fix abort defect

ufshcd_abort_all froce abort all on-going command and the host

will automatically fill in the OCS field of the corresponding response with OCS_ABORTED based on different working modes.

SDB mode: aborts a command using UTRLCLR. Task Management response which means a Transfer Request was aborted.

For these two cases, set a flag to notify SCSI to requeue the command after receiving response with OCS_ABORTED.

drivers/ufs/core/ufshcd.c +7 -0

@@ -7561,6 +7551,20 @@ int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
  		goto out;
  	}
 
+	/*
+	 * When the host software receives a "FUNCTION COMPLETE", set flag
+	 * to requeue command after receive response with OCS_ABORTED
+	 * SDB mode: UTRLCLR Task Management response which means a Transfer
+	 *           Request was aborted.
+	 * MCQ mode: Host will post to CQ with OCS_ABORTED after SQ cleanup
+	 * This flag is set because ufshcd_abort_all forcibly aborts all

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 1 line · On 9/10/24 12:30 AM, [email protected] wrote:

ufshcd_abort_all froce abort all on-going command and the host

^^^^^ ^^^^^ ^^^^^^^ ^^^^
forcibly? aborts? commands? host
controller?

Quoted 2 lines

will automatically fill in the OCS field of the corresponding response with OCS_ABORTED based on different working modes.

                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

The host controller only sets the OCS field to OCS_ABORTED in MCQ mode
if the host controller successfully aborted the command. If the abort TMF is submitted to the UFS device, the OCS field won't be changed
into OCS_ABORTED. In SDB mode, the host controller does not modify the
OCS field either.

Quoted 2 lines

SDB mode: aborts a command using UTRLCLR. Task Management response which means a Transfer Request was aborted.

Hmm ... my understanding is that clearing a bit from UTRLCLR is only allowed *after* a command has been aborted and also that clearing a bit
from this register does not abort a command but only frees the resources
in the host controller associated with the command.

Quoted 2 lines

For these two cases, set a flag to notify SCSI to requeue the command after receiving response with OCS_ABORTED.

I think there is only one case when the SCSI core needs to be requested
to requeue a command, namely when the UFS driver decided to initiate the
abort (ufshcd_abort_all()).

Review int ufshcd_try_to_abort_task(struct ufs_hba

Quoted 1 line

(quoted diff) +0 -0

@@ -7561,6 +7551,20 @@ int ufshcd_try_to_abort_task(struct ufs_hba

*hba, int tag)

drivers/ufs/core/ufshcd.c int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)

Quoted 7 lines

drivers/ufs/core/ufshcd.c +4 -0

  goto out;
  }
 
+/*
+ * When the host software receives a "FUNCTION COMPLETE", set flag
+ * to requeue command after receive response with OCS_ABORTED
+ * SDB mode: UTRLCLR Task Management response which means a

Transfer

Review int ufshcd_try_to_abort_task(struct ufs_hba · line 7559

Quoted 2 lines

(quoted diff) +2 -0

+ *           Request was aborted.
+ * MCQ mode: Host will post to CQ with OCS_ABORTED after SQ

cleanup

drivers/ufs/core/ufshcd.c int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)

Quoted 4 lines

drivers/ufs/core/ufshcd.c +4 -0

+ * This flag is set because ufshcd_abort_all forcibly aborts all
+ * commands, and the host will automatically fill in the OCS field
+ * of the corresponding response with OCS_ABORTED.
+ * Therefore, upon receiving this response, it needs to be

requeued.

drivers/ufs/core/ufshcd.c int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag) · line 7570

Quoted 7 lines

drivers/ufs/core/ufshcd.c +4 -0

+ */
+if (!err)
+lrbp->abort_initiated_by_err = true;
+
  err = ufshcd_clear_cmd(hba, tag);
  if (err)
  dev_err(hba->dev, "%s: Failed clearing cmd at tag %d, err %d\n",

The above change is misplaced. ufshcd_try_to_abort_task() can be called
when the SCSI core decides to abort a command while abort_initiated_by_err must not be set in that case. Please move the above code block into ufshcd_abort_one().

Regarding the word "host" in the above comment block: the host is the
Android device. I think that in the above comment "host" should be changed into "host controller".

include/ufs/ufshcd.h struct ufs_pm_lvl_states { · line 173

Quoted 6 lines

include/ufs/ufshcd.h +0 -0

index 0fd2aebac728..15b357672ca5 100644
@@ -173,6 +173,8 @@ struct ufs_pm_lvl_states {
   * @crypto_key_slot: the key slot to use for inline crypto (-1 if

none)

include/ufs/ufshcd.h struct ufs_pm_lvl_states { · line 174

Quoted 1 line

include/ufs/ufshcd.h +0 -0

   * @data_unit_num: the data unit number for the first block for

inline crypto

include/ufs/ufshcd.h struct ufs_pm_lvl_states { · line 176

Quoted 2 lines

include/ufs/ufshcd.h +1 -0

   * @req_abort_skip: skip request abort task flag
+ * @abort_initiated_by_err: The flag is specifically used to

handle aborts

include/ufs/ufshcd.h struct ufs_pm_lvl_states { · line 177

Quoted 1 line

include/ufs/ufshcd.h +1 -0

+ *                          caused by errors due to host/device

communication

The "abort_initiated_by_err" name still seems confusing to me. Please make it more clear that this flag is only set if the UFS error handler
decides to abort a command. How about "abort_initiated_by_eh"?

Please also make the description of this member variable more clear.

Thanks,

Bart.

This statement is not quite accurate becasue in UFSHIC2.1, SDB mode specification already have OCS: ABORTED (0x6) define. And it is used in below UTRLCLR description:
'which means a Transfer Request was "aborted"' Therefore, the host controller should follow the specification and fill the OCS field with OCS: ABORTED. If not so, at what point does your host controller use the
OCS: ABORTED status?

Quoted 13 lines

Quoted 4 lines

+/*
+ * When the host software receives a "FUNCTION COMPLETE", set flag
+ * to requeue command after receive response with OCS_ABORTED
+ * SDB mode: UTRLCLR Task Management response which means a

Transfer

Quoted 2 lines

+ * Request was aborted.
+ * MCQ mode: Host will post to CQ with OCS_ABORTED after SQ

cleanup

Quoted 4 lines

+ * This flag is set because ufshcd_abort_all forcibly aborts all
+ * commands, and the host will automatically fill in the OCS field
+ * of the corresponding response with OCS_ABORTED.
+ * Therefore, upon receiving this response, it needs to be

requeued.

Quoted 13 lines

drivers/ufs/core/ufshcd.c int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag) · line 7570

Quoted 7 lines

+ */

drivers/ufs/core/ufshcd.c +3 -0

+if (!err)
+lrbp->abort_initiated_by_err = true;
+
   err = ufshcd_clear_cmd(hba, tag);
   if (err)
   dev_err(hba->dev, "%s: Failed clearing cmd at tag %d, err %d\n",

The above change is misplaced. ufshcd_try_to_abort_task() can be called
when the SCSI core decides to abort a command while abort_initiated_by_err must not be set in that case. Please move the above code block into ufshcd_abort_one().

But move to ufshcd_abort_one may have race condition, beacause we need set this flag before ufshcd_clear_cmd host controller fill OCS_ABORTED to response. I will add check ufshcd_eh_in_progress.

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 8 lines · Peter Wang (王信友)

This statement is not quite accurate becasue in UFSHIC2.1, SDB mode specification already have OCS: ABORTED (0x6) define. And it is used in below UTRLCLR description:
'which means a Transfer Request was "aborted"' Therefore, the host controller should follow the specification and fill the OCS field with OCS: ABORTED. If not so, at what point does your host controller use the
OCS: ABORTED status?

Hmm ... I have not been able to find any explanation in the UFSHCI 2.1
specification that says when the OCS status is set to aborted. Did I perhaps overlook something?

This is what I found in the UTRLCLR description: "The host software shall use this field only when a UTP Transfer Request is expected to not be completed, e.g., when the host software receives a “FUNCTION COMPLETE” Task Management response which means a Transfer Request was aborted." This does not mean that the host controller is expected to set the OCS status to "ABORTED". I will send an email to the JC-64 mailing list to request clarification.

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

+/*
+ * When the host software receives a "FUNCTION COMPLETE", set

flag

Quoted 8 lines

Quoted 8 lines

Quoted 2 lines

+ * to requeue command after receive response with OCS_ABORTED
+ * SDB mode: UTRLCLR Task Management response which means a

Transfer

Quoted 2 lines

+ * Request was aborted.
+ * MCQ mode: Host will post to CQ with OCS_ABORTED after SQ

cleanup

Quoted 2 lines

+ * This flag is set because ufshcd_abort_all forcibly aborts all
+ * commands, and the host will automatically fill in the OCS

field

Quoted 10 lines

Quoted 10 lines

Quoted 2 lines

+ * of the corresponding response with OCS_ABORTED.
+ * Therefore, upon receiving this response, it needs to be

requeued.

drivers/ufs/core/ufshcd.c int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)

Quoted 7 lines

+ */

drivers/ufs/core/ufshcd.c +3 -0

+if (!err)
+lrbp->abort_initiated_by_err = true;
+
   err = ufshcd_clear_cmd(hba, tag);
   if (err)
   dev_err(hba->dev, "%s: Failed clearing cmd at tag %d, err

%d\n",

Quoted 5 lines

Quoted 5 lines

The above change is misplaced. ufshcd_try_to_abort_task() can be called
when the SCSI core decides to abort a command while abort_initiated_by_err must not be set in that case. Please move

the

Quoted 5 lines

Quoted 1 line

above code block into ufshcd_abort_one().

But move to ufshcd_abort_one may have race condition, beacause we need set this flag before ufshcd_clear_cmd host controller fill OCS_ABORTED to response. I will add check ufshcd_eh_in_progress.

Calling ufshcd_clear_cmd() does not affect the OCS status as far as I know. Did I perhaps overlook something?

Thanks,

Bart.

in SDB mode:
ufshcd_utrl_clear set UTRLC, Mediatek host controller (may not all host controller) will post response with OCS ABORTED.

In both cases, we have an interrupt sent to the host, and there may be a race condition before we set this flag for requeue. So I need to set this flag before ufshcd_clear_cmd.

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 7 lines · Peter Wang (王信友)

in SDB mode:
ufshcd_utrl_clear set UTRLC, Mediatek host controller (may not all host controller) will post response with OCS ABORTED.

In both cases, we have an interrupt sent to the host, and there may be a race condition before we set this flag for requeue. So I need to set this flag before ufshcd_clear_cmd.

If a completion interrupt is sent to the host if a command has been cleared in SDB mode (I doubt this is what happens), I think that's a severe controller bug. A UFSHCI controller is not allowed to send a completion interrupt to the host if a command is cleared by writing into
the UTRLCLR register.

Thanks,

Bart.

Because the MediaTek UFS controller uses UTRLCLR to clear commands and fills OCS with ABORTED.

Regarding the specification of UTRCS:
This bit is set to '1' by the host controller upon one of the following:
Overall command Status (OCS) of the completed command is not equal to 'SUCCESS' even if its UTRD Interrupt bit set to '0'

So, MediaTek host controller will send interrupt in this case.

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 10 lines · Peter Wang (王信友)

Because the MediaTek UFS controller uses UTRLCLR to clear commands and fills OCS with ABORTED.

Regarding the specification of UTRCS:
This bit is set to '1' by the host controller upon one of the following:
Overall command Status (OCS) of the completed command is not equal to 'SUCCESS' even if its UTRD Interrupt bit set to '0'

So, MediaTek host controller will send interrupt in this case.

