Skip to contentSkip to conversation

mmc: core: Add new flag to force hardware reset

PATCH In review Michael Wu 3 participants 6 messages Sep 14, 2023 – Jan 11, 2024 MMC, SD and SDIO

lore
Share & tools
lore

Michael Wu

author
#0 lore raw

PATCH 1/1 mmc: core: Add new flag to force hardware reset

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

---
 drivers/mmc/core/block.c | 2 +-
 include/linux/mmc/host.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
 	pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
 	err = mmc_cqe_recovery(host);
-	if (err)
+	if (err || host->cqe_recovery_reset_always)
 		mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
 	mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -0
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
 	int			cqe_qdepth;
 	bool			cqe_enabled;
 	bool			cqe_on;
+	bool			cqe_recovery_reset_always;
 
 	/* Inline encryption support */
 #ifdef CONFIG_MMC_CRYPTO
Signature · 1 line
2.29.0
#1 lore raw
Quoted 10 lines · Michael Wu

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

like: https://lore.kernel.org/linux-mmc/[email protected]/

You should enable it in the vendor host.

Quoted 33 lines
---
 drivers/mmc/core/block.c | 2 +-
 include/linux/mmc/host.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
        pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
        err = mmc_cqe_recovery(host);
-       if (err)
+       if (err || host->cqe_recovery_reset_always)
                mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
        mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -1
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
        int                     cqe_qdepth;
        bool                    cqe_enabled;
        bool                    cqe_on;
+       bool                    cqe_recovery_reset_always;
 
        /* Inline encryption support */
 #ifdef CONFIG_MMC_CRYPTO
--

2.29.0

#2 lore raw

- trimmed cc-list, + Sartak Garg

Quoted 11 lines · Wenchao Chen
Quoted 9 lines · Michael Wu

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

Quoted 3 lines
Quoted 1 line

like: https://lore.kernel.org/linux-mmc/[email protected]/

Looks like this series didn't make it. I was awaiting a rebase from Sartak to apply it, but apparently something got in his way for a new submission.

Quoted 2 lines

You should enable it in the vendor host.

Yes! We don't want unused code in the core. We need a user of it too.

May I suggest that you pick up Sartak's patch for the core and thus add another patch for the host driver you care about and then re-submit it as a small series.

Kind regards
Uffe

Quoted 34 lines
Quoted 34 lines
---
 drivers/mmc/core/block.c | 2 +-
 include/linux/mmc/host.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
        pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
        err = mmc_cqe_recovery(host);
-       if (err)
+       if (err || host->cqe_recovery_reset_always)
                mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
        mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -1
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
        int                     cqe_qdepth;
        bool                    cqe_enabled;
        bool                    cqe_on;
+       bool                    cqe_recovery_reset_always;
 
        /* Inline encryption support */
 #ifdef CONFIG_MMC_CRYPTO
--

2.29.0

author
#3 lore raw
Quoted 63 lines · Ulf Hansson

- trimmed cc-list, + Sartak Garg

Quoted 11 lines · Wenchao Chen
Quoted 9 lines · Michael Wu

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

Quoted 3 lines
Quoted 1 line

like: https://lore.kernel.org/linux-mmc/[email protected]/

Looks like this series didn't make it. I was awaiting a rebase from Sartak to apply it, but apparently something got in his way for a new submission.

Quoted 2 lines

You should enable it in the vendor host.

Yes! We don't want unused code in the core. We need a user of it too.

May I suggest that you pick up Sartak's patch for the core and thus add another patch for the host driver you care about and then re-submit it as a small series.

Kind regards
Uffe

Quoted 29 lines
Quoted 29 lines
---
  drivers/mmc/core/block.c | 2 +-
  include/linux/mmc/host.h | 1 +
  2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
         pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
         err = mmc_cqe_recovery(host);
-       if (err)
+       if (err || host->cqe_recovery_reset_always)
                 mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
         mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -0
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
         int                     cqe_qdepth;
         bool                    cqe_enabled;
         bool                    cqe_on;
+       bool                    cqe_recovery_reset_always;
 
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
include/linux/mmc/host.h +0 -1
         /* Inline encryption support */
  #ifdef CONFIG_MMC_CRYPTO
--

2.29.0

Dear Ulf,
I have tested Sartak's patch and it is also able to resolve the issue we are currently facing. Therefore, I would like to inquire about the expected timeline for merging Sartak's patch.

