Fix system resume for SCSI devices
v2 cover Fix system resume for SCSI devices
Hi Martin,
This patch series fixes a particular type of system resume failure for SCSI devices. Please consider this patch series for the next merge window.
Thanks,
Bart.
Changes compared to v1:
- Renamed SCMD_RETRY_PASST_ON_UA into SCMD_RETRY_PASSTHROUGH.
Bart Van Assche (2):
scsi: core: Retry passthrough commands if SCMD_RETRY_PASSTHROUGH is
set
sd: Retry START STOP UNIT commands
drivers/scsi/scsi_error.c | 5 ++++- drivers/scsi/sd.c | 1 + include/scsi/scsi_cmnd.h | 5 ++++- 3 files changed, 9 insertions(+), 2 deletions(-)
PATCH v2 1/2 scsi: core: Retry passthrough commands if SCMD_RETRY_PASSTHROUGH is set
The SCSI core does not retry passthrough commands even if the SCSI device reports a retryable unit attention condition. Support retrying in this case by introducing the SCMD_RETRY_PASSTHROUGH flag.
--- drivers/scsi/scsi_error.c | 5 ++++- include/scsi/scsi_cmnd.h | 5 ++++- 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/scsi_error.c +4 -1
@@ -1855,7 +1855,10 @@ bool scsi_noretry_cmd(struct scsi_cmnd *scmd)* assume caller has checked sense and determined * the check condition was retryable. */ -if (req->cmd_flags &REQ_FAILFAST_DEV ||blk_rq_is_passthrough (req)) +if (req->cmd_flags &REQ_FAILFAST_DEV ) +return true; +if (blk_rq_is_passthrough (req) && + !(scmd->flags &SCMD_RETRY_PASSTHROUGH ))return true;return false;
include/scsi/scsi_cmnd.h +4 -1
@@ -58,8 +58,11 @@ struct scsi_pointer {*/ #define SCMD_FORCE_EH_SUCCESS (1 << 3)#define SCMD_FAIL_IF_RECOVERING (1 << 4) +/* If set, retry a passthrough command in case of a unit attention. */ +#define SCMD_RETRY_PASSTHROUGH (1 << 5)/* flags preserved across unprep / reprep */ -#define SCMD_PRESERVED_FLAGS (SCMD_INITIALIZED | SCMD_FAIL_IF_RECOVERING) +#define SCMD_PRESERVED_FLAGS \ + (SCMD_INITIALIZED | SCMD_FAIL_IF_RECOVERING | SCMD_RETRY_PASSTHROUGH)/* for scmd->state */ #define SCMD_STATE_COMPLETE 0
Re: [PATCH v2 1/2] scsi: core: Retry passthrough commands if SCMD_RETRY_PASSTHROUGH is set
Quoted 18 lines · Bart Van Assche
The SCSI core does not retry passthrough commands even if the SCSI device reports a retryable unit attention condition. Support retrying in this case by introducing the
SCMD_RETRY_PASSTHROUGHflag.Cc Damien Le Moal <[email protected]>Cc Mike Christie <[email protected]>Signed-off-by Bart Van Assche <[email protected]>--- drivers/scsi/scsi_error.c | 5 ++++- include/scsi/scsi_cmnd.h | 5 ++++- 2 files changed, 8 insertions(+), 2 deletions(-)drivers/scsi/scsi_error.c +0
-0@@ -1855,7 +1855,10 @@ bool scsi_noretry_cmd(struct scsi_cmnd *scmd)* assume caller has checked sense and determined
drivers/scsi/scsi_error.c bool scsi_noretry_cmd(struct scsi_cmnd *scmd) · line 1864
Quoted 10 lines
drivers/scsi/scsi_error.c +4
-1* the check condition was retryable. */ -if (req->cmd_flags &REQ_FAILFAST_DEV ||blk_rq_is_passthrough (req)) +if (req->cmd_flags &REQ_FAILFAST_DEV ) +return true; +if (blk_rq_is_passthrough (req) && + !(scmd->flags &SCMD_RETRY_PASSTHROUGH ))return true;return false;
return (req->cmd_flags & REQ_FAILFAST_DEV) || (blk_rq_is_passthrough(req) && !(scmd->flags & SCMD_RETRY_PASSTHROUGH));
maybe ? Not necessarily more readable though.
include/scsi/scsi_cmnd.h struct scsi_pointer { · line 61
Quoted 9 lines
include/scsi/scsi_cmnd.h +1
-0@@ -58,8 +58,11 @@ struct scsi_pointer {*/ #define SCMD_FORCE_EH_SUCCESS (1 << 3)#define SCMD_FAIL_IF_RECOVERING (1 << 4) +/* If set, retry a passthrough command in case of a unit attention. */
"...in case of error". There are other sense keys than unit attention and there is no code checking that this flag applies only to unit attention.
With this comment fixed:
Quoted 8 lines
include/scsi/scsi_cmnd.h +3
-1+#define SCMD_RETRY_PASSTHROUGH (1 << 5)/* flags preserved across unprep / reprep */ -#define SCMD_PRESERVED_FLAGS (SCMD_INITIALIZED | SCMD_FAIL_IF_RECOVERING) +#define SCMD_PRESERVED_FLAGS \ + (SCMD_INITIALIZED | SCMD_FAIL_IF_RECOVERING | SCMD_RETRY_PASSTHROUGH)/* for scmd->state */ #define SCMD_STATE_COMPLETE 0
Signature · 2 lines
Damien Le Moal Western Digital Research
Re: [PATCH v2 1/2] scsi: core: Retry passthrough commands if SCMD_RETRY_PASSTHROUGH is set
Quoted 13 lines · Damien Le Moal
drivers/scsi/scsi_error.c bool scsi_noretry_cmd(struct scsi_cmnd *scmd) · line 1864
Quoted 5 lines · Bart Van Assche
drivers/scsi/scsi_error.c +2
-0+if (blk_rq_is_passthrough (req) && + !(scmd->flags &SCMD_RETRY_PASSTHROUGH ))return true;return false;return (req->cmd_flags & REQ_FAILFAST_DEV) || (blk_rq_is_passthrough(req) && !(scmd->flags & SCMD_RETRY_PASSTHROUGH));maybe ? Not necessarily more readable though.
Quoted 12 lines
include/scsi/scsi_cmnd.h struct scsi_pointer { · line 61
Quoted 9 lines
include/scsi/scsi_cmnd.h +1
-0@@ -58,8 +58,11 @@ struct scsi_pointer {*/ #define SCMD_FORCE_EH_SUCCESS (1 << 3)#define SCMD_FAIL_IF_RECOVERING (1 << 4) +/* If set, retry a passthrough command in case of a unit attention. */ "...in case of error". There are other sense keys than unit attention and there is no code checking that this flag applies only to unit attention.
Hi Damien,
Thanks for the detailed feedback. Since Martin replied to this series that he will reconcile his changes with mine, I plan to wait until either Martin posts his patches or until Martin asks me to repost my patches.
Thanks,
Bart.
Re: [PATCH v2 1/2] scsi: core: Retry passthrough commands if SCMD_RETRY_PASSTHROUGH is set
In reply to Bart Van AsscheHi Bart!
Sorry this took so long. I have made several attempts at automatically retrying commands originating from within SCSI since the lack of retries is a general problem.
However, in the end I prefer the approach of describing the specific conditions we want handled using scsi_failure at each call site. Works well.
Quoted 3 lines
The SCSI core does not retry passthrough commands even if the SCSI device reports a retryable unit attention condition. Support retrying in this case by introducing the
SCMD_RETRY_PASSTHROUGHflag.
FWIW, I find the use of "passthrough" to describe commands issued by the SCSI layer quite confusing.
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering
PATCH v2 2/2 sd: Retry START STOP UNIT commands
In reply to Bart Van AsscheDuring system resume, sd_start_stop_device() submits a START STOP UNIT command to the SCSI device that is being resumed. That command is not retried in case of a unit attention and hence may fail. An example:
[16575.983359] sd 0:0:0:3: [sdd] Starting disk [16575.983693] sd 0:0:0:3: [sdd] Start/Stop Unit failed: Result: hostbyte=0x00 driverbyte=DRIVER_OK [16575.983712] sd 0:0:0:3: [sdd] Sense Key : 0x6 [16575.983730] sd 0:0:0:3: [sdd] ASC=0x29 ASCQ=0x0 [16575.983738] sd 0:0:0:3: PM: dpm_run_callback(): scsi_bus_resume+0x0/0xa0 returns -5 [16575.983783] sd 0:0:0:3: PM: failed to resume async: error -5
Make the SCSI core retry the START STOP UNIT command if a retryable error is encountered.
--- drivers/scsi/sd.c | 1 + 1 file changed, 1 insertion(+)
drivers/scsi/sd.c +1 -0
@@ -4093,6 +4093,7 @@ static int sd_start_stop_device(struct scsi_disk *sdkp, int start)const struct scsi_exec_args exec_args = { .sshdr = &sshdr, .req_flags =BLK_MQ_REQ_PM , + .scmd_flags =SCMD_RETRY_PASSTHROUGH , };struct scsi_device *sdp = sdkp->device;int res;
Re: [PATCH v2 2/2] sd: Retry START STOP UNIT commands
Quoted 11 lines · Bart Van Assche
During system resume,
sd_start_stop_device()submits a START STOP UNIT command to the SCSI device that is being resumed. That command is not retried in case of a unit attention and hence may fail. An example:[16575.983359] sd 0:0:0:3: [sdd] Starting disk [16575.983693] sd 0:0:0:3: [sdd] Start/Stop Unit failed: Result: hostbyte=0x00 driverbyte=DRIVER_OK [16575.983712] sd 0:0:0:3: [sdd] Sense Key : 0x6 [16575.983730] sd 0:0:0:3: [sdd] ASC=0x29 ASCQ=0x0 [16575.983738] sd 0:0:0:3: PM: dpm_run_callback(): scsi_bus_resume+0x0/0xa0 returns -5 [16575.983783] sd 0:0:0:3: PM: failed to resume async: error -5
Quoted 6 lines
Make the SCSI core retry the START STOP UNIT command if a retryable error is encountered.
Cc Damien Le Moal <[email protected]>Cc Mike Christie <[email protected]>Signed-off-by Bart Van Assche <[email protected]>
Looks OK to me.
Signature · 2 lines
Damien Le Moal Western Digital Research
Hi Bart!
Quoted 3 lines
This patch series fixes a particular type of system resume failure for SCSI devices. Please consider this patch series for the next merge window.
I have been working on several tricky regressions where the remedy is similar in nature to yours. I.e. retrying in some situations. I ended up using the scsi_failure infrastructure to handle the cases I care about but now wonder if a more generic approach is warranted. I'll contemplate a bit and try to reconcile my changes with yours.
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering