android_kernel_xiaomi_sm8350/drivers/firewire/fw-sbp2.c

1652 lines
47 KiB
C
Raw Normal View History

/*
* SBP2 driver (SCSI over IEEE1394)
*
* Copyright (C) 2005-2007 Kristian Hoegsberg <krh@bitplanet.net>
*
* 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.
*/
/*
* The basic structure of this driver is based on the old storage driver,
* 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.
*/
#include <linux/blkdev.h>
#include <linux/bug.h>
#include <linux/delay.h>
#include <linux/device.h>
#include <linux/dma-mapping.h>
#include <linux/kernel.h>
#include <linux/mod_devicetable.h>
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/scatterlist.h>
#include <linux/string.h>
#include <linux/stringify.h>
#include <linux/timer.h>
#include <linux/workqueue.h>
#include <asm/system.h>
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include "fw-device.h"
#include "fw-topology.h"
#include "fw-transaction.h"
/*
* 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.
*/
static int sbp2_param_exclusive_login = 1;
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)");
/*
* 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.
*
* - delay inquiry
* Wait extra SBP2_INQUIRY_DELAY seconds after login before SCSI inquiry.
*
* - 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.
*
* - 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
#define SBP2_WORKAROUND_DELAY_INQUIRY 0x10
#define SBP2_INQUIRY_DELAY 12
#define SBP2_WORKAROUND_POWER_CONDITION 0x20
#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)
", delay inquiry = " __stringify(SBP2_WORKAROUND_DELAY_INQUIRY)
", set power condition in start stop unit = "
__stringify(SBP2_WORKAROUND_POWER_CONDITION)
", override internal blacklist = " __stringify(SBP2_WORKAROUND_OVERRIDE)
", or a combination)");
/* I don't know why the SCSI stack doesn't define something like this... */
typedef void (*scsi_done_fn_t)(struct scsi_cmnd *);
static const char sbp2_driver_name[] = "sbp2";
/*
* 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;
struct fw_address_handler address_handler;
struct list_head orb_list;
u64 command_block_agent_address;
u16 lun;
int login_id;
/*
* 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.
*/
int generation;
int retries;
struct delayed_work work;
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 10:37:28 -05:00
bool blocked;
};
/*
* We create one struct sbp2_target per IEEE 1212 Unit Directory
* and one struct Scsi_Host per sbp2_target.
*/
struct sbp2_target {
struct kref kref;
struct fw_unit *unit;
const char *bus_id;
struct list_head lu_list;
u64 management_agent_address;
u64 guid;
int directory_id;
int node_id;
int address_high;
unsigned int workarounds;
unsigned int mgt_orb_timeout;
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 10:37:28 -05:00
int dont_block; /* counter for each logical unit */
int blocked; /* ditto */
};
/* Impossible login_id, to detect logout attempt before successful login */
#define INVALID_LOGIN_ID 0x10000
/*
* 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.
*/
#define SBP2_MIN_LOGIN_ORB_TIMEOUT 5000U /* Timeout in ms */
#define SBP2_MAX_LOGIN_ORB_TIMEOUT 40000U /* Timeout in ms */
#define SBP2_ORB_TIMEOUT 2000U /* Timeout in ms */
#define SBP2_ORB_NULL 0x80000000
#define SBP2_RETRY_LIMIT 0xf /* 15 retries */
#define SBP2_CYCLE_LIMIT (0xc8 << 12) /* 200 125us cycles */
/*
* The default maximum s/g segment size of a FireWire controller is
* usually 0x10000, but SBP-2 only allows 0xffff. Since buffers have to
* be quadlet-aligned, we set the length limit to 0xffff & ~3.
*/
#define SBP2_MAX_SEG_SIZE 0xfffc
/* Unit directory keys */
#define SBP2_CSR_UNIT_CHARACTERISTICS 0x3a
#define SBP2_CSR_FIRMWARE_REVISION 0x3c
#define SBP2_CSR_LOGICAL_UNIT_NUMBER 0x14
#define SBP2_CSR_LOGICAL_UNIT_DIRECTORY 0xd4
/* 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
#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)
struct sbp2_status {
u32 status;
u32 orb_low;
u8 data[24];
};
struct sbp2_pointer {
__be32 high;
__be32 low;
};
struct sbp2_orb {
struct fw_transaction t;
struct kref kref;
dma_addr_t request_bus;
int rcode;
struct sbp2_pointer pointer;
void (*callback)(struct sbp2_orb * orb, struct sbp2_status * status);
struct list_head link;
};
#define MANAGEMENT_ORB_LUN(v) ((v))
#define MANAGEMENT_ORB_FUNCTION(v) ((v) << 16)
#define MANAGEMENT_ORB_RECONNECT(v) ((v) << 20)
#define MANAGEMENT_ORB_EXCLUSIVE(v) ((v) ? 1 << 28 : 0)
#define MANAGEMENT_ORB_REQUEST_FORMAT(v) ((v) << 29)
#define MANAGEMENT_ORB_NOTIFY ((1) << 31)
#define MANAGEMENT_ORB_RESPONSE_LENGTH(v) ((v))
#define MANAGEMENT_ORB_PASSWORD_LENGTH(v) ((v) << 16)
struct sbp2_management_orb {
struct sbp2_orb base;
struct {
struct sbp2_pointer password;
struct sbp2_pointer response;
__be32 misc;
__be32 length;
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 {
__be32 misc;
struct sbp2_pointer command_block_agent;
__be32 reconnect_hold;
};
#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)
#define COMMAND_ORB_DIRECTION ((1) << 27)
#define COMMAND_ORB_REQUEST_FORMAT(v) ((v) << 29)
#define COMMAND_ORB_NOTIFY ((1) << 31)
struct sbp2_command_orb {
struct sbp2_orb base;
struct {
struct sbp2_pointer next;
struct sbp2_pointer data_descriptor;
__be32 misc;
u8 command_block[12];
} request;
struct scsi_cmnd *cmd;
scsi_done_fn_t done;
struct sbp2_logical_unit *lu;
struct sbp2_pointer page_table[SG_ALL] __attribute__((aligned(8)));
dma_addr_t page_table_bus;
};
/*
* 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
* so far. We use ~0 as a wildcard, since the 24 bit values we get
* from the config rom can never match that.
*/
static const struct {
u32 firmware_revision;
u32 model;
unsigned int workarounds;
} sbp2_workarounds_table[] = {
/* DViCO Momobay CX-1 with TSB42AA9 bridge */ {
.firmware_revision = 0x002800,
.model = 0x001010,
.workarounds = SBP2_WORKAROUND_INQUIRY_36 |
SBP2_WORKAROUND_MODE_SENSE_8 |
SBP2_WORKAROUND_POWER_CONDITION,
},
/* DViCO Momobay FX-3A with TSB42AA9A bridge */ {
.firmware_revision = 0x002800,
.model = 0x000000,
.workarounds = SBP2_WORKAROUND_DELAY_INQUIRY |
SBP2_WORKAROUND_POWER_CONDITION,
},
/* Initio bridges, actually only needed for some older ones */ {
.firmware_revision = 0x000200,
.model = ~0,
.workarounds = SBP2_WORKAROUND_INQUIRY_36,
},
/* PL-3507 bridge with Prolific firmware */ {
.firmware_revision = 0x012800,
.model = ~0,
.workarounds = SBP2_WORKAROUND_POWER_CONDITION,
},
/* Symbios bridge */ {
.firmware_revision = 0xa0b800,
.model = ~0,
.workarounds = SBP2_WORKAROUND_128K_MAX_TRANS,
},
/* Datafab MD2-FW2 with Symbios/LSILogic SYM13FW500 bridge */ {
.firmware_revision = 0x002600,
.model = ~0,
.workarounds = SBP2_WORKAROUND_128K_MAX_TRANS,
},
/*
* There are iPods (2nd gen, 3rd gen) with model_id == 0, but
* these iPods do not feature the read_capacity bug according
* to one report. Read_capacity behaviour as well as model_id
* could change due to Apple-supplied firmware updates though.
*/
/* iPod 4th generation. */ {
.firmware_revision = 0x0a2700,
.model = 0x000021,
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
},
/* iPod mini */ {
.firmware_revision = 0x0a2700,
.model = 0x000022,
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
},
/* iPod mini */ {
.firmware_revision = 0x0a2700,
.model = 0x000023,
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
},
/* iPod Photo */ {
.firmware_revision = 0x0a2700,
.model = 0x00007e,
.workarounds = SBP2_WORKAROUND_FIX_CAPACITY,
}
};
static void
free_orb(struct kref *kref)
{
struct sbp2_orb *orb = container_of(kref, struct sbp2_orb, kref);
kfree(orb);
}
static void
sbp2_status_write(struct fw_card *card, struct fw_request *request,
int tcode, int destination, int source,
int generation, int speed,
unsigned long long offset,
void *payload, size_t length, void *callback_data)
{
struct sbp2_logical_unit *lu = callback_data;
struct sbp2_orb *orb;
struct sbp2_status status;
size_t header_size;
unsigned long flags;
if (tcode != TCODE_WRITE_BLOCK_REQUEST ||
length == 0 || length > sizeof(status)) {
fw_send_response(card, request, RCODE_TYPE_ERROR);
return;
}
header_size = min(length, 2 * sizeof(u32));
fw_memcpy_from_be32(&status, payload, header_size);
if (length > header_size)
memcpy(status.data, payload + 8, length - header_size);
if (STATUS_GET_SOURCE(status) == 2 || STATUS_GET_SOURCE(status) == 3) {
fw_notify("non-orb related status write, not handled\n");
fw_send_response(card, request, RCODE_COMPLETE);
return;
}
/* Lookup the orb corresponding to this status write. */
spin_lock_irqsave(&card->lock, flags);
list_for_each_entry(orb, &lu->orb_list, link) {
if (STATUS_GET_ORB_HIGH(status) == 0 &&
STATUS_GET_ORB_LOW(status) == orb->request_bus) {
orb->rcode = RCODE_COMPLETE;
list_del(&orb->link);
break;
}
}
spin_unlock_irqrestore(&card->lock, flags);
if (&orb->link != &lu->orb_list)
orb->callback(orb, &status);
else
fw_error("status write for unknown orb\n");
kref_put(&orb->kref, free_orb);
fw_send_response(card, request, RCODE_COMPLETE);
}
static void
complete_transaction(struct fw_card *card, int rcode,
void *payload, size_t length, void *data)
{
struct sbp2_orb *orb = data;
unsigned long flags;
/*
* 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;
if (orb->rcode != RCODE_COMPLETE) {
list_del(&orb->link);
spin_unlock_irqrestore(&card->lock, flags);
orb->callback(orb, NULL);
} else {
spin_unlock_irqrestore(&card->lock, flags);
}
kref_put(&orb->kref, free_orb);
}
static void
sbp2_send_orb(struct sbp2_orb *orb, struct sbp2_logical_unit *lu,
int node_id, int generation, u64 offset)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
unsigned long flags;
orb->pointer.high = 0;
orb->pointer.low = cpu_to_be32(orb->request_bus);
spin_lock_irqsave(&device->card->lock, flags);
list_add_tail(&orb->link, &lu->orb_list);
spin_unlock_irqrestore(&device->card->lock, flags);
/* Take a ref for the orb list and for the transaction callback. */
kref_get(&orb->kref);
kref_get(&orb->kref);
fw_send_request(device->card, &orb->t, TCODE_WRITE_BLOCK_REQUEST,
node_id, generation, device->max_speed, offset,
&orb->pointer, sizeof(orb->pointer),
complete_transaction, orb);
}
static int sbp2_cancel_orbs(struct sbp2_logical_unit *lu)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
struct sbp2_orb *orb, *next;
struct list_head list;
unsigned long flags;
int retval = -ENOENT;
INIT_LIST_HEAD(&list);
spin_lock_irqsave(&device->card->lock, flags);
list_splice_init(&lu->orb_list, &list);
spin_unlock_irqrestore(&device->card->lock, flags);
list_for_each_entry_safe(orb, next, &list, link) {
retval = 0;
if (fw_cancel_transaction(device->card, &orb->t) == 0)
continue;
orb->rcode = RCODE_CANCELLED;
orb->callback(orb, NULL);
}
return retval;
}
static void
complete_management_orb(struct sbp2_orb *base_orb, struct sbp2_status *status)
{
struct sbp2_management_orb *orb =
container_of(base_orb, struct sbp2_management_orb, base);
if (status)
memcpy(&orb->status, status, sizeof(*status));
complete(&orb->done);
}
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)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
struct sbp2_management_orb *orb;
unsigned int timeout;
int retval = -ENOMEM;
firewire: fw-sbp2: don't retry login or reconnect after unplug If a device is being unplugged while fw-sbp2 had a login or reconnect on schedule, it would take about half a minute to shut the fw_unit down: Jan 27 18:34:54 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 18:34:59 stein firewire_sbp2: sbp2_scsi_abort Jan 27 18:34:59 stein scsi 25:0:0:0: Device offlined - not ready after error recovery Jan 27 18:35:01 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:06 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:12 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:17 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:22 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:27 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: failed to login to fw2.0 LUN 0000 Jan 27 18:35:32 stein firewire_sbp2: released fw2.0 After this patch, typically only a few seconds spent in __scsi_add_device remain: Jan 27 19:05:50 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 19:05:56 stein firewire_sbp2: sbp2_scsi_abort Jan 27 19:05:56 stein scsi 33:0:0:0: Device offlined - not ready after error recovery Jan 27 19:05:56 stein firewire_sbp2: released fw2.0 The benefit of this is less noise in the syslog. It furthermore avoids a few wasted CPU cycles and needlessly prolonged lifetime of a few driver objects. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-01-27 13:14:44 -05:00
if (function == SBP2_LOGOUT_REQUEST && fw_device_is_shutdown(device))
return 0;
orb = kzalloc(sizeof(*orb), GFP_ATOMIC);
if (orb == NULL)
return -ENOMEM;
kref_init(&orb->base.kref);
orb->response_bus =
dma_map_single(device->card->device, &orb->response,
sizeof(orb->response), DMA_FROM_DEVICE);
dma-mapping: add the device argument to dma_mapping_error() Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER architecture does: This enables us to cleanly fix the Calgary IOMMU issue that some devices are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423). I think that per-device dma_mapping_ops support would be also helpful for KVM people to support PCI passthrough but Andi thinks that this makes it difficult to support the PCI passthrough (see the above thread). So I CC'ed this to KVM camp. Comments are appreciated. A pointer to dma_mapping_ops to struct dev_archdata is added. If the pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's NULL, the system-wide dma_ops pointer is used as before. If it's useful for KVM people, I plan to implement a mechanism to register a hook called when a new pci (or dma capable) device is created (it works with hot plugging). It enables IOMMUs to set up an appropriate dma_mapping_ops per device. The major obstacle is that dma_mapping_error doesn't take a pointer to the device unlike other DMA operations. So x86 can't have dma_mapping_ops per device. Note all the POWER IOMMUs use the same dma_mapping_error function so this is not a problem for POWER but x86 IOMMUs use different dma_mapping_error functions. The first patch adds the device argument to dma_mapping_error. The patch is trivial but large since it touches lots of drivers and dma-mapping.h in all the architecture. This patch: dma_mapping_error() doesn't take a pointer to the device unlike other DMA operations. So we can't have dma_mapping_ops per device. Note that POWER already has dma_mapping_ops per device but all the POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device argument. [akpm@linux-foundation.org: fix sge] [akpm@linux-foundation.org: fix svc_rdma] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix bnx2x] [akpm@linux-foundation.org: fix s2io] [akpm@linux-foundation.org: fix pasemi_mac] [akpm@linux-foundation.org: fix sdhci] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix sparc] [akpm@linux-foundation.org: fix ibmvscsi] Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Cc: Muli Ben-Yehuda <muli@il.ibm.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Avi Kivity <avi@qumranet.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 22:44:49 -04:00
if (dma_mapping_error(device->card->device, orb->response_bus))
goto fail_mapping_response;
orb->request.response.high = 0;
orb->request.response.low = cpu_to_be32(orb->response_bus);
orb->request.misc = cpu_to_be32(
MANAGEMENT_ORB_NOTIFY |
MANAGEMENT_ORB_FUNCTION(function) |
MANAGEMENT_ORB_LUN(lun_or_login_id));
orb->request.length = cpu_to_be32(
MANAGEMENT_ORB_RESPONSE_LENGTH(sizeof(orb->response)));
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);
if (function == SBP2_LOGIN_REQUEST) {
firewire: fw-sbp2: try to increase reconnect_hold (speed up reconnection) Ask the target to grant 4 seconds instead of the standard and minimum of 1 second window after bus reset for reconnection. This accelerates reconnection if there are more than one targets on the bus: If a login and inquiry to one target blocks the fw-sbp2 workqueue for more than 1s after bus reset, we now still can reconnect to the other target. Before that, fw-sbp2's reconnect attempts would be rejected with "error status: 0:9" (function rejected), and fw-sbp2 would finally re-login. All those futile reconnect attemps cost extra time until the target which needs re-login is ready for I/O again. The reconnect timeout field in the login ORB doesn't have to be honored by the target though. I found that we could get up to - allegedly 32768s from an old OXFW911 firmware - 256s from LSI bridges - 4s from OXUF922 and OXFW912 bridges, - 2s from TI bridges, - only the standard 1s from Initio and Prolific bridges and from Apple OpenFirmware in target mode. We just try to get 4 seconds which already covers the case of a few HDDs on the same bus quite nicely. A minor drawback occurs in the following (rare and impractical) border case: - two initiators are there, initiator 1 holds an exclusive login to a target, - initiator 1 goes off the bus, - target refuses login attempts from initiator 2 until reconnect_hold seconds after bus reset. An alternative approach to the issue at hand would be to parallelize fw-sbp2's reconnect and login work. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Acked-by: Jarod Wilson <jwilson@redhat.com>
2008-01-19 19:25:31 -05:00
/* Ask for 2^2 == 4 seconds reconnect grace period */
orb->request.misc |= cpu_to_be32(
firewire: fw-sbp2: try to increase reconnect_hold (speed up reconnection) Ask the target to grant 4 seconds instead of the standard and minimum of 1 second window after bus reset for reconnection. This accelerates reconnection if there are more than one targets on the bus: If a login and inquiry to one target blocks the fw-sbp2 workqueue for more than 1s after bus reset, we now still can reconnect to the other target. Before that, fw-sbp2's reconnect attempts would be rejected with "error status: 0:9" (function rejected), and fw-sbp2 would finally re-login. All those futile reconnect attemps cost extra time until the target which needs re-login is ready for I/O again. The reconnect timeout field in the login ORB doesn't have to be honored by the target though. I found that we could get up to - allegedly 32768s from an old OXFW911 firmware - 256s from LSI bridges - 4s from OXUF922 and OXFW912 bridges, - 2s from TI bridges, - only the standard 1s from Initio and Prolific bridges and from Apple OpenFirmware in target mode. We just try to get 4 seconds which already covers the case of a few HDDs on the same bus quite nicely. A minor drawback occurs in the following (rare and impractical) border case: - two initiators are there, initiator 1 holds an exclusive login to a target, - initiator 1 goes off the bus, - target refuses login attempts from initiator 2 until reconnect_hold seconds after bus reset. An alternative approach to the issue at hand would be to parallelize fw-sbp2's reconnect and login work. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Acked-by: Jarod Wilson <jwilson@redhat.com>
2008-01-19 19:25:31 -05:00
MANAGEMENT_ORB_RECONNECT(2) |
MANAGEMENT_ORB_EXCLUSIVE(sbp2_param_exclusive_login));
timeout = lu->tgt->mgt_orb_timeout;
} else {
timeout = SBP2_ORB_TIMEOUT;
}
init_completion(&orb->done);
orb->base.callback = complete_management_orb;
orb->base.request_bus =
dma_map_single(device->card->device, &orb->request,
sizeof(orb->request), DMA_TO_DEVICE);
dma-mapping: add the device argument to dma_mapping_error() Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER architecture does: This enables us to cleanly fix the Calgary IOMMU issue that some devices are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423). I think that per-device dma_mapping_ops support would be also helpful for KVM people to support PCI passthrough but Andi thinks that this makes it difficult to support the PCI passthrough (see the above thread). So I CC'ed this to KVM camp. Comments are appreciated. A pointer to dma_mapping_ops to struct dev_archdata is added. If the pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's NULL, the system-wide dma_ops pointer is used as before. If it's useful for KVM people, I plan to implement a mechanism to register a hook called when a new pci (or dma capable) device is created (it works with hot plugging). It enables IOMMUs to set up an appropriate dma_mapping_ops per device. The major obstacle is that dma_mapping_error doesn't take a pointer to the device unlike other DMA operations. So x86 can't have dma_mapping_ops per device. Note all the POWER IOMMUs use the same dma_mapping_error function so this is not a problem for POWER but x86 IOMMUs use different dma_mapping_error functions. The first patch adds the device argument to dma_mapping_error. The patch is trivial but large since it touches lots of drivers and dma-mapping.h in all the architecture. This patch: dma_mapping_error() doesn't take a pointer to the device unlike other DMA operations. So we can't have dma_mapping_ops per device. Note that POWER already has dma_mapping_ops per device but all the POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device argument. [akpm@linux-foundation.org: fix sge] [akpm@linux-foundation.org: fix svc_rdma] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix bnx2x] [akpm@linux-foundation.org: fix s2io] [akpm@linux-foundation.org: fix pasemi_mac] [akpm@linux-foundation.org: fix sdhci] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix sparc] [akpm@linux-foundation.org: fix ibmvscsi] Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Cc: Muli Ben-Yehuda <muli@il.ibm.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Avi Kivity <avi@qumranet.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 22:44:49 -04:00
if (dma_mapping_error(device->card->device, orb->base.request_bus))
goto fail_mapping_request;
sbp2_send_orb(&orb->base, lu, node_id, generation,
lu->tgt->management_agent_address);
wait_for_completion_timeout(&orb->done, msecs_to_jiffies(timeout));
retval = -EIO;
if (sbp2_cancel_orbs(lu) == 0) {
fw_error("%s: orb reply timed out, rcode=0x%02x\n",
lu->tgt->bus_id, orb->base.rcode);
goto out;
}
if (orb->base.rcode != RCODE_COMPLETE) {
fw_error("%s: management write failed, rcode 0x%02x\n",
lu->tgt->bus_id, orb->base.rcode);
goto out;
}
if (STATUS_GET_RESPONSE(orb->status) != 0 ||
STATUS_GET_SBP_STATUS(orb->status) != 0) {
fw_error("%s: error status: %d:%d\n", lu->tgt->bus_id,
STATUS_GET_RESPONSE(orb->status),
STATUS_GET_SBP_STATUS(orb->status));
goto out;
}
retval = 0;
out:
dma_unmap_single(device->card->device, orb->base.request_bus,
sizeof(orb->request), DMA_TO_DEVICE);
fail_mapping_request:
dma_unmap_single(device->card->device, orb->response_bus,
sizeof(orb->response), DMA_FROM_DEVICE);
fail_mapping_response:
if (response)
memcpy(response, orb->response, sizeof(orb->response));
kref_put(&orb->base.kref, free_orb);
return retval;
}
static void sbp2_agent_reset(struct sbp2_logical_unit *lu)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
__be32 d = 0;
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,
&d, sizeof(d));
}
static void
complete_agent_reset_write_no_wait(struct fw_card *card, int rcode,
void *payload, size_t length, void *data)
{
kfree(data);
}
static void sbp2_agent_reset_no_wait(struct sbp2_logical_unit *lu)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
struct fw_transaction *t;
static __be32 d;
t = kmalloc(sizeof(*t), GFP_ATOMIC);
if (t == NULL)
return;
fw_send_request(device->card, t, TCODE_WRITE_QUADLET_REQUEST,
lu->tgt->node_id, lu->generation, device->max_speed,
lu->command_block_agent_address + SBP2_AGENT_RESET,
&d, sizeof(d), complete_agent_reset_write_no_wait, t);
}
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 10:37:28 -05:00
static void sbp2_set_generation(struct sbp2_logical_unit *lu, int generation)
{
struct fw_card *card = fw_device(lu->tgt->unit->device.parent)->card;
unsigned long flags;
/* serialize with comparisons of lu->generation and card->generation */
spin_lock_irqsave(&card->lock, flags);
lu->generation = generation;
spin_unlock_irqrestore(&card->lock, flags);
}
static inline void sbp2_allow_block(struct sbp2_logical_unit *lu)
{
/*
* We may access dont_block without taking card->lock here:
* All callers of sbp2_allow_block() and all callers of sbp2_unblock()
* are currently serialized against each other.
* And a wrong result in sbp2_conditionally_block()'s access of
* dont_block is rather harmless, it simply misses its first chance.
*/
--lu->tgt->dont_block;
}
/*
* 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;
struct fw_card *card = fw_device(tgt->unit->device.parent)->card;
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;
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 10:37:28 -05:00
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;
struct fw_card *card = fw_device(tgt->unit->device.parent)->card;
struct Scsi_Host *shost =
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
unsigned long flags;
bool unblock = false;
spin_lock_irqsave(&card->lock, flags);
if (lu->blocked && lu->generation == card->generation) {
lu->blocked = false;
unblock = --tgt->blocked == 0;
}
spin_unlock_irqrestore(&card->lock, flags);
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 10:37:28 -05:00
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)
{
struct fw_card *card = fw_device(tgt->unit->device.parent)->card;
struct Scsi_Host *shost =
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
unsigned long flags;
spin_lock_irqsave(&card->lock, flags);
++tgt->dont_block;
spin_unlock_irqrestore(&card->lock, flags);
scsi_unblock_requests(shost);
}
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);
}
static void sbp2_release_target(struct kref *kref)
{
struct sbp2_target *tgt = container_of(kref, struct sbp2_target, kref);
struct sbp2_logical_unit *lu, *next;
struct Scsi_Host *shost =
container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
struct scsi_device *sdev;
firewire: fix crash in automatic module unloading "modprobe firewire-ohci; sleep .1; modprobe -r firewire-ohci" used to result in crashes like this: BUG: unable to handle kernel paging request at ffffffff8807b455 IP: [<ffffffff8807b455>] PGD 203067 PUD 207063 PMD 7c170067 PTE 0 Oops: 0010 [1] PREEMPT SMP CPU 0 Modules linked in: i915 drm cpufreq_ondemand acpi_cpufreq freq_table applesmc input_polldev led_class coretemp hwmon eeprom snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss button thermal processor sg snd_hda_intel snd_pcm snd_timer snd snd_page_alloc sky2 i2c_i801 rtc [last unloaded: crc_itu_t] Pid: 9, comm: events/0 Not tainted 2.6.25-rc2 #3 RIP: 0010:[<ffffffff8807b455>] [<ffffffff8807b455>] RSP: 0018:ffff81007dcdde88 EFLAGS: 00010246 RAX: ffff81007dc95040 RBX: ffff81007dee5390 RCX: 0000000000005e13 RDX: 0000000000008c8b RSI: 0000000000000001 RDI: ffff81007dee5388 RBP: ffff81007dc5eb40 R08: 0000000000000002 R09: ffffffff8022d05c R10: ffffffff8023b34c R11: ffffffff8041a353 R12: ffff81007dee5388 R13: ffffffff8807b455 R14: ffffffff80593bc0 R15: 0000000000000000 FS: 0000000000000000(0000) GS:ffffffff8055a000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: ffffffff8807b455 CR3: 0000000000201000 CR4: 00000000000006e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process events/0 (pid: 9, threadinfo ffff81007dcdc000, task ffff81007dc95040) Stack: ffffffff8023b396 ffffffff88082524 0000000000000000 ffffffff8807d9ae ffff81007dc5eb40 ffff81007dc9dce0 ffff81007dc5eb40 ffff81007dc5eb80 ffff81007dc9dce0 ffffffffffffffff ffffffff8023be87 0000000000000000 Call Trace: [<ffffffff8023b396>] ? run_workqueue+0xdf/0x1df [<ffffffff8023be87>] ? worker_thread+0xd8/0xe3 [<ffffffff8023e917>] ? autoremove_wake_function+0x0/0x2e [<ffffffff8023bdaf>] ? worker_thread+0x0/0xe3 [<ffffffff8023e813>] ? kthread+0x47/0x74 [<ffffffff804198e0>] ? trace_hardirqs_on_thunk+0x35/0x3a [<ffffffff8020c008>] ? child_rip+0xa/0x12 [<ffffffff8020b6e3>] ? restore_args+0x0/0x3d [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e7cc>] ? kthread+0x0/0x74 [<ffffffff8020bffe>] ? child_rip+0x0/0x12 Code: Bad RIP value. RIP [<ffffffff8807b455>] RSP <ffff81007dcdde88> CR2: ffffffff8807b455 ---[ end trace c7366c6657fe5bed ]--- Note that this crash happened _after_ firewire-core was unloaded. The shared workqueue tried to run firewire-core's device initialization jobs or similar jobs. The fix makes sure that firewire-ohci and hence firewire-core is not unloaded before all device shutdown jobs have been completed. This is determined by the count of device initializations minus device releases. Also skip useless retries in the node initialization job if the node is to be shut down. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-02-27 16:14:27 -05:00
struct fw_device *device = fw_device(tgt->unit->device.parent);
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 10:37:28 -05:00
/* prevent deadlocks */
sbp2_unblock(tgt);
list_for_each_entry_safe(lu, next, &tgt->lu_list, link) {
sdev = scsi_device_lookup(shost, 0, 0, sbp2_lun2int(lu->lun));
if (sdev) {
scsi_remove_device(sdev);
scsi_device_put(sdev);
}
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);
}
scsi_remove_host(shost);
fw_notify("released %s, target %d:0:0\n", tgt->bus_id, shost->host_no);
fw_unit_put(tgt->unit);
scsi_host_put(shost);
firewire: fix crash in automatic module unloading "modprobe firewire-ohci; sleep .1; modprobe -r firewire-ohci" used to result in crashes like this: BUG: unable to handle kernel paging request at ffffffff8807b455 IP: [<ffffffff8807b455>] PGD 203067 PUD 207063 PMD 7c170067 PTE 0 Oops: 0010 [1] PREEMPT SMP CPU 0 Modules linked in: i915 drm cpufreq_ondemand acpi_cpufreq freq_table applesmc input_polldev led_class coretemp hwmon eeprom snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss button thermal processor sg snd_hda_intel snd_pcm snd_timer snd snd_page_alloc sky2 i2c_i801 rtc [last unloaded: crc_itu_t] Pid: 9, comm: events/0 Not tainted 2.6.25-rc2 #3 RIP: 0010:[<ffffffff8807b455>] [<ffffffff8807b455>] RSP: 0018:ffff81007dcdde88 EFLAGS: 00010246 RAX: ffff81007dc95040 RBX: ffff81007dee5390 RCX: 0000000000005e13 RDX: 0000000000008c8b RSI: 0000000000000001 RDI: ffff81007dee5388 RBP: ffff81007dc5eb40 R08: 0000000000000002 R09: ffffffff8022d05c R10: ffffffff8023b34c R11: ffffffff8041a353 R12: ffff81007dee5388 R13: ffffffff8807b455 R14: ffffffff80593bc0 R15: 0000000000000000 FS: 0000000000000000(0000) GS:ffffffff8055a000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: ffffffff8807b455 CR3: 0000000000201000 CR4: 00000000000006e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process events/0 (pid: 9, threadinfo ffff81007dcdc000, task ffff81007dc95040) Stack: ffffffff8023b396 ffffffff88082524 0000000000000000 ffffffff8807d9ae ffff81007dc5eb40 ffff81007dc9dce0 ffff81007dc5eb40 ffff81007dc5eb80 ffff81007dc9dce0 ffffffffffffffff ffffffff8023be87 0000000000000000 Call Trace: [<ffffffff8023b396>] ? run_workqueue+0xdf/0x1df [<ffffffff8023be87>] ? worker_thread+0xd8/0xe3 [<ffffffff8023e917>] ? autoremove_wake_function+0x0/0x2e [<ffffffff8023bdaf>] ? worker_thread+0x0/0xe3 [<ffffffff8023e813>] ? kthread+0x47/0x74 [<ffffffff804198e0>] ? trace_hardirqs_on_thunk+0x35/0x3a [<ffffffff8020c008>] ? child_rip+0xa/0x12 [<ffffffff8020b6e3>] ? restore_args+0x0/0x3d [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e7cc>] ? kthread+0x0/0x74 [<ffffffff8020bffe>] ? child_rip+0x0/0x12 Code: Bad RIP value. RIP [<ffffffff8807b455>] RSP <ffff81007dcdde88> CR2: ffffffff8807b455 ---[ end trace c7366c6657fe5bed ]--- Note that this crash happened _after_ firewire-core was unloaded. The shared workqueue tried to run firewire-core's device initialization jobs or similar jobs. The fix makes sure that firewire-ohci and hence firewire-core is not unloaded before all device shutdown jobs have been completed. This is determined by the count of device initializations minus device releases. Also skip useless retries in the node initialization job if the node is to be shut down. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-02-27 16:14:27 -05:00
fw_device_put(device);
}
static struct workqueue_struct *sbp2_wq;
static void sbp2_target_put(struct sbp2_target *tgt)
{
kref_put(&tgt->kref, sbp2_release_target);
}
/*
* Always get the target's kref when scheduling work on one its units.
* Each workqueue job is responsible to call sbp2_target_put() upon return.
*/
static void sbp2_queue_work(struct sbp2_logical_unit *lu, unsigned long delay)
{
kref_get(&lu->tgt->kref);
if (!queue_delayed_work(sbp2_wq, &lu->work, delay))
sbp2_target_put(lu->tgt);
}
/*
* 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.
