
|
|
__be32 reconnect_hold;
|
2006-12-20 07:58:40 +07:00
|
|
|
};
|
2007-05-08 07:33:35 +07:00
|
|
|
#define COMMAND_ORB_DATA_SIZE(v) ((v))
|
|
|
|
#define COMMAND_ORB_PAGE_SIZE(v) ((v) << 16)
|
|
|
|
#define COMMAND_ORB_PAGE_TABLE_PRESENT ((1) << 19)
|
|
|
|
#define COMMAND_ORB_MAX_PAYLOAD(v) ((v) << 20)
|
|
|
|
#define COMMAND_ORB_SPEED(v) ((v) << 24)
|
2008-02-29 02:52:02 +07:00
|
|
|
#define COMMAND_ORB_DIRECTION ((1) << 27)
|
2007-05-08 07:33:35 +07:00
|
|
|
#define COMMAND_ORB_REQUEST_FORMAT(v) ((v) << 29)
|
|
|
|
#define COMMAND_ORB_NOTIFY ((1) << 31)
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
struct sbp2_command_orb {
|
|
|
|
struct sbp2_orb base;
|
|
|
|
struct {
|
|
|
|
struct sbp2_pointer next;
|
|
|
|
struct sbp2_pointer data_descriptor;
|
2008-02-29 02:51:11 +07:00
|
|
|
__be32 misc;
|
2009-07-01 01:27:59 +07:00
|
|
|
u8 command_block[SBP2_MAX_CDB_SIZE];
|
2006-12-20 07:58:40 +07:00
|
|
|
} request;
|
|
|
|
struct scsi_cmnd *cmd;
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-07-01 18:55:31 +07:00
|
|
|
struct sbp2_pointer page_table[SG_ALL] __attribute__((aligned(8)));
|
2006-12-20 07:58:40 +07:00
|
|
|
dma_addr_t page_table_bus;
|
|
|
|
};
|
|
|
|
|
2009-01-25 01:41:46 +07:00
|
|
|
#define SBP2_ROM_VALUE_WILDCARD ~0 /* match all */
|
|
|
|
#define SBP2_ROM_VALUE_MISSING 0xff000000 /* not present in the unit dir. */
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
/*
|
|
|
|
* List of devices with known bugs.
|
|
|
|
*
|
|
|
|
* The firmware_revision field, masked with 0xffff00, is the best
|
|
|
|
* indicator for the type of bridge chip of a device. It yields a few
|
|
|
|
* false positives but this did not break correctly behaving devices
|
2009-01-25 01:41:46 +07:00
|
|
|
* so far.
|
2006-12-20 07:58:40 +07:00
|
|
|
*/
|
|
|
|
static const struct {
|
|
|
|
u32 firmware_revision;
|
|
|
|
u32 model;
|
2008-01-26 23:42:45 +07:00
|
|
|
unsigned int workarounds;
|
2006-12-20 07:58:40 +07:00
|
|
|
} sbp2_workarounds_table[] = {
|
|
|
|
/* DViCO Momobay CX-1 with TSB42AA9 bridge */ {
|
|
|
|
.firmware_revision = 0x002800,
|
|
|
|
.model = 0x001010,
|
|
|
|
.workarounds = SBP2_WORKAROUND_INQUIRY_36 |
|
2008-05-11 05:35:04 +07:00
|
|
|
SBP2_WORKAROUND_MODE_SENSE_8 |
|
|
|
|
SBP2_WORKAROUND_POWER_CONDITION,
|
2006-12-20 07:58:40 +07:00
|
|
|
},
|
2008-02-04 05:04:38 +07:00
|
|
|
/* DViCO Momobay FX-3A with TSB42AA9A bridge */ {
|
|
|
|
.firmware_revision = 0x002800,
|
|
|
|
.model = 0x000000,
|
2009-09-07 00:33:50 +07:00
|
|
|
.workarounds = SBP2_WORKAROUND_POWER_CONDITION,
|
2008-02-04 05:04:38 +07:00
|
|
|
},
|
2006-12-20 07:58:40 +07:00
|
|
|
/* Initio bridges, actually only needed for some older ones */ {
|
|
|
|
.firmware_revision = 0x000200,
|
2009-01-25 01:41:46 +07:00
|
|
|
.model = SBP2_ROM_VALUE_WILDCARD,
|
2006-12-20 07:58:40 +07:00
|
|
|
.workarounds = SBP2_WORKAROUND_INQUIRY_36,
|
|
|
|
},
|
2008-05-11 05:35:04 +07:00
|
|
|
/* PL-3507 bridge with Prolific firmware */ {
|
|
|
|
.firmware_revision = 0x012800,
|
2009-01-25 01:41:46 +07:00
|
|
|
.model = SBP2_ROM_VALUE_WILDCARD,
|
2008-05-11 05:35:04 +07:00
|
|
|
.workarounds = SBP2_WORKAROUND_POWER_CONDITION,
|
|
|
|
},
|
2006-12-20 07:58:40 +07:00
|
|
|
/* Symbios bridge */ {
|
|
|
|
.firmware_revision = 0xa0b800,
|
2009-01-25 01:41:46 +07:00
|
|
|
.model = SBP2_ROM_VALUE_WILDCARD,
|
2006-12-20 07:58:40 +07:00
|
|
|
.workarounds = SBP2_WORKAROUND_128K_MAX_TRANS,
|
|
|
|
},
|
2008-03-12 04:32:03 +07:00
|
|
|
/* Datafab MD2-FW2 with Symbios/LSILogic SYM13FW500 bridge */ {
|
|
|
|
.firmware_revision = 0x002600,
|
2009-01-25 01:41:46 +07:00
|
|
|
.model = SBP2_ROM_VALUE_WILDCARD,
|
2008-03-12 04:32:03 +07:00
|
|
|
.workarounds = SBP2_WORKAROUND_128K_MAX_TRANS,
|
|
|
|
},
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
2009-01-29 06:11:59 +07:00
|
|
|
* iPod 2nd generation: needs 128k max transfer size workaround
|
|
|
|
* iPod 3rd generation: needs fix capacity workaround
|
2007-05-08 07:33:32 +07:00
|
|
|
*/
|
2009-01-29 06:11:59 +07:00
|
|
|
{
|
|
|
|
.firmware_revision = 0x0a2700,
|
|
|
|
.model = 0x000000,
|
|
|
|
.workarounds = SBP2_WORKAROUND_128K_MAX_TRANS |
|
|
|
|
SBP2_WORKAROUND_FIX_CAPACITY,
|
|
|
|
},
|
|
|
|
/* iPod 4th generation */ {
|
2006-12-20 07:58:40 +07:00
|
|
|
.firmware_revision = 0x0a2700,
|
|
|
|
.model = 0x000021,
|
|
|
|
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
|
|
|
|
},
|
2008-11-22 18:38:58 +07:00
|
|
|
/* iPod mini */ {
|
|
|
|
.firmware_revision = 0x0a2700,
|
|
|
|
.model = 0x000022,
|
|
|
|
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
|
|
|
|
},
|
2006-12-20 07:58:40 +07:00
|
|
|
/* iPod mini */ {
|
|
|
|
.firmware_revision = 0x0a2700,
|
|
|
|
.model = 0x000023,
|
|
|
|
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
|
|
|
|
},
|
|
|
|
/* iPod Photo */ {
|
|
|
|
.firmware_revision = 0x0a2700,
|
|
|
|
.model = 0x00007e,
|
|
|
|
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void free_orb(struct kref *kref)
|
2007-08-25 05:59:58 +07:00
|
|
|
{
|
|
|
|
struct sbp2_orb *orb = container_of(kref, struct sbp2_orb, kref);
|
|
|
|
|
|
|
|
kfree(orb);
|
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void sbp2_status_write(struct fw_card *card, struct fw_request *request,
|
|
|
|
int tcode, int destination, int source,
|
2010-06-21 03:50:35 +07:00
|
|
|
int generation, unsigned long long offset,
|
2008-12-15 03:47:04 +07:00
|
|
|
void *payload, size_t length, void *callback_data)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu = callback_data;
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_orb *orb;
|
|
|
|
struct sbp2_status status;
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
if (tcode != TCODE_WRITE_BLOCK_REQUEST ||
|
2009-09-06 23:51:27 +07:00
|
|
|
length < 8 || length > sizeof(status)) {
|
2006-12-20 07:58:40 +07:00
|
|
|
fw_send_response(card, request, RCODE_TYPE_ERROR);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2009-09-06 23:51:27 +07:00
|
|
|
status.status = be32_to_cpup(payload);
|
|
|
|
status.orb_low = be32_to_cpup(payload + 4);
|
|
|
|
memset(status.data, 0, sizeof(status.data));
|
|
|
|
if (length > 8)
|
|
|
|
memcpy(status.data, payload + 8, length - 8);
|
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
if (STATUS_GET_SOURCE(status) == 2 || STATUS_GET_SOURCE(status) == 3) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(lu_dev(lu),
|
|
|
|
"non-ORB related status write, not handled\n");
|
2006-12-20 07:58:40 +07:00
|
|
|
fw_send_response(card, request, RCODE_COMPLETE);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Lookup the orb corresponding to this status write. */
|
|
|
|
spin_lock_irqsave(&card->lock, flags);
|
2007-08-25 19:05:28 +07:00
|
|
|
list_for_each_entry(orb, &lu->orb_list, link) {
|
2007-05-08 07:33:35 +07:00
|
|
|
if (STATUS_GET_ORB_HIGH(status) == 0 &&
|
2007-08-25 05:59:58 +07:00
|
|
|
STATUS_GET_ORB_LOW(status) == orb->request_bus) {
|
|
|
|
orb->rcode = RCODE_COMPLETE;
|
2006-12-20 07:58:40 +07:00
|
|
|
list_del(&orb->link);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&card->lock, flags);
|
|
|
|
|
2009-09-04 04:07:35 +07:00
|
|
|
if (&orb->link != &lu->orb_list) {
|
2006-12-20 07:58:40 +07:00
|
|
|
orb->callback(orb, &status);
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
kref_put(&orb->kref, free_orb); /* orb callback reference */
|
2009-09-04 04:07:35 +07:00
|
|
|
} else {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(lu_dev(lu), "status write for unknown ORB\n");
|
2009-09-04 04:07:35 +07:00
|
|
|
}
|
2007-08-25 05:59:58 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
fw_send_response(card, request, RCODE_COMPLETE);
|
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void complete_transaction(struct fw_card *card, int rcode,
|
|
|
|
void *payload, size_t length, void *data)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
|
|
|
struct sbp2_orb *orb = data;
|
|
|
|
unsigned long flags;
|
|
|
|
|
2007-08-25 05:59:58 +07:00
|
|
|
/*
|
|
|
|
* This is a little tricky. We can get the status write for
|
|
|
|
* the orb before we get this callback. The status write
|
|
|
|
* handler above will assume the orb pointer transaction was
|
|
|
|
* successful and set the rcode to RCODE_COMPLETE for the orb.
|
|
|
|
* So this callback only sets the rcode if it hasn't already
|
|
|
|
* been set and only does the cleanup if the transaction
|
|
|
|
* failed and we didn't already get a status write.
|
|
|
|
*/
|
|
|
|
spin_lock_irqsave(&card->lock, flags);
|
|
|
|
|
|
|
|
if (orb->rcode == -1)
|
|
|
|
orb->rcode = rcode;
|
2011-03-15 06:04:42 +07:00
|
|
|
if (orb->rcode != RCODE_COMPLETE) {
|
2006-12-20 07:58:40 +07:00
|
|
|
list_del(&orb->link);
|
2007-08-25 15:40:42 +07:00
|
|
|
spin_unlock_irqrestore(&card->lock, flags);
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
orb->callback(orb, NULL);
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
kref_put(&orb->kref, free_orb); /* orb callback reference */
|
2007-08-25 15:40:42 +07:00
|
|
|
} else {
|
|
|
|
spin_unlock_irqrestore(&card->lock, flags);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
2007-08-25 05:59:58 +07:00
|
|
|
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
kref_put(&orb->kref, free_orb); /* transaction callback reference */
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void sbp2_send_orb(struct sbp2_orb *orb, struct sbp2_logical_unit *lu,
|
|
|
|
int node_id, int generation, u64 offset)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2011-05-02 02:06:42 +07:00
|
|
|
struct sbp2_pointer orb_pointer;
|
2006-12-20 07:58:40 +07:00
|
|
|
unsigned long flags;
|
|
|
|
|
2011-05-02 02:06:42 +07:00
|
|
|
orb_pointer.high = 0;
|
|
|
|
orb_pointer.low = cpu_to_be32(orb->request_bus);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
spin_lock_irqsave(&device->card->lock, flags);
|
2007-08-25 19:05:28 +07:00
|
|
|
list_add_tail(&orb->link, &lu->orb_list);
|
2006-12-20 07:58:40 +07:00
|
|
|
spin_unlock_irqrestore(&device->card->lock, flags);
|
|
|
|
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
kref_get(&orb->kref); /* transaction callback reference */
|
|
|
|
kref_get(&orb->kref); /* orb callback reference */
|
2007-08-25 05:59:58 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
fw_send_request(device->card, &orb->t, TCODE_WRITE_BLOCK_REQUEST,
|
2007-06-11 02:31:36 +07:00
|
|
|
node_id, generation, device->max_speed, offset,
|
2011-05-02 02:06:42 +07:00
|
|
|
&orb_pointer, 8, complete_transaction, orb);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
static int sbp2_cancel_orbs(struct sbp2_logical_unit *lu)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_orb *orb, *next;
|
|
|
|
struct list_head list;
|
2007-03-08 00:12:47 +07:00
|
|
|
int retval = -ENOENT;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
INIT_LIST_HEAD(&list);
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_lock_irq(&device->card->lock);
|
2007-08-25 19:05:28 +07:00
|
|
|
list_splice_init(&lu->orb_list, &list);
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_unlock_irq(&device->card->lock);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
list_for_each_entry_safe(orb, next, &list, link) {
|
2007-03-08 00:12:47 +07:00
|
|
|
retval = 0;
|
2011-03-15 06:04:42 +07:00
|
|
|
if (fw_cancel_transaction(device->card, &orb->t) == 0)
|
|
|
|
continue;
|
2007-02-07 02:49:32 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
orb->rcode = RCODE_CANCELLED;
|
|
|
|
orb->callback(orb, NULL);
|
firewire: sbp2: fix memory leak in sbp2_cancel_orbs or at send error
When an ORB was canceled (Command ORB i.e. SCSI request timed out, or
Management ORB timed out), or there was a send error in the initial
transaction, we missed to drop one of the ORB's references and thus
leaked memory.
Background:
In total, we hold 3 references to each Operation Request Block:
- 1 during sbp2_scsi_queuecommand() or sbp2_send_management_orb()
respectively,
- 1 for the duration of the write transaction to the ORB_Pointer or
Management_Agent register of the target,
- 1 for as long as the ORB stays within the lu->orb_list, until
the ORB is unlinked from the list and the orb->callback was
executed.
The latter one of these 3 references is finished
- normally by sbp2_status_write() when the target wrote status
for a pending ORB,
- or by sbp2_cancel_orbs() in case of an ORB time-out,
- or by complete_transaction() in case of a send error.
Of them, the latter two lacked the kref_put.
Add the missing kref_put()s. Add comments to the gets and puts of
references for transaction callbacks and ORB callbacks so that it is
easier to see what is supposed to happen.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2010-08-17 02:58:03 +07:00
|
|
|
kref_put(&orb->kref, free_orb); /* orb callback reference */
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2007-03-08 00:12:47 +07:00
|
|
|
return retval;
|
2007-02-07 02:49:33 +07:00
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void complete_management_orb(struct sbp2_orb *base_orb,
|
|
|
|
struct sbp2_status *status)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
|
|
|
struct sbp2_management_orb *orb =
|
2007-06-28 03:04:33 +07:00
|
|
|
container_of(base_orb, struct sbp2_management_orb, base);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
if (status)
|
2007-05-10 06:23:14 +07:00
|
|
|
memcpy(&orb->status, status, sizeof(*status));
|
2006-12-20 07:58:40 +07:00
|
|
|
complete(&orb->done);
|
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static int sbp2_send_management_orb(struct sbp2_logical_unit *lu, int node_id,
|
|
|
|
int generation, int function,
|
|
|
|
int lun_or_login_id, void *response)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_management_orb *orb;
|
2008-01-19 19:15:05 +07:00
|
|
|
unsigned int timeout;
|
2006-12-20 07:58:40 +07:00
|
|
|
int retval = -ENOMEM;
|
|
|
|
|
2008-01-28 01:14:44 +07:00
|
|
|
if (function == SBP2_LOGOUT_REQUEST && fw_device_is_shutdown(device))
|
|
|
|
return 0;
|
|
|
|
|
2012-03-04 20:23:00 +07:00
|
|
|
orb = kzalloc(sizeof(*orb), GFP_NOIO);
|
2006-12-20 07:58:40 +07:00
|
|
|
if (orb == NULL)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2007-08-25 05:59:58 +07:00
|
|
|
kref_init(&orb->base.kref);
|
2006-12-20 07:58:40 +07:00
|
|
|
orb->response_bus =
|
|
|
|
dma_map_single(device->card->device, &orb->response,
|
2007-05-10 06:23:14 +07:00
|
|
|
sizeof(orb->response), DMA_FROM_DEVICE);
|
2008-07-26 09:44:49 +07:00
|
|
|
if (dma_mapping_error(device->card->device, orb->response_bus))
|
2007-07-03 02:04:44 +07:00
|
|
|
goto fail_mapping_response;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.response.high = 0;
|
|
|
|
orb->request.response.low = cpu_to_be32(orb->response_bus);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.misc = cpu_to_be32(
|
2007-05-08 07:33:35 +07:00
|
|
|
MANAGEMENT_ORB_NOTIFY |
|
|
|
|
MANAGEMENT_ORB_FUNCTION(function) |
|
2008-02-29 02:51:11 +07:00
|
|
|
MANAGEMENT_ORB_LUN(lun_or_login_id));
|
|
|
|
orb->request.length = cpu_to_be32(
|
|
|
|
MANAGEMENT_ORB_RESPONSE_LENGTH(sizeof(orb->response)));
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.status_fifo.high =
|
|
|
|
cpu_to_be32(lu->address_handler.offset >> 32);
|
|
|
|
orb->request.status_fifo.low =
|
|
|
|
cpu_to_be32(lu->address_handler.offset);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
if (function == SBP2_LOGIN_REQUEST) {
|
2008-01-20 07:25:31 +07:00
|
|
|
/* Ask for 2^2 == 4 seconds reconnect grace period */
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.misc |= cpu_to_be32(
|
2008-01-20 07:25:31 +07:00
|
|
|
MANAGEMENT_ORB_RECONNECT(2) |
|
2008-02-29 02:51:11 +07:00
|
|
|
MANAGEMENT_ORB_EXCLUSIVE(sbp2_param_exclusive_login));
|
2008-01-26 11:31:12 +07:00
|
|
|
timeout = lu->tgt->mgt_orb_timeout;
|
2008-01-19 19:15:05 +07:00
|
|
|
} else {
|
|
|
|
timeout = SBP2_ORB_TIMEOUT;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
init_completion(&orb->done);
|
|
|
|
orb->base.callback = complete_management_orb;
|
2007-03-08 00:12:47 +07:00
|
|
|
|
2007-07-03 02:04:44 +07:00
|
|
|
orb->base.request_bus =
|
|
|
|
dma_map_single(device->card->device, &orb->request,
|
|
|
|
sizeof(orb->request), DMA_TO_DEVICE);
|
2008-07-26 09:44:49 +07:00
|
|
|
if (dma_mapping_error(device->card->device, orb->base.request_bus))
|
2007-07-03 02:04:44 +07:00
|
|
|
goto fail_mapping_request;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
sbp2_send_orb(&orb->base, lu, node_id, generation,
|
|
|
|
lu->tgt->management_agent_address);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-01-19 19:15:05 +07:00
|
|
|
wait_for_completion_timeout(&orb->done, msecs_to_jiffies(timeout));
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
retval = -EIO;
|
2007-08-25 19:05:28 +07:00
|
|
|
if (sbp2_cancel_orbs(lu) == 0) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(lu_dev(lu), "ORB reply timed out, rcode 0x%02x\n",
|
|
|
|
orb->base.rcode);
|
2006-12-20 07:58:40 +07:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2007-03-08 00:12:47 +07:00
|
|
|
if (orb->base.rcode != RCODE_COMPLETE) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(lu_dev(lu), "management write failed, rcode 0x%02x\n",
|
|
|
|
orb->base.rcode);
|
2006-12-20 07:58:40 +07:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
if (STATUS_GET_RESPONSE(orb->status) != 0 ||
|
|
|
|
STATUS_GET_SBP_STATUS(orb->status) != 0) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(lu_dev(lu), "error status: %d:%d\n",
|
2007-05-08 07:33:35 +07:00
|
|
|
STATUS_GET_RESPONSE(orb->status),
|
|
|
|
STATUS_GET_SBP_STATUS(orb->status));
|
2006-12-20 07:58:40 +07:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
retval = 0;
|
|
|
|
out:
|
|
|
|
dma_unmap_single(device->card->device, orb->base.request_bus,
|
2007-05-10 06:23:14 +07:00
|
|
|
sizeof(orb->request), DMA_TO_DEVICE);
|
2007-07-03 02:04:44 +07:00
|
|
|
fail_mapping_request:
|
2006-12-20 07:58:40 +07:00
|
|
|
dma_unmap_single(device->card->device, orb->response_bus,
|
2007-05-10 06:23:14 +07:00
|
|
|
sizeof(orb->response), DMA_FROM_DEVICE);
|
2007-07-03 02:04:44 +07:00
|
|
|
fail_mapping_response:
|
2006-12-20 07:58:40 +07:00
|
|
|
if (response)
|
2008-02-29 02:51:11 +07:00
|
|
|
memcpy(response, orb->response, sizeof(orb->response));
|
2007-08-25 05:59:58 +07:00
|
|
|
kref_put(&orb->base.kref, free_orb);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2008-02-04 05:08:58 +07:00
|
|
|
static void sbp2_agent_reset(struct sbp2_logical_unit *lu)
|
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2008-07-20 19:20:53 +07:00
|
|
|
__be32 d = 0;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-07-20 19:20:53 +07:00
|
|
|
fw_run_transaction(device->card, TCODE_WRITE_QUADLET_REQUEST,
|
|
|
|
lu->tgt->node_id, lu->generation, device->max_speed,
|
|
|
|
lu->command_block_agent_address + SBP2_AGENT_RESET,
|
2010-06-13 01:29:07 +07:00
|
|
|
&d, 4);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void complete_agent_reset_write_no_wait(struct fw_card *card,
|
|
|
|
int rcode, void *payload, size_t length, void *data)
|
2008-02-04 05:08:58 +07:00
|
|
|
{
|
|
|
|
kfree(data);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void sbp2_agent_reset_no_wait(struct sbp2_logical_unit *lu)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
struct fw_transaction *t;
|
2008-07-20 19:20:53 +07:00
|
|
|
static __be32 d;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-02-04 05:08:58 +07:00
|
|
|
t = kmalloc(sizeof(*t), GFP_ATOMIC);
|
2006-12-20 07:58:40 +07:00
|
|
|
if (t == NULL)
|
2008-02-04 05:08:58 +07:00
|
|
|
return;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
fw_send_request(device->card, t, TCODE_WRITE_QUADLET_REQUEST,
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->tgt->node_id, lu->generation, device->max_speed,
|
|
|
|
lu->command_block_agent_address + SBP2_AGENT_RESET,
|
2010-06-13 01:29:07 +07:00
|
|
|
&d, 4, complete_agent_reset_write_no_wait, t);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
static inline void sbp2_allow_block(struct sbp2_logical_unit *lu)
|
|
|
|
{
|
2014-03-04 05:22:35 +07:00
|
|
|
struct sbp2_target *tgt = lu->tgt;
|
|
|
|
struct fw_card *card = target_parent_device(tgt)->card;
|
|
|
|
|
|
|
|
spin_lock_irq(&card->lock);
|
|
|
|
--tgt->dont_block;
|
|
|
|
spin_unlock_irq(&card->lock);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Blocks lu->tgt if all of the following conditions are met:
|
|
|
|
* - Login, INQUIRY, and high-level SCSI setup of all of the target's
|
|
|
|
* logical units have been finished (indicated by dont_block == 0).