Hi Peter,

Thank you for having shared this information. Please consider introducing a quirk for ignoring completions triggered by clearing a command, e.g. as follows (there may be better approaches):
* In ufshcd_clear_cmd(), before a command is cleared, initialize the completion that will be used for waiting for the completion interrupt. After a command has been cleared, call wait_for_completion_timeout().
* In ufshcd_compl_one_cqe(), check whether the completion is the result of a command being cleared. If so, call complete() instead of executing the regular completion code.

Thanks,

Bart.

Basically, this patch currently only needs to handle requeueing for the error handler abort.
The approach for DBR mode and MCQ mode should be consistent. If receive an interrupt response (OCS:ABORTED or INVALID_OCS_VALUE), then set DID_REQUEUE. If there is no interrupt, it will also set SCSI DID_REQUEUE in ufshcd_err_handler through ufshcd_complete_requests
with force_compl = true.

The more problematic part is with MCQ mode. To imitate the DBR approach, we just need to set DID_REQUEUE upon receiving an interrupt. Everything else remains the same. This would make things simpler.

Moving forward, if we want to simplify things and we have also taken stock of the two or three scenarios where OCS: ABORTED occurs, do we even need a flag? Couldn't we just set DID_REQUEUE directly for OCS: ABORTED?
What do you think?

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 4 lines · Peter Wang (王信友)

Basically, this patch currently only needs to handle requeueing for the error handler abort.
The approach for DBR mode and MCQ mode should be consistent. If receive an interrupt response (OCS:ABORTED or

INVALID_OCS_VALUE),

Quoted 4 lines

then set DID_REQUEUE. If there is no interrupt, it will also set SCSI DID_REQUEUE in ufshcd_err_handler through ufshcd_complete_requests
with force_compl = true.

Reporting a completion for commands cleared by writing into the legacy
UTRLCLR register is not compliant with any version of the UFSHCI standard. Reporting a completion for commands cleared by writing into that register is problematic because it causes ufshcd_release_scsi_cmd()
to be called as follows:

ufshcd_sl_intr()
ufshcd_transfer_req_compl()

     ufshcd_poll()
       __ufshcd_transfer_req_compl()
         ufshcd_compl_one_cqe()
           cmd->result = ...
           ufshcd_release_scsi_cmd()
           scsi_done()

Calling ufshcd_release_scsi_cmd() if a command has been cleared is problematic because the SCSI core does not expect this. If ufshcd_try_to_abort_task() clears a SCSI command, ufshcd_release_scsi_cmd() must not be called until the SCSI core decides to release the command. This is why I wrote in a previous mail
that I think that a quirk should be introduced to suppress the completions generated by clearing a SCSI command.

Quoted 2 lines

The more problematic part is with MCQ mode. To imitate the DBR approach, we just need to set DID_REQUEUE upon receiving an

interrupt.

Quoted 4 lines

Everything else remains the same. This would make things simpler.

Moving forward, if we want to simplify things and we have also taken stock of the two or three scenarios where OCS: ABORTED

occurs,

Quoted 3 lines

do we even need a flag? Couldn't we just set DID_REQUEUE directly for OCS: ABORTED?
What do you think?

How about making ufshcd_compl_one_cqe() skip entries with status OCS_ABORTED? That would make ufshcd_compl_one_cqe() behave as the SCSI core expects, namely not freeing any command resources if a SCSI command is aborted successfully.

This approach may require further changes to ufshcd_abort_all(). In that function there are separate code paths for legacy and MCQ mode. This is less than ideal. Would it be possible to combine these code paths by removing the ufshcd_complete_requests() call from ufshcd_abort_all() and by handling completions from inside ufshcd_abort_one()?

Thanks,

Bart.

The four case flows for abort are as follows:
----------------------------------------------------------------

Case1: DBR ufshcd_abort

In this case, you can see that ufshcd_release_scsi_cmd will definitely be called.

ufshcd_abort()
ufshcd_try_to_abort_task() // It should trigger an
interrupt, but the tensor might not
get outstanding_lock
clear outstanding_reqs tag
ufshcd_release_scsi_cmd()
release outstanding_lock

ufshcd_intr()
ufshcd_sl_intr()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock
__ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE // mediatek may need quirk
change DID_ABORT to DID_REQUEUE

           ufshcd_release_scsi_cmd()
           scsi_done();

In most cases, ufshcd_intr will not reach scsi_done because the outstanding_reqs tag is cleared by the original thread. Therefore, whether there is an interrupt or not doesn't affect the result because the ISR will do nothing in most cases.

In a very low chance, the ISR will reach scsi_done and notify SCSI to requeue, and the original thread will not call ufshcd_release_scsi_cmd.
MediaTek may need to change DID_ABORT to DID_REQUEUE in this situation, or perhaps not handle this ISR at all.

----------------------------------------------------------------

Case2: MCQ ufshcd_abort

In the case of MCQ ufshcd_abort, you can also see that ufshcd_release_scsi_cmd will definitely be called too. However, there seems to be a problem here, as ufshcd_release_scsi_cmd might be called twice. This is because cmd is not null in ufshcd_release_scsi_cmd, which the previous version would set cmd to null. Skipping OCS: ABORTED in ufshcd_compl_one_cqe indeed can avoid this problem. This part needs further consideration on how to handle it.

ufshcd_abort()
ufshcd_mcq_abort()
ufshcd_try_to_abort_task() // will trigger ISR
ufshcd_release_scsi_cmd()

ufs_mtk_mcq_intr()
ufshcd_mcq_poll_cqe_lock()

     ufshcd_mcq_process_cqe()
       ufshcd_compl_one_cqe()
         cmd->result = DID_ABORT
         ufshcd_release_scsi_cmd() // will release twice
         scsi_done()

----------------------------------------------------------------

Case3: DBR ufshcd_err_handler

In the case of the DBR mode error handler, it's the same; ufshcd_release_scsi_cmd will also be executed, and scsi_done will definitely be used to notify SCSI to requeue.

ufshcd_err_handler()
ufshcd_abort_all()
ufshcd_abort_one()
ufshcd_try_to_abort_task() // It should trigger an
interrupt, but the tensor might not
ufshcd_complete_requests()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock __ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE // mediatek may need quirk
change DID_ABORT to DID_REQUEUE

               ufshcd_release_scsi_cmd()
               scsi_done()

ufshcd_intr()
ufshcd_sl_intr()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock __ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE // mediatek may need quirk change
DID_ABORT to DID_REQUEUE

           ufshcd_release_scsi_cmd()
           scsi_done();

At this time, the same actions are taken regardless of whether there is an ISR, and with the protection of outstanding_lock, only one thread will execute ufshcd_release_scsi_cmd and scsi_done.
----------------------------------------------------------------

Case4: MCQ ufshcd_err_handler

It's the same with MCQ mode; there is protection from the cqe lock, so only one thread will execute. What my patch 2 aims to do is to change DID_ABORT to DID_REQUEUE in this situation.

ufshcd_err_handler()
ufshcd_abort_all()

     ufshcd_abort_one()
       ufshcd_try_to_abort_task()	// will trigger irq thread
     ufshcd_complete_requests()
       ufshcd_mcq_compl_pending_transfer()
         ufshcd_mcq_poll_cqe_lock()
           ufshcd_mcq_process_cqe()
             ufshcd_compl_one_cqe()
               cmd->result = DID_ABORT // should change to DID_REQUEUE
               ufshcd_release_scsi_cmd()
               scsi_done()

ufs_mtk_mcq_intr()
ufshcd_mcq_poll_cqe_lock()

     ufshcd_mcq_process_cqe()
       ufshcd_compl_one_cqe()
         cmd->result = DID_ABORT  // should change to DID_REQUEUE
         ufshcd_release_scsi_cmd()
         scsi_done()

External email : Please do not click links or open attachments until you have verified the sender or the content.
On 9/19/24 5:16 AM, Peter Wang (王信友) wrote:

Quoted 4 lines

The four case flows for abort are as follows:
----------------------------------------------------------------

Case1: DBR ufshcd_abort

Please follow the terminology from the UFSHCI 4.0 standard and use the
word "legacy" instead of "DBR".

