scsi: libsas: set data_dir as DMA_NONE if libata mark qc as NODATA
Subsystems (2)
PATCH 1/1 scsi: libsas: set data_dir as DMA_NONE if libata mark qc as NODATA
We found that it will fail every time when set feature to SATA disk by "sdparm -s WCE=0 /dev/sde".
After checking protocol, we know that MODE SELECT is the SCSI command for setting WCE, and it do not exist in the SATA protocol. Therefore, this commands are encapsulated in the SET FEATURE command in SATA protocol. The difference is that the MODE SELECT command sent to SAS disk contains data and is sent through the DMA. But when send to SATA disk through SET FEATURE command, it does not contain data.
I think libsas was not thorough enough in dealing with the situation, at sas_ata_qc_issue(), task->dma_dir is inherited from qc->dma_dir(qc->dma_dir is also inherited from SCSI). However, in ATA driver, if qc->tf.protocol is set to ATA_PROT_NODATA, ata_sg_setup() will not invoked by ata_qc_issue(). It mean that ATA didn't use dma_dir and it's not reliable. So, if libsas still inherits from qc->dma_dir when there is no data need to be send. As a result, task with no data are mistakenly considered as carrying data and it will make LLDD report an error on IO completion.
So, When ATA driver mark tf->protocol as NODATA, dma_dir should be set as DMA_NONE. And we can see WCE is successfully disable for SATA disk then.
Euler:~ # sdparm -s WCE=0 /dev/sde
/dev/sde: ATA ST4000NM0035-1V4 TN03
Euler:~ # sdparm /dev/sde
/dev/sde: ATA ST4000NM0035-1V4 TN03
Read write error recovery mode page:
AWRE 1 [cha: n, def: 1]
ARRE 0 [cha: n, def: 0]
PER 0 [cha: n, def: 0]
Caching (SBC) mode page:
WCE 0 [cha: y, def: 0]
RCD 0 [cha: n, def: 0]Control mode page:
SWP 0 [cha: n, def: 0]
--- drivers/scsi/libsas/sas_ata.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/scsi/libsas/sas_ata.c +4 -1
@@ -209,7 +209,10 @@ static unsigned int sas_ata_qc_issue(struct ata_queued_cmd *qc) task->num_scatter = si; } - task->data_dir = qc->dma_dir; +if (qc->tf.protocol ==ATA_PROT_NODATA ) + task->data_dir =DMA_NONE ; +else + task->data_dir = qc->dma_dir; task->scatter = qc->sg; task->ata_task.retry_count =1 ; task->task_state_flags =SAS_TASK_STATE_PENDING ;
Signature · 1 line
2.7.4
Quoted 51 lines · Luo Jiaxing
We found that it will fail every time when set feature to SATA disk by "sdparm -s WCE=0
/dev/sde".After checking protocol, we know that MODE SELECT is the SCSI command for setting WCE, and it do not exist in the SATA protocol. Therefore, this commands are encapsulated in the SET FEATURE command in SATA protocol. The difference is that the MODE SELECT command sent to SAS disk contains data and is sent through the DMA. But when send to SATA disk through SET FEATURE command, it does not contain data.
I think libsas was not thorough enough in dealing with the situation, at
sas_ata_qc_issue(),task->dma_diris inherited fromqc->dma_dir(qc->dma_dir is also inherited from SCSI). However, in ATA driver, ifqc->tf.protocolis set toATA_PROT_NODATA,ata_sg_setup()will not invoked byata_qc_issue(). It mean that ATA didn't usedma_dirand it's not reliable. So, if libsas still inherits fromqc->dma_dirwhen there is no data need to be send. As a result, task with no data are mistakenly considered as carrying data and it will make LLDD report an error on IO completion.So, When ATA driver mark
tf->protocolas NODATA,dma_dirshould be set asDMA_NONE. And we can see WCE is successfully disable for SATA disk then.Euler:~ # sdparm -s WCE=0
/dev/sde/dev/sde: ATA ST4000NM0035-1V4 TN03 Euler:~ # sdparm /dev/sde /dev/sde: ATA ST4000NM0035-1V4 TN03 Read write error recovery mode page: AWRE 1 [cha: n, def: 1] ARRE 0 [cha: n, def: 0] PER 0 [cha: n, def: 0] Caching (SBC) mode page: WCE 0 [cha: y, def: 0] RCD 0 [cha: n, def: 0]Control mode page:
SWP 0 [cha: n, def: 0]Fixes fa1c1e8f1ece ("[SCSI] Add SATA support to libsas")Signed-off-by Luo Jiaxing <[email protected]>Reviewed-by John Garry <[email protected]>--- drivers/scsi/libsas/sas_ata.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)drivers/scsi/libsas/sas_ata.c +0
-0@@ -209,7 +209,10 @@ static unsigned int sas_ata_qc_issue(struct ata_queued_cmd *qc) task->num_scatter = si;
drivers/scsi/libsas/sas_ata.c static unsigned int sas_ata_qc_issue(struct ata_queued_cmd *qc) · line 218
Quoted 10 lines
drivers/scsi/libsas/sas_ata.c +4
-1} - task->data_dir = qc->dma_dir; +if (qc->tf.protocol ==ATA_PROT_NODATA ) + task->data_dir =DMA_NONE ; +else + task->data_dir = qc->dma_dir; task->scatter = qc->sg; task->ata_task.retry_count =1 ; task->task_state_flags =SAS_TASK_STATE_PENDING ;
Martin K. Petersen
AppliedQuoted 11 lines · Luo Jiaxing
We found that it will fail every time when set feature to SATA disk by "sdparm -s WCE=0
/dev/sde".After checking protocol, we know that MODE SELECT is the SCSI command for setting WCE, and it do not exist in the SATA protocol. Therefore, this commands are encapsulated in the SET FEATURE command in SATA protocol. The difference is that the MODE SELECT command sent to SAS disk contains data and is sent through the DMA. But when send to SATA disk through SET FEATURE command, it does not contain data.
[...]
Applied to 5.9/scsi-fixes, thanks!
[1/1] scsi: libsas: Set data_dir as DMA_NONE if libata marks qc as NODATA
https://git.kernel.org/mkp/scsi/c/53de092f47ff
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering