* [PATCH 01/14] scsi: scsi_error: Define framework for LUN/target based error handle
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 02/14] scsi: scsi_error: Move complete variable eh_action from shost to sdevice JiangJianJun
` (15 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
The old scsi error handle logic is based on host, once a scsi command
in one LUN of this host is classfied as failed, SCSI mid-level would
set the whole host to recovery state, and no IO can be submitted to
all LUNs of this host any more before recovery finished, while the
recovery process might take a long time to finish.
It's unreasonable when there are a lot of LUNs in one host.
This change introduce a way for driver to implement its own
error handle logic which can be based on scsi LUN or scsi target
as minimum unit.
scsi_device_eh is defined for error handle based on scsi LUN, and
pointer struct scsi_device_eh "eh" is added in scsi_device, which
is NULL by default.
LLDs can initialize the sdev->eh in hostt->slave_alloc to implement an
scsi LUN based error handle. If this member is not NULL, SCSI mid-level
would branch to drivers' error handler rather than the old one which
block whole host's IO.
scsi_target_eh is defined for error handle based on scsi target, and
pointer struct scsi_target_eh "eh" is added in scsi_target, which is NULL
by default.
LLDs can initialize the starget->eh in hostt->target_alloc to implement
an scsi target based error handle. If this member is not NULL, SCSI
mid-level would branch to drivers' error handler rather than the
old one which block whole host's IO.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 81 ++++++++++++++++++++++++++++++++++++--
drivers/scsi/scsi_lib.c | 8 ++++
drivers/scsi/scsi_priv.h | 15 +++++++
include/scsi/scsi_device.h | 57 +++++++++++++++++++++++++++
4 files changed, 157 insertions(+), 4 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 746ff6a1f309..135b63d89d72 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -271,8 +271,9 @@ scsi_abort_command(struct scsi_cmnd *scmd)
*/
static void scsi_eh_reset(struct scsi_cmnd *scmd)
{
+ struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
+
if (!blk_rq_is_passthrough(scsi_cmd_to_rq(scmd))) {
- struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
if (sdrv->eh_reset)
sdrv->eh_reset(scmd);
}
@@ -291,11 +292,46 @@ static void scsi_eh_inc_host_failed(struct rcu_head *head)
spin_unlock_irqrestore(shost->host_lock, flags);
}
+static bool scsi_eh_scmd_add_sdev(struct scsi_cmnd *scmd)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_device_eh *eh = sdev->eh;
+
+ if (!eh || !eh->add_cmnd)
+ return true;
+
+ scsi_eh_reset(scmd);
+ eh->add_cmnd(scmd);
+
+ if (eh->wakeup)
+ eh->wakeup(sdev);
+
+ return false;
+}
+
+static bool scsi_eh_scmd_add_starget(struct scsi_cmnd *scmd)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_target *starget = scsi_target(sdev);
+ struct scsi_target_eh *eh = starget->eh;
+
+ if (!eh || !eh->add_cmnd)
+ return true;
+
+ scsi_eh_reset(scmd);
+ eh->add_cmnd(scmd);
+
+ if (eh->wakeup)
+ eh->wakeup(starget);
+
+ return false;
+}
+
/**
- * scsi_eh_scmd_add - add scsi cmd to error handling.
- * @scmd: scmd to run eh on.
+ * Add scsi cmd to error handling of Scsi_Host.
+ * This is default action of error handle.
*/
-void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
+static void scsi_eh_scmd_add_shost(struct scsi_cmnd *scmd)
{
struct Scsi_Host *shost = scmd->device->host;
unsigned long flags;
@@ -322,6 +358,43 @@ void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed);
}
+/**
+ * scsi_eh_scmd_add - add scsi cmd to error handling.
+ * @scmd: scmd to run eh on.
+ */
+void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_target *starget = scsi_target(sdev);
+ struct Scsi_Host *shost = sdev->host;
+
+ if (unlikely(scsi_host_in_recovery(shost))) {
+ scsi_eh_scmd_add_shost(scmd);
+ return;
+ }
+
+ if (unlikely(scsi_target_in_recovery(starget))) {
+ if (scsi_eh_scmd_add_starget(scmd))
+ scsi_eh_scmd_add_shost(scmd);
+ return;
+ }
+
+ if (scsi_eh_scmd_add_sdev(scmd))
+ if (scsi_eh_scmd_add_starget(scmd))
+ scsi_eh_scmd_add_shost(scmd);
+}
+
+void scsi_eh_try_wakeup(struct scsi_device *sdev)
+{
+ struct scsi_target *starget = scsi_target(sdev);
+
+ if (sdev->eh && sdev->eh->wakeup)
+ sdev->eh->wakeup(sdev);
+
+ if (starget->eh && starget->eh->wakeup)
+ starget->eh->wakeup(starget);
+}
+
/**
* scsi_timeout - Timeout function for normal scsi commands.
* @req: request that is timing out.
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 0c65ecfedfbd..63bd33af336b 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -398,6 +398,8 @@ void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
sbitmap_put(&sdev->budget_map, cmd->budget_token);
cmd->budget_token = -1;
+
+ scsi_eh_try_wakeup(sdev);
}
/*
@@ -1360,6 +1362,9 @@ static inline int scsi_dev_queue_ready(struct request_queue *q,
{
int token;
+ if (scsi_device_in_recovery(sdev))
+ return -1;
+
token = sbitmap_get(&sdev->budget_map);
if (token < 0)
return -1;
@@ -1393,6 +1398,9 @@ static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
struct scsi_target *starget = scsi_target(sdev);
unsigned int busy;
+ if (scsi_target_in_recovery(starget))
+ return 0;
+
if (starget->single_lun) {
spin_lock_irq(shost->host_lock);
if (starget->starget_sdev_user &&
diff --git a/drivers/scsi/scsi_priv.h b/drivers/scsi/scsi_priv.h
index 5b2b19f5e8ec..cd98f71830f4 100644
--- a/drivers/scsi/scsi_priv.h
+++ b/drivers/scsi/scsi_priv.h
@@ -92,6 +92,7 @@ extern int scsi_error_handler(void *host);
extern enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *cmd);
extern void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy);
extern void scsi_eh_scmd_add(struct scsi_cmnd *);
+extern void scsi_eh_try_wakeup(struct scsi_device *sdev);
void scsi_eh_ready_devs(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q);
@@ -191,6 +192,20 @@ static inline void scsi_dh_add_device(struct scsi_device *sdev) { }
static inline void scsi_dh_release_device(struct scsi_device *sdev) { }
#endif
+static inline bool scsi_device_in_recovery(struct scsi_device *sdev)
+{
+ struct scsi_device_eh *eh = sdev->eh;
+
+ return eh && eh->is_busy && eh->is_busy(sdev);
+}
+
+static inline bool scsi_target_in_recovery(struct scsi_target *starget)
+{
+ struct scsi_target_eh *eh = starget->eh;
+
+ return eh && eh->is_busy && eh->is_busy(starget);
+}
+
struct bsg_device *scsi_bsg_register_queue(struct scsi_device *sdev);
extern int scsi_device_max_queue_depth(struct scsi_device *sdev);
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index 6d6500148c4b..6a58b297e74c 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -100,6 +100,61 @@ struct scsi_vpd {
unsigned char data[];
};
+struct scsi_device;
+struct scsi_target;
+
+struct scsi_device_eh {
+ /*
+ * add scsi command to error handler so it would be handuled by
+ * driver's error handle strategy
+ */
+ void (*add_cmnd)(struct scsi_cmnd *scmd);
+
+ /*
+ * to judge if the device is busy handling errors, called before
+ * dispatch scsi cmnd
+ *
+ * return 0 if it's ready to accepy scsi cmnd
+ * return 1 if it's in error handle, command's would not be dispatched
+ */
+ bool (*is_busy)(struct scsi_device *sdev);
+
+ /*
+ * wakeup device's error handle
+ *
+ * usually the error handler strategy would not run at once when
+ * error command is added. This function would be called when any
+ * scsi cmnd is finished or when scsi cmnd is added.
+ */
+ void (*wakeup)(struct scsi_device *sdev);
+};
+
+struct scsi_target_eh {
+ /*
+ * add scsi command to error handler so it would be handuled by
+ * driver's error handle strategy
+ */
+ void (*add_cmnd)(struct scsi_cmnd *scmd);
+
+ /*
+ * to judge if the device is busy handling errors, called before
+ * dispatch scsi cmnd
+ *
+ * return 0 if it's ready to accepy scsi cmnd
+ * return 1 if it's in error handle, command's would not be dispatched
+ */
+ bool (*is_busy)(struct scsi_target *starget);
+
+ /*
+ * wakeup device's error handle
+ *
+ * usually the error handler strategy would not run at once when
+ * error command is added. This function would be called when any
+ * scsi cmnd is finished or when scsi cmnd is added.
+ */
+ void (*wakeup)(struct scsi_target *starget);
+};
+
struct scsi_device {
struct Scsi_Host *host;
struct request_queue *request_queue;
@@ -289,6 +344,7 @@ struct scsi_device {
struct mutex state_mutex;
enum scsi_device_state sdev_state;
struct task_struct *quiesced_by;
+ struct scsi_device_eh *eh;
unsigned long sdev_data[];
} __attribute__((aligned(sizeof(unsigned long))));
@@ -375,6 +431,7 @@ struct scsi_target {
char scsi_level;
enum scsi_target_state state;
void *hostdata; /* available to low-level driver */
+ struct scsi_target_eh *eh;
unsigned long starget_data[]; /* for the transport */
/* starget_data must be the last element!!!! */
} __attribute__((aligned(sizeof(unsigned long))));
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 02/14] scsi: scsi_error: Move complete variable eh_action from shost to sdevice
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
2025-08-16 11:24 ` [PATCH 01/14] scsi: scsi_error: Define framework for LUN/target based error handle JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 03/14] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset JiangJianJun
` (14 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
eh_action is used to wait for error handle command's completion if scsi
command is send in error handle. Now the error handler might based on
scsi_device, so move it to scsi_device.
This is preparation for a genernal LUN/target based error handle
strategy.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 6 +++---
include/scsi/scsi_device.h | 2 ++
include/scsi/scsi_host.h | 2 --
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 135b63d89d72..6020f20c7f07 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -954,7 +954,7 @@ void scsi_eh_done(struct scsi_cmnd *scmd)
SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
"%s result: %x\n", __func__, scmd->result));
- eh_action = scmd->device->host->eh_action;
+ eh_action = scmd->device->eh_action;
if (eh_action)
complete(eh_action);
}
@@ -1243,7 +1243,7 @@ static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
retry:
scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
- shost->eh_action = &done;
+ sdev->eh_action = &done;
scsi_log_send(scmd);
scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
@@ -1287,7 +1287,7 @@ static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
rtn = SUCCESS;
}
- shost->eh_action = NULL;
+ sdev->eh_action = NULL;
scsi_log_completion(scmd, rtn);
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index 6a58b297e74c..6f47a7a74cd1 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -345,6 +345,8 @@ struct scsi_device {
enum scsi_device_state sdev_state;
struct task_struct *quiesced_by;
struct scsi_device_eh *eh;
+ struct completion *eh_action; /* Wait for specific actions */
+ /* on the device. */
unsigned long sdev_data[];
} __attribute__((aligned(sizeof(unsigned long))));
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index c53812b9026f..46f57fe78505 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -558,8 +558,6 @@ struct Scsi_Host {
struct list_head eh_abort_list;
struct list_head eh_cmd_q;
struct task_struct * ehandler; /* Error recovery thread. */
- struct completion * eh_action; /* Wait for specific actions on the
- host. */
wait_queue_head_t host_wait;
const struct scsi_host_template *hostt;
struct scsi_transport_template *transportt;
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 03/14] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
2025-08-16 11:24 ` [PATCH 01/14] scsi: scsi_error: Define framework for LUN/target based error handle JiangJianJun
2025-08-16 11:24 ` [PATCH 02/14] scsi: scsi_error: Move complete variable eh_action from shost to sdevice JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 04/14] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT JiangJianJun
` (13 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
This is preparation for a genernal LUN/target based error handle
strategy, the strategy would reuse some error handler APIs,
but some steps of these function should not be performed. For
example, we should not perform target reset if we just stop IOs
on one single LUN.
This change add checks in scsi_try_xxx_reset to make sure
the reset operations would not be performed only if the condition
is not satisfied.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 37 +++++++++++++++++++++++++++++++------
1 file changed, 31 insertions(+), 6 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 6020f20c7f07..2da876a928c7 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -963,7 +963,7 @@ void scsi_eh_done(struct scsi_cmnd *scmd)
* scsi_try_host_reset - ask host adapter to reset itself
* @scmd: SCSI cmd to send host reset.
*/
-static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
+static enum scsi_disposition __scsi_try_host_reset(struct scsi_cmnd *scmd)
{
unsigned long flags;
enum scsi_disposition rtn;
@@ -989,11 +989,19 @@ static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
return rtn;
}
+static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
+{
+ if (!scsi_host_in_recovery(scmd->device->host))
+ return FAILED;
+
+ return __scsi_try_host_reset(scmd);
+}
+
/**
* scsi_try_bus_reset - ask host to perform a bus reset
* @scmd: SCSI cmd to send bus reset.
*/
-static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
+static enum scsi_disposition __scsi_try_bus_reset(struct scsi_cmnd *scmd)
{
unsigned long flags;
enum scsi_disposition rtn;
@@ -1019,6 +1027,14 @@ static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
return rtn;
}
+static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
+{
+ if (!scsi_host_in_recovery(scmd->device->host))
+ return FAILED;
+
+ return __scsi_try_bus_reset(scmd);
+}
+
static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
{
sdev->was_reset = 1;
@@ -1035,7 +1051,7 @@ static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
* timer on it, and set the host back to a consistent state prior to
* returning.
*/
-static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
+static enum scsi_disposition __scsi_try_target_reset(struct scsi_cmnd *scmd)
{
unsigned long flags;
enum scsi_disposition rtn;
@@ -1056,6 +1072,15 @@ static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
return rtn;
}
+static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd)
+{
+ if (!(scsi_target_in_recovery(scsi_target(scmd->device)) ||
+ scsi_host_in_recovery(scmd->device->host)))
+ return FAILED;
+
+ return __scsi_try_target_reset(scmd);
+}
+
/**
* scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
* @scmd: SCSI cmd used to send BDR
@@ -2579,17 +2604,17 @@ scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
break;
fallthrough;
case SG_SCSI_RESET_TARGET:
- rtn = scsi_try_target_reset(scmd);
+ rtn = __scsi_try_target_reset(scmd);
if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
break;
fallthrough;
case SG_SCSI_RESET_BUS:
- rtn = scsi_try_bus_reset(scmd);
+ rtn = __scsi_try_bus_reset(scmd);
if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
break;
fallthrough;
case SG_SCSI_RESET_HOST:
- rtn = scsi_try_host_reset(scmd);
+ rtn = __scsi_try_host_reset(scmd);
if (rtn == SUCCESS)
break;
fallthrough;
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 04/14] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (2 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 03/14] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 05/14] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset JiangJianJun
` (12 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper function scsi_eh_sdev_stu() to perform START_UNIT and check
if to finish some error commands.
This is preparation for a genernal LUN/target based error handle
strategy and did not change original logic.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 50 +++++++++++++++++++++++----------------
1 file changed, 29 insertions(+), 21 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 2da876a928c7..24c695a36a76 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -1605,6 +1605,31 @@ static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
return 1;
}
+static int scsi_eh_sdev_stu(struct scsi_cmnd *scmd,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_cmnd *next;
+
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: Sending START_UNIT\n", current->comm));
+
+ if (scsi_eh_try_stu(scmd)) {
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: START_UNIT failed\n", current->comm));
+ return 0;
+ }
+
+ if (!scsi_device_online(sdev) || !scsi_eh_tur(scmd))
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry)
+ if (scmd->device == sdev &&
+ scsi_eh_action(scmd, SUCCESS) == SUCCESS)
+ scsi_eh_finish_cmd(scmd, done_q);
+
+ return list_empty(work_q);
+}
+
/**
* scsi_eh_stu - send START_UNIT if needed
* @shost: &scsi host being recovered.
@@ -1619,7 +1644,7 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q)
{
- struct scsi_cmnd *scmd, *stu_scmd, *next;
+ struct scsi_cmnd *scmd, *stu_scmd;
struct scsi_device *sdev;
shost_for_each_device(sdev, shost) {
@@ -1642,26 +1667,9 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
if (!stu_scmd)
continue;
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: Sending START_UNIT\n",
- current->comm));
-
- if (!scsi_eh_try_stu(stu_scmd)) {
- if (!scsi_device_online(sdev) ||
- !scsi_eh_tur(stu_scmd)) {
- list_for_each_entry_safe(scmd, next,
- work_q, eh_entry) {
- if (scmd->device == sdev &&
- scsi_eh_action(scmd, SUCCESS) == SUCCESS)
- scsi_eh_finish_cmd(scmd, done_q);
- }
- }
- } else {
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: START_UNIT failed\n",
- current->comm));
+ if (scsi_eh_sdev_stu(stu_scmd, work_q, done_q)) {
+ scsi_device_put(sdev);
+ break;
}
}
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 05/14] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (3 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 04/14] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 06/14] scsi: scsi_error: Add flags to mark error handle steps has done JiangJianJun
` (11 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper function scsi_eh_sdev_reset() to perform lun reset and check
if to finish some error commands.
This is preparation for a genernal LUN/target based error handle
strategy and did not change original logic.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 54 +++++++++++++++++++++++----------------
1 file changed, 32 insertions(+), 22 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 24c695a36a76..02f6ef6393ed 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -1676,6 +1676,34 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
return list_empty(work_q);
}
+static int scsi_eh_sdev_reset(struct scsi_cmnd *scmd,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_cmnd *next;
+ struct scsi_device *sdev = scmd->device;
+ enum scsi_disposition rtn;
+
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: Sending BDR\n", current->comm));
+
+ rtn = scsi_try_bus_device_reset(scmd);
+ if (rtn != SUCCESS && rtn != FAST_IO_FAIL) {
+ SCSI_LOG_ERROR_RECOVERY(3,
+ sdev_printk(KERN_INFO, sdev,
+ "%s: BDR failed\n", current->comm));
+ return 0;
+ }
+
+ if (!scsi_device_online(sdev) || rtn == FAST_IO_FAIL ||
+ !scsi_eh_tur(scmd))
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry)
+ if (scmd->device == sdev &&
+ scsi_eh_action(scmd, rtn) != FAILED)
+ scsi_eh_finish_cmd(scmd, done_q);
+
+ return list_empty(work_q);
+}
/**
* scsi_eh_bus_device_reset - send bdr if needed
@@ -1693,9 +1721,8 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q)
{
- struct scsi_cmnd *scmd, *bdr_scmd, *next;
+ struct scsi_cmnd *scmd, *bdr_scmd;
struct scsi_device *sdev;
- enum scsi_disposition rtn;
shost_for_each_device(sdev, shost) {
if (scsi_host_eh_past_deadline(shost)) {
@@ -1716,26 +1743,9 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
if (!bdr_scmd)
continue;
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: Sending BDR\n", current->comm));
- rtn = scsi_try_bus_device_reset(bdr_scmd);
- if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
- if (!scsi_device_online(sdev) ||
- rtn == FAST_IO_FAIL ||
- !scsi_eh_tur(bdr_scmd)) {
- list_for_each_entry_safe(scmd, next,
- work_q, eh_entry) {
- if (scmd->device == sdev &&
- scsi_eh_action(scmd, rtn) != FAILED)
- scsi_eh_finish_cmd(scmd,
- done_q);
- }
- }
- } else {
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: BDR failed\n", current->comm));
+ if (scsi_eh_sdev_reset(bdr_scmd, work_q, done_q)) {
+ scsi_device_put(sdev);
+ break;
}
}
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 06/14] scsi: scsi_error: Add flags to mark error handle steps has done
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (4 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 05/14] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 07/14] scsi: scsi_error: Add helper to handle scsi device's error command list JiangJianJun
` (10 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
LUN based error handle would mainly do three steps to recover
commands which are check sense, start unit, and reset lun. It might
fallback to target/host based error handle which would do these steps
too.
Target based error handle would reset target, it would also fallback
to host based error handle.
Add some flags to mark these steps are done to avoid repeating
these steps.
The flags should be cleared when LUN/target based error handler is
waked up or when target/host based error handle finished, and set
when fallback to target/host based error handle.
scsi_eh_get_sense, scsi_eh_stu, scsi_eh_bus_device_reset and
scsi_eh_target_reset would check these flags before actually action.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 55 ++++++++++++++++++++++++++++++++++++++
include/scsi/scsi_device.h | 28 +++++++++++++++++++
2 files changed, 83 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 02f6ef6393ed..cd53e2744a4f 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -57,10 +57,50 @@
#define BUS_RESET_SETTLE_TIME (10)
#define HOST_RESET_SETTLE_TIME (10)
+#define sdev_flags_done(flag) \
+static inline int sdev_##flag(struct scsi_device *sdev) \
+{ \
+ struct scsi_device_eh *eh = sdev->eh; \
+ if (!eh) \
+ return 0; \
+ return eh->flag; \
+}
+
static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *,
struct scsi_cmnd *);
+sdev_flags_done(get_sense_done);
+sdev_flags_done(stu_done);
+sdev_flags_done(reset_done);
+
+static inline int starget_reset_done(struct scsi_target *starget)
+{
+ struct scsi_target_eh *eh = starget->eh;
+
+ if (!eh)
+ return 0;
+ return eh->reset_done;
+}
+
+static inline void shost_clear_eh_done(struct Scsi_Host *shost)
+{
+ struct scsi_device *sdev;
+ struct scsi_target *starget;
+
+ list_for_each_entry(starget, &shost->__targets, siblings)
+ if (starget->eh)
+ starget->eh->reset_done = 0;
+
+ shost_for_each_device(sdev, shost) {
+ if (!sdev->eh)
+ continue;
+ sdev->eh->get_sense_done = 0;
+ sdev->eh->stu_done = 0;
+ sdev->eh->reset_done = 0;
+ }
+}
+
void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy)
{
lockdep_assert_held(shost->host_lock);
@@ -1443,6 +1483,9 @@ int scsi_eh_get_sense(struct list_head *work_q,
current->comm));
break;
}
+ if (sdev_get_sense_done(scmd->device) ||
+ starget_reset_done(scsi_target(scmd->device)))
+ continue;
if (!scsi_status_is_check_condition(scmd->result))
/*
* don't request sense if there's no check condition
@@ -1656,6 +1699,9 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
scsi_device_put(sdev);
break;
}
+ if (sdev_stu_done(sdev) ||
+ starget_reset_done(scsi_target(sdev)))
+ continue;
stu_scmd = NULL;
list_for_each_entry(scmd, work_q, eh_entry)
if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
@@ -1739,6 +1785,9 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
bdr_scmd = scmd;
break;
}
+ if (sdev_reset_done(sdev) ||
+ starget_reset_done(scsi_target(sdev)))
+ continue;
if (!bdr_scmd)
continue;
@@ -1787,6 +1836,11 @@ static int scsi_eh_target_reset(struct Scsi_Host *shost,
}
scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
+ if (starget_reset_done(scsi_target(scmd->device))) {
+ /* push back on work queue for further processing */
+ list_move(&scmd->eh_entry, work_q);
+ continue;
+ }
id = scmd_id(scmd);
SCSI_LOG_ERROR_RECOVERY(3,
@@ -2403,6 +2457,7 @@ static void scsi_unjam_host(struct Scsi_Host *shost)
if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
+ shost_clear_eh_done(shost);
spin_lock_irqsave(shost->host_lock, flags);
if (shost->eh_deadline != -1)
shost->last_reset = 0;
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index 6f47a7a74cd1..0f88b176cdf9 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -104,6 +104,24 @@ struct scsi_device;
struct scsi_target;
struct scsi_device_eh {
+ /*
+ * LUN rebased error handle would mainly do three
+ * steps to recovery commands which are
+ * check sense
+ * start unit
+ * reset lun
+ * While we would fallback to target or host based error handle
+ * which would do these steps too. Add flags to mark thes steps
+ * are done to avoid repeating these steps.
+ *
+ * The flags should be cleared when LUN based error handler is
+ * wakedup or when target/host based error handle finished,
+ * set when fallback to target or host based error handle.
+ */
+ unsigned get_sense_done:1;
+ unsigned stu_done:1;
+ unsigned reset_done:1;
+
/*
* add scsi command to error handler so it would be handuled by
* driver's error handle strategy
@@ -130,6 +148,16 @@ struct scsi_device_eh {
};
struct scsi_target_eh {
+ /*
+ * flag to mark target reset is done to avoid repeating
+ * these steps when fallback to host based error handle
+ *
+ * The flag should be cleared when target based error handler
+ * is * wakedup or when host based error handle finished,
+ * set when fallback to host based error handle.
+ */
+ unsigned reset_done:1;
+
/*
* add scsi command to error handler so it would be handuled by
* driver's error handle strategy
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 07/14] scsi: scsi_error: Add helper to handle scsi device's error command list
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (5 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 06/14] scsi: scsi_error: Add flags to mark error handle steps has done JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 08/14] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
` (9 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper scsi_sdev_eh() to handle scsi device's error command list,
it would perform some steps which can be done with LUN's IO blocked,
including check sense, start unit and reset lun.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 37 +++++++++++++++++++++++++++++++++++++
include/scsi/scsi_eh.h | 2 ++
2 files changed, 39 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index cd53e2744a4f..b09f721d4d93 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2554,6 +2554,43 @@ int scsi_error_handler(void *data)
return 0;
}
+/*
+ * Single LUN error handle
+ *
+ * @work_q: list of scsi commands need to recovery
+ * @done_q: list of scsi commands handled
+ *
+ * return: return 1 if all commands in work_q is recoveryed, else 0 is returned
+ */
+int scsi_sdev_eh(struct scsi_device *sdev,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ int ret = 0;
+ struct scsi_cmnd *scmd;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: checking sense\n", current->comm));
+ ret = scsi_eh_get_sense(work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: start unit\n", current->comm));
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+ ret = scsi_eh_sdev_stu(scmd, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: reset LUN\n", current->comm));
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+ ret = scsi_eh_sdev_reset(scmd, work_q, done_q);
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(scsi_sdev_eh);
+
/**
* scsi_report_bus_reset() - report bus reset observed
*
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index 1ae08e81339f..5ce791063baf 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -18,6 +18,8 @@ extern int scsi_block_when_processing_errors(struct scsi_device *);
extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
struct scsi_sense_hdr *sshdr);
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
+extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
+ struct list_head *doneq);
static inline bool scsi_sense_is_deferred(const struct scsi_sense_hdr *sshdr)
{
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 08/14] scsi: scsi_error: Add a general LUN based error handler
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (6 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 07/14] scsi: scsi_error: Add helper to handle scsi device's error command list JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-17 9:18 ` Markus Elfring
2025-08-16 11:24 ` [PATCH 09/14] scsi: core: increase/decrease target_busy if set " JiangJianJun
` (8 subsequent siblings)
16 siblings, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add a general LUN based error handler which can be used by drivers
directly. This error handler implements an scsi_device_eh, when handling
error commands, it would call helper function scsi_sdev_eh() added before
to try recover error commands.
The behavior if scsi_sdev_eh() can not recover all error commands
depends on fallback flag, which is initialized when scsi_device is
allocated. If fallback is set, it would fallback to further error
recover strategy like old host based error handle; else it would
mark this scsi device offline and flush all error commands. Add a
flag for controlling rollback in scsi_host_template.
Add interface sdev_setup_eh/sdev_clear_eh in scsi_host_template, used
for setup/clear LUN based error handler. Drivers can implements them
custom, or use inner implements:
scsi_device_setup_eh/scsi_device_clear_eh.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 179 ++++++++++++++++++++++++++++++++++++++
drivers/scsi/scsi_scan.c | 7 ++
drivers/scsi/scsi_sysfs.c | 2 +
include/scsi/scsi_eh.h | 2 +
include/scsi/scsi_host.h | 16 ++++
5 files changed, 206 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index b09f721d4d93..49344f30bab9 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2804,3 +2804,182 @@ bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
}
}
EXPORT_SYMBOL(scsi_get_sense_info_fld);
+
+static inline void *scsi_eh_device_priv(struct scsi_device_eh *eh)
+{
+ return eh + 1;
+}
+
+struct scsi_lun_eh {
+ spinlock_t eh_lock;
+ unsigned int eh_num;
+ struct list_head eh_cmd_q;
+ struct scsi_device *sdev;
+ struct work_struct eh_handle_work;
+ unsigned int fallback:1;
+};
+
+/*
+ * error handle strategy based on LUN, following steps
+ * is applied to recovery error commands in list:
+ * check sense data
+ * send start unit
+ * reset lun
+ * if there are still error commands, it would fallback to
+ * target based or host based error handle for further recovery.
+ */
+static void sdev_eh_work(struct work_struct *work)
+{
+ unsigned long flags;
+ struct scsi_lun_eh *luneh =
+ container_of(work, struct scsi_lun_eh, eh_handle_work);
+ struct scsi_device *sdev = luneh->sdev;
+ struct scsi_device_eh *eh = sdev->eh;
+ struct Scsi_Host *shost = sdev->host;
+ struct scsi_cmnd *scmd, *next;
+ LIST_HEAD(eh_work_q);
+ LIST_HEAD(eh_done_q);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ list_splice_init(&luneh->eh_cmd_q, &eh_work_q);
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ if (scsi_sdev_eh(sdev, &eh_work_q, &eh_done_q))
+ goto out_flush_done;
+
+ if (!luneh->fallback) {
+ list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry)
+ scsi_eh_finish_cmd(scmd, &eh_done_q);
+
+ sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: Device offlined - not ready after error recovery\n",
+ current->comm);
+
+ mutex_lock(&sdev->state_mutex);
+ scsi_device_set_state(sdev, SDEV_OFFLINE);
+ mutex_unlock(&sdev->state_mutex);
+
+ goto out_flush_done;
+ }
+
+ /*
+ * fallback to target or host based error handle
+ */
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh fallback to further recovery\n", current->comm));
+ list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry) {
+ list_del_init(&scmd->eh_entry);
+
+ if (scsi_host_in_recovery(shost) ||
+ scsi_eh_scmd_add_starget(scmd))
+ scsi_eh_scmd_add_shost(scmd);
+ }
+
+ eh->get_sense_done = 1;
+ eh->stu_done = 1;
+ eh->reset_done = 1;
+
+out_flush_done:
+ scsi_eh_flush_done_q(&eh_done_q);
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ luneh->eh_num = 0;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+}
+static void sdev_eh_add_cmnd(struct scsi_cmnd *scmd)
+{
+ unsigned long flags;
+ struct scsi_lun_eh *luneh;
+ struct scsi_device *sdev = scmd->device;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ list_add_tail(&scmd->eh_entry, &luneh->eh_cmd_q);
+ luneh->eh_num++;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+}
+static bool sdev_eh_is_busy(struct scsi_device *sdev)
+{
+ int ret = 0;
+ unsigned long flags;
+ struct scsi_lun_eh *luneh;
+
+ if (!sdev->eh)
+ return false;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ ret = luneh->eh_num;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ return ret != 0;
+}
+static void sdev_eh_wakeup(struct scsi_device *sdev)
+{
+ unsigned long flags;
+ unsigned int nr_error;
+ unsigned int nr_busy;
+ struct scsi_lun_eh *luneh;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ nr_error = luneh->eh_num;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ nr_busy = scsi_device_busy(sdev);
+
+ if (!nr_error || nr_busy != nr_error) {
+ SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: do not wake up, busy/error: %d/%d\n",
+ current->comm, nr_busy, nr_error));
+ return;
+ }
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: waking up, busy/error: %d/%d\n",
+ current->comm, nr_busy, nr_error));
+
+ schedule_work(&luneh->eh_handle_work);
+}
+
+/*
+ * This is default implement of Scsi_Host.sdev_setup_eh.
+ */
+int scsi_device_setup_eh(struct scsi_device *sdev)
+{
+ struct scsi_device_eh *eh;
+ struct scsi_lun_eh *luneh;
+
+ eh = kzalloc(sizeof(struct scsi_device_eh) + sizeof(struct scsi_lun_eh),
+ GFP_KERNEL);
+ if (!eh)
+ return -ENOMEM;
+ luneh = scsi_eh_device_priv(eh);
+
+ eh->add_cmnd = sdev_eh_add_cmnd;
+ eh->is_busy = sdev_eh_is_busy;
+ eh->wakeup = sdev_eh_wakeup;
+
+ luneh->fallback = sdev->host->hostt->sdev_eh_fallback;
+ luneh->sdev = sdev;
+ spin_lock_init(&luneh->eh_lock);
+ INIT_LIST_HEAD(&luneh->eh_cmd_q);
+ INIT_WORK(&luneh->eh_handle_work, sdev_eh_work);
+
+ sdev->eh = eh;
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(scsi_device_setup_eh);
+
+/*
+ * This is default implement of Scsi_Host.sdev_clear_eh.
+ */
+void scsi_device_clear_eh(struct scsi_device *sdev)
+{
+ kfree(sdev->eh);
+ sdev->eh = NULL;
+}
+EXPORT_SYMBOL_GPL(scsi_device_clear_eh);
diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 3c6e089e80c3..1c5cb77dfb22 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -377,9 +377,16 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
goto out_device_destroy;
}
}
+ if (shost->hostt->sdev_setup_eh) {
+ ret = shost->hostt->sdev_setup_eh(sdev);
+ if (ret)
+ goto out_device_eh;
+ }
return sdev;
+out_device_eh:
+ shost->hostt->sdev_destroy(sdev);
out_device_destroy:
__scsi_remove_device(sdev);
out:
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
index 169af7d47ce7..d859371c01bb 100644
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -1513,6 +1513,8 @@ void __scsi_remove_device(struct scsi_device *sdev)
kref_put(&sdev->host->tagset_refcnt, scsi_mq_free_tags);
cancel_work_sync(&sdev->requeue_work);
+ if (sdev->host->hostt->sdev_clear_eh)
+ sdev->host->hostt->sdev_clear_eh(sdev);
if (sdev->host->hostt->sdev_destroy)
sdev->host->hostt->sdev_destroy(sdev);
transport_destroy_device(dev);
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index 5ce791063baf..d8e4475ff004 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -20,6 +20,8 @@ extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
struct list_head *doneq);
+extern int scsi_device_setup_eh(struct scsi_device *sdev);
+extern void scsi_device_clear_eh(struct scsi_device *sdev);
static inline bool scsi_sense_is_deferred(const struct scsi_sense_hdr *sshdr)
{
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index 46f57fe78505..9cc34bfff3f7 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -225,6 +225,19 @@ struct scsi_host_template {
*/
void (* sdev_destroy)(struct scsi_device *);
+ /*
+ * Setup or clear error handler field scsi_device.eh.
+ * This error handler is working on designated device, it will only
+ * operate the designated device, do not affect other devices.
+ * If not set, error handle will fallback.
+ * LLDD can use custom error handler, or use inner defined:
+ * scsi_device_setup_eh/scsi_device_clear_eh
+ *
+ * Status: OPTIONAL
+ */
+ int (*sdev_setup_eh)(struct scsi_device *sdev);
+ void (*sdev_clear_eh)(struct scsi_device *sdev);
+
/*
* Before the mid layer attempts to scan for a new device attached
* to a target where no target currently exists, it will call this
@@ -468,6 +481,9 @@ struct scsi_host_template {
/* The queuecommand callback may block. See also BLK_MQ_F_BLOCKING. */
unsigned queuecommand_may_block:1;
+ /* The error handle of scsi_device will fallback when failed. */
+ unsigned sdev_eh_fallback:1;
+
/*
* Countdown for host blocking with no commands outstanding.
*/
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* Re: [PATCH 08/14] scsi: scsi_error: Add a general LUN based error handler
2025-08-16 11:24 ` [PATCH 08/14] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
@ 2025-08-17 9:18 ` Markus Elfring
0 siblings, 0 replies; 38+ messages in thread
From: Markus Elfring @ 2025-08-17 9:18 UTC (permalink / raw)
To: JiangJianJun, Wenchao Hao, linux-scsi, James Bottomley,
Martin K. Petersen
Cc: LKML, Bart Van Assche, Bo Wu, Christoph Hellwig, Hannes Reinecke,
hewenliang4, John Garry, Mike Christie, Wenchao Hao, wuyifeng10,
yangxingui, yangyun50
…
> +++ b/drivers/scsi/scsi_error.c
> @@ -2804,3 +2804,182 @@ bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
…
> +static void sdev_eh_add_cmnd(struct scsi_cmnd *scmd)
> +{
…
> + spin_lock_irqsave(&luneh->eh_lock, flags);
> + list_add_tail(&scmd->eh_entry, &luneh->eh_cmd_q);
> + luneh->eh_num++;
> + spin_unlock_irqrestore(&luneh->eh_lock, flags);
> +}
…
Under which circumstances would you become interested to apply a statement
like “guard(spinlock_irqsave)(&luneh->eh_lock);”?
https://elixir.bootlin.com/linux/v6.17-rc1/source/include/linux/spinlock.h#L585-L588
Regards,
Markus
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH 09/14] scsi: core: increase/decrease target_busy if set error handler
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (7 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 08/14] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list JiangJianJun
` (7 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: JiangJianJun <jiangjianjun3@h-partners.com>
This is preparation for a genernal target based error handle strategy
to check if to wake up actual error handler, we should enable counter
field scsi_target.target_busy in this case.
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_lib.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 63bd33af336b..5115ab5d0942 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -393,7 +393,7 @@ void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
scsi_dec_host_busy(shost, cmd);
- if (starget->can_queue > 0)
+ if (starget->can_queue > 0 || starget->eh)
atomic_dec(&starget->target_busy);
sbitmap_put(&sdev->budget_map, cmd->budget_token);
@@ -1412,8 +1412,11 @@ static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
spin_unlock_irq(shost->host_lock);
}
- if (starget->can_queue <= 0)
+ if (starget->can_queue <= 0) {
+ if (starget->eh)
+ atomic_inc(&starget->target_busy);
return 1;
+ }
busy = atomic_inc_return(&starget->target_busy) - 1;
if (atomic_read(&starget->target_blocked) > 0) {
@@ -1885,7 +1888,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
out_dec_host_busy:
scsi_dec_host_busy(shost, cmd);
out_dec_target_busy:
- if (scsi_target(sdev)->can_queue > 0)
+ if (scsi_target(sdev)->can_queue > 0 || scsi_target(sdev)->eh)
atomic_dec(&scsi_target(sdev)->target_busy);
out_put_budget:
scsi_mq_put_budget(q, cmd->budget_token);
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (8 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 09/14] scsi: core: increase/decrease target_busy if set " JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 23:19 ` kernel test robot
2025-08-16 11:24 ` [PATCH 11/14] scsi: scsi_error: Add a general target based error handler JiangJianJun
` (6 subsequent siblings)
16 siblings, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper scsi_starget_eh() to handle scsi target's error command list,
it would perform some steps which can be done with target's IO blocked,
including check sense, start unit, reset lun and reset target.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 130 ++++++++++++++++++++++++++++++++++++++
include/scsi/scsi_eh.h | 2 +
2 files changed, 132 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 49344f30bab9..bc9ca7f38580 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2591,6 +2591,136 @@ int scsi_sdev_eh(struct scsi_device *sdev,
}
EXPORT_SYMBOL_GPL(scsi_sdev_eh);
+static int starget_eh_stu(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev;
+ struct scsi_cmnd *scmd, *stu_scmd;
+
+ list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
+ if (sdev_stu_done(sdev))
+ continue;
+
+ stu_scmd = NULL;
+ list_for_each_entry(scmd, work_q, eh_entry)
+ if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
+ scsi_check_sense(scmd) == FAILED) {
+ stu_scmd = scmd;
+ break;
+ }
+ if (!stu_scmd)
+ continue;
+
+ if (scsi_eh_sdev_stu(stu_scmd, work_q, done_q))
+ return 1;
+ }
+
+ return 0;
+}
+
+static int starget_eh_reset_lun(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev;
+ struct scsi_cmnd *scmd, *bdr_scmd;
+
+ list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
+ if (sdev_reset_done(sdev))
+ continue;
+
+ bdr_scmd = NULL;
+ list_for_each_entry(scmd, work_q, eh_entry)
+ if (scmd->device) {
+ bdr_scmd = scmd;
+ break;
+ }
+ if (!bdr_scmd)
+ continue;
+
+ if (scsi_eh_sdev_reset(bdr_scmd, work_q, done_q))
+ return 1;
+ }
+
+ return 0;
+}
+
+static int starget_eh_reset_target(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ enum scsi_disposition rtn;
+ struct scsi_cmnd *scmd, *next;
+ LIST_HEAD(check_list);
+
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Sending target reset\n", current->comm));
+
+ rtn = scsi_try_target_reset(scmd);
+ if (rtn != SUCCESS && rtn != FAST_IO_FAIL) {
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Target reset failed\n",
+ current->comm));
+ return 0;
+ }
+
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Target reset %s\n", current->comm,
+ scsi_mlreturn_string(rtn)));
+
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+ if (rtn == SUCCESS)
+ list_move_tail(&scmd->eh_entry, &check_list);
+ else if (rtn == FAST_IO_FAIL)
+ scsi_eh_finish_cmd(scmd, done_q);
+ }
+
+ return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
+}
+
+/*
+ * Target based error handle
+ *
+ * @work_q: list of scsi commands need to recovery
+ * @done_q: list of scsi commands handled
+ *
+ * return: return 1 if all commands in work_q is recoveryed, else 0 is returned
+ */
+int scsi_starget_eh(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ int ret = 0;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: checking sense\n", current->comm));
+ ret = scsi_eh_get_sense(work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: start unit\n", current->comm));
+ ret = starget_eh_stu(starget, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: reset LUN\n", current->comm));
+ ret = starget_eh_reset_lun(starget, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: reset target\n", current->comm));
+ ret = starget_eh_reset_target(starget, work_q, done_q);
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(scsi_starget_eh);
+
/**
* scsi_report_bus_reset() - report bus reset observed
*
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index d8e4475ff004..cda0b962fc47 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -20,6 +20,8 @@ extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
struct list_head *doneq);
+extern int scsi_starget_eh(struct scsi_target *starget,
+ struct list_head *workq, struct list_head *doneq);
extern int scsi_device_setup_eh(struct scsi_device *sdev);
extern void scsi_device_clear_eh(struct scsi_device *sdev);
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* Re: [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list
2025-08-16 11:24 ` [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list JiangJianJun
@ 2025-08-16 23:19 ` kernel test robot
2025-08-17 2:46 ` JiangJianJun
0 siblings, 1 reply; 38+ messages in thread
From: kernel test robot @ 2025-08-16 23:19 UTC (permalink / raw)
To: JiangJianJun, James.Bottomley, martin.petersen, linux-scsi
Cc: oe-kbuild-all, linux-kernel, hare, bvanassche, michael.christie,
hch, haowenchao22, john.g.garry, hewenliang4, yangyun50,
wuyifeng10, wubo40, yangxingui
Hi JiangJianJun,
kernel test robot noticed the following build warnings:
[auto build test WARNING on jejb-scsi/for-next]
[also build test WARNING on mkp-scsi/for-next linus/master v6.17-rc1 next-20250815]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/JiangJianJun/scsi-scsi_error-Define-framework-for-LUN-target-based-error-handle/20250816-185707
base: https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next
patch link: https://lore.kernel.org/r/20250816112417.3581253-11-jiangjianjun3%40huawei.com
patch subject: [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list
config: sh-randconfig-002-20250817 (https://download.01.org/0day-ci/archive/20250817/202508170715.5Q0ZpgmO-lkp@intel.com/config)
compiler: sh4-linux-gcc (GCC) 12.5.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250817/202508170715.5Q0ZpgmO-lkp@intel.com/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202508170715.5Q0ZpgmO-lkp@intel.com/
All warnings (new ones prefixed by >>):
In file included from drivers/scsi/scsi_error.c:46:
drivers/scsi/scsi_error.c: In function 'starget_eh_reset_target':
>> drivers/scsi/scsi_error.c:2671:30: warning: '%s' directive argument is null [-Wformat-overflow=]
2671 | "%s: Target reset %s\n", current->comm,
| ^~~~~~~~~~~~~~~~~~~~~~~
drivers/scsi/scsi_logging.h:51:25: note: in definition of macro 'SCSI_CHECK_LOGGING'
51 | CMD; \
| ^~~
drivers/scsi/scsi_error.c:2670:9: note: in expansion of macro 'SCSI_LOG_ERROR_RECOVERY'
2670 | SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
| ^~~~~~~~~~~~~~~~~~~~~~~
include/scsi/scsi_device.h:478:9: note: in expansion of macro 'dev_printk'
478 | dev_printk(prefix, &(starget)->dev, fmt, ##a)
| ^~~~~~~~~~
drivers/scsi/scsi_error.c:2670:36: note: in expansion of macro 'starget_printk'
2670 | SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
| ^~~~~~~~~~~~~~
drivers/scsi/scsi_error.c:2671:48: note: format string is defined here
2671 | "%s: Target reset %s\n", current->comm,
| ^~
vim +2671 drivers/scsi/scsi_error.c
2648
2649 static int starget_eh_reset_target(struct scsi_target *starget,
2650 struct list_head *work_q,
2651 struct list_head *done_q)
2652 {
2653 enum scsi_disposition rtn;
2654 struct scsi_cmnd *scmd, *next;
2655 LIST_HEAD(check_list);
2656
2657 scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
2658
2659 SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
2660 "%s: Sending target reset\n", current->comm));
2661
2662 rtn = scsi_try_target_reset(scmd);
2663 if (rtn != SUCCESS && rtn != FAST_IO_FAIL) {
2664 SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
2665 "%s: Target reset failed\n",
2666 current->comm));
2667 return 0;
2668 }
2669
2670 SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
> 2671 "%s: Target reset %s\n", current->comm,
2672 scsi_mlreturn_string(rtn)));
2673
2674 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
2675 if (rtn == SUCCESS)
2676 list_move_tail(&scmd->eh_entry, &check_list);
2677 else if (rtn == FAST_IO_FAIL)
2678 scsi_eh_finish_cmd(scmd, done_q);
2679 }
2680
2681 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
2682 }
2683
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
^ permalink raw reply [flat|nested] 38+ messages in thread* [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list
2025-08-16 23:19 ` kernel test robot
@ 2025-08-17 2:46 ` JiangJianJun
0 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-17 2:46 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper scsi_starget_eh() to handle scsi target's error command list,
it would perform some steps which can be done with target's IO blocked,
including check sense, start unit, reset lun and reset target.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 130 ++++++++++++++++++++++++++++++++++++++
include/scsi/scsi_eh.h | 2 +
2 files changed, 132 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 49344f30bab9..15be739135b7 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2591,6 +2591,136 @@ int scsi_sdev_eh(struct scsi_device *sdev,
}
EXPORT_SYMBOL_GPL(scsi_sdev_eh);
+static int starget_eh_stu(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev;
+ struct scsi_cmnd *scmd, *stu_scmd;
+
+ list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
+ if (sdev_stu_done(sdev))
+ continue;
+
+ stu_scmd = NULL;
+ list_for_each_entry(scmd, work_q, eh_entry)
+ if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
+ scsi_check_sense(scmd) == FAILED) {
+ stu_scmd = scmd;
+ break;
+ }
+ if (!stu_scmd)
+ continue;
+
+ if (scsi_eh_sdev_stu(stu_scmd, work_q, done_q))
+ return 1;
+ }
+
+ return 0;
+}
+
+static int starget_eh_reset_lun(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev;
+ struct scsi_cmnd *scmd, *bdr_scmd;
+
+ list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
+ if (sdev_reset_done(sdev))
+ continue;
+
+ bdr_scmd = NULL;
+ list_for_each_entry(scmd, work_q, eh_entry)
+ if (scmd->device) {
+ bdr_scmd = scmd;
+ break;
+ }
+ if (!bdr_scmd)
+ continue;
+
+ if (scsi_eh_sdev_reset(bdr_scmd, work_q, done_q))
+ return 1;
+ }
+
+ return 0;
+}
+
+static int starget_eh_reset_target(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ enum scsi_disposition rtn;
+ struct scsi_cmnd *scmd, *next;
+ LIST_HEAD(check_list);
+
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Sending target reset\n", current->comm));
+
+ rtn = scsi_try_target_reset(scmd);
+ if (rtn != SUCCESS && rtn != FAST_IO_FAIL) {
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Target reset failed\n",
+ current->comm));
+ return 0;
+ }
+
+ SCSI_LOG_ERROR_RECOVERY(3, starget_printk(KERN_INFO, starget,
+ "%s: Target reset success: rtn=%#x\n",
+ current->comm, rtn));
+
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
+ if (rtn == SUCCESS)
+ list_move_tail(&scmd->eh_entry, &check_list);
+ else if (rtn == FAST_IO_FAIL)
+ scsi_eh_finish_cmd(scmd, done_q);
+ }
+
+ return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
+}
+
+/*
+ * Target based error handle
+ *
+ * @work_q: list of scsi commands need to recovery
+ * @done_q: list of scsi commands handled
+ *
+ * return: return 1 if all commands in work_q is recoveryed, else 0 is returned
+ */
+int scsi_starget_eh(struct scsi_target *starget,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ int ret = 0;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: checking sense\n", current->comm));
+ ret = scsi_eh_get_sense(work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: start unit\n", current->comm));
+ ret = starget_eh_stu(starget, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: reset LUN\n", current->comm));
+ ret = starget_eh_reset_lun(starget, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: reset target\n", current->comm));
+ ret = starget_eh_reset_target(starget, work_q, done_q);
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(scsi_starget_eh);
+
/**
* scsi_report_bus_reset() - report bus reset observed
*
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index d8e4475ff004..cda0b962fc47 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -20,6 +20,8 @@ extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
struct list_head *doneq);
+extern int scsi_starget_eh(struct scsi_target *starget,
+ struct list_head *workq, struct list_head *doneq);
extern int scsi_device_setup_eh(struct scsi_device *sdev);
extern void scsi_device_clear_eh(struct scsi_device *sdev);
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* [PATCH 11/14] scsi: scsi_error: Add a general target based error handler
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (9 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 10/14] scsi: scsi_error: Add helper to handle scsi target's error command list JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 12/14] scsi: scsi_debug: Add params for configuring the " JiangJianJun
` (5 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add a general target based error handler which can be used by drivers
directly. This error handler implements an scsi_target_eh, when handling
error commands, it would call helper function scsi_starget_eh() added
before to try recover error commands.
The behavior if scsi_starget_eh() can not recover all error commands
depends on fallback flag, which is initialized when scsi_target is
allocated. If fallback is set, it would fallback to further error
recover strategy like old host based error handle; else it would
mark this scsi devices of this target offline and flush all error
commands. Add a flag for controlling rollback in scsi_host_template.
Add interface target_setup_eh/target_clear_eh in scsi_host_template,
used for setup/clear target based error handler. Drivers can
implements them custom, or use inner implements:
scsi_target_setup_eh/scsi_target_clear_eh.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@h-partners.com>
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_error.c | 171 ++++++++++++++++++++++++++++++++++++++
drivers/scsi/scsi_scan.c | 12 +++
include/scsi/scsi_eh.h | 2 +
include/scsi/scsi_host.h | 15 ++++
4 files changed, 200 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index bc9ca7f38580..e2adf0349f81 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -3113,3 +3113,174 @@ void scsi_device_clear_eh(struct scsi_device *sdev)
sdev->eh = NULL;
}
EXPORT_SYMBOL_GPL(scsi_device_clear_eh);
+
+static inline void starget_clear_eh_done(struct scsi_target *starget)
+{
+ struct scsi_device *sdev;
+
+ list_for_each_entry(sdev, &starget->devices, same_target_siblings) {
+ if (!sdev->eh)
+ continue;
+ sdev->eh->get_sense_done = 0;
+ sdev->eh->stu_done = 0;
+ sdev->eh->reset_done = 0;
+ }
+}
+
+static inline void *scsi_eh_starget_priv(struct scsi_target_eh *eh)
+{
+ return eh + 1;
+}
+
+struct starget_eh {
+ spinlock_t eh_lock;
+ unsigned int eh_num;
+ struct list_head eh_cmd_q;
+ struct scsi_target *starget;
+ struct work_struct eh_handle_work;
+ unsigned int fallback:1;
+};
+
+static void starget_eh_work(struct work_struct *work)
+{
+ struct scsi_cmnd *scmd, *next;
+ unsigned long flags;
+ LIST_HEAD(eh_work_q);
+ LIST_HEAD(eh_done_q);
+ struct starget_eh *stargeteh =
+ container_of(work, struct starget_eh, eh_handle_work);
+ struct scsi_target *starget = stargeteh->starget;
+ struct scsi_target_eh *eh = starget->eh;
+
+ spin_lock_irqsave(&stargeteh->eh_lock, flags);
+ list_splice_init(&stargeteh->eh_cmd_q, &eh_work_q);
+ spin_unlock_irqrestore(&stargeteh->eh_lock, flags);
+
+ if (scsi_starget_eh(starget, &eh_work_q, &eh_done_q))
+ goto out_clear_flag;
+
+ if (!stargeteh->fallback) {
+ scsi_eh_offline_sdevs(&eh_work_q, &eh_done_q);
+ goto out_clear_flag;
+ }
+
+ /*
+ * fallback to host based error handle
+ */
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh fallback to further recovery\n", current->comm));
+ eh->reset_done = 1;
+ list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry) {
+ list_del_init(&scmd->eh_entry);
+ scsi_eh_scmd_add_shost(scmd);
+ }
+ goto out_flush_done;
+
+out_clear_flag:
+ starget_clear_eh_done(starget);
+
+out_flush_done:
+ scsi_eh_flush_done_q(&eh_done_q);
+ spin_lock_irqsave(&stargeteh->eh_lock, flags);
+ stargeteh->eh_num = 0;
+ spin_unlock_irqrestore(&stargeteh->eh_lock, flags);
+}
+
+static void starget_eh_add_cmnd(struct scsi_cmnd *scmd)
+{
+ unsigned long flags;
+ struct scsi_target *starget = scmd->device->sdev_target;
+ struct starget_eh *eh;
+
+ eh = scsi_eh_starget_priv(starget->eh);
+
+ spin_lock_irqsave(&eh->eh_lock, flags);
+ list_add_tail(&scmd->eh_entry, &eh->eh_cmd_q);
+ eh->eh_num++;
+ spin_unlock_irqrestore(&eh->eh_lock, flags);
+}
+
+static bool starget_eh_is_busy(struct scsi_target *starget)
+{
+ int ret = 0;
+ unsigned long flags;
+ struct starget_eh *eh;
+
+ eh = scsi_eh_starget_priv(starget->eh);
+
+ spin_lock_irqsave(&eh->eh_lock, flags);
+ ret = eh->eh_num;
+ spin_unlock_irqrestore(&eh->eh_lock, flags);
+
+ return ret != 0;
+}
+
+static void starget_eh_wakeup(struct scsi_target *starget)
+{
+ unsigned long flags;
+ unsigned int nr_error;
+ unsigned int nr_busy;
+ struct starget_eh *eh;
+
+ eh = scsi_eh_starget_priv(starget->eh);
+
+ spin_lock_irqsave(&eh->eh_lock, flags);
+ nr_error = eh->eh_num;
+ spin_unlock_irqrestore(&eh->eh_lock, flags);
+
+ nr_busy = atomic_read(&starget->target_busy);
+
+ if (!nr_error || nr_busy != nr_error) {
+ SCSI_LOG_ERROR_RECOVERY(5, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: do not wake up, busy/error is %d/%d\n",
+ current->comm, nr_busy, nr_error));
+ return;
+ }
+
+ SCSI_LOG_ERROR_RECOVERY(2, starget_printk(KERN_INFO, starget,
+ "%s:targeteh: waking up, busy/error is %d/%d\n",
+ current->comm, nr_busy, nr_error));
+
+ schedule_work(&eh->eh_handle_work);
+}
+
+/*
+ * This is default implement of Scsi_Host.target_setup_eh.
+ */
+int scsi_target_setup_eh(struct scsi_target *starget)
+{
+ struct Scsi_Host *shost = dev_to_shost(starget->dev.parent);
+ struct scsi_target_eh *eh;
+ struct starget_eh *stargeteh;
+
+ eh = kzalloc(sizeof(struct scsi_target_eh) + sizeof(struct starget_eh),
+ GFP_KERNEL);
+ if (!eh)
+ return -ENOMEM;
+ stargeteh = scsi_eh_starget_priv(eh);
+
+ eh->add_cmnd = starget_eh_add_cmnd;
+ eh->is_busy = starget_eh_is_busy;
+ eh->wakeup = starget_eh_wakeup;
+ stargeteh->starget = starget;
+ stargeteh->fallback = shost->hostt->target_eh_fallback;
+
+ spin_lock_init(&stargeteh->eh_lock);
+ INIT_LIST_HEAD(&stargeteh->eh_cmd_q);
+ INIT_WORK(&stargeteh->eh_handle_work, starget_eh_work);
+
+ starget->eh = eh;
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(scsi_target_setup_eh);
+
+/*
+ * This is default implement of Scsi_Host.target_clear_eh.
+ */
+void scsi_target_clear_eh(struct scsi_target *starget)
+{
+ kfree(starget->eh);
+ starget->eh = NULL;
+}
+EXPORT_SYMBOL_GPL(scsi_target_clear_eh);
diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 1c5cb77dfb22..15848beb2b93 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -405,6 +405,8 @@ static void scsi_target_destroy(struct scsi_target *starget)
starget->state = STARGET_DEL;
transport_destroy_device(dev);
spin_lock_irqsave(shost->host_lock, flags);
+ if (shost->hostt->target_clear_eh)
+ shost->hostt->target_clear_eh(starget);
if (shost->hostt->target_destroy)
shost->hostt->target_destroy(starget);
list_del_init(&starget->siblings);
@@ -553,6 +555,16 @@ static struct scsi_target *scsi_alloc_target(struct device *parent,
return NULL;
}
}
+ if (shost->hostt->target_setup_eh) {
+ error = shost->hostt->target_setup_eh(starget);
+ if (error) {
+ dev_err(dev, "target setup error handler failed, error %d\n",
+ error);
+ scsi_target_destroy(starget);
+ return NULL;
+ }
+ }
+
get_device(dev);
return starget;
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index cda0b962fc47..4d302be8d6d8 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -24,6 +24,8 @@ extern int scsi_starget_eh(struct scsi_target *starget,
struct list_head *workq, struct list_head *doneq);
extern int scsi_device_setup_eh(struct scsi_device *sdev);
extern void scsi_device_clear_eh(struct scsi_device *sdev);
+extern int scsi_target_setup_eh(struct scsi_target *starget);
+extern void scsi_target_clear_eh(struct scsi_target *starget);
static inline bool scsi_sense_is_deferred(const struct scsi_sense_hdr *sshdr)
{
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index 9cc34bfff3f7..cac661cd9b5b 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -265,6 +265,19 @@ struct scsi_host_template {
*/
void (* target_destroy)(struct scsi_target *);
+ /*
+ * Setup or clear error handler field scsi_target.eh.
+ * This error handler is working on designated target, it will only
+ * operate the designated target, do not affect other targets.
+ * If not set, error handle will fallback.
+ * LLDD can use custom error handler, or use inner defined:
+ * scsi_target_setup_eh/scsi_target_clear_eh
+ *
+ * Status: OPTIONAL
+ */
+ int (*target_setup_eh)(struct scsi_target *starget);
+ void (*target_clear_eh)(struct scsi_target *starget);
+
/*
* If a host has the ability to discover targets on its own instead
* of scanning the entire bus, it can fill in this function and
@@ -483,6 +496,8 @@ struct scsi_host_template {
/* The error handle of scsi_device will fallback when failed. */
unsigned sdev_eh_fallback:1;
+ /* The error handle of scsi_target will fallback when failed. */
+ unsigned target_eh_fallback:1;
/*
* Countdown for host blocking with no commands outstanding.
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 12/14] scsi: scsi_debug: Add params for configuring the error handler
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (10 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 11/14] scsi: scsi_error: Add a general target based error handler JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 13/14] scsi: virtio_scsi: enable LUN based error handlers JiangJianJun
` (4 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: JiangJianJun <jiangjianjun3@h-partners.com>
Add a new module parameter to configure error handlers based on
LUN/targets and toggle the enable/disable fallback functionality.
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/scsi_debug.c | 26 +++++++++++++++++++++++++-
1 file changed, 25 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
index 0847767d4d43..64d5b85d961b 100644
--- a/drivers/scsi/scsi_debug.c
+++ b/drivers/scsi/scsi_debug.c
@@ -961,6 +961,10 @@ static bool write_since_sync;
static bool sdebug_statistics = DEF_STATISTICS;
static bool sdebug_wp;
static bool sdebug_allow_restart;
+static bool sdebug_lun_eh;
+static bool sdebug_lun_eh_fallback;
+static bool sdebug_target_eh;
+static bool sdebug_target_eh_fallback;
static enum {
BLK_ZONED_NONE = 0,
BLK_ZONED_HA = 1,
@@ -7362,6 +7366,10 @@ module_param_named(zone_max_open, sdeb_zbc_max_open, int, S_IRUGO);
module_param_named(zone_nr_conv, sdeb_zbc_nr_conv, int, S_IRUGO);
module_param_named(zone_size_mb, sdeb_zbc_zone_size_mb, int, S_IRUGO);
module_param_named(allow_restart, sdebug_allow_restart, bool, S_IRUGO | S_IWUSR);
+module_param_named(lun_eh, sdebug_lun_eh, bool, 0444);
+module_param_named(lun_eh_fallback, sdebug_lun_eh_fallback, bool, 0444);
+module_param_named(target_eh, sdebug_target_eh, bool, 0444);
+module_param_named(target_eh_fallback, sdebug_target_eh_fallback, bool, 0444);
MODULE_AUTHOR("Eric Youngdale + Douglas Gilbert");
MODULE_DESCRIPTION("SCSI debug adapter driver");
@@ -7441,6 +7449,10 @@ MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones; [0] for no limit
MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones (def=1)");
MODULE_PARM_DESC(zone_size_mb, "Zone size in MiB (def=auto)");
MODULE_PARM_DESC(allow_restart, "Set scsi_device's allow_restart flag(def=0)");
+MODULE_PARM_DESC(lun_eh, "LUN based error handle (def=0)");
+MODULE_PARM_DESC(lun_eh_fallback, "Fallback to further recovery if LUN recovery failed (def=0)");
+MODULE_PARM_DESC(target_eh, "target based error handle (def=0)");
+MODULE_PARM_DESC(target_eh_fallback, "Fallback to further recovery if target recovery failed (def=0)");
#define SDEBUG_INFO_LEN 256
static char sdebug_info[SDEBUG_INFO_LEN];
@@ -8427,6 +8439,7 @@ static struct attribute *sdebug_drv_attrs[] = {
ATTRIBUTE_GROUPS(sdebug_drv);
static struct device *pseudo_primary;
+static struct scsi_host_template sdebug_driver_template;
static int __init scsi_debug_init(void)
{
@@ -8435,6 +8448,17 @@ static int __init scsi_debug_init(void)
int k, ret, hosts_to_add;
int idx = -1;
+ if (sdebug_lun_eh) {
+ sdebug_driver_template.sdev_setup_eh = scsi_device_setup_eh;
+ sdebug_driver_template.sdev_clear_eh = scsi_device_clear_eh;
+ sdebug_driver_template.sdev_eh_fallback = sdebug_lun_eh_fallback;
+ }
+ if (sdebug_target_eh) {
+ sdebug_driver_template.target_setup_eh = scsi_target_setup_eh;
+ sdebug_driver_template.target_clear_eh = scsi_target_clear_eh;
+ sdebug_driver_template.target_eh_fallback = sdebug_target_eh_fallback;
+ }
+
if (sdebug_ndelay >= 1000 * 1000 * 1000) {
pr_warn("ndelay must be less than 1 second, ignored\n");
sdebug_ndelay = 0;
@@ -9412,7 +9436,7 @@ static int sdebug_init_cmd_priv(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
return 0;
}
-static const struct scsi_host_template sdebug_driver_template = {
+static struct scsi_host_template sdebug_driver_template = {
.show_info = scsi_debug_show_info,
.write_info = scsi_debug_write_info,
.proc_name = sdebug_proc_name,
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 13/14] scsi: virtio_scsi: enable LUN based error handlers
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (11 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 12/14] scsi: scsi_debug: Add params for configuring the " JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-16 11:24 ` [PATCH 14/14] scsi: iscsi_tcp: enable LUN-based and target-based " JiangJianJun
` (3 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: JiangJianJun <jiangjianjun3@h-partners.com>
Enable the virtio_scsi driver to support LUN-based error handlers.
Since virtio_scsi does not define any further reset callbacks,
there is no need to fallback for further recover, so we have
disabled the fallback functionality.
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/virtio_scsi.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index 96a69edddbe5..3c525521351c 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -28,6 +28,7 @@
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_tcq.h>
#include <scsi/scsi_devinfo.h>
+#include <scsi/scsi_eh.h>
#include <linux/seqlock.h>
#include "sd.h"
@@ -801,6 +802,8 @@ static const struct scsi_host_template virtscsi_host_template = {
.eh_device_reset_handler = virtscsi_device_reset,
.eh_timed_out = virtscsi_eh_timed_out,
.sdev_init = virtscsi_device_alloc,
+ .sdev_setup_eh = scsi_device_setup_eh,
+ .sdev_clear_eh = scsi_device_clear_eh,
.dma_boundary = UINT_MAX,
.map_queues = virtscsi_map_queues,
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 14/14] scsi: iscsi_tcp: enable LUN-based and target-based error handlers
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (12 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 13/14] scsi: virtio_scsi: enable LUN based error handlers JiangJianJun
@ 2025-08-16 11:24 ` JiangJianJun
2025-08-17 8:46 ` [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (2 subsequent siblings)
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-16 11:24 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
From: JiangJianJun <jiangjianjun3@h-partners.com>
The iSCSI TCP driver now supports resetting LUNs or targets,
allowing us to enable LUN-based error handlers and enable them
to fall back to target-based error handlers.
Signed-off-by: JiangJianJun <jiangjianjun3@h-partners.com>
---
drivers/scsi/iscsi_tcp.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c
index 7b4fe0e6afb2..328e76219b1c 100644
--- a/drivers/scsi/iscsi_tcp.c
+++ b/drivers/scsi/iscsi_tcp.c
@@ -33,6 +33,7 @@
#include <scsi/scsi_device.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi.h>
+#include <scsi/scsi_eh.h>
#include <scsi/scsi_transport_iscsi.h>
#include <trace/events/iscsi.h>
#include <trace/events/sock.h>
@@ -1040,6 +1041,9 @@ static const struct scsi_host_template iscsi_sw_tcp_sht = {
.eh_target_reset_handler = iscsi_eh_recover_target,
.dma_boundary = PAGE_SIZE - 1,
.sdev_configure = iscsi_sw_tcp_sdev_configure,
+ .sdev_setup_eh = scsi_device_setup_eh,
+ .sdev_clear_eh = scsi_device_clear_eh,
+ .sdev_eh_fallback = 1,
.proc_name = "iscsi_tcp",
.this_id = -1,
.track_queue_depth = 1,
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (13 preceding siblings ...)
2025-08-16 11:24 ` [PATCH 14/14] scsi: iscsi_tcp: enable LUN-based and target-based " JiangJianJun
@ 2025-08-17 8:46 ` JiangJianJun
2025-08-20 12:36 ` Hannes Reinecke
2025-08-22 7:39 ` [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism Damien Le Moal
16 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-08-17 8:46 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
I apologize for taking too long to test and modify these patches.Here
are the revision points, with the first two being the main ones:
1. The scsi_target.can_queue value can control the increment and
decrement operations of scsi_target.target_busy. The new error
handler also needs to adjust target_busy. Both Bart and Mike opposed
removing this control condition. In this version, I have only added
the error handler check when scsi_target.can_queue <= 0.
link: https://lore.kernel.org/linux-scsi/daba5c92-2395-4eee-b212-978fbe83b56f@oracle.com/
2. I have added callbacks for setting and clearing the error handler
in scsi_host_template. Drivers can support device or target error
handlers by setting these callbacks. I believe the advantage of this
approach is that driver developers will be aware of this feature
when they see the callback prototypes and comments, and it can be
used even without device or target initialization callbacks.
However, this means that the modparam controlling the enablement can
only be removed. I have considered adding configurations only for
the virtio_scsi and iscsi_tcp drivers.
link: https://lore.kernel.org/linux-scsi/b8350de1-6ac8-4d5f-aaa7-7b03e2f7aa93@oracle.com/
3. In scsi_eh_scmd_add, a return statement was added under the
condition of xxx_in_recovery because each branch ultimately calls
scsi_eh_scmd_add_shost, which does not fail. Therefore, continuing
further would result in duplicate additions.
4. The return type of ->is_busy was changed to bool.
I have retained the original content of Wenchao's email in cover letter.
^ permalink raw reply [flat|nested] 38+ messages in thread* Re: [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (14 preceding siblings ...)
2025-08-17 8:46 ` [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
@ 2025-08-20 12:36 ` Hannes Reinecke
2025-09-02 5:56 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
2025-08-22 7:39 ` [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism Damien Le Moal
16 siblings, 2 replies; 38+ messages in thread
From: Hannes Reinecke @ 2025-08-20 12:36 UTC (permalink / raw)
To: JiangJianJun, James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, bvanassche, michael.christie, hch, haowenchao22,
john.g.garry, hewenliang4, yangyun50, wuyifeng10, wubo40,
yangxingui
On 8/16/25 13:24, JiangJianJun wrote:
> It's unbearable for systems with large scale scsi devices share HBAs to
> block all devices' IOs when handle error commands, we need a new error
> handle mechanism to address this issue.
>
> I consulted about this issue a year ago, the discuss link can be found in
> refenence. Hannes replied about why we have to block the SCSI host
> then perform error recovery kindly. I think it's unnecessary to block
> SCSI host for all drivers and can try a small level recovery(LUN based for
> example) first to avoid block the SCSI host.
>
> The new error handle mechanism introduced in this patchset has been
> developed and tested with out self developed hardware since one year
> ago, now we want this mechanism can be used by more drivers.
>
> Drivers can decide if using the new error handle mechanism and how to
> handle error commands when scsi_device are scanned,the new mechanism
> makes SCSI error handle more flexible.
>
Hmm. Yes, and no.
I fully agree that SCSI EH is in need of reworking. But adding
another layer of complexity on top of the existing one ... not sure.
Additionally: TARGET RESET TMF is dead, and has been removed from SAM
since several years. It really is not worthwhile implementing.
Can't we take a simple step, and just try to have a non-blocking version
of device reset?
I think that should cover quite some issues already.
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@suse.de +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
^ permalink raw reply [flat|nested] 38+ messages in thread* [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-08-20 12:36 ` Hannes Reinecke
@ 2025-09-02 5:56 ` JiangJianJun
2025-09-02 6:37 ` Hannes Reinecke
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
1 sibling, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-09-02 5:56 UTC (permalink / raw)
To: hare, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
>I fully agree that SCSI EH is in need of reworking. But adding
>another layer of complexity on top of the existing one ... not sure.
Perhaps it would have been better to use only the error handler on the
device from the start. Users might wonder why a single disk failure
could cause other disks to become blocking.
>Additionally: TARGET RESET TMF is dead, and has been removed from SAM
>since several years. It really is not worthwhile implementing.
Hmm.
>Can't we take a simple step, and just try to have a non-blocking version
>of device reset?
>I think that should cover quite some issues already.
Do you think it's necessary to escalate the issue after the device reset
fails? Should we reset the bus or the host?
Moreover, a failed device reset does not necessarily indicate a fault
with the target or host.
And what means of "non-blocking"?
^ permalink raw reply [flat|nested] 38+ messages in thread
* Re: [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-09-02 5:56 ` JiangJianJun
@ 2025-09-02 6:37 ` Hannes Reinecke
0 siblings, 0 replies; 38+ messages in thread
From: Hannes Reinecke @ 2025-09-02 6:37 UTC (permalink / raw)
To: JiangJianJun, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
On 9/2/25 07:56, JiangJianJun wrote:
>> I fully agree that SCSI EH is in need of reworking. But adding
>> another layer of complexity on top of the existing one ... not sure.
>
> Perhaps it would have been better to use only the error handler on the
> device from the start. Users might wonder why a single disk failure
> could cause other disks to become blocking.
>
>> Additionally: TARGET RESET TMF is dead, and has been removed from SAM
>> since several years. It really is not worthwhile implementing.
>
> Hmm.
>
>> Can't we take a simple step, and just try to have a non-blocking version
>> of device reset?
>> I think that should cover quite some issues already.
>
> Do you think it's necessary to escalate the issue after the device reset
> fails? Should we reset the bus or the host?
> Moreover, a failed device reset does not necessarily indicate a fault
> with the target or host.
> And what means of "non-blocking"?
>
On the contrary, a failed device reset _always_ needs to be escalated.
The problem is that all EH issues start with a failed command (ignoring
the sg_reset case for now).
And a command typically is associated with data buffers / memory areas.
So when a command is failed we need to know when these buffers can be
released. If the device reset fails the command could not be reset,
and the buffers cannot be released. And without further escalation the
buffers remain locked until the next reboot.
That's why host reset is so important: that typically resets the entire
HBA (via a PCI-level reset or similar), so we can be sure that
afterwards all buffers are released and the command can be completed.
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@suse.de +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
^ permalink raw reply [flat|nested] 38+ messages in thread
* [RFC PATCH v4 0/9] scsi: scsi_error: Introduce new error handle mechanism
2025-08-20 12:36 ` Hannes Reinecke
2025-09-02 5:56 ` JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 1/9] scsi: scsi_error: Define framework for LUN based error handle JiangJianJun
` (8 more replies)
1 sibling, 9 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
> Additionally: TARGET RESET TMF is dead, and has been removed from SAM
> since several years. It really is not worthwhile implementing.
Hannes suggested removing the faulty handler on the target, so I revised
a version and submitted it as a proposal.
> Also, was this all tested with libata and libsas attached devices as well ?
> They all depend on scsi EH.
And I removed the settings for iscsi_tcp and virtio_scsi, maybe that
adding this proposal to specific drivers before the plan is finalized is
not appropriate. Keep the setting of scsi_debug for test the solution.
The introduction can be simplified as follows: This tmf only blocks IO
on the current device and does not interfere with IO or tmf on other
devices.
JiangJianJun (1):
scsi: scsi_debug: Add params for configuring the error handler
Wenchao Hao (8):
scsi: scsi_error: Define framework for LUN based error handle
scsi: scsi_error: Move complete variable eh_action from shost to
sdevice
scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset
scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT
scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset
scsi: scsi_error: Add flags to mark error handle steps has done
scsi: scsi_error: Add helper to handle scsi device's error command
list
scsi: scsi_error: Add a general LUN based error handler
drivers/scsi/scsi_debug.c | 15 +-
drivers/scsi/scsi_error.c | 422 ++++++++++++++++++++++++++++++++-----
drivers/scsi/scsi_lib.c | 7 +
drivers/scsi/scsi_priv.h | 8 +
drivers/scsi/scsi_scan.c | 7 +
drivers/scsi/scsi_sysfs.c | 2 +
include/scsi/scsi_device.h | 49 +++++
include/scsi/scsi_eh.h | 4 +
include/scsi/scsi_host.h | 18 +-
9 files changed, 476 insertions(+), 56 deletions(-)
--
2.33.0
^ permalink raw reply [flat|nested] 38+ messages in thread* [RFC PATCH v4 1/9] scsi: scsi_error: Define framework for LUN based error handle
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 2/9] scsi: scsi_error: Move complete variable eh_action from shost to sdevice JiangJianJun
` (7 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
The old scsi error handle logic is based on host, once a scsi command
in one LUN of this host is classfied as failed, SCSI mid-level would
set the whole host to recovery state, and no IO can be submitted to
all LUNs of this host any more before recovery finished, while the
recovery process might take a long time to finish.
It's unreasonable when there are a lot of LUNs in one host.
This change introduce a way for driver to implement its own
error handle logic which can be based on scsi LUN as minimum unit.
scsi_device_eh is defined for error handle based on scsi LUN, and
pointer struct scsi_device_eh "eh" is added in scsi_device, which
is NULL by default.
LLDs can initialize the sdev->eh in hostt->slave_alloc to implement an
scsi LUN based error handle. If this member is not NULL, SCSI mid-level
would branch to drivers' error handler rather than the old one which
block whole host's IO.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 42 +++++++++++++++++++++++++++++++++++---
drivers/scsi/scsi_lib.c | 7 +++++++
drivers/scsi/scsi_priv.h | 8 ++++++++
include/scsi/scsi_device.h | 29 ++++++++++++++++++++++++++
4 files changed, 83 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 746ff6a1f309..b5b04f2c5d62 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -291,11 +291,28 @@ static void scsi_eh_inc_host_failed(struct rcu_head *head)
spin_unlock_irqrestore(shost->host_lock, flags);
}
+static bool scsi_eh_scmd_add_sdev(struct scsi_cmnd *scmd)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_device_eh *eh = sdev->eh;
+
+ if (!eh || !eh->add_cmnd)
+ return true;
+
+ scsi_eh_reset(scmd);
+ eh->add_cmnd(scmd);
+
+ if (eh->wakeup)
+ eh->wakeup(sdev);
+
+ return false;
+}
+
/**
- * scsi_eh_scmd_add - add scsi cmd to error handling.
- * @scmd: scmd to run eh on.
+ * Add scsi cmd to error handling of Scsi_Host.
+ * This is default action of error handle.
*/
-void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
+static void scsi_eh_scmd_add_shost(struct scsi_cmnd *scmd)
{
struct Scsi_Host *shost = scmd->device->host;
unsigned long flags;
@@ -322,6 +339,25 @@ void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
call_rcu_hurry(&scmd->rcu, scsi_eh_inc_host_failed);
}
+/**
+ * scsi_eh_scmd_add - add scsi cmd to error handling.
+ * @scmd: scmd to run eh on.
+ */
+void scsi_eh_scmd_add(struct scsi_cmnd *scmd)
+{
+ struct Scsi_Host *shost = scmd->device->host;
+
+ if (unlikely(scsi_host_in_recovery(shost)) ||
+ scsi_eh_scmd_add_sdev(scmd))
+ scsi_eh_scmd_add_shost(scmd);
+}
+
+void scsi_eh_try_wakeup_sdev(struct scsi_device *sdev)
+{
+ if (sdev->eh && sdev->eh->wakeup)
+ sdev->eh->wakeup(sdev);
+}
+
/**
* scsi_timeout - Timeout function for normal scsi commands.
* @req: request that is timing out.
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 0c65ecfedfbd..ac8fb801aa82 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -398,6 +398,8 @@ void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
sbitmap_put(&sdev->budget_map, cmd->budget_token);
cmd->budget_token = -1;
+
+ scsi_eh_try_wakeup_sdev(sdev);
}
/*
@@ -1360,6 +1362,9 @@ static inline int scsi_dev_queue_ready(struct request_queue *q,
{
int token;
+ if (scsi_device_in_recovery(sdev))
+ return -1;
+
token = sbitmap_get(&sdev->budget_map);
if (token < 0)
return -1;
@@ -1374,6 +1379,7 @@ static inline int scsi_dev_queue_ready(struct request_queue *q,
if (scsi_device_busy(sdev) > 1 ||
atomic_dec_return(&sdev->device_blocked) > 0) {
sbitmap_put(&sdev->budget_map, token);
+ scsi_eh_try_wakeup_sdev(sdev);
return -1;
}
@@ -1882,6 +1888,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
out_put_budget:
scsi_mq_put_budget(q, cmd->budget_token);
cmd->budget_token = -1;
+ scsi_eh_try_wakeup_sdev(sdev);
switch (ret) {
case BLK_STS_OK:
break;
diff --git a/drivers/scsi/scsi_priv.h b/drivers/scsi/scsi_priv.h
index 5b2b19f5e8ec..906f090d917c 100644
--- a/drivers/scsi/scsi_priv.h
+++ b/drivers/scsi/scsi_priv.h
@@ -92,6 +92,7 @@ extern int scsi_error_handler(void *host);
extern enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *cmd);
extern void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy);
extern void scsi_eh_scmd_add(struct scsi_cmnd *);
+extern void scsi_eh_try_wakeup_sdev(struct scsi_device *sdev);
void scsi_eh_ready_devs(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q);
@@ -191,6 +192,13 @@ static inline void scsi_dh_add_device(struct scsi_device *sdev) { }
static inline void scsi_dh_release_device(struct scsi_device *sdev) { }
#endif
+static inline bool scsi_device_in_recovery(struct scsi_device *sdev)
+{
+ struct scsi_device_eh *eh = sdev->eh;
+
+ return eh && eh->is_busy && eh->is_busy(sdev);
+}
+
struct bsg_device *scsi_bsg_register_queue(struct scsi_device *sdev);
extern int scsi_device_max_queue_depth(struct scsi_device *sdev);
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index 6d6500148c4b..c21b0a84bbd2 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -100,6 +100,34 @@ struct scsi_vpd {
unsigned char data[];
};
+struct scsi_device;
+
+struct scsi_device_eh {
+ /*
+ * add scsi command to error handler so it would be handuled by
+ * driver's error handle strategy
+ */
+ void (*add_cmnd)(struct scsi_cmnd *scmd);
+
+ /*
+ * to judge if the device is busy handling errors, called before
+ * dispatch scsi cmnd
+ *
+ * return 0 if it's ready to accepy scsi cmnd
+ * return 1 if it's in error handle, command's would not be dispatched
+ */
+ bool (*is_busy)(struct scsi_device *sdev);
+
+ /*
+ * wakeup device's error handle
+ *
+ * usually the error handler strategy would not run at once when
+ * error command is added. This function would be called when any
+ * scsi cmnd is finished or when scsi cmnd is added.
+ */
+ void (*wakeup)(struct scsi_device *sdev);
+};
+
struct scsi_device {
struct Scsi_Host *host;
struct request_queue *request_queue;
@@ -289,6 +317,7 @@ struct scsi_device {
struct mutex state_mutex;
enum scsi_device_state sdev_state;
struct task_struct *quiesced_by;
+ struct scsi_device_eh *eh;
unsigned long sdev_data[];
} __attribute__((aligned(sizeof(unsigned long))));
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 2/9] scsi: scsi_error: Move complete variable eh_action from shost to sdevice
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 1/9] scsi: scsi_error: Define framework for LUN based error handle JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 3/9] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset JiangJianJun
` (6 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
eh_action is used to wait for error handle command's completion if scsi
command is send in error handle. Now the error handler might based on
scsi_device, so move it to scsi_device.
This is preparation for a genernal LUN based error handle strategy.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 6 +++---
include/scsi/scsi_device.h | 2 ++
include/scsi/scsi_host.h | 2 --
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index b5b04f2c5d62..d8b3f5b0fd47 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -917,7 +917,7 @@ void scsi_eh_done(struct scsi_cmnd *scmd)
SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
"%s result: %x\n", __func__, scmd->result));
- eh_action = scmd->device->host->eh_action;
+ eh_action = scmd->device->eh_action;
if (eh_action)
complete(eh_action);
}
@@ -1206,7 +1206,7 @@ static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
retry:
scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
- shost->eh_action = &done;
+ sdev->eh_action = &done;
scsi_log_send(scmd);
scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER;
@@ -1250,7 +1250,7 @@ static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd,
rtn = SUCCESS;
}
- shost->eh_action = NULL;
+ sdev->eh_action = NULL;
scsi_log_completion(scmd, rtn);
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index c21b0a84bbd2..9d42858035ed 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -318,6 +318,8 @@ struct scsi_device {
enum scsi_device_state sdev_state;
struct task_struct *quiesced_by;
struct scsi_device_eh *eh;
+ struct completion *eh_action; /* Wait for specific actions */
+ /* on the device. */
unsigned long sdev_data[];
} __attribute__((aligned(sizeof(unsigned long))));
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index c53812b9026f..46f57fe78505 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -558,8 +558,6 @@ struct Scsi_Host {
struct list_head eh_abort_list;
struct list_head eh_cmd_q;
struct task_struct * ehandler; /* Error recovery thread. */
- struct completion * eh_action; /* Wait for specific actions on the
- host. */
wait_queue_head_t host_wait;
const struct scsi_host_template *hostt;
struct scsi_transport_template *transportt;
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 3/9] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 1/9] scsi: scsi_error: Define framework for LUN based error handle JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 2/9] scsi: scsi_error: Move complete variable eh_action from shost to sdevice JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 4/9] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT JiangJianJun
` (5 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
This is preparation for a genernal LUN based error handle
strategy, the strategy would reuse some error handler APIs,
but some steps of these function should not be performed. For
example, we should not perform bus/host reset if we just stop IOs
on one single LUN.
This change add checks in scsi_try_xxx_reset to make sure
the reset operations would not be performed only if the condition
is not satisfied.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 24 ++++++++++++++++++++----
1 file changed, 20 insertions(+), 4 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index d8b3f5b0fd47..80a85b387068 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -926,7 +926,7 @@ void scsi_eh_done(struct scsi_cmnd *scmd)
* scsi_try_host_reset - ask host adapter to reset itself
* @scmd: SCSI cmd to send host reset.
*/
-static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
+static enum scsi_disposition __scsi_try_host_reset(struct scsi_cmnd *scmd)
{
unsigned long flags;
enum scsi_disposition rtn;
@@ -952,11 +952,19 @@ static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
return rtn;
}
+static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd)
+{
+ if (!scsi_host_in_recovery(scmd->device->host))
+ return FAILED;
+
+ return __scsi_try_host_reset(scmd);
+}
+
/**
* scsi_try_bus_reset - ask host to perform a bus reset
* @scmd: SCSI cmd to send bus reset.
*/
-static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
+static enum scsi_disposition __scsi_try_bus_reset(struct scsi_cmnd *scmd)
{
unsigned long flags;
enum scsi_disposition rtn;
@@ -982,6 +990,14 @@ static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
return rtn;
}
+static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd)
+{
+ if (!scsi_host_in_recovery(scmd->device->host))
+ return FAILED;
+
+ return __scsi_try_bus_reset(scmd);
+}
+
static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
{
sdev->was_reset = 1;
@@ -2547,12 +2563,12 @@ scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
break;
fallthrough;
case SG_SCSI_RESET_BUS:
- rtn = scsi_try_bus_reset(scmd);
+ rtn = __scsi_try_bus_reset(scmd);
if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
break;
fallthrough;
case SG_SCSI_RESET_HOST:
- rtn = scsi_try_host_reset(scmd);
+ rtn = __scsi_try_host_reset(scmd);
if (rtn == SUCCESS)
break;
fallthrough;
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 4/9] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (2 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 3/9] scsi: scsi_error: Check if to do reset in scsi_try_xxx_reset JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 5/9] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset JiangJianJun
` (4 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper function scsi_eh_sdev_stu() to perform START_UNIT and check
if to finish some error commands.
This is preparation for a genernal LUN based error handle
strategy and did not change original logic.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 50 +++++++++++++++++++++++----------------
1 file changed, 29 insertions(+), 21 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 80a85b387068..a1b700dbb0ec 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -1559,6 +1559,31 @@ static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
return 1;
}
+static int scsi_eh_sdev_stu(struct scsi_cmnd *scmd,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_device *sdev = scmd->device;
+ struct scsi_cmnd *next;
+
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: Sending START_UNIT\n", current->comm));
+
+ if (scsi_eh_try_stu(scmd)) {
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: START_UNIT failed\n", current->comm));
+ return 0;
+ }
+
+ if (!scsi_device_online(sdev) || !scsi_eh_tur(scmd))
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry)
+ if (scmd->device == sdev &&
+ scsi_eh_action(scmd, SUCCESS) == SUCCESS)
+ scsi_eh_finish_cmd(scmd, done_q);
+
+ return list_empty(work_q);
+}
+
/**
* scsi_eh_stu - send START_UNIT if needed
* @shost: &scsi host being recovered.
@@ -1573,7 +1598,7 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q)
{
- struct scsi_cmnd *scmd, *stu_scmd, *next;
+ struct scsi_cmnd *scmd, *stu_scmd;
struct scsi_device *sdev;
shost_for_each_device(sdev, shost) {
@@ -1596,26 +1621,9 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
if (!stu_scmd)
continue;
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: Sending START_UNIT\n",
- current->comm));
-
- if (!scsi_eh_try_stu(stu_scmd)) {
- if (!scsi_device_online(sdev) ||
- !scsi_eh_tur(stu_scmd)) {
- list_for_each_entry_safe(scmd, next,
- work_q, eh_entry) {
- if (scmd->device == sdev &&
- scsi_eh_action(scmd, SUCCESS) == SUCCESS)
- scsi_eh_finish_cmd(scmd, done_q);
- }
- }
- } else {
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: START_UNIT failed\n",
- current->comm));
+ if (scsi_eh_sdev_stu(stu_scmd, work_q, done_q)) {
+ scsi_device_put(sdev);
+ break;
}
}
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 5/9] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (3 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 4/9] scsi: scsi_error: Add helper scsi_eh_sdev_stu to do START_UNIT JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 6/9] scsi: scsi_error: Add flags to mark error handle steps has done JiangJianJun
` (3 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper function scsi_eh_sdev_reset() to perform lun reset and check
if to finish some error commands.
This is preparation for a genernal LUN based error handle
strategy and did not change original logic.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 54 +++++++++++++++++++++++----------------
1 file changed, 32 insertions(+), 22 deletions(-)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index a1b700dbb0ec..f58aad351463 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -1630,6 +1630,34 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
return list_empty(work_q);
}
+static int scsi_eh_sdev_reset(struct scsi_cmnd *scmd,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ struct scsi_cmnd *next;
+ struct scsi_device *sdev = scmd->device;
+ enum scsi_disposition rtn;
+
+ SCSI_LOG_ERROR_RECOVERY(3, sdev_printk(KERN_INFO, sdev,
+ "%s: Sending BDR\n", current->comm));
+
+ rtn = scsi_try_bus_device_reset(scmd);
+ if (rtn != SUCCESS && rtn != FAST_IO_FAIL) {
+ SCSI_LOG_ERROR_RECOVERY(3,
+ sdev_printk(KERN_INFO, sdev,
+ "%s: BDR failed\n", current->comm));
+ return 0;
+ }
+
+ if (!scsi_device_online(sdev) || rtn == FAST_IO_FAIL ||
+ !scsi_eh_tur(scmd))
+ list_for_each_entry_safe(scmd, next, work_q, eh_entry)
+ if (scmd->device == sdev &&
+ scsi_eh_action(scmd, rtn) != FAILED)
+ scsi_eh_finish_cmd(scmd, done_q);
+
+ return list_empty(work_q);
+}
/**
* scsi_eh_bus_device_reset - send bdr if needed
@@ -1647,9 +1675,8 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
struct list_head *work_q,
struct list_head *done_q)
{
- struct scsi_cmnd *scmd, *bdr_scmd, *next;
+ struct scsi_cmnd *scmd, *bdr_scmd;
struct scsi_device *sdev;
- enum scsi_disposition rtn;
shost_for_each_device(sdev, shost) {
if (scsi_host_eh_past_deadline(shost)) {
@@ -1670,26 +1697,9 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
if (!bdr_scmd)
continue;
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: Sending BDR\n", current->comm));
- rtn = scsi_try_bus_device_reset(bdr_scmd);
- if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
- if (!scsi_device_online(sdev) ||
- rtn == FAST_IO_FAIL ||
- !scsi_eh_tur(bdr_scmd)) {
- list_for_each_entry_safe(scmd, next,
- work_q, eh_entry) {
- if (scmd->device == sdev &&
- scsi_eh_action(scmd, rtn) != FAILED)
- scsi_eh_finish_cmd(scmd,
- done_q);
- }
- }
- } else {
- SCSI_LOG_ERROR_RECOVERY(3,
- sdev_printk(KERN_INFO, sdev,
- "%s: BDR failed\n", current->comm));
+ if (scsi_eh_sdev_reset(bdr_scmd, work_q, done_q)) {
+ scsi_device_put(sdev);
+ break;
}
}
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 6/9] scsi: scsi_error: Add flags to mark error handle steps has done
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (4 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 5/9] scsi: scsi_error: Add helper scsi_eh_sdev_reset to do lun reset JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 7/9] scsi: scsi_error: Add helper to handle scsi device's error command list JiangJianJun
` (2 subsequent siblings)
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
LUN based error handle would mainly do three steps to recover
commands which are check sense, start unit, and reset lun. It might
fallback to host based error handler which would do these steps
too.
Add some flags to mark these steps are done to avoid repeating
these steps.
The flags should be cleared when LUN based error handler is waked up
or when host based error handler finished, and set when fallback to
host based error handle.
scsi_eh_get_sense, scsi_eh_stu, scsi_eh_bus_device_reset would check
these flags before actually action.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 33 +++++++++++++++++++++++++++++++++
include/scsi/scsi_device.h | 18 ++++++++++++++++++
2 files changed, 51 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index f58aad351463..239f8231c3ff 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -57,10 +57,36 @@
#define BUS_RESET_SETTLE_TIME (10)
#define HOST_RESET_SETTLE_TIME (10)
+#define sdev_flags_done(flag) \
+static inline int sdev_##flag(struct scsi_device *sdev) \
+{ \
+ struct scsi_device_eh *eh = sdev->eh; \
+ if (!eh) \
+ return 0; \
+ return eh->flag; \
+}
+
static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *,
struct scsi_cmnd *);
+sdev_flags_done(get_sense_done);
+sdev_flags_done(stu_done);
+sdev_flags_done(reset_done);
+
+static inline void shost_clear_eh_done(struct Scsi_Host *shost)
+{
+ struct scsi_device *sdev;
+
+ shost_for_each_device(sdev, shost) {
+ if (!sdev->eh)
+ continue;
+ sdev->eh->get_sense_done = 0;
+ sdev->eh->stu_done = 0;
+ sdev->eh->reset_done = 0;
+ }
+}
+
void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy)
{
lockdep_assert_held(shost->host_lock);
@@ -1397,6 +1423,8 @@ int scsi_eh_get_sense(struct list_head *work_q,
current->comm));
break;
}
+ if (sdev_get_sense_done(scmd->device))
+ continue;
if (!scsi_status_is_check_condition(scmd->result))
/*
* don't request sense if there's no check condition
@@ -1610,6 +1638,8 @@ static int scsi_eh_stu(struct Scsi_Host *shost,
scsi_device_put(sdev);
break;
}
+ if (sdev_stu_done(sdev))
+ continue;
stu_scmd = NULL;
list_for_each_entry(scmd, work_q, eh_entry)
if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
@@ -1693,6 +1723,8 @@ static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
bdr_scmd = scmd;
break;
}
+ if (sdev_reset_done(sdev))
+ continue;
if (!bdr_scmd)
continue;
@@ -2357,6 +2389,7 @@ static void scsi_unjam_host(struct Scsi_Host *shost)
if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
+ shost_clear_eh_done(shost);
spin_lock_irqsave(shost->host_lock, flags);
if (shost->eh_deadline != -1)
shost->last_reset = 0;
diff --git a/include/scsi/scsi_device.h b/include/scsi/scsi_device.h
index 9d42858035ed..9fc052e48a3b 100644
--- a/include/scsi/scsi_device.h
+++ b/include/scsi/scsi_device.h
@@ -103,6 +103,24 @@ struct scsi_vpd {
struct scsi_device;
struct scsi_device_eh {
+ /*
+ * LUN rebased error handle would mainly do three
+ * steps to recovery commands which are
+ * check sense
+ * start unit
+ * reset lun
+ * While we would fallback to host based error handler which would
+ * do these steps too. Add flags to mark thes steps are done to
+ * avoid repeating these steps.
+ *
+ * The flags should be cleared when LUN based error handler is
+ * wakedup or when host based error handler finished, set when
+ * fallback to host based error handle.
+ */
+ unsigned get_sense_done:1;
+ unsigned stu_done:1;
+ unsigned reset_done:1;
+
/*
* add scsi command to error handler so it would be handuled by
* driver's error handle strategy
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 7/9] scsi: scsi_error: Add helper to handle scsi device's error command list
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (5 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 6/9] scsi: scsi_error: Add flags to mark error handle steps has done JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
2025-09-14 10:41 ` [RFC PATCH v4 9/9] scsi: scsi_debug: Add params for configuring the " JiangJianJun
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add helper scsi_sdev_eh() to handle scsi device's error command list,
it would perform some steps which can be done with LUN's IO blocked,
including check sense, start unit and reset lun.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 37 +++++++++++++++++++++++++++++++++++++
include/scsi/scsi_eh.h | 2 ++
2 files changed, 39 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 239f8231c3ff..1f2b3deace32 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2486,6 +2486,43 @@ int scsi_error_handler(void *data)
return 0;
}
+/*
+ * Single LUN error handle
+ *
+ * @work_q: list of scsi commands need to recovery
+ * @done_q: list of scsi commands handled
+ *
+ * return: return 1 if all commands in work_q is recoveryed, else 0 is returned
+ */
+int scsi_sdev_eh(struct scsi_device *sdev,
+ struct list_head *work_q,
+ struct list_head *done_q)
+{
+ int ret = 0;
+ struct scsi_cmnd *scmd;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: checking sense\n", current->comm));
+ ret = scsi_eh_get_sense(work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: start unit\n", current->comm));
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+ ret = scsi_eh_sdev_stu(scmd, work_q, done_q);
+ if (ret)
+ return ret;
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: reset LUN\n", current->comm));
+ scmd = list_first_entry(work_q, struct scsi_cmnd, eh_entry);
+ ret = scsi_eh_sdev_reset(scmd, work_q, done_q);
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(scsi_sdev_eh);
+
/**
* scsi_report_bus_reset() - report bus reset observed
*
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index 1ae08e81339f..5ce791063baf 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -18,6 +18,8 @@ extern int scsi_block_when_processing_errors(struct scsi_device *);
extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
struct scsi_sense_hdr *sshdr);
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
+extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
+ struct list_head *doneq);
static inline bool scsi_sense_is_deferred(const struct scsi_sense_hdr *sshdr)
{
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (6 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 7/9] scsi: scsi_error: Add helper to handle scsi device's error command list JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
2025-09-14 11:59 ` kernel test robot
2025-09-14 10:41 ` [RFC PATCH v4 9/9] scsi: scsi_debug: Add params for configuring the " JiangJianJun
8 siblings, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
From: Wenchao Hao <haowenchao2@huawei.com>
Add a general LUN based error handler which can be used by drivers
directly. This error handler implements an scsi_device_eh, when handling
error commands, it would call helper function scsi_sdev_eh() added before
to try recover error commands.
The behavior if scsi_sdev_eh() can not recover all error commands
depends on fallback flag, which is initialized when scsi_device is
allocated. If fallback is set, it would fallback to further error
recover strategy like old host based error handle; else it would
mark this scsi device offline and flush all error commands. Add a
flag for controlling rollback in scsi_host_template.
Add interface sdev_setup_eh/sdev_clear_eh in scsi_host_template, used
for setup/clear LUN based error handler. Drivers can implements them
custom, or use inner implements:
scsi_device_setup_eh/scsi_device_clear_eh.
Signed-off-by: Wenchao Hao <haowenchao2@huawei.com>
Co-developed-by: JiangJianJun <jiangjianjun3@huawei.com>
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_error.c | 176 ++++++++++++++++++++++++++++++++++++++
drivers/scsi/scsi_scan.c | 7 ++
drivers/scsi/scsi_sysfs.c | 2 +
include/scsi/scsi_eh.h | 2 +
include/scsi/scsi_host.h | 16 ++++
5 files changed, 203 insertions(+)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index 1f2b3deace32..c1b4cd10216b 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -2736,3 +2736,179 @@ bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len,
}
}
EXPORT_SYMBOL(scsi_get_sense_info_fld);
+
+static inline void *scsi_eh_device_priv(struct scsi_device_eh *eh)
+{
+ return eh + 1;
+}
+
+struct scsi_lun_eh {
+ spinlock_t eh_lock;
+ unsigned int eh_num;
+ struct list_head eh_cmd_q;
+ struct scsi_device *sdev;
+ struct work_struct eh_handle_work;
+ unsigned int fallback:1;
+};
+
+/*
+ * error handle strategy based on LUN, following steps
+ * is applied to recovery error commands in list:
+ * check sense data
+ * send start unit
+ * reset lun
+ * if there are still error commands, it would fallback to
+ * host based error handler for further recovery.
+ */
+static void sdev_eh_work(struct work_struct *work)
+{
+ unsigned long flags;
+ struct scsi_lun_eh *luneh =
+ container_of(work, struct scsi_lun_eh, eh_handle_work);
+ struct scsi_device *sdev = luneh->sdev;
+ struct scsi_device_eh *eh = sdev->eh;
+ struct Scsi_Host *shost = sdev->host;
+ struct scsi_cmnd *scmd, *next;
+ LIST_HEAD(eh_work_q);
+ LIST_HEAD(eh_done_q);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ list_splice_init(&luneh->eh_cmd_q, &eh_work_q);
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ if (scsi_sdev_eh(sdev, &eh_work_q, &eh_done_q))
+ goto out_flush_done;
+
+ if (!luneh->fallback) {
+ list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry)
+ scsi_eh_finish_cmd(scmd, &eh_done_q);
+
+ sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: Device offlined - not ready after error recovery\n",
+ current->comm);
+
+ mutex_lock(&sdev->state_mutex);
+ scsi_device_set_state(sdev, SDEV_OFFLINE);
+ mutex_unlock(&sdev->state_mutex);
+
+ goto out_flush_done;
+ }
+
+ /*
+ * fallback to host based error handler
+ */
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh fallback to further recovery\n", current->comm));
+ list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry) {
+ list_del_init(&scmd->eh_entry);
+ scsi_eh_scmd_add_shost(scmd);
+ }
+
+ eh->get_sense_done = 1;
+ eh->stu_done = 1;
+ eh->reset_done = 1;
+
+out_flush_done:
+ scsi_eh_flush_done_q(&eh_done_q);
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ luneh->eh_num = 0;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+}
+static void sdev_eh_add_cmnd(struct scsi_cmnd *scmd)
+{
+ unsigned long flags;
+ struct scsi_lun_eh *luneh;
+ struct scsi_device *sdev = scmd->device;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ list_add_tail(&scmd->eh_entry, &luneh->eh_cmd_q);
+ luneh->eh_num++;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+}
+static bool sdev_eh_is_busy(struct scsi_device *sdev)
+{
+ int ret = 0;
+ unsigned long flags;
+ struct scsi_lun_eh *luneh;
+
+ if (!sdev->eh)
+ return false;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ ret = luneh->eh_num;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ return ret != 0;
+}
+static void sdev_eh_wakeup(struct scsi_device *sdev)
+{
+ unsigned long flags;
+ unsigned int nr_error;
+ unsigned int nr_busy;
+ struct scsi_lun_eh *luneh;
+
+ luneh = scsi_eh_device_priv(sdev->eh);
+
+ spin_lock_irqsave(&luneh->eh_lock, flags);
+ nr_error = luneh->eh_num;
+ spin_unlock_irqrestore(&luneh->eh_lock, flags);
+
+ nr_busy = scsi_device_busy(sdev);
+
+ if (!nr_error || nr_busy != nr_error) {
+ SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: do not wake up, busy/error: %d/%d\n",
+ current->comm, nr_busy, nr_error));
+ return;
+ }
+
+ SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
+ "%s:luneh: waking up, busy/error: %d/%d\n",
+ current->comm, nr_busy, nr_error));
+
+ schedule_work(&luneh->eh_handle_work);
+}
+
+/*
+ * This is default implement of Scsi_Host.sdev_setup_eh.
+ */
+int scsi_device_setup_eh(struct scsi_device *sdev)
+{
+ struct scsi_device_eh *eh;
+ struct scsi_lun_eh *luneh;
+
+ eh = kzalloc(sizeof(struct scsi_device_eh) + sizeof(struct scsi_lun_eh),
+ GFP_KERNEL);
+ if (!eh)
+ return -ENOMEM;
+ luneh = scsi_eh_device_priv(eh);
+
+ eh->add_cmnd = sdev_eh_add_cmnd;
+ eh->is_busy = sdev_eh_is_busy;
+ eh->wakeup = sdev_eh_wakeup;
+
+ luneh->fallback = sdev->host->hostt->sdev_eh_fallback;
+ luneh->sdev = sdev;
+ spin_lock_init(&luneh->eh_lock);
+ INIT_LIST_HEAD(&luneh->eh_cmd_q);
+ INIT_WORK(&luneh->eh_handle_work, sdev_eh_work);
+
+ sdev->eh = eh;
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(scsi_device_setup_eh);
+
+/*
+ * This is default implement of Scsi_Host.sdev_clear_eh.
+ */
+void scsi_device_clear_eh(struct scsi_device *sdev)
+{
+ kfree(sdev->eh);
+ sdev->eh = NULL;
+}
+EXPORT_SYMBOL_GPL(scsi_device_clear_eh);
diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 3c6e089e80c3..1c5cb77dfb22 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -377,9 +377,16 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
goto out_device_destroy;
}
}
+ if (shost->hostt->sdev_setup_eh) {
+ ret = shost->hostt->sdev_setup_eh(sdev);
+ if (ret)
+ goto out_device_eh;
+ }
return sdev;
+out_device_eh:
+ shost->hostt->sdev_destroy(sdev);
out_device_destroy:
__scsi_remove_device(sdev);
out:
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
index 15ba493d2138..76e788450345 100644
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -1513,6 +1513,8 @@ void __scsi_remove_device(struct scsi_device *sdev)
kref_put(&sdev->host->tagset_refcnt, scsi_mq_free_tags);
cancel_work_sync(&sdev->requeue_work);
+ if (sdev->host->hostt->sdev_clear_eh)
+ sdev->host->hostt->sdev_clear_eh(sdev);
if (sdev->host->hostt->sdev_destroy)
sdev->host->hostt->sdev_destroy(sdev);
transport_destroy_device(dev);
diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
index 5ce791063baf..d8e4475ff004 100644
--- a/include/scsi/scsi_eh.h
+++ b/include/scsi/scsi_eh.h
@@ -20,6 +20,8 @@ extern bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
extern enum scsi_disposition scsi_check_sense(struct scsi_cmnd *);
extern int scsi_sdev_eh(struct scsi_device *sdev, struct list_head *workq,
struct list_head *doneq);
+extern int scsi_device_setup_eh(struct scsi_device *sdev);
+extern void scsi_device_clear_eh(struct scsi_device *sdev);
static inline bool scsi_sense_is_deferred(const struct scsi_sense_hdr *sshdr)
{
diff --git a/include/scsi/scsi_host.h b/include/scsi/scsi_host.h
index 46f57fe78505..9cc34bfff3f7 100644
--- a/include/scsi/scsi_host.h
+++ b/include/scsi/scsi_host.h
@@ -225,6 +225,19 @@ struct scsi_host_template {
*/
void (* sdev_destroy)(struct scsi_device *);
+ /*
+ * Setup or clear error handler field scsi_device.eh.
+ * This error handler is working on designated device, it will only
+ * operate the designated device, do not affect other devices.
+ * If not set, error handle will fallback.
+ * LLDD can use custom error handler, or use inner defined:
+ * scsi_device_setup_eh/scsi_device_clear_eh
+ *
+ * Status: OPTIONAL
+ */
+ int (*sdev_setup_eh)(struct scsi_device *sdev);
+ void (*sdev_clear_eh)(struct scsi_device *sdev);
+
/*
* Before the mid layer attempts to scan for a new device attached
* to a target where no target currently exists, it will call this
@@ -468,6 +481,9 @@ struct scsi_host_template {
/* The queuecommand callback may block. See also BLK_MQ_F_BLOCKING. */
unsigned queuecommand_may_block:1;
+ /* The error handle of scsi_device will fallback when failed. */
+ unsigned sdev_eh_fallback:1;
+
/*
* Countdown for host blocking with no commands outstanding.
*/
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread* Re: [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler
2025-09-14 10:41 ` [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
@ 2025-09-14 11:59 ` kernel test robot
0 siblings, 0 replies; 38+ messages in thread
From: kernel test robot @ 2025-09-14 11:59 UTC (permalink / raw)
To: JiangJianJun; +Cc: oe-kbuild-all
Hi JiangJianJun,
[This is a private test report for your RFC patch.]
kernel test robot noticed the following build warnings:
[auto build test WARNING on jejb-scsi/for-next]
[also build test WARNING on mkp-scsi/for-next linus/master v6.17-rc5 next-20250912]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/JiangJianJun/scsi-scsi_error-Define-framework-for-LUN-based-error-handle/20250914-181340
base: https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next
patch link: https://lore.kernel.org/r/20250914104145.2239901-9-jiangjianjun3%40huawei.com
patch subject: [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler
config: arc-randconfig-002-20250914 (https://download.01.org/0day-ci/archive/20250914/202509141933.GikctXJJ-lkp@intel.com/config)
compiler: arc-linux-gcc (GCC) 10.5.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250914/202509141933.GikctXJJ-lkp@intel.com/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202509141933.GikctXJJ-lkp@intel.com/
All warnings (new ones prefixed by >>):
drivers/scsi/scsi_error.c: In function 'sdev_eh_work':
>> drivers/scsi/scsi_error.c:2770:20: warning: unused variable 'shost' [-Wunused-variable]
2770 | struct Scsi_Host *shost = sdev->host;
| ^~~~~
vim +/shost +2770 drivers/scsi/scsi_error.c
2753
2754 /*
2755 * error handle strategy based on LUN, following steps
2756 * is applied to recovery error commands in list:
2757 * check sense data
2758 * send start unit
2759 * reset lun
2760 * if there are still error commands, it would fallback to
2761 * host based error handler for further recovery.
2762 */
2763 static void sdev_eh_work(struct work_struct *work)
2764 {
2765 unsigned long flags;
2766 struct scsi_lun_eh *luneh =
2767 container_of(work, struct scsi_lun_eh, eh_handle_work);
2768 struct scsi_device *sdev = luneh->sdev;
2769 struct scsi_device_eh *eh = sdev->eh;
> 2770 struct Scsi_Host *shost = sdev->host;
2771 struct scsi_cmnd *scmd, *next;
2772 LIST_HEAD(eh_work_q);
2773 LIST_HEAD(eh_done_q);
2774
2775 spin_lock_irqsave(&luneh->eh_lock, flags);
2776 list_splice_init(&luneh->eh_cmd_q, &eh_work_q);
2777 spin_unlock_irqrestore(&luneh->eh_lock, flags);
2778
2779 if (scsi_sdev_eh(sdev, &eh_work_q, &eh_done_q))
2780 goto out_flush_done;
2781
2782 if (!luneh->fallback) {
2783 list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry)
2784 scsi_eh_finish_cmd(scmd, &eh_done_q);
2785
2786 sdev_printk(KERN_INFO, sdev,
2787 "%s:luneh: Device offlined - not ready after error recovery\n",
2788 current->comm);
2789
2790 mutex_lock(&sdev->state_mutex);
2791 scsi_device_set_state(sdev, SDEV_OFFLINE);
2792 mutex_unlock(&sdev->state_mutex);
2793
2794 goto out_flush_done;
2795 }
2796
2797 /*
2798 * fallback to host based error handler
2799 */
2800 SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
2801 "%s:luneh fallback to further recovery\n", current->comm));
2802 list_for_each_entry_safe(scmd, next, &eh_work_q, eh_entry) {
2803 list_del_init(&scmd->eh_entry);
2804 scsi_eh_scmd_add_shost(scmd);
2805 }
2806
2807 eh->get_sense_done = 1;
2808 eh->stu_done = 1;
2809 eh->reset_done = 1;
2810
2811 out_flush_done:
2812 scsi_eh_flush_done_q(&eh_done_q);
2813 spin_lock_irqsave(&luneh->eh_lock, flags);
2814 luneh->eh_num = 0;
2815 spin_unlock_irqrestore(&luneh->eh_lock, flags);
2816 }
2817 static void sdev_eh_add_cmnd(struct scsi_cmnd *scmd)
2818 {
2819 unsigned long flags;
2820 struct scsi_lun_eh *luneh;
2821 struct scsi_device *sdev = scmd->device;
2822
2823 luneh = scsi_eh_device_priv(sdev->eh);
2824
2825 spin_lock_irqsave(&luneh->eh_lock, flags);
2826 list_add_tail(&scmd->eh_entry, &luneh->eh_cmd_q);
2827 luneh->eh_num++;
2828 spin_unlock_irqrestore(&luneh->eh_lock, flags);
2829 }
2830 static bool sdev_eh_is_busy(struct scsi_device *sdev)
2831 {
2832 int ret = 0;
2833 unsigned long flags;
2834 struct scsi_lun_eh *luneh;
2835
2836 if (!sdev->eh)
2837 return false;
2838
2839 luneh = scsi_eh_device_priv(sdev->eh);
2840
2841 spin_lock_irqsave(&luneh->eh_lock, flags);
2842 ret = luneh->eh_num;
2843 spin_unlock_irqrestore(&luneh->eh_lock, flags);
2844
2845 return ret != 0;
2846 }
2847 static void sdev_eh_wakeup(struct scsi_device *sdev)
2848 {
2849 unsigned long flags;
2850 unsigned int nr_error;
2851 unsigned int nr_busy;
2852 struct scsi_lun_eh *luneh;
2853
2854 luneh = scsi_eh_device_priv(sdev->eh);
2855
2856 spin_lock_irqsave(&luneh->eh_lock, flags);
2857 nr_error = luneh->eh_num;
2858 spin_unlock_irqrestore(&luneh->eh_lock, flags);
2859
2860 nr_busy = scsi_device_busy(sdev);
2861
2862 if (!nr_error || nr_busy != nr_error) {
2863 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
2864 "%s:luneh: do not wake up, busy/error: %d/%d\n",
2865 current->comm, nr_busy, nr_error));
2866 return;
2867 }
2868
2869 SCSI_LOG_ERROR_RECOVERY(2, sdev_printk(KERN_INFO, sdev,
2870 "%s:luneh: waking up, busy/error: %d/%d\n",
2871 current->comm, nr_busy, nr_error));
2872
2873 schedule_work(&luneh->eh_handle_work);
2874 }
2875
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
^ permalink raw reply [flat|nested] 38+ messages in thread
* [RFC PATCH v4 9/9] scsi: scsi_debug: Add params for configuring the error handler
2025-09-14 10:41 ` [RFC PATCH v4 0/9] " JiangJianJun
` (7 preceding siblings ...)
2025-09-14 10:41 ` [RFC PATCH v4 8/9] scsi: scsi_error: Add a general LUN based error handler JiangJianJun
@ 2025-09-14 10:41 ` JiangJianJun
8 siblings, 0 replies; 38+ messages in thread
From: JiangJianJun @ 2025-09-14 10:41 UTC (permalink / raw)
To: James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, dlemoal, hewenliang4, yangyun50, wuyifeng10,
yangxingui
Add a new module parameter to configure error handlers based on
LUN and toggle the enable/disable fallback functionality.
Signed-off-by: JiangJianJun <jiangjianjun3@huawei.com>
---
drivers/scsi/scsi_debug.c | 15 ++++++++++++++-
1 file changed, 14 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
index 353cb60e1abe..f221ad31b44d 100644
--- a/drivers/scsi/scsi_debug.c
+++ b/drivers/scsi/scsi_debug.c
@@ -961,6 +961,8 @@ static bool write_since_sync;
static bool sdebug_statistics = DEF_STATISTICS;
static bool sdebug_wp;
static bool sdebug_allow_restart;
+static bool sdebug_lun_eh;
+static bool sdebug_lun_eh_fallback;
static enum {
BLK_ZONED_NONE = 0,
BLK_ZONED_HA = 1,
@@ -7385,6 +7387,8 @@ module_param_named(zone_max_open, sdeb_zbc_max_open, int, S_IRUGO);
module_param_named(zone_nr_conv, sdeb_zbc_nr_conv, int, S_IRUGO);
module_param_named(zone_size_mb, sdeb_zbc_zone_size_mb, int, S_IRUGO);
module_param_named(allow_restart, sdebug_allow_restart, bool, S_IRUGO | S_IWUSR);
+module_param_named(lun_eh, sdebug_lun_eh, bool, 0444);
+module_param_named(lun_eh_fallback, sdebug_lun_eh_fallback, bool, 0444);
MODULE_AUTHOR("Eric Youngdale + Douglas Gilbert");
MODULE_DESCRIPTION("SCSI debug adapter driver");
@@ -7464,6 +7468,8 @@ MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones; [0] for no limit
MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones (def=1)");
MODULE_PARM_DESC(zone_size_mb, "Zone size in MiB (def=auto)");
MODULE_PARM_DESC(allow_restart, "Set scsi_device's allow_restart flag(def=0)");
+MODULE_PARM_DESC(lun_eh, "LUN based error handle (def=0)");
+MODULE_PARM_DESC(lun_eh_fallback, "Fallback to further recovery if LUN recovery failed (def=0)");
#define SDEBUG_INFO_LEN 256
static char sdebug_info[SDEBUG_INFO_LEN];
@@ -8450,6 +8456,7 @@ static struct attribute *sdebug_drv_attrs[] = {
ATTRIBUTE_GROUPS(sdebug_drv);
static struct device *pseudo_primary;
+static struct scsi_host_template sdebug_driver_template;
static int __init scsi_debug_init(void)
{
@@ -8458,6 +8465,12 @@ static int __init scsi_debug_init(void)
int k, ret, hosts_to_add;
int idx = -1;
+ if (sdebug_lun_eh) {
+ sdebug_driver_template.sdev_setup_eh = scsi_device_setup_eh;
+ sdebug_driver_template.sdev_clear_eh = scsi_device_clear_eh;
+ sdebug_driver_template.sdev_eh_fallback = sdebug_lun_eh_fallback;
+ }
+
if (sdebug_ndelay >= 1000 * 1000 * 1000) {
pr_warn("ndelay must be less than 1 second, ignored\n");
sdebug_ndelay = 0;
@@ -9435,7 +9448,7 @@ static int sdebug_init_cmd_priv(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
return 0;
}
-static const struct scsi_host_template sdebug_driver_template = {
+static struct scsi_host_template sdebug_driver_template = {
.show_info = scsi_debug_show_info,
.write_info = scsi_debug_write_info,
.proc_name = sdebug_proc_name,
--
2.33.0
^ permalink raw reply related [flat|nested] 38+ messages in thread
* Re: [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-08-16 11:24 [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism JiangJianJun
` (15 preceding siblings ...)
2025-08-20 12:36 ` Hannes Reinecke
@ 2025-08-22 7:39 ` Damien Le Moal
2025-09-02 5:30 ` JiangJianJun
16 siblings, 1 reply; 38+ messages in thread
From: Damien Le Moal @ 2025-08-22 7:39 UTC (permalink / raw)
To: JiangJianJun, James.Bottomley, martin.petersen, linux-scsi
Cc: linux-kernel, hare, bvanassche, michael.christie, hch,
haowenchao22, john.g.garry, hewenliang4, yangyun50, wuyifeng10,
wubo40, yangxingui
On 8/16/25 20:24, JiangJianJun wrote:
> It's unbearable for systems with large scale scsi devices share HBAs to
> block all devices' IOs when handle error commands, we need a new error
> handle mechanism to address this issue.
>
> I consulted about this issue a year ago, the discuss link can be found in
> refenence. Hannes replied about why we have to block the SCSI host
> then perform error recovery kindly. I think it's unnecessary to block
> SCSI host for all drivers and can try a small level recovery(LUN based for
> example) first to avoid block the SCSI host.
>
> The new error handle mechanism introduced in this patchset has been
> developed and tested with out self developed hardware since one year
> ago, now we want this mechanism can be used by more drivers.
>
> Drivers can decide if using the new error handle mechanism and how to
> handle error commands when scsi_device are scanned,the new mechanism
> makes SCSI error handle more flexible.
Barely half of your emails have made it through for me and they landed in my
spam folder. So please check your email setup.
Also, was this all tested with libata and libsas attached devices as well ?
They all depend on scsi EH.
--
Damien Le Moal
Western Digital Research
^ permalink raw reply [flat|nested] 38+ messages in thread* [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-08-22 7:39 ` [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism Damien Le Moal
@ 2025-09-02 5:30 ` JiangJianJun
2025-09-02 5:08 ` Damien Le Moal
0 siblings, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-09-02 5:30 UTC (permalink / raw)
To: dlemoal, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
>Barely half of your emails have made it through for me and they landed in my
>spam folder. So please check your email setup.
I also find it strange, but my colleague can receive it. Maybe i reset email
and send again?
>Also, was this all tested with libata and libsas attached devices as well ?
>They all depend on scsi EH.
There is currently no tool available for injecting faults into hard drives,
but we have implemented this solution in our company's products. So i just
test with scsi_debug.
^ permalink raw reply [flat|nested] 38+ messages in thread
* Re: [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-09-02 5:30 ` JiangJianJun
@ 2025-09-02 5:08 ` Damien Le Moal
2025-09-02 6:03 ` JiangJianJun
0 siblings, 1 reply; 38+ messages in thread
From: Damien Le Moal @ 2025-09-02 5:08 UTC (permalink / raw)
To: JiangJianJun, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
On 9/2/25 2:30 PM, JiangJianJun wrote:
>> Barely half of your emails have made it through for me and they landed in my
>> spam folder. So please check your email setup.
>
> I also find it strange, but my colleague can receive it. Maybe i reset email
> and send again?
>
>> Also, was this all tested with libata and libsas attached devices as well ?
>> They all depend on scsi EH.
>
> There is currently no tool available for injecting faults into hard drives,
> but we have implemented this solution in our company's products. So i just
> test with scsi_debug.
Use write long command to "destroy" sectors. Then try to read them. That will
generate uncorrectable read errors.
See sg_write_long (sg3utils).
--
Damien Le Moal
Western Digital Research
^ permalink raw reply [flat|nested] 38+ messages in thread
* [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-09-02 5:08 ` Damien Le Moal
@ 2025-09-02 6:03 ` JiangJianJun
2025-09-02 5:30 ` Damien Le Moal
0 siblings, 1 reply; 38+ messages in thread
From: JiangJianJun @ 2025-09-02 6:03 UTC (permalink / raw)
To: dlemoal, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
>Use write long command to "destroy" sectors. Then try to read them. That will
>generate uncorrectable read errors.
There is a misunderstanding here, the condition that triggers this error
handler is when the device is slow or unresponsive, or fails to start;
bad blocks refer to data errors rather than faults.
^ permalink raw reply [flat|nested] 38+ messages in thread
* Re: [PATCH 00/14] scsi: scsi_error: Introduce new error handle mechanism
2025-09-02 6:03 ` JiangJianJun
@ 2025-09-02 5:30 ` Damien Le Moal
0 siblings, 0 replies; 38+ messages in thread
From: Damien Le Moal @ 2025-09-02 5:30 UTC (permalink / raw)
To: JiangJianJun, linux-scsi
Cc: linux-kernel, hewenliang4, yangyun50, wuyifeng10, yangxingui
On 9/2/25 3:03 PM, JiangJianJun wrote:
>> Use write long command to "destroy" sectors. Then try to read them. That will
>> generate uncorrectable read errors.
>
> There is a misunderstanding here, the condition that triggers this error
> handler is when the device is slow or unresponsive, or fails to start;
> bad blocks refer to data errors rather than faults.
You probably easilly can simulate an unresponsive device using either qemu or
tcmu-runner.
--
Damien Le Moal
Western Digital Research
^ permalink raw reply [flat|nested] 38+ messages in thread