*/
static void sbp2_set_busy_timeout(struct sbp2_logical_unit *lu)
{
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
__be32 d = cpu_to_be32(SBP2_CYCLE_LIMIT | SBP2_RETRY_LIMIT);
fw_run_transaction(device->card, TCODE_WRITE_QUADLET_REQUEST,
lu->tgt->node_id, lu->generation, device->max_speed,
CSR_REGISTER_BASE + CSR_BUSY_TIMEOUT,
&d, sizeof(d));
}
static void sbp2_reconnect(struct work_struct *work);
static void sbp2_login(struct work_struct *work)
{
struct sbp2_logical_unit *lu =
container_of(work, struct sbp2_logical_unit, work.work);
struct sbp2_target *tgt = lu->tgt;
struct fw_device *device = fw_device(tgt->unit->device.parent);
struct Scsi_Host *shost;
struct scsi_device *sdev;
struct sbp2_login_response response;
int generation, node_id, local_node_id;
firewire: fw-sbp2: don't retry login or reconnect after unplug If a device is being unplugged while fw-sbp2 had a login or reconnect on schedule, it would take about half a minute to shut the fw_unit down: Jan 27 18:34:54 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 18:34:59 stein firewire_sbp2: sbp2_scsi_abort Jan 27 18:34:59 stein scsi 25:0:0:0: Device offlined - not ready after error recovery Jan 27 18:35:01 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:06 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:12 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:17 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:22 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:27 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: failed to login to fw2.0 LUN 0000 Jan 27 18:35:32 stein firewire_sbp2: released fw2.0 After this patch, typically only a few seconds spent in __scsi_add_device remain: Jan 27 19:05:50 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 19:05:56 stein firewire_sbp2: sbp2_scsi_abort Jan 27 19:05:56 stein scsi 33:0:0:0: Device offlined - not ready after error recovery Jan 27 19:05:56 stein firewire_sbp2: released fw2.0 The benefit of this is less noise in the syslog. It furthermore avoids a few wasted CPU cycles and needlessly prolonged lifetime of a few driver objects. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-01-27 13:14:44 -05:00
if (fw_device_is_shutdown(device))
goto out;
generation = device->generation;
smp_rmb(); /* node_id must not be older than generation */
node_id = device->node_id;
local_node_id = device->card->node_id;
/* If this is a re-login attempt, log out, or we might be rejected. */
if (lu->has_sdev)
sbp2_send_management_orb(lu, device->node_id, generation,
SBP2_LOGOUT_REQUEST, lu->login_id, NULL);
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 10:37:28 -05:00
if (lu->retries++ < 5) {
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 10:37:28 -05:00
} else {
fw_error("%s: failed to login to LUN %04x\n",
tgt->bus_id, 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 10:37:28 -05:00
/* Let any waiting I/O fail from now on. */
sbp2_unblock(lu->tgt);
}
goto out;
}
tgt->node_id = node_id;
tgt->address_high = local_node_id << 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 10:37:28 -05:00
sbp2_set_generation(lu, generation);
lu->command_block_agent_address =
((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;
fw_notify("%s: logged in to LUN %04x (%d retries)\n",
tgt->bus_id, lu->lun, lu->retries);
/* set appropriate retry limit(s) in BUSY_TIMEOUT register */
sbp2_set_busy_timeout(lu);
PREPARE_DELAYED_WORK(&lu->work, sbp2_reconnect);
sbp2_agent_reset(lu);
/* This was a re-login. */
if (lu->has_sdev) {
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 10:37:28 -05:00
sbp2_conditionally_unblock(lu);
goto out;
}
if (lu->tgt->workarounds & SBP2_WORKAROUND_DELAY_INQUIRY)
ssleep(SBP2_INQUIRY_DELAY);
shost = container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
sdev = __scsi_add_device(shost, 0, 0, sbp2_lun2int(lu->lun), lu);
/*
* 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.
*/
/* 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);
scsi_device_put(sdev);
goto out_logout_login;
}
/* No error during __scsi_add_device() */
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 10:37:28 -05:00
sbp2_allow_block(lu);
goto out;
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.
*/
PREPARE_DELAYED_WORK(&lu->work, sbp2_login);
out:
sbp2_target_put(tgt);
}
static int sbp2_add_logical_unit(struct sbp2_target *tgt, int lun_entry)
{
struct sbp2_logical_unit *lu;
lu = kmalloc(sizeof(*lu), GFP_KERNEL);
if (!lu)
return -ENOMEM;
lu->address_handler.length = 0x100;
lu->address_handler.address_callback = sbp2_status_write;
lu->address_handler.callback_data = lu;
if (fw_core_add_address_handler(&lu->address_handler,
&fw_high_memory_region) < 0) {
kfree(lu);
return -ENOMEM;
}
lu->tgt = tgt;
lu->lun = lun_entry & 0xffff;
lu->login_id = INVALID_LOGIN_ID;
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 10:37:28 -05:00
++tgt->dont_block;
INIT_LIST_HEAD(&lu->orb_list);
INIT_DELAYED_WORK(&lu->work, sbp2_login);
list_add_tail(&lu->link, &tgt->lu_list);
return 0;
}
static int sbp2_scan_logical_unit_dir(struct sbp2_target *tgt, u32 *directory)
{
struct fw_csr_iterator ci;
int key, value;
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;
}
static int sbp2_scan_unit_dir(struct sbp2_target *tgt, u32 *directory,
u32 *model, u32 *firmware_revision)
{
struct fw_csr_iterator ci;
int key, value;
unsigned int timeout;
fw_csr_iterator_init(&ci, directory);
while (fw_csr_iterator_next(&ci, &key, &value)) {
switch (key) {
case CSR_DEPENDENT_INFO | CSR_OFFSET:
tgt->management_agent_address =
CSR_REGISTER_BASE + 4 * value;
break;
case CSR_DIRECTORY_ID:
tgt->directory_id = value;
break;
case CSR_MODEL:
*model = value;
break;
case SBP2_CSR_FIRMWARE_REVISION:
*firmware_revision = value;
break;
case SBP2_CSR_UNIT_CHARACTERISTICS:
/* the timeout value is stored in 500ms units */
timeout = ((unsigned int) value >> 8 & 0xff) * 500;
timeout = max(timeout, SBP2_MIN_LOGIN_ORB_TIMEOUT);
tgt->mgt_orb_timeout =
min(timeout, SBP2_MAX_LOGIN_ORB_TIMEOUT);
if (timeout > tgt->mgt_orb_timeout)
fw_notify("%s: config rom contains %ds "
"management ORB timeout, limiting "
"to %ds\n", tgt->bus_id,
timeout / 1000,
tgt->mgt_orb_timeout / 1000);
break;
case SBP2_CSR_LOGICAL_UNIT_NUMBER:
if (sbp2_add_logical_unit(tgt, value) < 0)
return -ENOMEM;
break;
case SBP2_CSR_LOGICAL_UNIT_DIRECTORY:
/* Adjust for the increment in the iterator */
if (sbp2_scan_logical_unit_dir(tgt, ci.p - 1 + value) < 0)
return -ENOMEM;
break;
}
}
return 0;
}
static void sbp2_init_workarounds(struct sbp2_target *tgt, u32 model,
u32 firmware_revision)
{
int i;
unsigned int w = sbp2_param_workarounds;
if (w)
fw_notify("Please notify linux1394-devel@lists.sourceforge.net "
"if you need the workarounds parameter for %s\n",
tgt->bus_id);
if (w & SBP2_WORKAROUND_OVERRIDE)
goto out;
for (i = 0; i < ARRAY_SIZE(sbp2_workarounds_table); i++) {
if (sbp2_workarounds_table[i].firmware_revision !=
(firmware_revision & 0xffffff00))
continue;
if (sbp2_workarounds_table[i].model != model &&
sbp2_workarounds_table[i].model != ~0)
continue;
w |= sbp2_workarounds_table[i].workarounds;
break;
}
out:
if (w)
fw_notify("Workarounds for %s: 0x%x "
"(firmware_revision 0x%06x, model_id 0x%06x)\n",
tgt->bus_id, w, firmware_revision, model);
tgt->workarounds = w;
}
static struct scsi_host_template scsi_driver_template;
static int sbp2_probe(struct device *dev)
{
struct fw_unit *unit = fw_unit(dev);
struct fw_device *device = fw_device(unit->device.parent);
struct sbp2_target *tgt;
struct sbp2_logical_unit *lu;
struct Scsi_Host *shost;
u32 model, firmware_revision;
if (dma_get_max_seg_size(device->card->device) > SBP2_MAX_SEG_SIZE)
BUG_ON(dma_set_max_seg_size(device->card->device,
SBP2_MAX_SEG_SIZE));
shost = scsi_host_alloc(&scsi_driver_template, sizeof(*tgt));
if (shost == NULL)
return -ENOMEM;
tgt = (struct sbp2_target *)shost->hostdata;
unit->device.driver_data = tgt;
tgt->unit = unit;
kref_init(&tgt->kref);
INIT_LIST_HEAD(&tgt->lu_list);
tgt->bus_id = dev_name(&unit->device);
tgt->guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
if (fw_device_enable_phys_dma(device) < 0)
goto fail_shost_put;
if (scsi_add_host(shost, &unit->device) < 0)
goto fail_shost_put;
firewire: fix crash in automatic module unloading "modprobe firewire-ohci; sleep .1; modprobe -r firewire-ohci" used to result in crashes like this: BUG: unable to handle kernel paging request at ffffffff8807b455 IP: [<ffffffff8807b455>] PGD 203067 PUD 207063 PMD 7c170067 PTE 0 Oops: 0010 [1] PREEMPT SMP CPU 0 Modules linked in: i915 drm cpufreq_ondemand acpi_cpufreq freq_table applesmc input_polldev led_class coretemp hwmon eeprom snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss button thermal processor sg snd_hda_intel snd_pcm snd_timer snd snd_page_alloc sky2 i2c_i801 rtc [last unloaded: crc_itu_t] Pid: 9, comm: events/0 Not tainted 2.6.25-rc2 #3 RIP: 0010:[<ffffffff8807b455>] [<ffffffff8807b455>] RSP: 0018:ffff81007dcdde88 EFLAGS: 00010246 RAX: ffff81007dc95040 RBX: ffff81007dee5390 RCX: 0000000000005e13 RDX: 0000000000008c8b RSI: 0000000000000001 RDI: ffff81007dee5388 RBP: ffff81007dc5eb40 R08: 0000000000000002 R09: ffffffff8022d05c R10: ffffffff8023b34c R11: ffffffff8041a353 R12: ffff81007dee5388 R13: ffffffff8807b455 R14: ffffffff80593bc0 R15: 0000000000000000 FS: 0000000000000000(0000) GS:ffffffff8055a000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: ffffffff8807b455 CR3: 0000000000201000 CR4: 00000000000006e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process events/0 (pid: 9, threadinfo ffff81007dcdc000, task ffff81007dc95040) Stack: ffffffff8023b396 ffffffff88082524 0000000000000000 ffffffff8807d9ae ffff81007dc5eb40 ffff81007dc9dce0 ffff81007dc5eb40 ffff81007dc5eb80 ffff81007dc9dce0 ffffffffffffffff ffffffff8023be87 0000000000000000 Call Trace: [<ffffffff8023b396>] ? run_workqueue+0xdf/0x1df [<ffffffff8023be87>] ? worker_thread+0xd8/0xe3 [<ffffffff8023e917>] ? autoremove_wake_function+0x0/0x2e [<ffffffff8023bdaf>] ? worker_thread+0x0/0xe3 [<ffffffff8023e813>] ? kthread+0x47/0x74 [<ffffffff804198e0>] ? trace_hardirqs_on_thunk+0x35/0x3a [<ffffffff8020c008>] ? child_rip+0xa/0x12 [<ffffffff8020b6e3>] ? restore_args+0x0/0x3d [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e7cc>] ? kthread+0x0/0x74 [<ffffffff8020bffe>] ? child_rip+0x0/0x12 Code: Bad RIP value. RIP [<ffffffff8807b455>] RSP <ffff81007dcdde88> CR2: ffffffff8807b455 ---[ end trace c7366c6657fe5bed ]--- Note that this crash happened _after_ firewire-core was unloaded. The shared workqueue tried to run firewire-core's device initialization jobs or similar jobs. The fix makes sure that firewire-ohci and hence firewire-core is not unloaded before all device shutdown jobs have been completed. This is determined by the count of device initializations minus device releases. Also skip useless retries in the node initialization job if the node is to be shut down. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-02-27 16:14:27 -05:00
fw_device_get(device);
fw_unit_get(unit);
firewire: fix crash in automatic module unloading "modprobe firewire-ohci; sleep .1; modprobe -r firewire-ohci" used to result in crashes like this: BUG: unable to handle kernel paging request at ffffffff8807b455 IP: [<ffffffff8807b455>] PGD 203067 PUD 207063 PMD 7c170067 PTE 0 Oops: 0010 [1] PREEMPT SMP CPU 0 Modules linked in: i915 drm cpufreq_ondemand acpi_cpufreq freq_table applesmc input_polldev led_class coretemp hwmon eeprom snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss button thermal processor sg snd_hda_intel snd_pcm snd_timer snd snd_page_alloc sky2 i2c_i801 rtc [last unloaded: crc_itu_t] Pid: 9, comm: events/0 Not tainted 2.6.25-rc2 #3 RIP: 0010:[<ffffffff8807b455>] [<ffffffff8807b455>] RSP: 0018:ffff81007dcdde88 EFLAGS: 00010246 RAX: ffff81007dc95040 RBX: ffff81007dee5390 RCX: 0000000000005e13 RDX: 0000000000008c8b RSI: 0000000000000001 RDI: ffff81007dee5388 RBP: ffff81007dc5eb40 R08: 0000000000000002 R09: ffffffff8022d05c R10: ffffffff8023b34c R11: ffffffff8041a353 R12: ffff81007dee5388 R13: ffffffff8807b455 R14: ffffffff80593bc0 R15: 0000000000000000 FS: 0000000000000000(0000) GS:ffffffff8055a000(0000) knlGS:0000000000000000 CS: 0010 DS: 0018 ES: 0018 CR0: 000000008005003b CR2: ffffffff8807b455 CR3: 0000000000201000 CR4: 00000000000006e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400 Process events/0 (pid: 9, threadinfo ffff81007dcdc000, task ffff81007dc95040) Stack: ffffffff8023b396 ffffffff88082524 0000000000000000 ffffffff8807d9ae ffff81007dc5eb40 ffff81007dc9dce0 ffff81007dc5eb40 ffff81007dc5eb80 ffff81007dc9dce0 ffffffffffffffff ffffffff8023be87 0000000000000000 Call Trace: [<ffffffff8023b396>] ? run_workqueue+0xdf/0x1df [<ffffffff8023be87>] ? worker_thread+0xd8/0xe3 [<ffffffff8023e917>] ? autoremove_wake_function+0x0/0x2e [<ffffffff8023bdaf>] ? worker_thread+0x0/0xe3 [<ffffffff8023e813>] ? kthread+0x47/0x74 [<ffffffff804198e0>] ? trace_hardirqs_on_thunk+0x35/0x3a [<ffffffff8020c008>] ? child_rip+0xa/0x12 [<ffffffff8020b6e3>] ? restore_args+0x0/0x3d [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e68a>] ? kthreadd+0x14c/0x171 [<ffffffff8023e7cc>] ? kthread+0x0/0x74 [<ffffffff8020bffe>] ? child_rip+0x0/0x12 Code: Bad RIP value. RIP [<ffffffff8807b455>] RSP <ffff81007dcdde88> CR2: ffffffff8807b455 ---[ end trace c7366c6657fe5bed ]--- Note that this crash happened _after_ firewire-core was unloaded. The shared workqueue tried to run firewire-core's device initialization jobs or similar jobs. The fix makes sure that firewire-ohci and hence firewire-core is not unloaded before all device shutdown jobs have been completed. This is determined by the count of device initializations minus device releases. Also skip useless retries in the node initialization job if the node is to be shut down. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-02-27 16:14:27 -05:00
/* Initialize to values that won't match anything in our table. */
firmware_revision = 0xff000000;
model = 0xff000000;
/* implicit directory ID */
tgt->directory_id = ((unit->directory - device->config_rom) * 4
+ CSR_CONFIG_ROM) & 0xffffff;
if (sbp2_scan_unit_dir(tgt, unit->directory, &model,
&firmware_revision) < 0)
goto fail_tgt_put;
sbp2_init_workarounds(tgt, model, firmware_revision);
/* Do the login in a workqueue so we can easily reschedule retries. */
list_for_each_entry(lu, &tgt->lu_list, link)
sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
return 0;
fail_tgt_put:
sbp2_target_put(tgt);
return -ENOMEM;
fail_shost_put:
scsi_host_put(shost);
return -ENOMEM;
}
static int sbp2_remove(struct device *dev)
{
struct fw_unit *unit = fw_unit(dev);
struct sbp2_target *tgt = unit->device.driver_data;
sbp2_target_put(tgt);
return 0;
}
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;
struct fw_device *device = fw_device(tgt->unit->device.parent);
int generation, node_id, local_node_id;
firewire: fw-sbp2: don't retry login or reconnect after unplug If a device is being unplugged while fw-sbp2 had a login or reconnect on schedule, it would take about half a minute to shut the fw_unit down: Jan 27 18:34:54 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 18:34:59 stein firewire_sbp2: sbp2_scsi_abort Jan 27 18:34:59 stein scsi 25:0:0:0: Device offlined - not ready after error recovery Jan 27 18:35:01 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:06 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:12 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:17 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:22 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:27 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 27 18:35:32 stein firewire_sbp2: failed to login to fw2.0 LUN 0000 Jan 27 18:35:32 stein firewire_sbp2: released fw2.0 After this patch, typically only a few seconds spent in __scsi_add_device remain: Jan 27 19:05:50 stein firewire_sbp2: logged in to fw2.0 LUN 0000 (0 retries) <unplug> Jan 27 19:05:56 stein firewire_sbp2: sbp2_scsi_abort Jan 27 19:05:56 stein scsi 33:0:0:0: Device offlined - not ready after error recovery Jan 27 19:05:56 stein firewire_sbp2: released fw2.0 The benefit of this is less noise in the syslog. It furthermore avoids a few wasted CPU cycles and needlessly prolonged lifetime of a few driver objects. Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> Signed-off-by: Jarod Wilson <jwilson@redhat.com>
2008-01-27 13:14:44 -05:00
if (fw_device_is_shutdown(device))
goto out;
generation = device->generation;
smp_rmb(); /* node_id 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) {
fw_error("%s: failed to reconnect\n", tgt->bus_id);
lu->retries = 0;
PREPARE_DELAYED_WORK(&lu->work, sbp2_login);
}
sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
goto out;
}
tgt->node_id = node_id;
tgt->address_high = local_node_id << 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 10:37:28 -05:00
sbp2_set_generation(lu, generation);
fw_notify("%s: reconnected to LUN %04x (%d retries)\n",
tgt->bus_id, lu->lun, lu->retries);
sbp2_agent_reset(lu);
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 10:37:28 -05:00
sbp2_conditionally_unblock(lu);
out:
sbp2_target_put(tgt);
}
static void sbp2_update(struct fw_unit *unit)
{
struct sbp2_target *tgt = unit->device.driver_data;
struct sbp2_logical_unit *lu;
fw_device_enable_phys_dma(fw_device(unit->device.parent));
/*
* 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) {
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 10:37:28 -05:00
sbp2_conditionally_block(lu);
lu->retries = 0;
sbp2_queue_work(lu, 0);
}
}
#define SBP2_UNIT_SPEC_ID_ENTRY 0x0000609e
#define SBP2_SW_VERSION_ENTRY 0x00010483
static const struct fw_device_id sbp2_id_table[] = {
{
.match_flags = FW_MATCH_SPECIFIER_ID | FW_MATCH_VERSION,
.specifier_id = SBP2_UNIT_SPEC_ID_ENTRY,
.version = SBP2_SW_VERSION_ENTRY,
},
{ }
};
static struct fw_driver sbp2_driver = {
.driver = {
.owner = THIS_MODULE,
.name = sbp2_driver_name,
.bus = &fw_bus_type,
.probe = sbp2_probe,
.remove = sbp2_remove,
},
.update = sbp2_update,
.id_table = sbp2_id_table,
};
static unsigned int
sbp2_status_to_sense_data(u8 *sbp2_status, u8 *sense_data)
{
int sam_status;
sense_data[0] = 0x70;
sense_data[1] = 0x0;
sense_data[2] = sbp2_status[1];
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];
sam_status = sbp2_status[0] & 0x3f;
switch (sam_status) {
case SAM_STAT_GOOD:
case SAM_STAT_CHECK_CONDITION:
case SAM_STAT_CONDITION_MET:
case SAM_STAT_BUSY:
case SAM_STAT_RESERVATION_CONFLICT:
case SAM_STAT_COMMAND_TERMINATED:
return DID_OK << 16 | sam_status;
default:
return DID_ERROR << 16;
}
}
static void
complete_command_orb(struct sbp2_orb *base_orb, struct sbp2_status *status)
{
struct sbp2_command_orb *orb =
container_of(base_orb, struct sbp2_command_orb, base);
struct fw_device *device = fw_device(orb->lu->tgt->unit->device.parent);
int result;
if (status != NULL) {
if (STATUS_GET_DEAD(*status))
sbp2_agent_reset_no_wait(orb->lu);
switch (STATUS_GET_RESPONSE(*status)) {
case SBP2_STATUS_REQUEST_COMPLETE:
result = DID_OK << 16;
break;
case SBP2_STATUS_TRANSPORT_FAILURE:
result = DID_BUS_BUSY << 16;
break;
case SBP2_STATUS_ILLEGAL_REQUEST:
case SBP2_STATUS_VENDOR_DEPENDENT:
default:
result = DID_ERROR << 16;
break;
}
if (result == DID_OK << 16 && STATUS_GET_LEN(*status) > 1)
result = sbp2_status_to_sense_data(STATUS_GET_DATA(*status),
orb->cmd->sense_buffer);
} else {
/*
* If the orb completes with status == NULL, something
* went wrong, typically a bus reset happened mid-orb
* or when sending the write (less likely).
*/
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 10:37:28 -05:00
sbp2_conditionally_block(orb->lu);
}
dma_unmap_single(device->card->device, orb->base.request_bus,
sizeof(orb->request), DMA_TO_DEVICE);
if (scsi_sg_count(orb->cmd) > 0)
dma_unmap_sg(device->card->device, scsi_sglist(orb->cmd),
scsi_sg_count(orb->cmd),
orb->cmd->sc_data_direction);
if (orb->page_table_bus != 0)
dma_unmap_single(device->card->device, orb->page_table_bus,
sizeof(orb->page_table), DMA_TO_DEVICE);
orb->cmd->result = result;
orb->done(orb->cmd);
}
static int
sbp2_map_scatterlist(struct sbp2_command_orb *orb, struct fw_device *device,
struct sbp2_logical_unit *lu)
{
struct scatterlist *sg = scsi_sglist(orb->cmd);
int i, n;
n = dma_map_sg(device->card->device, sg, scsi_sg_count(orb->cmd),
orb->cmd->sc_data_direction);
if (n == 0)
goto fail;
/*
* Handle the special case where there is only one element in
* 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
* tables.
*/
if (n == 1) {
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)));
return 0;
}
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));
}
orb->page_table_bus =
dma_map_single(device->card->device, orb->page_table,
sizeof(orb->page_table), DMA_TO_DEVICE);
dma-mapping: add the device argument to dma_mapping_error() Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER architecture does: This enables us to cleanly fix the Calgary IOMMU issue that some devices are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423). I think that per-device dma_mapping_ops support would be also helpful for KVM people to support PCI passthrough but Andi thinks that this makes it difficult to support the PCI passthrough (see the above thread). So I CC'ed this to KVM camp. Comments are appreciated. A pointer to dma_mapping_ops to struct dev_archdata is added. If the pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's NULL, the system-wide dma_ops pointer is used as before. If it's useful for KVM people, I plan to implement a mechanism to register a hook called when a new pci (or dma capable) device is created (it works with hot plugging). It enables IOMMUs to set up an appropriate dma_mapping_ops per device. The major obstacle is that dma_mapping_error doesn't take a pointer to the device unlike other DMA operations. So x86 can't have dma_mapping_ops per device. Note all the POWER IOMMUs use the same dma_mapping_error function so this is not a problem for POWER but x86 IOMMUs use different dma_mapping_error functions. The first patch adds the device argument to dma_mapping_error. The patch is trivial but large since it touches lots of drivers and dma-mapping.h in all the architecture. This patch: dma_mapping_error() doesn't take a pointer to the device unlike other DMA operations. So we can't have dma_mapping_ops per device. Note that POWER already has dma_mapping_ops per device but all the POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device argument. [akpm@linux-foundation.org: fix sge] [akpm@linux-foundation.org: fix svc_rdma] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix bnx2x] [akpm@linux-foundation.org: fix s2io] [akpm@linux-foundation.org: fix pasemi_mac] [akpm@linux-foundation.org: fix sdhci] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix sparc] [akpm@linux-foundation.org: fix ibmvscsi] Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Cc: Muli Ben-Yehuda <muli@il.ibm.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Avi Kivity <avi@qumranet.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 22:44:49 -04:00
if (dma_mapping_error(device->card->device, orb->page_table_bus))
goto fail_page_table;
/*
* The data_descriptor pointer is the one case where we need
* 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
* on other nodes so we need to put our ID in descriptor.high.
*/
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 |
COMMAND_ORB_DATA_SIZE(n));
return 0;
fail_page_table:
dma_unmap_sg(device->card->device, scsi_sglist(orb->cmd),
scsi_sg_count(orb->cmd), orb->cmd->sc_data_direction);
fail:
return -ENOMEM;
}
/* SCSI stack integration */
static int sbp2_scsi_queuecommand(struct scsi_cmnd *cmd, scsi_done_fn_t done)
{
struct sbp2_logical_unit *lu = cmd->device->hostdata;
struct fw_device *device = fw_device(lu->tgt->unit->device.parent);
struct sbp2_command_orb *orb;
unsigned int max_payload;
int generation, retval = SCSI_MLQUEUE_HOST_BUSY;
/*
* Bidirectional commands are not yet implemented, and unknown
* transfer direction not handled.
*/
if (cmd->sc_data_direction == DMA_BIDIRECTIONAL) {
fw_error("Can't handle DMA_BIDIRECTIONAL, rejecting command\n");
cmd->result = DID_ERROR << 16;
done(cmd);
return 0;
}
orb = kzalloc(sizeof(*orb), GFP_ATOMIC);
if (orb == NULL) {
fw_notify("failed to alloc orb\n");
return SCSI_MLQUEUE_HOST_BUSY;
}
/* Initialize rcode to something not RCODE_COMPLETE. */
orb->base.rcode = -1;
kref_init(&orb->base.kref);
orb->lu = lu;
orb->done = done;
orb->cmd = cmd;
orb->request.next.high = cpu_to_be32(SBP2_ORB_NULL);
/*
* At speed 100 we can do 512 bytes per packet, at speed 200,
* 1024 bytes per packet etc. 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.