|
|
|
|
* - lu->generation is stale.
|
|
|
|
*
|
|
|
|
* Note, scsi_block_requests() must be called while holding card->lock,
|
|
|
|
* otherwise it might foil sbp2_[conditionally_]unblock()'s attempt to
|
|
|
|
* unblock the target.
|
|
|
|
*/
|
|
|
|
static void sbp2_conditionally_block(struct sbp2_logical_unit *lu)
|
|
|
|
{
|
|
|
|
struct sbp2_target *tgt = lu->tgt;
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_card *card = target_parent_device(tgt)->card;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
struct Scsi_Host *shost =
|
|
|
|
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&card->lock, flags);
|
|
|
|
if (!tgt->dont_block && !lu->blocked &&
|
|
|
|
lu->generation != card->generation) {
|
|
|
|
lu->blocked = true;
|
2008-02-29 02:53:45 +07:00
|
|
|
if (++tgt->blocked == 1)
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
scsi_block_requests(shost);
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&card->lock, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Unblocks lu->tgt as soon as all its logical units can be unblocked.
|
|
|
|
* Note, it is harmless to run scsi_unblock_requests() outside the
|
|
|
|
* card->lock protected section. On the other hand, running it inside
|
|
|
|
* the section might clash with shost->host_lock.
|
|
|
|
*/
|
|
|
|
static void sbp2_conditionally_unblock(struct sbp2_logical_unit *lu)
|
|
|
|
{
|
|
|
|
struct sbp2_target *tgt = lu->tgt;
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_card *card = target_parent_device(tgt)->card;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
struct Scsi_Host *shost =
|
|
|
|
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
|
|
|
|
bool unblock = false;
|
|
|
|
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_lock_irq(&card->lock);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
if (lu->blocked && lu->generation == card->generation) {
|
|
|
|
lu->blocked = false;
|
|
|
|
unblock = --tgt->blocked == 0;
|
|
|
|
}
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_unlock_irq(&card->lock);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
|
2008-02-29 02:53:45 +07:00
|
|
|
if (unblock)
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
scsi_unblock_requests(shost);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Prevents future blocking of tgt and unblocks it.
|
|
|
|
* Note, it is harmless to run scsi_unblock_requests() outside the
|
|
|
|
* card->lock protected section. On the other hand, running it inside
|
|
|
|
* the section might clash with shost->host_lock.
|
|
|
|
*/
|
|
|
|
static void sbp2_unblock(struct sbp2_target *tgt)
|
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_card *card = target_parent_device(tgt)->card;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
struct Scsi_Host *shost =
|
|
|
|
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
|
|
|
|
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_lock_irq(&card->lock);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
++tgt->dont_block;
|
firewire: sbp2: replace some spin_lock_irqsave by spin_lock_irq
Users of card->lock Calling context
------------------------------------------------------------------------
sbp2_status_write AR-req handler, tasklet
complete_transaction AR-resp or AT-req handler, tasklet
sbp2_send_orb among else scsi host .queuecommand, which may
be called in some sort of atomic context
sbp2_cancel_orbs sbp2_send_management_orb/
sbp2_{login,reconnect,remove},
worklet or process
sbp2_scsi_abort, scsi eh thread
sbp2_allow_block sbp2_login, worklet
sbp2_conditionally_block among else complete_command_orb, tasklet
sbp2_conditionally_unblock sbp2_{login,reconnect}, worklet
sbp2_unblock sbp2_{login,remove}, worklet or process
Drop the IRQ flags saving from sbp2_cancel_orbs,
sbp2_conditionally_unblock, and sbp2_unblock.
It was already omitted in sbp2_allow_block.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2014-03-04 05:23:22 +07:00
|
|
|
spin_unlock_irq(&card->lock);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
|
|
|
|
scsi_unblock_requests(shost);
|
|
|
|
}
|
|
|
|
|
2008-02-27 05:30:02 +07:00
|
|
|
static int sbp2_lun2int(u16 lun)
|
|
|
|
{
|
|
|
|
struct scsi_lun eight_bytes_lun;
|
|
|
|
|
|
|
|
memset(&eight_bytes_lun, 0, sizeof(eight_bytes_lun));
|
|
|
|
eight_bytes_lun.scsi_lun[0] = (lun >> 8) & 0xff;
|
|
|
|
eight_bytes_lun.scsi_lun[1] = lun & 0xff;
|
|
|
|
|
|
|
|
return scsilun_to_int(&eight_bytes_lun);
|
|
|
|
}
|
|
|
|
|
2008-03-07 13:43:31 +07:00
|
|
|
/*
|
|
|
|
* Write retransmit retry values into the BUSY_TIMEOUT register.
|
|
|
|
* - The single-phase retry protocol is supported by all SBP-2 devices, but the
|
|
|
|
* default retry_limit value is 0 (i.e. never retry transmission). We write a
|
|
|
|
* saner value after logging into the device.
|
|
|
|
* - The dual-phase retry protocol is optional to implement, and if not
|
|
|
|
* supported, writes to the dual-phase portion of the register will be
|
|
|
|
* ignored. We try to write the original 1394-1995 default here.
|
|
|
|
* - In the case of devices that are also SBP-3-compliant, all writes are
|
|
|
|
* ignored, as the register is read-only, but contains single-phase retry of
|
|
|
|
* 15, which is what we're trying to set for all SBP-2 device anyway, so this
|
|
|
|
* write attempt is safe and yields more consistent behavior for all devices.
|
|
|
|
*
|
|
|
|
* See section 8.3.2.3.5 of the 1394-1995 spec, section 6.2 of the SBP-2 spec,
|
|
|
|
* and section 6.4 of the SBP-3 spec for further details.
|
|
|
|
*/
|
firewire: fw-sbp2: set single-phase retry_limit
Per the SBP-2 specification, all SBP-2 target devices must have a BUSY_TIMEOUT
register. Per the 1394-1995 specification, the retry_limt portion of the
register should be set to 0x0 initially, and set on the target by a logged in
initiator (i.e., a Linux host w/firewire controller(s)).
Well, as it turns out, lots of devices these days have actually moved on to
starting to implement SBP-3 compliance, which says that retry_limit should
default to 0xf instead (yes, SBP-3 stomps directly on 1394-1995, oops).
Prior to this change, the firewire driver stack didn't touch retry_limit, and
any SBP-3 compliant device worked fine, while SBP-2 compliant ones were unable
to retransmit when the host returned an ack_busy_X, which resulted in stalled
out I/O, eventually causing the SCSI layer to give up and offline the device.
The simple fix is for us to set retry_limit to 0xf in the register for all
devices (which actually matches what the old ieee1394 stack did).
Prior to this change, a hard disk behind an SBP-2 Prolific PL-3507 bridge chip
would routinely encounter buffer I/O errors and wind up offlined by the SCSI
layer. With this change, I've encountered zero I/O failures moving tens of GB
of data around.
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-03-07 13:43:01 +07:00
|
|
|
static void sbp2_set_busy_timeout(struct sbp2_logical_unit *lu)
|
|
|
|
{
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2008-07-20 19:20:53 +07:00
|
|
|
__be32 d = cpu_to_be32(SBP2_CYCLE_LIMIT | SBP2_RETRY_LIMIT);
|
firewire: fw-sbp2: set single-phase retry_limit
Per the SBP-2 specification, all SBP-2 target devices must have a BUSY_TIMEOUT
register. Per the 1394-1995 specification, the retry_limt portion of the
register should be set to 0x0 initially, and set on the target by a logged in
initiator (i.e., a Linux host w/firewire controller(s)).
Well, as it turns out, lots of devices these days have actually moved on to
starting to implement SBP-3 compliance, which says that retry_limit should
default to 0xf instead (yes, SBP-3 stomps directly on 1394-1995, oops).
Prior to this change, the firewire driver stack didn't touch retry_limit, and
any SBP-3 compliant device worked fine, while SBP-2 compliant ones were unable
to retransmit when the host returned an ack_busy_X, which resulted in stalled
out I/O, eventually causing the SCSI layer to give up and offline the device.
The simple fix is for us to set retry_limit to 0xf in the register for all
devices (which actually matches what the old ieee1394 stack did).
Prior to this change, a hard disk behind an SBP-2 Prolific PL-3507 bridge chip
would routinely encounter buffer I/O errors and wind up offlined by the SCSI
layer. With this change, I've encountered zero I/O failures moving tens of GB
of data around.