Quoted 30 lines

In this case, you can see that ufshcd_release_scsi_cmd will definitely be called.

ufshcd_abort()
ufshcd_try_to_abort_task()// It should trigger an
interrupt, but the tensor might not
get outstanding_lock
clear outstanding_reqs tag ufshcd_release_scsi_cmd()
release outstanding_lock

ufshcd_intr()
ufshcd_sl_intr()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock __ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE// mediatek may need quirk
change DID_ABORT to DID_REQUEUE

           ufshcd_release_scsi_cmd()
           scsi_done();

In most cases, ufshcd_intr will not reach scsi_done because the outstanding_reqs tag is cleared by the original thread. Therefore, whether there is an interrupt or not doesn't affect the result because the ISR will do nothing in most cases.

Quoted 5 lines

In a very low chance, the ISR will reach scsi_done and notify SCSI to requeue, and the original thread will not call ufshcd_release_scsi_cmd.
MediaTek may need to change DID_ABORT to DID_REQUEUE in this situation, or perhaps not handle this ISR at all.

Please modify ufshcd_compl_one_cqe() such that it ignores commands with status OCS_ABORTED. This will make the UFSHCI driver behave in the same way for all UFSHCI controllers, whether or not clearing a command triggers a completion interrupt.

Quoted 19 lines

----------------------------------------------------------------

Case2: MCQ ufshcd_abort

In the case of MCQ ufshcd_abort, you can also see that ufshcd_release_scsi_cmd will definitely be called too. However, there seems to be a problem here, as ufshcd_release_scsi_cmd might be called twice. This is because cmd is not null in ufshcd_release_scsi_cmd, which the previous version would set cmd to null. Skipping OCS: ABORTED in ufshcd_compl_one_cqe indeed can avoid this problem. This part needs further consideration on how to handle it.

ufshcd_abort()
ufshcd_mcq_abort()
ufshcd_try_to_abort_task()// will trigger ISR ufshcd_release_scsi_cmd()

Quoted 7 lines

ufs_mtk_mcq_intr()
ufshcd_mcq_poll_cqe_lock()

     ufshcd_mcq_process_cqe()
       ufshcd_compl_one_cqe()
         cmd->result = DID_ABORT
         ufshcd_release_scsi_cmd() // will release twice
         scsi_done()

Do you agree that this case can be addressed with the ufshcd_compl_one_cqe() change proposed above?

Quoted 26 lines

----------------------------------------------------------------

Case3: DBR ufshcd_err_handler

In the case of the DBR mode error handler, it's the same; ufshcd_release_scsi_cmd will also be executed, and scsi_done will definitely be used to notify SCSI to requeue.

ufshcd_err_handler()
ufshcd_abort_all()
ufshcd_abort_one()
ufshcd_try_to_abort_task()// It should trigger an
interrupt, but the tensor might not
ufshcd_complete_requests()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock __ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE // mediatek may need quirk
change DID_ABORT to DID_REQUEUE

               ufshcd_release_scsi_cmd()
               scsi_done()

Quoted 10 lines

ufshcd_intr()
ufshcd_sl_intr()
ufshcd_transfer_req_compl()
ufshcd_poll()
get outstanding_lock
clear outstanding_reqs tag
release outstanding_lock __ufshcd_transfer_req_compl()
ufshcd_compl_one_cqe()
cmd->result = DID_REQUEUE // mediatek may need quirk

change

Quoted 15 lines

DID_ABORT to DID_REQUEUE

           ufshcd_release_scsi_cmd()
           scsi_done();

At this time, the same actions are taken regardless of whether there is an ISR, and with the protection of outstanding_lock, only one thread will execute ufshcd_release_scsi_cmd and scsi_done.
----------------------------------------------------------------

Case4: MCQ ufshcd_err_handler

It's the same with MCQ mode; there is protection from the cqe lock, so only one thread will execute. What my patch 2 aims to do is to change DID_ABORT to DID_REQUEUE in this situation.

Quoted 10 lines

ufshcd_err_handler()
ufshcd_abort_all()
ufshcd_abort_one()
ufshcd_try_to_abort_task()// will trigger irq thread
ufshcd_complete_requests()
ufshcd_mcq_compl_pending_transfer()
ufshcd_mcq_poll_cqe_lock()
ufshcd_mcq_process_cqe()
ufshcd_compl_one_cqe()
cmd->result = DID_ABORT // should change to

DID_REQUEUE

Quoted 10 lines

               ufshcd_release_scsi_cmd()
               scsi_done()

ufs_mtk_mcq_intr()
ufshcd_mcq_poll_cqe_lock()

     ufshcd_mcq_process_cqe()
       ufshcd_compl_one_cqe()
         cmd->result = DID_ABORT  // should change to DID_REQUEUE
         ufshcd_release_scsi_cmd()
         scsi_done()

For legacy and MCQ mode, I prefer the following behavior for ufshcd_abort_all():
* ufshcd_compl_one_cqe() ignores commands with status OCS_ABORTED.
* ufshcd_release_scsi_cmd() is called either by ufshcd_abort_one() or by ufshcd_abort_all().

Do you agree with making the changes proposed above?

Thank you,

Bar

Quoted 7 lines · Bart Van Assche

For legacy and MCQ mode, I prefer the following behavior for ufshcd_abort_all():
* ufshcd_compl_one_cqe() ignores commands with status OCS_ABORTED.
* ufshcd_release_scsi_cmd() is called either by ufshcd_abort_one() or by ufshcd_abort_all().

Do you agree with making the changes proposed above?

This might not work, as SDB mode doesn't ignore
OCS: INVALID_OCS_VALUE but rather notifies SCSI to requeue.

So what we need to correct is to notify SCSI to requeue when MCQ mode receives OCS: ABORTED as well.

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 5 lines · Peter Wang (王信友)

Quoted 4 lines · Bart Van Assche

For legacy and MCQ mode, I prefer the following behavior for ufshcd_abort_all():
* ufshcd_compl_one_cqe() ignores commands with status OCS_ABORTED.
* ufshcd_release_scsi_cmd() is called either by ufshcd_abort_one()

or

Quoted 2 lines

Quoted 2 lines

by ufshcd_abort_all().

Quoted 4 lines

Quoted 1 line

Do you agree with making the changes proposed above?

This might not work, as SDB mode doesn't ignore
OCS: INVALID_OCS_VALUE but rather notifies SCSI to requeue.

cmd->result should be ignored for aborted commands. Hence, how OCS_INVALID_COMMAND_STATUS is translated by ufshcd_transfer_rsp_status() is not relevant for aborted commands.

Quoted 2 lines

So what we need to correct is to notify SCSI to requeue when MCQ mode receives OCS: ABORTED as well.

Unless the host controller violates the UFSHCI specification, the command status is not set for aborted commands in legacy mode. Let's keep the code uniform for legacy mode, MCQ mode, compliant and non- ompliant controllers and not rely on the command status for aborted commands.

Thanks,

Bart.

This statement is not quite accurate becasue in UFSHIC2.1, SDB mode specification already have OCS: ABORTED (0x6) define.

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 2 lines · Peter Wang (王信友)

This statement is not quite accurate becasue in UFSHIC2.1, SDB mode specification already have OCS: ABORTED (0x6) define.

I think I found why that status code is defined in the UFSHCI 2.1 standard. From the UFS 2.0 standard: "TASK ABORTED - This status shall
be returned when a command is aborted by a command or task management function on another I_T nexus and the Control mode page TAS bit is set
to one. Since in UFS there is only one I_T nexus and TAS bit is zero TASK ABORTED status codes will never occur."

Bart.