Signature · 2 lines
Regards,
Michael Wu
#4 lore raw
Quoted 65 lines · Michael Wu
Quoted 63 lines · Ulf Hansson

- trimmed cc-list, + Sartak Garg

Quoted 11 lines · Wenchao Chen
Quoted 9 lines · Michael Wu

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

Quoted 3 lines
Quoted 1 line

like: https://lore.kernel.org/linux-mmc/[email protected]/

Looks like this series didn't make it. I was awaiting a rebase from Sartak to apply it, but apparently something got in his way for a new submission.

Quoted 2 lines

You should enable it in the vendor host.

Yes! We don't want unused code in the core. We need a user of it too.

May I suggest that you pick up Sartak's patch for the core and thus add another patch for the host driver you care about and then re-submit it as a small series.

Kind regards
Uffe

Quoted 29 lines
Quoted 29 lines
---
  drivers/mmc/core/block.c | 2 +-
  include/linux/mmc/host.h | 1 +
  2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
         pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
         err = mmc_cqe_recovery(host);
-       if (err)
+       if (err || host->cqe_recovery_reset_always)
                 mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
         mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -0
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
         int                     cqe_qdepth;
         bool                    cqe_enabled;
         bool                    cqe_on;
+       bool                    cqe_recovery_reset_always;
 
Quoted 9 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
include/linux/mmc/host.h +0 -1
         /* Inline encryption support */
  #ifdef CONFIG_MMC_CRYPTO
--

2.29.0

Dear Ulf,
I have tested Sartak's patch and it is also able to resolve the issue we are currently facing. Therefore, I would like to inquire about the expected timeline for merging Sartak's patch.

Hi Michael,

There is another series [1] that Adrian is working on that is related to the problem in $subject patch. Perhaps you can take a look and try it out instead of Sartak's patch, which seems to have gotten stalled.

Kind regards
Uffe

[1]
https://lore.kernel.org/all/[email protected]/

#5 lore raw
Quoted 67 lines · Ulf Hansson
Quoted 65 lines · Michael Wu
Quoted 63 lines · Ulf Hansson

- trimmed cc-list, + Sartak Garg

Quoted 11 lines · Wenchao Chen
Quoted 9 lines · Michael Wu

Entering the recovery system itself indicates a transmission error. In this situation, we intend to execute the mmc_blk_reset function to clear any anomalies that may be caused by errors. We have previously discussed with several MMC device manufacturers, and they expressed their desire for us to reset the device when errors occur to ensure stable operation. We aim to make this code compatible with all devices and ensure its stable performance, so we would like to add this patch

Quoted 3 lines
Quoted 1 line

like: https://lore.kernel.org/linux-mmc/[email protected]/

Looks like this series didn't make it. I was awaiting a rebase from Sartak to apply it, but apparently something got in his way for a new submission.

Quoted 2 lines

You should enable it in the vendor host.

Yes! We don't want unused code in the core. We need a user of it too.

May I suggest that you pick up Sartak's patch for the core and thus add another patch for the host driver you care about and then re-submit it as a small series.

Kind regards
Uffe

Quoted 29 lines
Quoted 29 lines
---
  drivers/mmc/core/block.c | 2 +-
  include/linux/mmc/host.h | 1 +
  2 files changed, 2 insertions(+), 1 deletion(-)
drivers/mmc/core/block.c +1 -1
index b5b414a71e0b..29fbe0ddeadb 100644
@@ -1503,7 +1503,7 @@ void mmc_blk_cqe_recovery(struct mmc_queue *mq)
         pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
 
         err = mmc_cqe_recovery(host);
-       if (err)
+       if (err || host->cqe_recovery_reset_always)
                 mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
         mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
 
include/linux/mmc/host.h +1 -0
index 62a6847a3b6f..f578541a06b5 100644
@@ -518,6 +518,7 @@ struct mmc_host {
         int                     cqe_qdepth;
         bool                    cqe_enabled;
         bool                    cqe_on;
+       bool                    cqe_recovery_reset_always;
 
Quoted 10 lines
Quoted 9 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
include/linux/mmc/host.h +0 -1
         /* Inline encryption support */
  #ifdef CONFIG_MMC_CRYPTO
--

2.29.0

Dear Ulf,
I have tested Sartak's patch and it is also able to resolve the issue we are currently facing. Therefore, I would like to inquire about the expected timeline for merging Sartak's patch.

Hi Sartak

Any updates?

Quoted 11 lines

Hi Michael,

There is another series [1] that Adrian is working on that is related to the problem in $subject patch. Perhaps you can take a look and try it out instead of Sartak's patch, which seems to have gotten stalled.

Kind regards
Uffe

[1]
https://lore.kernel.org/all/[email protected]/