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-03-07 13:43:01 +07:00
|
|
|
|
2008-07-20 19:20:53 +07:00
|
|
|
fw_run_transaction(device->card, TCODE_WRITE_QUADLET_REQUEST,
|
|
|
|
lu->tgt->node_id, lu->generation, device->max_speed,
|
2010-06-13 01:29:07 +07:00
|
|
|
CSR_REGISTER_BASE + CSR_BUSY_TIMEOUT, &d, 4);
|
firewire: fw-sbp2: set single-phase retry_limit
Per the SBP-2 specification, all SBP-2 target devices must have a BUSY_TIMEOUT
register. Per the 1394-1995 specification, the retry_limt portion of the
register should be set to 0x0 initially, and set on the target by a logged in
initiator (i.e., a Linux host w/firewire controller(s)).
Well, as it turns out, lots of devices these days have actually moved on to
starting to implement SBP-3 compliance, which says that retry_limit should
default to 0xf instead (yes, SBP-3 stomps directly on 1394-1995, oops).
Prior to this change, the firewire driver stack didn't touch retry_limit, and
any SBP-3 compliant device worked fine, while SBP-2 compliant ones were unable
to retransmit when the host returned an ack_busy_X, which resulted in stalled
out I/O, eventually causing the SCSI layer to give up and offline the device.
The simple fix is for us to set retry_limit to 0xf in the register for all
devices (which actually matches what the old ieee1394 stack did).
Prior to this change, a hard disk behind an SBP-2 Prolific PL-3507 bridge chip
would routinely encounter buffer I/O errors and wind up offlined by the SCSI
layer. With this change, I've encountered zero I/O failures moving tens of GB
of data around.
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-03-07 13:43:01 +07:00
|
|
|
}
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
static void sbp2_reconnect(struct work_struct *work);
|
|
|
|
|
2007-02-07 02:49:34 +07:00
|
|
|
static void sbp2_login(struct work_struct *work)
|
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu =
|
|
|
|
container_of(work, struct sbp2_logical_unit, work.work);
|
2008-02-04 05:09:50 +07:00
|
|
|
struct sbp2_target *tgt = lu->tgt;
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(tgt);
|
2008-02-04 05:09:50 +07:00
|
|
|
struct Scsi_Host *shost;
|
2007-08-25 19:05:28 +07:00
|
|
|
struct scsi_device *sdev;
|
2007-02-07 02:49:34 +07:00
|
|
|
struct sbp2_login_response response;
|
2007-08-25 19:05:28 +07:00
|
|
|
int generation, node_id, local_node_id;
|
2007-02-07 02:49:34 +07:00
|
|
|
|
2008-01-28 01:14:44 +07:00
|
|
|
if (fw_device_is_shutdown(device))
|
2011-08-27 20:33:34 +07:00
|
|
|
return;
|
2008-01-28 01:14:44 +07:00
|
|
|
|
2008-01-24 07:53:19 +07:00
|
|
|
generation = device->generation;
|
2008-10-26 17:04:20 +07:00
|
|
|
smp_rmb(); /* node IDs must not be older than generation */
|
2008-01-24 07:53:19 +07:00
|
|
|
node_id = device->node_id;
|
|
|
|
local_node_id = device->card->node_id;
|
2007-02-07 02:49:34 +07:00
|
|
|
|
2008-02-04 05:11:39 +07:00
|
|
|
/* If this is a re-login attempt, log out, or we might be rejected. */
|
2008-02-27 05:30:02 +07:00
|
|
|
if (lu->has_sdev)
|
2008-02-04 05:11:39 +07:00
|
|
|
sbp2_send_management_orb(lu, device->node_id, generation,
|
|
|
|
SBP2_LOGOUT_REQUEST, lu->login_id, NULL);
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (sbp2_send_management_orb(lu, node_id, generation,
|
|
|
|
SBP2_LOGIN_REQUEST, lu->lun, &response) < 0) {
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
if (lu->retries++ < 5) {
|
2007-11-07 07:12:51 +07:00
|
|
|
sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
} else {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(tgt_dev(tgt), "failed to login to LUN %04x\n",
|
|
|
|
lu->lun);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
/* Let any waiting I/O fail from now on. */
|
|
|
|
sbp2_unblock(lu->tgt);
|
|
|
|
}
|
2011-08-27 20:33:34 +07:00
|
|
|
return;
|
2007-02-07 02:49:34 +07:00
|
|
|
}
|
|
|
|
|
2008-02-04 05:09:50 +07:00
|
|
|
tgt->node_id = node_id;
|
|
|
|
tgt->address_high = local_node_id << 16;
|
2008-10-26 17:04:20 +07:00
|
|
|
smp_wmb(); /* node IDs must not be older than generation */
|
|
|
|
lu->generation = generation;
|
2007-02-07 02:49:34 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->command_block_agent_address =
|
2008-02-29 02:51:11 +07:00
|
|
|
((u64)(be32_to_cpu(response.command_block_agent.high) & 0xffff)
|
|
|
|
<< 32) | be32_to_cpu(response.command_block_agent.low);
|
|
|
|
lu->login_id = be32_to_cpu(response.misc) & 0xffff;
|
2007-02-07 02:49:34 +07:00
|
|
|
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(tgt_dev(tgt), "logged in to LUN %04x (%d retries)\n",
|
|
|
|
lu->lun, lu->retries);
|
2007-02-07 02:49:34 +07:00
|
|
|
|
firewire: fw-sbp2: set single-phase retry_limit
Per the SBP-2 specification, all SBP-2 target devices must have a BUSY_TIMEOUT
register. Per the 1394-1995 specification, the retry_limt portion of the
register should be set to 0x0 initially, and set on the target by a logged in
initiator (i.e., a Linux host w/firewire controller(s)).
Well, as it turns out, lots of devices these days have actually moved on to
starting to implement SBP-3 compliance, which says that retry_limit should
default to 0xf instead (yes, SBP-3 stomps directly on 1394-1995, oops).
Prior to this change, the firewire driver stack didn't touch retry_limit, and
any SBP-3 compliant device worked fine, while SBP-2 compliant ones were unable
to retransmit when the host returned an ack_busy_X, which resulted in stalled
out I/O, eventually causing the SCSI layer to give up and offline the device.
The simple fix is for us to set retry_limit to 0xf in the register for all
devices (which actually matches what the old ieee1394 stack did).
Prior to this change, a hard disk behind an SBP-2 Prolific PL-3507 bridge chip
would routinely encounter buffer I/O errors and wind up offlined by the SCSI
layer. With this change, I've encountered zero I/O failures moving tens of GB
of data around.
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-03-07 13:43:01 +07:00
|
|
|
/* set appropriate retry limit(s) in BUSY_TIMEOUT register */
|
|
|
|
sbp2_set_busy_timeout(lu);
|
2007-02-07 02:49:34 +07:00
|
|
|
|
2014-03-07 22:19:57 +07:00
|
|
|
lu->workfn = sbp2_reconnect;
|
2007-08-25 19:05:28 +07:00
|
|
|
sbp2_agent_reset(lu);
|
|
|
|
|
2008-02-04 05:10:47 +07:00
|
|
|
/* This was a re-login. */
|
2008-02-27 05:30:02 +07:00
|
|
|
if (lu->has_sdev) {
|
2008-02-04 05:10:47 +07:00
|
|
|
sbp2_cancel_orbs(lu);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
sbp2_conditionally_unblock(lu);
|
2011-08-27 20:33:34 +07:00
|
|
|
|
|
|
|
return;
|
2008-02-04 05:10:47 +07:00
|
|
|
}
|
|
|
|
|
2008-02-04 05:04:38 +07:00
|
|
|
if (lu->tgt->workarounds & SBP2_WORKAROUND_DELAY_INQUIRY)
|
|
|
|
ssleep(SBP2_INQUIRY_DELAY);
|
|
|
|
|
2008-02-04 05:09:50 +07:00
|
|
|
shost = container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
|
2008-02-27 05:30:02 +07:00
|
|
|
sdev = __scsi_add_device(shost, 0, 0, sbp2_lun2int(lu->lun), lu);
|
2008-02-16 03:29:02 +07:00
|
|
|
/*
|
|
|
|
* FIXME: We are unable to perform reconnects while in sbp2_login().
|
|
|
|
* Therefore __scsi_add_device() will get into trouble if a bus reset
|
|
|
|
* happens in parallel. It will either fail or leave us with an
|
|
|
|
* unusable sdev. As a workaround we check for this and retry the
|
|
|
|
* whole login and SCSI probing.
|
|
|
|
*/
|
2008-01-26 23:43:23 +07:00
|
|
|
|
2008-02-16 03:29:02 +07:00
|
|
|
/* Reported error during __scsi_add_device() */
|
|
|
|
if (IS_ERR(sdev))
|
|
|
|
goto out_logout_login;
|
|
|
|
|
|
|
|
/* Unreported error during __scsi_add_device() */
|
|
|
|
smp_rmb(); /* get current card generation */
|
|
|
|
if (generation != device->card->generation) {
|
|
|
|
scsi_remove_device(sdev);
|
2008-02-19 15:05:49 +07:00
|
|
|
scsi_device_put(sdev);
|
2008-02-16 03:29:02 +07:00
|
|
|
goto out_logout_login;
|
2007-02-07 02:49:34 +07:00
|
|
|
}
|
2008-02-16 03:29:02 +07:00
|
|
|
|
|
|
|
/* No error during __scsi_add_device() */
|
2008-02-27 05:30:02 +07:00
|
|
|
lu->has_sdev = true;
|
|
|
|
scsi_device_put(sdev);
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
sbp2_allow_block(lu);
|
2011-08-27 20:33:34 +07:00
|
|
|
|
|
|
|
return;
|
2008-02-16 03:29:02 +07:00
|
|
|
|
|
|
|
out_logout_login:
|
|
|
|
smp_rmb(); /* generation may have changed */
|
|
|
|
generation = device->generation;
|
|
|
|
smp_rmb(); /* node_id must not be older than generation */
|
|
|
|
|
|
|
|
sbp2_send_management_orb(lu, device->node_id, generation,
|
|
|
|
SBP2_LOGOUT_REQUEST, lu->login_id, NULL);
|
|
|
|
/*
|
|
|
|
* If a bus reset happened, sbp2_update will have requeued
|
|
|
|
* lu->work already. Reset the work from reconnect to login.
|
|
|
|
*/
|
2014-03-07 22:19:57 +07:00
|
|
|
lu->workfn = sbp2_login;
|
2007-02-07 02:49:34 +07:00
|
|
|
}
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
static void sbp2_reconnect(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct sbp2_logical_unit *lu =
|
|
|
|
container_of(work, struct sbp2_logical_unit, work.work);
|
|
|
|
struct sbp2_target *tgt = lu->tgt;
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(tgt);
|
2011-08-27 20:34:32 +07:00
|
|
|
int generation, node_id, local_node_id;
|
|
|
|
|
|
|
|
if (fw_device_is_shutdown(device))
|
|
|
|
return;
|
|
|
|
|
|
|
|
generation = device->generation;
|
|
|
|
smp_rmb(); /* node IDs must not be older than generation */
|
|
|
|
node_id = device->node_id;
|
|
|
|
local_node_id = device->card->node_id;
|
|
|
|
|
|
|
|
if (sbp2_send_management_orb(lu, node_id, generation,
|
|
|
|
SBP2_RECONNECT_REQUEST,
|
|
|
|
lu->login_id, NULL) < 0) {
|
|
|
|
/*
|
|
|
|
* If reconnect was impossible even though we are in the
|
|
|
|
* current generation, fall back and try to log in again.
|
|
|
|
*
|
|
|
|
* We could check for "Function rejected" status, but
|
|
|
|
* looking at the bus generation as simpler and more general.
|
|
|
|
*/
|
|
|
|
smp_rmb(); /* get current card generation */
|
|
|
|
if (generation == device->card->generation ||
|
|
|
|
lu->retries++ >= 5) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(tgt_dev(tgt), "failed to reconnect\n");
|
2011-08-27 20:34:32 +07:00
|
|
|
lu->retries = 0;
|
2014-03-07 22:19:57 +07:00
|
|
|
lu->workfn = sbp2_login;
|
2011-08-27 20:34:32 +07:00
|
|
|
}
|
|
|
|
sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
|
|
|
|
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
tgt->node_id = node_id;
|
|
|
|
tgt->address_high = local_node_id << 16;
|
|
|
|
smp_wmb(); /* node IDs must not be older than generation */
|
|
|
|
lu->generation = generation;
|
|
|
|
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(tgt_dev(tgt), "reconnected to LUN %04x (%d retries)\n",
|
|
|
|
lu->lun, lu->retries);
|
2011-08-27 20:34:32 +07:00
|
|
|
|
|
|
|
sbp2_agent_reset(lu);
|
|
|
|
sbp2_cancel_orbs(lu);
|
|
|
|
sbp2_conditionally_unblock(lu);
|
|
|
|
}
|
|
|
|
|
2014-03-07 22:19:57 +07:00
|
|
|
static void sbp2_lu_workfn(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct sbp2_logical_unit *lu = container_of(to_delayed_work(work),
|
|
|
|
struct sbp2_logical_unit, work);
|
|
|
|
lu->workfn(work);
|
|
|
|
}
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
static int sbp2_add_logical_unit(struct sbp2_target *tgt, int lun_entry)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
lu = kmalloc(sizeof(*lu), GFP_KERNEL);
|
|
|
|
if (!lu)
|
|
|
|
return -ENOMEM;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->address_handler.length = 0x100;
|
|
|
|
lu->address_handler.address_callback = sbp2_status_write;
|
|
|
|
lu->address_handler.callback_data = lu;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (fw_core_add_address_handler(&lu->address_handler,
|
|
|
|
&fw_high_memory_region) < 0) {
|
|
|
|
kfree(lu);
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-02-27 05:30:02 +07:00
|
|
|
lu->tgt = tgt;
|
|
|
|
lu->lun = lun_entry & 0xffff;
|
2008-10-25 02:26:20 +07:00
|
|
|
lu->login_id = INVALID_LOGIN_ID;
|
2008-02-27 05:30:02 +07:00
|
|
|
lu->retries = 0;
|
|
|
|
lu->has_sdev = false;
|
|
|
|
lu->blocked = false;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
++tgt->dont_block;
|
2007-08-25 19:05:28 +07:00
|
|
|
INIT_LIST_HEAD(&lu->orb_list);
|
2014-03-07 22:19:57 +07:00
|
|
|
lu->workfn = sbp2_login;
|
|
|
|
INIT_DELAYED_WORK(&lu->work, sbp2_lu_workfn);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
list_add_tail(&lu->link, &tgt->lu_list);
|
|
|
|
return 0;
|
|
|
|
}
|
2007-05-10 06:23:07 +07:00
|
|
|
|
2012-02-15 21:59:08 +07:00
|
|
|
static void sbp2_get_unit_unique_id(struct sbp2_target *tgt,
|
|
|
|
const u32 *leaf)
|
|
|
|
{
|
|
|
|
if ((leaf[0] & 0xffff0000) == 0x00020000)
|
|
|
|
tgt->guid = (u64)leaf[1] << 32 | leaf[2];
|
|
|
|
}
|
|
|
|
|
2009-12-26 07:44:10 +07:00
|
|
|
static int sbp2_scan_logical_unit_dir(struct sbp2_target *tgt,
|
|
|
|
const u32 *directory)
|
2007-08-25 19:05:28 +07:00
|
|
|
{
|
|
|
|
struct fw_csr_iterator ci;
|
|
|
|
int key, value;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
fw_csr_iterator_init(&ci, directory);
|
|
|
|
while (fw_csr_iterator_next(&ci, &key, &value))
|
|
|
|
if (key == SBP2_CSR_LOGICAL_UNIT_NUMBER &&
|
|
|
|
sbp2_add_logical_unit(tgt, value) < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-12-26 07:44:10 +07:00
|
|
|
static int sbp2_scan_unit_dir(struct sbp2_target *tgt, const u32 *directory,
|
2007-08-25 19:05:28 +07:00
|
|
|
u32 *model, u32 *firmware_revision)
|
|
|
|
{
|
|
|
|
struct fw_csr_iterator ci;
|
|
|
|
int key, value;
|
|
|
|
|
|
|
|
fw_csr_iterator_init(&ci, directory);
|
2006-12-20 07:58:40 +07:00
|
|
|
while (fw_csr_iterator_next(&ci, &key, &value)) {
|
|
|
|
switch (key) {
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
case CSR_DEPENDENT_INFO | CSR_OFFSET:
|
2007-08-25 19:05:28 +07:00
|
|
|
tgt->management_agent_address =
|
|
|
|
CSR_REGISTER_BASE + 4 * value;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
|
|
|
case CSR_DIRECTORY_ID:
|
|
|
|
tgt->directory_id = value;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
case CSR_MODEL:
|
2007-08-25 19:05:28 +07:00
|
|
|
*model = value;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case SBP2_CSR_FIRMWARE_REVISION:
|
|
|
|
*firmware_revision = value;
|
|
|
|
break;
|
|
|
|
|
2008-01-26 11:31:12 +07:00
|
|
|
case SBP2_CSR_UNIT_CHARACTERISTICS:
|
|
|
|
/* the timeout value is stored in 500ms units */
|
2009-10-08 05:39:31 +07:00
|
|
|
tgt->mgt_orb_timeout = (value >> 8 & 0xff) * 500;
|
2008-01-26 11:31:12 +07:00
|
|
|
break;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
case SBP2_CSR_LOGICAL_UNIT_NUMBER:
|
|
|
|
if (sbp2_add_logical_unit(tgt, value) < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
break;
|
|
|
|
|
2012-02-15 21:59:08 +07:00
|
|
|
case SBP2_CSR_UNIT_UNIQUE_ID:
|
|
|
|
sbp2_get_unit_unique_id(tgt, ci.p - 1 + value);
|
|
|
|
break;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
case SBP2_CSR_LOGICAL_UNIT_DIRECTORY:
|
2008-06-25 09:11:13 +07:00
|
|
|
/* Adjust for the increment in the iterator */
|
|
|
|
if (sbp2_scan_logical_unit_dir(tgt, ci.p - 1 + value) < 0)
|
2007-08-25 19:05:28 +07:00
|
|
|
return -ENOMEM;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2007-08-25 19:05:28 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-10-08 05:39:31 +07:00
|
|
|
/*
|
|
|
|
* Per section 7.4.8 of the SBP-2 spec, a mgt_ORB_timeout value can be
|
|
|
|
* provided in the config rom. Most devices do provide a value, which
|
|
|
|
* we'll use for login management orbs, but with some sane limits.
|
|
|
|
*/
|
|
|
|
static void sbp2_clamp_management_orb_timeout(struct sbp2_target *tgt)
|
|
|
|
{
|
|
|
|
unsigned int timeout = tgt->mgt_orb_timeout;
|
|
|
|
|
|
|
|
if (timeout > 40000)
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(tgt_dev(tgt), "%ds mgt_ORB_timeout limited to 40s\n",
|
|
|
|
timeout / 1000);
|
2009-10-08 05:39:31 +07:00
|
|
|
|
|
|
|
tgt->mgt_orb_timeout = clamp_val(timeout, 5000, 40000);
|
|
|
|
}
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
static void sbp2_init_workarounds(struct sbp2_target *tgt, u32 model,
|
|
|
|
u32 firmware_revision)
|
|
|
|
{
|
|
|
|
int i;
|
2008-01-26 23:42:45 +07:00
|
|
|
unsigned int w = sbp2_param_workarounds;
|
2007-08-13 22:48:25 +07:00
|
|
|
|
|
|
|
if (w)
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(tgt_dev(tgt),
|
|
|
|
"Please notify linux1394-devel@lists.sf.net "
|
|
|
|
"if you need the workarounds parameter\n");
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2007-08-13 22:48:25 +07:00
|
|
|
if (w & SBP2_WORKAROUND_OVERRIDE)
|
|
|
|
goto out;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
for (i = 0; i < ARRAY_SIZE(sbp2_workarounds_table); i++) {
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
if (sbp2_workarounds_table[i].firmware_revision !=
|
|
|
|
(firmware_revision & 0xffffff00))
|
|
|
|
continue;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
if (sbp2_workarounds_table[i].model != model &&
|
2009-01-25 01:41:46 +07:00
|
|
|
sbp2_workarounds_table[i].model != SBP2_ROM_VALUE_WILDCARD)
|
2006-12-20 07:58:40 +07:00
|
|
|
continue;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2007-08-13 22:48:25 +07:00
|
|
|
w |= sbp2_workarounds_table[i].workarounds;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
|
|
|
}
|
2007-08-13 22:48:25 +07:00
|
|
|
out:
|
|
|
|
if (w)
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(tgt_dev(tgt), "workarounds 0x%x "
|
|
|
|
"(firmware_revision 0x%06x, model_id 0x%06x)\n",
|
|
|
|
w, firmware_revision, model);
|
2007-08-13 22:48:25 +07:00
|
|
|
tgt->workarounds = w;
|
2007-08-25 19:05:28 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct scsi_host_template scsi_driver_template;
|
2013-06-09 23:15:00 +07:00
|
|
|
static void sbp2_remove(struct fw_unit *unit);
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2013-06-09 23:15:00 +07:00
|
|
|
static int sbp2_probe(struct fw_unit *unit, const struct ieee1394_device_id *id)
|
2007-08-25 19:05:28 +07:00
|
|
|
{
|
2009-06-06 23:35:27 +07:00
|
|
|
struct fw_device *device = fw_parent_device(unit);
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_target *tgt;
|
|
|
|
struct sbp2_logical_unit *lu;
|
|
|
|
struct Scsi_Host *shost;
|
|
|
|
u32 model, firmware_revision;
|
|
|
|
|
2012-02-15 21:59:09 +07:00
|
|
|
/* cannot (or should not) handle targets on the local node */
|
|
|
|
if (device->is_local)
|
|
|
|
return -ENODEV;
|
|
|
|
|
2008-08-15 02:47:21 +07:00
|
|
|
if (dma_get_max_seg_size(device->card->device) > SBP2_MAX_SEG_SIZE)
|
2013-03-24 23:31:38 +07:00
|
|
|
WARN_ON(dma_set_max_seg_size(device->card->device,
|
|
|
|
SBP2_MAX_SEG_SIZE));
|
2008-08-15 02:47:21 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
shost = scsi_host_alloc(&scsi_driver_template, sizeof(*tgt));
|
|
|
|
if (shost == NULL)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
tgt = (struct sbp2_target *)shost->hostdata;
|
2009-05-01 04:43:31 +07:00
|
|
|
dev_set_drvdata(&unit->device, tgt);
|
2007-08-25 19:05:28 +07:00
|
|
|
tgt->unit = unit;
|
|
|
|
INIT_LIST_HEAD(&tgt->lu_list);
|
2008-03-25 02:54:28 +07:00
|
|
|
tgt->guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
|
2007-08-25 19:05:28 +07:00
|
|
|
|
|
|
|
if (fw_device_enable_phys_dma(device) < 0)
|
|
|
|
goto fail_shost_put;
|
|
|
|
|
2009-07-01 01:27:59 +07:00
|
|
|
shost->max_cmd_len = SBP2_MAX_CDB_SIZE;
|
|
|
|
|
2012-05-18 23:39:39 +07:00
|
|
|
if (scsi_add_host_with_dma(shost, &unit->device,
|
|
|
|
device->card->device) < 0)
|
2007-08-25 19:05:28 +07:00
|
|
|
goto fail_shost_put;
|
|
|
|
|
|
|
|
/* implicit directory ID */
|
|
|
|
tgt->directory_id = ((unit->directory - device->config_rom) * 4
|
|
|
|
+ CSR_CONFIG_ROM) & 0xffffff;
|
|
|
|
|
2009-01-25 01:41:46 +07:00
|
|
|
firmware_revision = SBP2_ROM_VALUE_MISSING;
|
|
|
|
model = SBP2_ROM_VALUE_MISSING;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (sbp2_scan_unit_dir(tgt, unit->directory, &model,
|
|
|
|
&firmware_revision) < 0)
|
2011-08-27 20:35:23 +07:00
|
|
|
goto fail_remove;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2009-10-08 05:39:31 +07:00
|
|
|
sbp2_clamp_management_orb_timeout(tgt);
|
2007-08-25 19:05:28 +07:00
|
|
|
sbp2_init_workarounds(tgt, model, firmware_revision);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2009-01-25 01:41:46 +07:00
|
|
|
/*
|
|
|
|
* At S100 we can do 512 bytes per packet, at S200 1024 bytes,
|
|
|
|
* and so on up to 4096 bytes. The SBP-2 max_payload field
|
|
|
|
* specifies the max payload size as 2 ^ (max_payload + 2), so
|
|
|
|
* if we set this to max_speed + 7, we get the right value.
