mmc: fix OCR Polling

PATCH New Andrew Gabbasov 3 participants 5 messages Mar 20, 2015 – Mar 30, 2015

lore

Share & tools

lore

Andrew Gabbasov

author

2015-03-20 07:19 UTC #0 lore raw

PATCH 1/1 mmc: fix OCR Polling

Hi Peng,

Quoted 27 lines

From: Peng.Fan at freescale.com (Peng Fan)
Date: Thu, 19 Mar 2015 16:22:46 +0800
Subject: [U-Boot] [PATCH] mmc: fix OCR Polling
If in mmc_send_op_cond, OCR_BUSY is set in CMD1's response, then state is transfered to Ready state, and there is no need to send CMD1 again. Otherwise following CMD1 will recieve no response, or timeour error from driver such as fsl_esdhc.c.
If not into Ready state in previous CMD1, then continue CMD1 command.
In mmc_complete_op_cond, we use the value mmc->op_cond_response from mmc_send_op_cond, since there should be no CMD1 command between mmc_send_op_cond and mmc_complete_op_cond
Before fixing this, uboot log shows:
"
CMD_SEND:0
                ARG                      0x00000000
                MMC_RSP_NONE
CMD_SEND:8
                ARG                      0x000001AA
                MMC_RSP_R1,5,6,7         0x18EC1504
CMD_SEND:55
                ARG                      0x00000000
                MMC_RSP_R1,5,6,7         0x18EC1504
CMD_SEND:0

Quoted 8 lines

                ARG                      0x00000000
                MMC_RSP_NONE
CMD_SEND:1
                ARG                      0x00000000
                MMC_RSP_R3,4             0x00FF8080
CMD_SEND:1
                ARG                      0x40300000
                MMC_RSP_R3,4             0xC0FF8080 --> Already OCR_BUSY

set

Quoted 19 lines

CMD_SEND:1
                ARG                      0x40300000
                MMC_RSP_R3,4             0x0096850A --> Failed CMD1
MMC init failed
"
Using this patch, this issue is fixed, emmc can be detected correctly.
Signed-off-by Peng Fan <[email protected]>
---
 drivers/mmc/mmc.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/mmc/mmc.c +0 -0
index a13769e..43a9a8a 100644

drivers/mmc/mmc.c static int mmc_complete_op_cond(struct mmc *mmc) · line 412

Quoted 8 lines

drivers/mmc/mmc.c +4 -1

@@ -406,14 +406,23 @@ static int mmc_complete_op_cond(struct mmc *mmc)
 
     mmc->op_cond_pending = 0;
     start = get_timer(0);
-    do {
+    /*
+     * If in mmc_send_op_cond, OCR_BUSY is set in CMD1's response, then
+     * state is transfered to Ready state, and there is no need to
+     * send CMD1 again. Otherwise following CMD1 will recieve no

response,

Quoted 19 lines

drivers/mmc/mmc.c +7 -1

+     * or timeour error from driver such as fsl_esdhc.c.
+     *
+     * If not into Ready state in previous CMD1, then continue CMD1
+     * command.
+     */
+    while (!(mmc->op_cond_response & OCR_BUSY)) {
         err = mmc_send_op_cond_iter(mmc, &cmd, 1);
         if (err)
             return err;
         if (get_timer(start) > timeout)
             return UNUSABLE_ERR;
         udelay(100);
-    } while (!(mmc->op_cond_response & OCR_BUSY));
+    }
 
     if (mmc_host_is_spi(mmc)) { /* read OCR for spi */
         cmd.cmdidx = MMC_CMD_SPI_READ_OCR;

Signature · 1 line

1.8.4

After this patch, if the busy flag is indeed already set (so that the loop body
is not executed), and it is not an SPI case (mmc_host_is_spi(mmc) is false), the "cmd" structure (which is local to mmc_complete_op_cond() function) is left uninitialized, and using cmd.response[0] later in the function becomes
incorrect. And the OCR register value and the high capacity flag may be set incorrectly.

So, this patch is not enough by itself, it needs to be modified (appended) in order to get the correct code.

