Skip to contentSkip to conversation

mmc: sdhci: fix up CMD12 sending

PATCH Adrian Hunter 3 participants 5 messages Nov 29, 2019 – Jan 17, 2020 1 message not in the archive MMC, SD and SDIO

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Adrian Hunter

author Acked-by
#1 lore raw
Quoted 6 lines · Yangbo Lu

The STOP command is disabled for multiple blocks r/w commands with auto CMD12, when start to send. However, if there is data error, software still needs to send CMD12 according to SD spec. This patch is to allow software CMD12 sending for this case.

Sorry for the delay. This looks good to me. Sending a STOP command on the error path in the auto-CMD12 case should be fine whether it has been sent already or not.

Quoted 43 lines
---
 drivers/mmc/host/sdhci.c | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)
drivers/mmc/host/sdhci.c +3 -14
index 09cdbe8..3041c39 100644
@@ -1326,12 +1326,12 @@ static void sdhci_finish_data(struct sdhci_host *host)
 
 	/*
 	 * Need to send CMD12 if -
-	 * a) open-ended multiblock transfer (no CMD23)
+	 * a) open-ended multiblock transfer not using auto CMD12 (no CMD23)
 	 * b) error in multiblock transfer
 	 */
 	if (data->stop &&
-	    (data->error ||
-	     !data->mrq->sbc)) {
+	    ((!data->mrq->sbc && !sdhci_auto_cmd12(host, data->mrq)) ||
+	     data->error)) {
 		/*
 		 * 'cap_cmd_during_tfr' request must not use the command line
 		 * after mmc_command_done() has been called. It is upper layer's
@@ -1825,17 +1825,6 @@ void sdhci_request(struct mmc_host *mmc, struct mmc_request *mrq)
 
 	sdhci_led_activate(host);
 
-	/*
-	 * Ensure we don't send the STOP for non-SET_BLOCK_COUNTED
-	 * requests if Auto-CMD12 is enabled.
-	 */
-	if (sdhci_auto_cmd12(host, mrq)) {
-		if (mrq->stop) {
-			mrq->data->stop = NULL;
-			mrq->stop = NULL;
-		}
-	}
-
 	if (!present || host->flags & SDHCI_DEVICE_DEAD) {
 		mrq->cmd->error = -ENOMEDIUM;
 		sdhci_finish_mrq(host, mrq);
#2 lore raw
In reply to a message not in the archive
Quoted 7 lines · Yangbo Lu

The STOP command is disabled for multiple blocks r/w commands with auto CMD12, when start to send. However, if there is data error, software still needs to send CMD12 according to SD spec. This patch is to allow software CMD12 sending for this case.

Applied for next, thanks!

Let's see how things go, then we can decide whether to add stable tag as well.

Kind regards
Uffe

Quoted 45 lines
---
 drivers/mmc/host/sdhci.c | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)
drivers/mmc/host/sdhci.c +3 -15
index 09cdbe8..3041c39 100644
@@ -1326,12 +1326,12 @@ static void sdhci_finish_data(struct sdhci_host *host)
 
        /*
         * Need to send CMD12 if -
-        * a) open-ended multiblock transfer (no CMD23)
+        * a) open-ended multiblock transfer not using auto CMD12 (no CMD23)
         * b) error in multiblock transfer
         */
        if (data->stop &&
-           (data->error ||
-            !data->mrq->sbc)) {
+           ((!data->mrq->sbc && !sdhci_auto_cmd12(host, data->mrq)) ||
+            data->error)) {
                /*
                 * 'cap_cmd_during_tfr' request must not use the command line
                 * after mmc_command_done() has been called. It is upper layer's
@@ -1825,17 +1825,6 @@ void sdhci_request(struct mmc_host *mmc, struct mmc_request *mrq)
 
        sdhci_led_activate(host);
 
-       /*
-        * Ensure we don't send the STOP for non-SET_BLOCK_COUNTED
-        * requests if Auto-CMD12 is enabled.
-        */
-       if (sdhci_auto_cmd12(host, mrq)) {
-               if (mrq->stop) {
-                       mrq->data->stop = NULL;
-                       mrq->stop = NULL;
-               }
-       }
-
        if (!present || host->flags & SDHCI_DEVICE_DEAD) {
                mrq->cmd->error = -ENOMEDIUM;
                sdhci_finish_mrq(host, mrq);
--

2.7.4

#3 lore raw

Hi Uffe and Adrian,

Back again on this topic. Actually we are trying to make the error recovery work after data error of CMD18 in linux-4.14.
With this patch, when CMD18 data error got, manual CMD12 would be sent. And then went into mmc_blk_cmd_recovery(). (Should be mmc_blk_mq_rw_recovery() in latest linux-5.5-rc2.) In mmc_blk_cmd_recovery(), re-tuning would fail before sending CMD13 on our specific board.
This may be some issue related to specific eMMC card we are investigating.

The above is just background introduction, and you may not care about that:)
I'd like to have some queries on CMD12 usage in MMC driver.
1. It seems CMD12 is always not using ABORT type for sending in sdhci.c. The SDHCI_CMD_ABORTCMD hasn't been used. Is this issue?
2. In block.c, CMD12 uses R1 response for data reading and R1B response for writing. Is it ok to use R1 response for SD? The SD spec mentions only R1B response for CMD12.

Appreciate your helps.
Thanks.

Best regards,
Yangbo Lu

Quoted 74 lines

-----Original Message-----
From: Ulf Hansson <[email protected]>
Sent: Tuesday, December 10, 2019 5:52 PM
To: Y.b. Lu <[email protected]>
Cc: [email protected]; Adrian Hunter <[email protected]>
Subject: Re: [PATCH] mmc: sdhci: fix up CMD12 sending

Quoted 7 lines · Yangbo Lu

The STOP command is disabled for multiple blocks r/w commands with auto CMD12, when start to send. However, if there is data error, software still needs to send CMD12 according to SD spec. This patch is to allow software CMD12 sending for this case.

Applied for next, thanks!

Let's see how things go, then we can decide whether to add stable tag as well.

Kind regards
Uffe

drivers/mmc/host/sdhci.c static void sdhci_finish_data(struct sdhci_host

Quoted 9 lines
---
 drivers/mmc/host/sdhci.c | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)
drivers/mmc/host/sdhci.c +0 -0
index 09cdbe8..3041c39 100644
@@ -1326,12 +1326,12 @@ static void sdhci_finish_data(struct sdhci_host

*host)

drivers/mmc/host/sdhci.c static void sdhci_finish_data(struct sdhci_host · line 1328

Quoted 5 lines
drivers/mmc/host/sdhci.c +1 -1
        /*
         * Need to send CMD12 if -
-        * a) open-ended multiblock transfer (no CMD23)
+        * a) open-ended multiblock transfer not using auto CMD12 (no

CMD23)

drivers/mmc/host/sdhci.c static void sdhci_finish_data(struct sdhci_host · line 1335

Quoted 9 lines
drivers/mmc/host/sdhci.c +2 -2
         * b) error in multiblock transfer
         */
        if (data->stop &&
-           (data->error ||
-            !data->mrq->sbc)) {
+           ((!data->mrq->sbc && !sdhci_auto_cmd12(host, data->mrq)) ||
+            data->error)) {
                /*
                 * 'cap_cmd_during_tfr' request must not use the

command line

drivers/mmc/host/sdhci.c static void sdhci_finish_data(struct sdhci_host · line 1336

Quoted 1 line
drivers/mmc/host/sdhci.c +0 -0
                 * after mmc_command_done() has been called. It is

upper layer's

drivers/mmc/host/sdhci.c void sdhci_request(struct mmc_host *mmc,

Quoted 1 line
drivers/mmc/host/sdhci.c +0 -0
@@ -1825,17 +1825,6 @@ void sdhci_request(struct mmc_host *mmc,

struct mmc_request *mrq)