*/
max_payload = min(device->max_speed + 7,
device->card->max_receive - 1);
orb->request.misc = cpu_to_be32(
COMMAND_ORB_MAX_PAYLOAD(max_payload) |
COMMAND_ORB_SPEED(device->max_speed) |
COMMAND_ORB_NOTIFY);
if (cmd->sc_data_direction == DMA_FROM_DEVICE)
orb->request.misc |= cpu_to_be32(COMMAND_ORB_DIRECTION);
generation = device->generation;
smp_rmb(); /* sbp2_map_scatterlist looks at tgt->address_high */
if (scsi_sg_count(cmd) && sbp2_map_scatterlist(orb, device, lu) < 0)
goto out;
[SCSI] Let scsi_cmnd->cmnd use request->cmd buffer - struct scsi_cmnd had a 16 bytes command buffer of its own. This is an unnecessary duplication and copy of request's cmd. It is probably left overs from the time that scsi_cmnd could function without a request attached. So clean that up. - Once above is done, few places, apart from scsi-ml, needed adjustments due to changing the data type of scsi_cmnd->cmnd. - Lots of drivers still use MAX_COMMAND_SIZE. So I have left that #define but equate it to BLK_MAX_CDB. The way I see it and is reflected in the patch below is. MAX_COMMAND_SIZE - means: The longest fixed-length (*) SCSI CDB as per the SCSI standard and is not related to the implementation. BLK_MAX_CDB. - The allocated space at the request level - I have audit all ISA drivers and made sure none use ->cmnd in a DMA Operation. Same audit was done by Andi Kleen. (*)fixed-length here means commands that their size can be determined by their opcode and the CDB does not carry a length specifier, (unlike the VARIABLE_LENGTH_CMD(0x7f) command). This is actually not exactly true and the SCSI standard also defines extended commands and vendor specific commands that can be bigger than 16 bytes. The kernel will support these using the same infrastructure used for VARLEN CDB's. So in effect MAX_COMMAND_SIZE means the maximum size command scsi-ml supports without specifying a cmd_len by ULD's Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-30 04:19:47 -04:00
memcpy(orb->request.command_block, cmd->cmnd, cmd->cmd_len);
orb->base.callback = complete_command_orb;
orb->base.request_bus =
dma_map_single(device->card->device, &orb->request,
sizeof(orb->request), DMA_TO_DEVICE);
dma-mapping: add the device argument to dma_mapping_error() Add per-device dma_mapping_ops support for CONFIG_X86_64 as POWER architecture does: This enables us to cleanly fix the Calgary IOMMU issue that some devices are not behind the IOMMU (http://lkml.org/lkml/2008/5/8/423). I think that per-device dma_mapping_ops support would be also helpful for KVM people to support PCI passthrough but Andi thinks that this makes it difficult to support the PCI passthrough (see the above thread). So I CC'ed this to KVM camp. Comments are appreciated. A pointer to dma_mapping_ops to struct dev_archdata is added. If the pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If it's NULL, the system-wide dma_ops pointer is used as before. If it's useful for KVM people, I plan to implement a mechanism to register a hook called when a new pci (or dma capable) device is created (it works with hot plugging). It enables IOMMUs to set up an appropriate dma_mapping_ops per device. The major obstacle is that dma_mapping_error doesn't take a pointer to the device unlike other DMA operations. So x86 can't have dma_mapping_ops per device. Note all the POWER IOMMUs use the same dma_mapping_error function so this is not a problem for POWER but x86 IOMMUs use different dma_mapping_error functions. The first patch adds the device argument to dma_mapping_error. The patch is trivial but large since it touches lots of drivers and dma-mapping.h in all the architecture. This patch: dma_mapping_error() doesn't take a pointer to the device unlike other DMA operations. So we can't have dma_mapping_ops per device. Note that POWER already has dma_mapping_ops per device but all the POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use device argument. [akpm@linux-foundation.org: fix sge] [akpm@linux-foundation.org: fix svc_rdma] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix bnx2x] [akpm@linux-foundation.org: fix s2io] [akpm@linux-foundation.org: fix pasemi_mac] [akpm@linux-foundation.org: fix sdhci] [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: fix sparc] [akpm@linux-foundation.org: fix ibmvscsi] Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Cc: Muli Ben-Yehuda <muli@il.ibm.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Avi Kivity <avi@qumranet.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 22:44:49 -04:00
if (dma_mapping_error(device->card->device, orb->base.request_bus))
goto out;
sbp2_send_orb(&orb->base, lu, lu->tgt->node_id, generation,
lu->command_block_agent_address + SBP2_ORB_POINTER);
retval = 0;
out:
kref_put(&orb->base.kref, free_orb);
return retval;
}
static int sbp2_scsi_slave_alloc(struct scsi_device *sdev)
{
struct sbp2_logical_unit *lu = sdev->hostdata;
/* (Re-)Adding logical units via the SCSI stack is not supported. */
if (!lu)
return -ENOSYS;
sdev->allow_restart = 1;
/* SBP-2 requires quadlet alignment of the data buffers. */
blk_queue_update_dma_alignment(sdev->request_queue, 4 - 1);
if (lu->tgt->workarounds & SBP2_WORKAROUND_INQUIRY_36)
sdev->inquiry_len = 36;
return 0;
}
static int sbp2_scsi_slave_configure(struct scsi_device *sdev)
{
struct sbp2_logical_unit *lu = sdev->hostdata;
sdev->use_10_for_rw = 1;
if (sbp2_param_exclusive_login)
sdev->manage_start_stop = 1;
if (sdev->type == TYPE_ROM)
sdev->use_10_for_ms = 1;
if (sdev->type == TYPE_DISK &&
lu->tgt->workarounds & SBP2_WORKAROUND_MODE_SENSE_8)
sdev->skip_ms_page_8 = 1;
if (lu->tgt->workarounds & SBP2_WORKAROUND_FIX_CAPACITY)
sdev->fix_capacity = 1;
if (lu->tgt->workarounds & SBP2_WORKAROUND_POWER_CONDITION)
sdev->start_stop_pwr_cond = 1;
if (lu->tgt->workarounds & SBP2_WORKAROUND_128K_MAX_TRANS)
blk_queue_max_sectors(sdev->request_queue, 128 * 1024 / 512);
blk_queue_max_segment_size(sdev->request_queue, SBP2_MAX_SEG_SIZE);
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)
{
struct sbp2_logical_unit *lu = cmd->device->hostdata;
fw_notify("%s: sbp2_scsi_abort\n", lu->tgt->bus_id);
sbp2_agent_reset(lu);
sbp2_cancel_orbs(lu);
return SUCCESS;
}
/*
* 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.
*/
static ssize_t
sbp2_sysfs_ieee1394_id_show(struct device *dev, struct device_attribute *attr,
char *buf)
{
struct scsi_device *sdev = to_scsi_device(dev);
struct sbp2_logical_unit *lu;
if (!sdev)
return 0;
lu = sdev->hostdata;
return sprintf(buf, "%016llx:%06x:%04x\n",
(unsigned long long)lu->tgt->guid,
lu->tgt->directory_id, lu->lun);
}
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
};
static struct scsi_host_template scsi_driver_template = {
.module = THIS_MODULE,
.name = "SBP-2 IEEE-1394",
.proc_name = sbp2_driver_name,
.queuecommand = sbp2_scsi_queuecommand,
.slave_alloc = sbp2_scsi_slave_alloc,
.slave_configure = sbp2_scsi_slave_configure,
.eh_abort_handler = sbp2_scsi_abort,
.this_id = -1,
.sg_tablesize = SG_ALL,
.use_clustering = ENABLE_CLUSTERING,
.cmd_per_lun = 1,
.can_queue = 1,
.sdev_attrs = sbp2_scsi_sysfs_attrs,
};
MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
MODULE_DESCRIPTION("SCSI over IEEE1394");
MODULE_LICENSE("GPL");
MODULE_DEVICE_TABLE(ieee1394, sbp2_id_table);
/* Provide a module alias so root-on-sbp2 initrds don't break. */
#ifndef CONFIG_IEEE1394_SBP2_MODULE
MODULE_ALIAS("sbp2");
#endif
static int __init sbp2_init(void)
{
sbp2_wq = create_singlethread_workqueue(KBUILD_MODNAME);
if (!sbp2_wq)
return -ENOMEM;
return driver_register(&sbp2_driver.driver);
}
static void __exit sbp2_cleanup(void)
{
driver_unregister(&sbp2_driver.driver);
destroy_workqueue(sbp2_wq);
}
module_init(sbp2_init);
module_exit(sbp2_cleanup);