Coincidentally, the same day I submitted a patch series ("[U-Boot] [PATCH 0/6] mmc: Fix OCR polling and splitted initialization"), that, among other changes, fixes the OCR polling problem too ("[U-Boot] [PATCH 4/6] mmc: Continue polling MMC card for OCR only if it is still not ready"),
but in a slightly different way, that hopefully doesn't have the mentioned issue.

Thanks.

Best regards,
Andrew

Peng Fan

2015-03-21 10:33 UTC #1 lore raw

PATCH 1/1 mmc: fix OCR Polling

Hi, Andrew

Quoted 85 lines · Andrew Gabbasov

Hi Peng,
Quoted 27 lines
From: Peng.Fan at freescale.com (Peng Fan)
Date: Thu, 19 Mar 2015 16:22:46 +0800
Subject: [U-Boot] [PATCH] mmc: fix OCR Polling
If in mmc_send_op_cond, OCR_BUSY is set in CMD1's response, then state is transfered to Ready state, and there is no need to send CMD1 again. Otherwise following CMD1 will recieve no response, or timeour error from driver such as fsl_esdhc.c.
If not into Ready state in previous CMD1, then continue CMD1 command.
In mmc_complete_op_cond, we use the value mmc->op_cond_response from mmc_send_op_cond, since there should be no CMD1 command between mmc_send_op_cond and mmc_complete_op_cond
Before fixing this, uboot log shows:
"
CMD_SEND:0
                 ARG                      0x00000000
                 MMC_RSP_NONE
CMD_SEND:8
                 ARG                      0x000001AA
                 MMC_RSP_R1,5,6,7         0x18EC1504
CMD_SEND:55
                 ARG                      0x00000000
                 MMC_RSP_R1,5,6,7         0x18EC1504
CMD_SEND:0
Quoted 8 lines
                 ARG                      0x00000000
                 MMC_RSP_NONE
CMD_SEND:1
                 ARG                      0x00000000
                 MMC_RSP_R3,4             0x00FF8080
CMD_SEND:1
                 ARG                      0x40300000
                 MMC_RSP_R3,4             0xC0FF8080 --> Already OCR_BUSY
set
Quoted 19 lines
CMD_SEND:1
                 ARG                      0x40300000
                 MMC_RSP_R3,4             0x0096850A --> Failed CMD1
MMC init failed
"
Using this patch, this issue is fixed, emmc can be detected correctly.
Signed-off-by Peng Fan <[email protected]>
---
  drivers/mmc/mmc.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
drivers/mmc/mmc.c +0 -0
index a13769e..43a9a8a 100644
drivers/mmc/mmc.c static int mmc_complete_op_cond(struct mmc *mmc) · line 412
Quoted 8 lines
drivers/mmc/mmc.c +4 -1
@@ -406,14 +406,23 @@ static int mmc_complete_op_cond(struct mmc *mmc)
 
      mmc->op_cond_pending = 0;
      start = get_timer(0);
-    do {
+    /*
+     * If in mmc_send_op_cond, OCR_BUSY is set in CMD1's response, then
+     * state is transfered to Ready state, and there is no need to
+     * send CMD1 again. Otherwise following CMD1 will recieve no
response,
Quoted 19 lines
drivers/mmc/mmc.c +7 -1
+     * or timeour error from driver such as fsl_esdhc.c.
+     *
+     * If not into Ready state in previous CMD1, then continue CMD1
+     * command.
+     */
+    while (!(mmc->op_cond_response & OCR_BUSY)) {
          err = mmc_send_op_cond_iter(mmc, &cmd, 1);
          if (err)
              return err;
          if (get_timer(start) > timeout)
              return UNUSABLE_ERR;
          udelay(100);
-    } while (!(mmc->op_cond_response & OCR_BUSY));
+    }
 
      if (mmc_host_is_spi(mmc)) { /* read OCR for spi */
          cmd.cmdidx = MMC_CMD_SPI_READ_OCR;
Signature · 1 line
1.8.4

Quoted 8 lines

After this patch, if the busy flag is indeed already set (so that the loop body
is not executed), and it is not an SPI case (mmc_host_is_spi(mmc) is false), the "cmd" structure (which is local to mmc_complete_op_cond() function) is left uninitialized, and using cmd.response[0] later in the function becomes
incorrect. And the OCR register value and the high capacity flag may be set incorrectly.

Yeah. you are right.
Maybe the following piece of code should be added to replace mmc->ocr = cmd.response[0]:
"
if (mmc_host_is_spi(mmc))

     mmc->ocr = cmd.response[0];
else
     mmc->ocr = mmc->op_cond_response;

"
Thanks for correcting me.

Quoted 11 lines

So, this patch is not enough by itself, it needs to be modified (appended) in order to get the correct code.
Coincidentally, the same day I submitted a patch series ("[U-Boot] [PATCH 0/6] mmc: Fix OCR polling and splitted initialization"), that, among other changes, fixes the OCR polling problem too ("[U-Boot] [PATCH 4/6] mmc: Continue polling MMC card for OCR only if it is still not ready"),
but in a slightly different way, that hopefully doesn't have the mentioned issue.

Your's does not have such issue:)

Quoted 8 lines

Thanks.
Best regards,
Andrew
.

Thanks
Peng.

Andrew Gabbasov

author

2015-03-23 08:23 UTC #2 lore raw

PATCH 1/1 mmc: fix OCR Polling

Hi Peng,

Quoted 18 lines

From: Peng Fan [mailto:Peng.Fan at freescale.com]
Sent: Saturday, March 21, 2015 1:34 PM
To: Gabbasov, Andrew; u-boot at lists.denx.de
Subject: Re: [U-Boot] [PATCH] mmc: fix OCR Polling
[skipped]
Quoted 6 lines
After this patch, if the busy flag is indeed already set (so that the loop body is not executed), and it is not an SPI case (mmc_host_is_spi(mmc) is false), the "cmd" structure (which is local to mmc_complete_op_cond() function) is left uninitialized, and using cmd.response[0] later in the function becomes incorrect. And the OCR register value and the high capacity flag may be set incorrectly.
Yeah. you are right.
Maybe the following piece of code should be added to replace mmc->ocr = cmd.response[0]:
"
if (mmc_host_is_spi(mmc))

Quoted 5 lines

     mmc->ocr = cmd.response[0];
else
     mmc->ocr = mmc->op_cond_response;

"
Thanks for correcting me.

Well, there can be several ways to correct that. The easiest would be something
similar to what you propose, but, just to avoid extra "if", we could add

    mmc->op_cond_response = cmd.response[0];
to the end of existing "if(mmc_host_is_spi(mmc))" and change
    mmc->ocr = cmd.response[0];
to
    mmc->ocr = mmc->op_cond_response;

at the end of function. Since op_cond_response should be already set from the function beginning, this can be used immediately.

And, going further, since op_cond_response is actually the same contents as mmc->ocr, we could combine them and use mmc->ocr at once, from the beginning of polling loops. This is a little more complex, but makes the code cleaner. This is what is done in one of other patches in my series ;-)

Thanks.

Best regards,
Andrew

Pantelis Antoniou

2015-03-27 18:31 UTC #3 lore raw

PATCH 1/1 mmc: fix OCR Polling

Hi Andrew, Peng,

Quoted 39 lines · Andrew Gabbasov

Hi Peng,
Quoted 18 lines
From: Peng Fan [mailto:Peng.Fan at freescale.com]
Sent: Saturday, March 21, 2015 1:34 PM
To: Gabbasov, Andrew; u-boot at lists.denx.de
Subject: Re: [U-Boot] [PATCH] mmc: fix OCR Polling
[skipped]
Quoted 6 lines
After this patch, if the busy flag is indeed already set (so that the loop body is not executed), and it is not an SPI case (mmc_host_is_spi(mmc) is false), the "cmd" structure (which is local to mmc_complete_op_cond() function) is left uninitialized, and using cmd.response[0] later in the function becomes incorrect. And the OCR register value and the high capacity flag may be set incorrectly.
Yeah. you are right.
Maybe the following piece of code should be added to replace mmc->ocr = cmd.response[0]:
"
if (mmc_host_is_spi(mmc))
Quoted 5 lines
    mmc->ocr = cmd.response[0];
else
    mmc->ocr = mmc->op_cond_response;
"
Thanks for correcting me.
Well, there can be several ways to correct that. The easiest would be something
similar to what you propose, but, just to avoid extra "if", we could add
mmc->op_cond_response = cmd.response[0];
to the end of existing "if(mmc_host_is_spi(mmc))" and change
mmc->ocr = cmd.response[0];
to
mmc->ocr = mmc->op_cond_response;
at the end of function. Since op_cond_response should be already set from the function beginning, this can be used immediately.