|
|
|
|
*/
|
2011-09-19 05:20:48 +07:00
|
|
|
tgt->max_payload = min3(device->max_speed + 7, 10U,
|
|
|
|
device->card->max_receive - 1);
|
2009-01-25 01:41:46 +07:00
|
|
|
|
2007-11-07 07:12:51 +07:00
|
|
|
/* Do the login in a workqueue so we can easily reschedule retries. */
|
2007-08-25 19:05:28 +07:00
|
|
|
list_for_each_entry(lu, &tgt->lu_list, link)
|
2008-10-22 05:28:36 +07:00
|
|
|
sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
|
2011-08-27 20:33:34 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
return 0;
|
2007-05-10 06:23:07 +07:00
|
|
|
|
2011-08-27 20:35:23 +07:00
|
|
|
fail_remove:
|
2013-06-09 23:15:00 +07:00
|
|
|
sbp2_remove(unit);
|
2007-08-25 19:05:28 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
fail_shost_put:
|
|
|
|
scsi_host_put(shost);
|
|
|
|
return -ENOMEM;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
static void sbp2_update(struct fw_unit *unit)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2009-05-01 04:43:31 +07:00
|
|
|
struct sbp2_target *tgt = dev_get_drvdata(&unit->device);
|
2011-08-23 03:38:38 +07:00
|
|
|
struct sbp2_logical_unit *lu;
|
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
fw_device_enable_phys_dma(fw_parent_device(unit));
|
2011-08-27 20:33:34 +07:00
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
/*
|
|
|
|
* Fw-core serializes sbp2_update() against sbp2_remove().
|
|
|
|
* Iteration over tgt->lu_list is therefore safe here.
|
|
|
|
*/
|
|
|
|
list_for_each_entry(lu, &tgt->lu_list, link) {
|
|
|
|
sbp2_conditionally_block(lu);
|
|
|
|
lu->retries = 0;
|
|
|
|
sbp2_queue_work(lu, 0);
|
|
|
|
}
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2013-06-09 23:15:00 +07:00
|
|
|
static void sbp2_remove(struct fw_unit *unit)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2011-08-27 20:35:23 +07:00
|
|
|
struct fw_device *device = fw_parent_device(unit);
|
|
|
|
struct sbp2_target *tgt = dev_get_drvdata(&unit->device);
|
2011-08-27 20:34:32 +07:00
|
|
|
struct sbp2_logical_unit *lu, *next;
|
|
|
|
struct Scsi_Host *shost =
|
|
|
|
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
|
|
|
|
struct scsi_device *sdev;
|
2008-01-28 01:14:44 +07:00
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
/* prevent deadlocks */
|
|
|
|
sbp2_unblock(tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
list_for_each_entry_safe(lu, next, &tgt->lu_list, link) {
|
2011-08-27 20:35:23 +07:00
|
|
|
cancel_delayed_work_sync(&lu->work);
|
2011-08-27 20:34:32 +07:00
|
|
|
sdev = scsi_device_lookup(shost, 0, 0, sbp2_lun2int(lu->lun));
|
|
|
|
if (sdev) {
|
|
|
|
scsi_remove_device(sdev);
|
|
|
|
scsi_device_put(sdev);
|
2007-02-07 02:49:34 +07:00
|
|
|
}
|
2011-08-27 20:34:32 +07:00
|
|
|
if (lu->login_id != INVALID_LOGIN_ID) {
|
|
|
|
int generation, node_id;
|
|
|
|
/*
|
|
|
|
* tgt->node_id may be obsolete here if we failed
|
|
|
|
* during initial login or after a bus reset where
|
|
|
|
* the topology changed.
|
|
|
|
*/
|
|
|
|
generation = device->generation;
|
|
|
|
smp_rmb(); /* node_id vs. generation */
|
|
|
|
node_id = device->node_id;
|
|
|
|
sbp2_send_management_orb(lu, node_id, generation,
|
|
|
|
SBP2_LOGOUT_REQUEST,
|
|
|
|
lu->login_id, NULL);
|
|
|
|
}
|
|
|
|
fw_core_remove_address_handler(&lu->address_handler);
|
|
|
|
list_del(&lu->link);
|
|
|
|
kfree(lu);
|
2007-02-07 02:49:34 +07:00
|
|
|
}
|
2011-08-27 20:34:32 +07:00
|
|
|
scsi_remove_host(shost);
|
2013-06-09 23:15:00 +07:00
|
|
|
dev_notice(&unit->device, "released target %d:0:0\n", shost->host_no);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2011-08-27 20:34:32 +07:00
|
|
|
scsi_host_put(shost);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
#define SBP2_UNIT_SPEC_ID_ENTRY 0x0000609e
|
|
|
|
#define SBP2_SW_VERSION_ENTRY 0x00010483
|
|
|
|
|
2009-02-16 05:12:34 +07:00
|
|
|
static const struct ieee1394_device_id sbp2_id_table[] = {
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2009-02-16 05:12:34 +07:00
|
|
|
.match_flags = IEEE1394_MATCH_SPECIFIER_ID |
|
|
|
|
IEEE1394_MATCH_VERSION,
|
2006-12-20 07:58:40 +07:00
|
|
|
.specifier_id = SBP2_UNIT_SPEC_ID_ENTRY,
|
2007-01-22 02:45:32 +07:00
|
|
|
.version = SBP2_SW_VERSION_ENTRY,
|
2006-12-20 07:58:40 +07:00
|
|
|
},
|
|
|
|
{ }
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct fw_driver sbp2_driver = {
|
|
|
|
.driver = {
|
|
|
|
.owner = THIS_MODULE,
|
2012-02-19 04:01:14 +07:00
|
|
|
.name = KBUILD_MODNAME,
|
2006-12-20 07:58:40 +07:00
|
|
|
.bus = &fw_bus_type,
|
|
|
|
},
|
2013-06-09 23:15:00 +07:00
|
|
|
.probe = sbp2_probe,
|
2006-12-20 07:58:40 +07:00
|
|
|
.update = sbp2_update,
|
2013-06-09 23:15:00 +07:00
|
|
|
.remove = sbp2_remove,
|
2006-12-20 07:58:40 +07:00
|
|
|
.id_table = sbp2_id_table,
|
|
|
|
};
|
|
|
|
|
2009-01-28 07:03:34 +07:00
|
|
|
static void sbp2_unmap_scatterlist(struct device *card_device,
|
|
|
|
struct sbp2_command_orb *orb)
|
|
|
|
{
|
2012-05-18 23:40:19 +07:00
|
|
|
scsi_dma_unmap(orb->cmd);
|
2009-01-28 07:03:34 +07:00
|
|
|
|
|
|
|
if (orb->request.misc & cpu_to_be32(COMMAND_ORB_PAGE_TABLE_PRESENT))
|
|
|
|
dma_unmap_single(card_device, orb->page_table_bus,
|
|
|
|
sizeof(orb->page_table), DMA_TO_DEVICE);
|
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static unsigned int sbp2_status_to_sense_data(u8 *sbp2_status, u8 *sense_data)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2007-04-11 05:11:18 +07:00
|
|
|
int sam_status;
|
2012-02-15 21:59:10 +07:00
|
|
|
int sfmt = (sbp2_status[0] >> 6) & 0x03;
|
2007-04-11 05:11:18 +07:00
|
|
|
|
2012-02-15 21:59:10 +07:00
|
|
|
if (sfmt == 2 || sfmt == 3) {
|
|
|
|
/*
|
|
|
|
* Reserved for future standardization (2) or
|
|
|
|
* Status block format vendor-dependent (3)
|
|
|
|
*/
|
|
|
|
return DID_ERROR << 16;
|
|
|
|
}
|
2007-04-11 05:11:18 +07:00
|
|
|
|
2012-02-15 21:59:10 +07:00
|
|
|
sense_data[0] = 0x70 | sfmt | (sbp2_status[1] & 0x80);
|
2006-12-20 07:58:40 +07:00
|
|
|
sense_data[1] = 0x0;
|
2012-02-15 21:59:10 +07:00
|
|
|
sense_data[2] = ((sbp2_status[1] << 1) & 0xe0) | (sbp2_status[1] & 0x0f);
|
2006-12-20 07:58:40 +07:00
|
|
|
sense_data[3] = sbp2_status[4];
|
|
|
|
sense_data[4] = sbp2_status[5];
|
|
|
|
sense_data[5] = sbp2_status[6];
|
|
|
|
sense_data[6] = sbp2_status[7];
|
|
|
|
sense_data[7] = 10;
|
|
|
|
sense_data[8] = sbp2_status[8];
|
|
|
|
sense_data[9] = sbp2_status[9];
|
|
|
|
sense_data[10] = sbp2_status[10];
|
|
|
|
sense_data[11] = sbp2_status[11];
|
|
|
|
sense_data[12] = sbp2_status[2];
|
|
|
|
sense_data[13] = sbp2_status[3];
|
|
|
|
sense_data[14] = sbp2_status[12];
|
|
|
|
sense_data[15] = sbp2_status[13];
|
|
|
|
|
2007-04-11 05:11:18 +07:00
|
|
|
sam_status = sbp2_status[0] & 0x3f;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-04-11 05:11:18 +07:00
|
|
|
switch (sam_status) {
|
|
|
|
case SAM_STAT_GOOD:
|
2006-12-20 07:58:40 +07:00
|
|
|
case SAM_STAT_CHECK_CONDITION:
|
|
|
|
case SAM_STAT_CONDITION_MET:
|
2007-04-11 05:11:18 +07:00
|
|
|
case SAM_STAT_BUSY:
|
2006-12-20 07:58:40 +07:00
|
|
|
case SAM_STAT_RESERVATION_CONFLICT:
|
|
|
|
case SAM_STAT_COMMAND_TERMINATED:
|
2007-04-11 05:11:18 +07:00
|
|
|
return DID_OK << 16 | sam_status;
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
default:
|
2007-04-11 05:11:18 +07:00
|
|
|
return DID_ERROR << 16;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static void complete_command_orb(struct sbp2_orb *base_orb,
|
|
|
|
struct sbp2_status *status)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2007-06-28 03:04:33 +07:00
|
|
|
struct sbp2_command_orb *orb =
|
|
|