Quoted 20 lines
drivers/mmc/host/sdhci.c +0 -12
        sdhci_led_activate(host);
 
-       /*
-        * Ensure we don't send the STOP for non-SET_BLOCK_COUNTED
-        * requests if Auto-CMD12 is enabled.
-        */
-       if (sdhci_auto_cmd12(host, mrq)) {
-               if (mrq->stop) {
-                       mrq->data->stop = NULL;
-                       mrq->stop = NULL;
-               }
-       }
-
        if (!present || host->flags & SDHCI_DEVICE_DEAD) {
                mrq->cmd->error = -ENOMEDIUM;
                sdhci_finish_mrq(host, mrq);
--

2.7.4

#4 lore raw
Quoted 11 lines · Y.b. Lu

Hi Uffe and Adrian,

Back again on this topic. Actually we are trying to make the error recovery work after data error of CMD18 in linux-4.14.
With this patch, when CMD18 data error got, manual CMD12 would be sent. And then went into mmc_blk_cmd_recovery(). (Should be mmc_blk_mq_rw_recovery() in latest linux-5.5-rc2.) In mmc_blk_cmd_recovery(), re-tuning would fail before sending CMD13 on our specific board.
This may be some issue related to specific eMMC card we are investigating.

The above is just background introduction, and you may not care about that:)
I'd like to have some queries on CMD12 usage in MMC driver.
1. It seems CMD12 is always not using ABORT type for sending in sdhci.c. The SDHCI_CMD_ABORTCMD hasn't been used. Is this issue?

I defer that question to Adrian.

Quoted 1 line

2. In block.c, CMD12 uses R1 response for data reading and R1B response for writing. Is it ok to use R1 response for SD? The SD spec mentions only R1B response for CMD12.

I think the specs isn't that clear on this. In this case, the R1B, is an R1 with an *optional* busy signaling on DAT0, unless it has been changed lately.

Additionally, as far as can tell, there have been no reports about problems with the current approach for "reads". Are you saying there is?

[...]

Kind regards
Uffe

#5 lore raw
Quoted 4 lines · Ulf Hansson
Quoted 3 lines · Y.b. Lu

Hi Uffe and Adrian,

Quoted 9 lines
Quoted 8 lines

Back again on this topic. Actually we are trying to make the error recovery work after data error of CMD18 in linux-4.14.
With this patch, when CMD18 data error got, manual CMD12 would be sent. And then went into mmc_blk_cmd_recovery(). (Should be mmc_blk_mq_rw_recovery() in latest linux-5.5-rc2.) In mmc_blk_cmd_recovery(), re-tuning would fail before sending CMD13 on our specific board.
This may be some issue related to specific eMMC card we are investigating.

The above is just background introduction, and you may not care about that:)
I'd like to have some queries on CMD12 usage in MMC driver.
1. It seems CMD12 is always not using ABORT type for sending in sdhci.c. The SDHCI_CMD_ABORTCMD hasn't been used. Is this issue?

AFAIK it is not an issue, but support for it can be added if you need it.

Quoted 6 lines
Quoted 2 lines

2. In block.c, CMD12 uses R1 response for data reading and R1B response for writing. Is it ok to use R1 response for SD? The SD spec mentions only R1B response for CMD12.

I think the specs isn't that clear on this. In this case, the R1B, is an R1 with an *optional* busy signaling on DAT0, unless it has been changed lately.

Also SDHCI offers SDHCI_QUIRK2_STOP_WITH_TC to turn all CMD12 responses into R1B

Quoted 10 lines

Additionally, as far as can tell, there have been no reports about problems with the current approach for "reads". Are you saying there is?

[...]

Kind regards
Uffe