Quoted 6 lines

And, going further, since op_cond_response is actually the same contents as mmc->ocr, we could combine them and use mmc->ocr at once, from the beginning of polling loops. This is a little more complex, but makes the code cleaner. This is what is done in one of other patches in my series ;-)

This does seem like a case where a simple accessor structure would help until we figure out how to process.

Something like mmc_get_ocr() as a private API perhaps?

Quoted 5 lines

Thanks.
Best regards,
Andrew

Regards

? Pantelis

Quoted 5 lines

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

Andrew Gabbasov

author

2015-03-30 11:15 UTC #4 lore raw

PATCH 1/1 mmc: fix OCR Polling

Hi Pantelis,

Quoted 55 lines

From: Pantelis Antoniou [mailto:pantelis.antoniou at gmail.com]
Sent: Friday, March 27, 2015 9:32 PM
To: Gabbasov, Andrew
Cc: Peng Fan; u-boot at lists.denx.de
Subject: Re: [U-Boot] [PATCH] mmc: fix OCR Polling
Hi Andrew, Peng,
Quoted 1 line
On Mar 23, 2015, at 01:23 , Andrew Gabbasov
Quoted 38 lines · <[email protected]> wrote:
Hi Peng,
Quoted 18 lines
From: Peng Fan [mailto:Peng.Fan at freescale.com]
Sent: Saturday, March 21, 2015 1:34 PM
To: Gabbasov, Andrew; u-boot at lists.denx.de
Subject: Re: [U-Boot] [PATCH] mmc: fix OCR Polling
[skipped]
Quoted 6 lines
After this patch, if the busy flag is indeed already set (so that the loop body is not executed), and it is not an SPI case (mmc_host_is_spi(mmc) is false), the "cmd" structure (which is local to mmc_complete_op_cond() function) is left uninitialized, and using cmd.response[0] later in the function becomes incorrect. And the OCR register value and the high capacity flag may be set incorrectly.
Yeah. you are right.
Maybe the following piece of code should be added to replace mmc->ocr = cmd.response[0]:
"
if (mmc_host_is_spi(mmc))
Quoted 5 lines
    mmc->ocr = cmd.response[0];
else
    mmc->ocr = mmc->op_cond_response;
"
Thanks for correcting me.
Well, there can be several ways to correct that. The easiest would be something
similar to what you propose, but, just to avoid extra "if", we could add
mmc->op_cond_response = cmd.response[0];
to the end of existing "if(mmc_host_is_spi(mmc))" and change
mmc->ocr = cmd.response[0];
to
mmc->ocr = mmc->op_cond_response;
at the end of function. Since op_cond_response should be already set from the function beginning, this can be used immediately.
Quoted 6 lines
And, going further, since op_cond_response is actually the same contents as mmc->ocr, we could combine them and use mmc->ocr at once, from the beginning of polling loops. This is a little more complex, but makes the code cleaner. This is what is done in one of other patches in my series ;-)

Quoted 4 lines

This does seem like a case where a simple accessor structure would help until we figure out how to process.
Something like mmc_get_ocr() as a private API perhaps?

Actually I don't see any need to introduce a new accessor or any other API here. Currently the 2 fields are used to store exactly the same data (OCR, which is a response to send_op_cond command). But the 'op_cond_response' is used while OCR polling and 'ocr' is filled in after its completion.
The correct solution from my point of view is to just use a single field (ocr) from the very beginning. I have already submitted this solution in one of my patches ("[PATCH 2/6] mmc: Avoid extra duplicate entry in mmc device structure" in "[PATCH 0/6] mmc: Fix OCR polling and splitted initialization" series).

Thanks.

Best regards,
Andrew