|
container_of(base_orb, struct sbp2_command_orb, base);
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(orb->lu->tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
int result;
|
|
|
|
|
|
|
|
if (status != NULL) {
|
2007-05-08 07:33:35 +07:00
|
|
|
if (STATUS_GET_DEAD(*status))
|
2008-02-04 05:08:58 +07:00
|
|
|
sbp2_agent_reset_no_wait(orb->lu);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
switch (STATUS_GET_RESPONSE(*status)) {
|
2006-12-20 07:58:40 +07:00
|
|
|
case SBP2_STATUS_REQUEST_COMPLETE:
|
2007-04-11 05:11:18 +07:00
|
|
|
result = DID_OK << 16;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
|
|
|
case SBP2_STATUS_TRANSPORT_FAILURE:
|
2007-04-11 05:11:18 +07:00
|
|
|
result = DID_BUS_BUSY << 16;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
|
|
|
case SBP2_STATUS_ILLEGAL_REQUEST:
|
|
|
|
case SBP2_STATUS_VENDOR_DEPENDENT:
|
|
|
|
default:
|
2007-04-11 05:11:18 +07:00
|
|
|
result = DID_ERROR << 16;
|
2006-12-20 07:58:40 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2007-05-08 07:33:35 +07:00
|
|
|
if (result == DID_OK << 16 && STATUS_GET_LEN(*status) > 1)
|
|
|
|
result = sbp2_status_to_sense_data(STATUS_GET_DATA(*status),
|
2006-12-20 07:58:40 +07:00
|
|
|
orb->cmd->sense_buffer);
|
|
|
|
} else {
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* If the orb completes with status == NULL, something
|
2006-12-20 07:58:40 +07:00
|
|
|
* went wrong, typically a bus reset happened mid-orb
|
2007-05-08 07:33:32 +07:00
|
|
|
* or when sending the write (less likely).
|
|
|
|
*/
|
2007-04-11 05:11:18 +07:00
|
|
|
result = DID_BUS_BUSY << 16;
|
firewire: fw-sbp2: (try to) avoid I/O errors during reconnect
While fw-sbp2 takes the necessary time to reconnect to a logical unit
after bus reset, the SCSI core keeps sending new commands. They are all
immediately completed with host busy status, and application clients or
filesystems will break quickly. The SCSI device might even be taken
offline: http://bugzilla.kernel.org/show_bug.cgi?id=9734
The only remedy seems to be to block the SCSI device until reconnect.
Alas the SCSI core has no useful API to block only one logical unit i.e.
the scsi_device, therefore we block the entire Scsi_Host. This
currently corresponds to an SBP-2 target. In case of targets with
multiple logical units, we need to satisfy the dependencies between
logical units by carefully tracking the blocking state of the target and
its units. We block all logical units of a target as soon as one of
them needs to be blocked, and keep them blocked until all of them are
ready to be unblocked.
Furthermore, as the history of the old sbp2 driver has shown, the
scsi_block_requests() API is a minefield with high potential of
deadlocks. We therefore take extra measures to keep logical units
unblocked during __scsi_add_device() and during shutdown.
This avoids I/O errors during reconnect in many but alas not in all
cases. There may still be errors after a re-login had to be performed.
Also, some bridges have been seen to cease fetching management ORBs if
I/O went on up until a bus reset. In these cases, all management ORBs
time out after mgt_orb_timeout. The old sbp2 driver is less vulnerable
or maybe not vulnerable to this, for as yet unknown reasons.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
2008-02-16 22:37:28 +07:00
|
|
|
sbp2_conditionally_block(orb->lu);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
dma_unmap_single(device->card->device, orb->base.request_bus,
|
2007-05-10 06:23:14 +07:00
|
|
|
sizeof(orb->request), DMA_TO_DEVICE);
|
2009-01-28 07:03:34 +07:00
|
|
|
sbp2_unmap_scatterlist(device->card->device, orb);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-04-11 05:11:18 +07:00
|
|
|
orb->cmd->result = result;
|
2011-04-22 17:21:44 +07:00
|
|
|
orb->cmd->scsi_done(orb->cmd);
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2008-12-15 03:47:04 +07:00
|
|
|
static int sbp2_map_scatterlist(struct sbp2_command_orb *orb,
|
|
|
|
struct fw_device *device, struct sbp2_logical_unit *lu)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2008-08-15 02:47:21 +07:00
|
|
|
struct scatterlist *sg = scsi_sglist(orb->cmd);
|
|
|
|
int i, n;
|
|
|
|
|
2012-05-18 23:40:19 +07:00
|
|
|
n = scsi_dma_map(orb->cmd);
|
|
|
|
if (n <= 0)
|
2007-05-10 06:23:08 +07:00
|
|
|
goto fail;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* Handle the special case where there is only one element in
|
2006-12-20 07:58:40 +07:00
|
|
|
* the scatter list by converting it to an immediate block
|
|
|
|
* request. This is also a workaround for broken devices such
|
|
|
|
* as the second generation iPod which doesn't support page
|
2007-05-08 07:33:32 +07:00
|
|
|
* tables.
|
|
|
|
*/
|
2008-08-15 02:47:21 +07:00
|
|
|
if (n == 1) {
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.data_descriptor.high =
|
|
|
|
cpu_to_be32(lu->tgt->address_high);
|
|
|
|
orb->request.data_descriptor.low =
|
|
|
|
cpu_to_be32(sg_dma_address(sg));
|
|
|
|
orb->request.misc |=
|
|
|
|
cpu_to_be32(COMMAND_ORB_DATA_SIZE(sg_dma_len(sg)));
|
2007-05-10 06:23:08 +07:00
|
|
|
return 0;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2008-08-15 02:47:21 +07:00
|
|
|
for_each_sg(sg, sg, n, i) {
|
|
|
|
orb->page_table[i].high = cpu_to_be32(sg_dma_len(sg) << 16);
|
|
|
|
orb->page_table[i].low = cpu_to_be32(sg_dma_address(sg));
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2007-07-03 03:07:34 +07:00
|
|
|
orb->page_table_bus =
|
|
|
|
dma_map_single(device->card->device, orb->page_table,
|
|
|
|
sizeof(orb->page_table), DMA_TO_DEVICE);
|
2008-07-26 09:44:49 +07:00
|
|
|
if (dma_mapping_error(device->card->device, orb->page_table_bus))
|
2007-07-03 03:07:34 +07:00
|
|
|
goto fail_page_table;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* The data_descriptor pointer is the one case where we need
|
2006-12-20 07:58:40 +07:00
|
|
|
* to fill in the node ID part of the address. All other
|
|
|
|
* pointers assume that the data referenced reside on the
|
|
|
|
* initiator (i.e. us), but data_descriptor can refer to data
|
2007-05-08 07:33:32 +07:00
|
|
|
* on other nodes so we need to put our ID in descriptor.high.
|
|
|
|
*/
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.data_descriptor.high = cpu_to_be32(lu->tgt->address_high);
|
|
|
|
orb->request.data_descriptor.low = cpu_to_be32(orb->page_table_bus);
|
|
|
|
orb->request.misc |= cpu_to_be32(COMMAND_ORB_PAGE_TABLE_PRESENT |
|
2008-08-15 02:47:21 +07:00
|
|
|
COMMAND_ORB_DATA_SIZE(n));
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-10 06:23:08 +07:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail_page_table:
|
2012-05-18 23:40:19 +07:00
|
|
|
scsi_dma_unmap(orb->cmd);
|
2007-05-10 06:23:08 +07:00
|
|
|
fail:
|
|
|
|
return -ENOMEM;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* SCSI stack integration */
|
|
|
|
|
2011-04-22 17:21:44 +07:00
|
|
|
static int sbp2_scsi_queuecommand(struct Scsi_Host *shost,
|
|
|
|
struct scsi_cmnd *cmd)
|
2006-12-20 07:58:40 +07:00
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu = cmd->device->hostdata;
|
2011-12-21 03:34:12 +07:00
|
|
|
struct fw_device *device = target_parent_device(lu->tgt);
|
2006-12-20 07:58:40 +07:00
|
|
|
struct sbp2_command_orb *orb;
|
2008-08-10 01:22:17 +07:00
|
|
|
int generation, retval = SCSI_MLQUEUE_HOST_BUSY;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-05-08 07:33:32 +07:00
|
|
|
/*
|
|
|
|
* Bidirectional commands are not yet implemented, and unknown
|
|
|
|
* transfer direction not handled.
|
|
|
|
*/
|
2006-12-20 07:58:40 +07:00
|
|
|
if (cmd->sc_data_direction == DMA_BIDIRECTIONAL) {
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_err(lu_dev(lu), "cannot handle bidirectional command\n");
|
2007-05-10 06:23:09 +07:00
|
|
|
cmd->result = DID_ERROR << 16;
|
2011-04-22 17:21:44 +07:00
|
|
|
cmd->scsi_done(cmd);
|
2007-05-10 06:23:09 +07:00
|
|
|
return 0;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2007-05-10 06:23:14 +07:00
|
|
|
orb = kzalloc(sizeof(*orb), GFP_ATOMIC);
|
2013-03-24 23:32:00 +07:00
|
|
|
if (orb == NULL)
|
2007-08-25 19:05:28 +07:00
|
|
|
return SCSI_MLQUEUE_HOST_BUSY;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-04-11 05:11:20 +07:00
|
|
|
/* Initialize rcode to something not RCODE_COMPLETE. */
|
|
|
|
orb->base.rcode = -1;
|
2007-08-25 05:59:58 +07:00
|
|
|
kref_init(&orb->base.kref);
|
2011-04-22 17:21:44 +07:00
|
|
|
orb->lu = lu;
|
|
|
|
orb->cmd = cmd;
|
2009-01-25 01:41:46 +07:00
|
|
|
orb->request.next.high = cpu_to_be32(SBP2_ORB_NULL);
|
2008-02-29 02:51:11 +07:00
|
|
|
orb->request.misc = cpu_to_be32(
|
2009-01-25 01:41:46 +07:00
|
|
|
COMMAND_ORB_MAX_PAYLOAD(lu->tgt->max_payload) |
|
2007-06-11 02:31:36 +07:00
|
|
|
COMMAND_ORB_SPEED(device->max_speed) |
|
2008-02-29 02:51:11 +07:00
|
|
|
COMMAND_ORB_NOTIFY);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
if (cmd->sc_data_direction == DMA_FROM_DEVICE)
|
2008-02-29 02:52:02 +07:00
|
|
|
orb->request.misc |= cpu_to_be32(COMMAND_ORB_DIRECTION);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-08-10 01:22:17 +07:00
|
|
|
generation = device->generation;
|
|
|
|
smp_rmb(); /* sbp2_map_scatterlist looks at tgt->address_high */
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (scsi_sg_count(cmd) && sbp2_map_scatterlist(orb, device, lu) < 0)
|
|
|
|
goto out;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2008-04-30 15:19:47 +07:00
|
|
|
memcpy(orb->request.command_block, cmd->cmnd, cmd->cmd_len);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
orb->base.callback = complete_command_orb;
|
2007-07-03 02:04:08 +07:00
|
|
|
orb->base.request_bus =
|
|
|
|
dma_map_single(device->card->device, &orb->request,
|
|
|
|
sizeof(orb->request), DMA_TO_DEVICE);
|
2009-01-28 07:03:34 +07:00
|
|
|
if (dma_mapping_error(device->card->device, orb->base.request_bus)) {
|
|
|
|
sbp2_unmap_scatterlist(device->card->device, orb);
|
2007-08-25 19:05:28 +07:00
|
|
|
goto out;
|
2009-01-28 07:03:34 +07:00
|
|
|
}
|
2007-02-07 02:49:40 +07:00
|
|
|
|
2008-08-10 01:22:17 +07:00
|
|
|
sbp2_send_orb(&orb->base, lu, lu->tgt->node_id, generation,
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->command_block_agent_address + SBP2_ORB_POINTER);
|
|
|
|
retval = 0;
|
|
|
|
out:
|
2007-08-25 05:59:58 +07:00
|
|
|
kref_put(&orb->base.kref, free_orb);
|
2007-08-25 19:05:28 +07:00
|
|
|
return retval;
|
2006-12-20 07:58:40 +07:00
|
|
|
}
|
|
|
|
|
2007-01-24 03:20:08 +07:00
|
|
|
static int sbp2_scsi_slave_alloc(struct scsi_device *sdev)
|
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu = sdev->hostdata;
|
2007-01-24 03:20:08 +07:00
|
|
|
|
2008-02-17 20:56:19 +07:00
|
|
|
/* (Re-)Adding logical units via the SCSI stack is not supported. */
|
|
|
|
if (!lu)
|
|
|
|
return -ENOSYS;
|
|
|
|
|
2007-01-24 03:20:08 +07:00
|
|
|
sdev->allow_restart = 1;
|
|
|
|
|
2012-05-19 03:26:21 +07:00
|
|
|
/*
|
|
|
|
* SBP-2 does not require any alignment, but we set it anyway
|
|
|
|
* for compatibility with earlier versions of this driver.
|
|
|
|
*/
|
2008-01-28 04:31:27 +07:00
|
|
|
blk_queue_update_dma_alignment(sdev->request_queue, 4 - 1);
|
2008-01-01 23:00:10 +07:00
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (lu->tgt->workarounds & SBP2_WORKAROUND_INQUIRY_36)
|
2007-01-24 03:20:08 +07:00
|
|
|
sdev->inquiry_len = 36;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2007-01-24 03:20:08 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
static int sbp2_scsi_slave_configure(struct scsi_device *sdev)
|
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu = sdev->hostdata;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2007-01-24 03:20:08 +07:00
|
|
|
sdev->use_10_for_rw = 1;
|
|
|
|
|
2008-05-11 05:36:47 +07:00
|
|
|
if (sbp2_param_exclusive_login)
|
|
|
|
sdev->manage_start_stop = 1;
|
|
|
|
|
2007-01-24 03:20:08 +07:00
|
|
|
if (sdev->type == TYPE_ROM)
|
|
|
|
sdev->use_10_for_ms = 1;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
if (sdev->type == TYPE_DISK &&
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->tgt->workarounds & SBP2_WORKAROUND_MODE_SENSE_8)
|
2006-12-20 07:58:40 +07:00
|
|
|
sdev->skip_ms_page_8 = 1;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
|
|
|
if (lu->tgt->workarounds & SBP2_WORKAROUND_FIX_CAPACITY)
|
2006-12-20 07:58:40 +07:00
|
|
|
sdev->fix_capacity = 1;
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2008-05-11 05:35:04 +07:00
|
|
|
if (lu->tgt->workarounds & SBP2_WORKAROUND_POWER_CONDITION)
|
|
|
|
sdev->start_stop_pwr_cond = 1;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
if (lu->tgt->workarounds & SBP2_WORKAROUND_128K_MAX_TRANS)
|
2010-02-26 12:20:38 +07:00
|
|
|
blk_queue_max_hw_sectors(sdev->request_queue, 128 * 1024 / 512);
|
2007-08-25 19:05:28 +07:00
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Called by scsi stack when something has really gone wrong. Usually
|
|
|
|
* called when a command has timed-out for some reason.
|
|
|
|
*/
|
|
|
|
static int sbp2_scsi_abort(struct scsi_cmnd *cmd)
|
|
|
|
{
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu = cmd->device->hostdata;
|
2006-12-20 07:58:40 +07:00
|
|
|
|
2011-12-21 03:34:12 +07:00
|
|
|
dev_notice(lu_dev(lu), "sbp2_scsi_abort\n");
|
2007-08-25 19:05:28 +07:00
|
|
|
sbp2_agent_reset(lu);
|
|
|
|
sbp2_cancel_orbs(lu);
|
2006-12-20 07:58:40 +07:00
|
|
|
|
|
|
|
return SUCCESS;
|
|
|
|
}
|
|
|
|
|
2007-05-27 18:18:27 +07:00
|
|
|
/*
|
|
|
|
* Format of /sys/bus/scsi/devices/.../ieee1394_id:
|
|
|
|
* u64 EUI-64 : u24 directory_ID : u16 LUN (all printed in hexadecimal)
|
|
|
|
*
|
|
|
|
* This is the concatenation of target port identifier and logical unit
|
|
|
|
* identifier as per SAM-2...SAM-4 annex A.
|
|
|
|
*/
|
2008-12-15 03:47:04 +07:00
|
|
|
static ssize_t sbp2_sysfs_ieee1394_id_show(struct device *dev,
|
|
|
|
struct device_attribute *attr, char *buf)
|
2007-05-27 18:18:27 +07:00
|
|
|
{
|
|
|
|
struct scsi_device *sdev = to_scsi_device(dev);
|
2007-08-25 19:05:28 +07:00
|
|
|
struct sbp2_logical_unit *lu;
|
2007-05-27 18:18:27 +07:00
|
|
|
|
|
|
|
if (!sdev)
|
|
|
|
return 0;
|
|
|
|
|
2007-08-25 19:05:28 +07:00
|
|
|
lu = sdev->hostdata;
|
2007-05-27 18:18:27 +07:00
|
|
|
|
2008-03-25 02:54:28 +07:00
|
|
|
return sprintf(buf, "%016llx:%06x:%04x\n",
|
|
|
|
(unsigned long long)lu->tgt->guid,
|
2007-08-25 19:05:28 +07:00
|
|
|
lu->tgt->directory_id, lu->lun);
|
2007-05-27 18:18:27 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
static DEVICE_ATTR(ieee1394_id, S_IRUGO, sbp2_sysfs_ieee1394_id_show, NULL);
|
|
|
|
|
|
|
|
static struct device_attribute *sbp2_scsi_sysfs_attrs[] = {
|
|
|
|
&dev_attr_ieee1394_id,
|
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
static struct scsi_host_template scsi_driver_template = {
|
|
|
|
.module = THIS_MODULE,
|
|
|
|
.name = "SBP-2 IEEE-1394",
|
2012-02-19 04:01:14 +07:00
|
|
|
.proc_name = "sbp2",
|
2006-12-20 07:58:40 +07:00
|
|
|
.queuecommand = sbp2_scsi_queuecommand,
|
2007-01-24 03:20:08 +07:00
|
|
|
.slave_alloc = sbp2_scsi_slave_alloc,
|
2006-12-20 07:58:40 +07:00
|
|
|
.slave_configure = sbp2_scsi_slave_configure,
|
|
|
|
.eh_abort_handler = sbp2_scsi_abort,
|
|
|
|
.this_id = -1,
|
|
|
|
.sg_tablesize = SG_ALL,
|
|
|
|
.use_clustering = ENABLE_CLUSTERING,
|
2007-01-22 02:50:11 +07:00
|
|
|
.cmd_per_lun = 1,
|
|
|
|
.can_queue = 1,
|
2007-05-27 18:18:27 +07:00
|
|
|
.sdev_attrs = sbp2_scsi_sysfs_attrs,
|
2006-12-20 07:58:40 +07:00
|
|
|
};
|
|
|
|
|
|
|
|
MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
|
|
|
|
MODULE_DESCRIPTION("SCSI over IEEE1394");
|
|
|
|
MODULE_LICENSE("GPL");
|
|
|
|
MODULE_DEVICE_TABLE(ieee1394, sbp2_id_table);
|
|
|
|
|
2007-05-06 04:17:13 +07:00
|
|
|
/* Provide a module alias so root-on-sbp2 initrds don't break. */
|
|
|
|
MODULE_ALIAS("sbp2");
|
|
|
|
|
2006-12-20 07:58:40 +07:00
|
|
|
static int __init sbp2_init(void)
|
|
|
|
{
|
|
|
|
return driver_register(&sbp2_driver.driver);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __exit sbp2_cleanup(void)
|
|
|
|
{
|
|
|
|
driver_unregister(&sbp2_driver.driver);
|
|
|
|
}
|
|
|
|
|
|
|
|
module_init(sbp2_init);
|
|
|
|
module_exit(sbp2_cleanup);
|