-----Original Message-----
From: Bough Chen <haibo.chen@nxp.com> 
Sent: Wednesday, January 27, 2021 3:36 PM
To: Ulf Hansson <redacted>; Wang, Xiaolei <redacted>; Fabio Estevam <festevam@gmail.com>
Cc: Pali Rohár <pali@kernel.org>; Lee Jones <redacted>; linux-mmc@vger.kernel.org; Linux Kernel Mailing List <redacted>
Subject: RE: [PATCH] mmc: core: Apply trim broken quirk to R1J57L

[Please note this e-mail is from an EXTERNAL e-mail address]

-----Original Message-----
From: Ulf Hansson [mailto:ulf.hansson@linaro.org]
Sent: 2021年1月21日 21:26
To: Wang, Xiaolei <redacted>; Fabio Estevam 
[off-list ref]; Bough Chen [off-list ref]
Cc: Pali Rohár <pali@kernel.org>; Lee Jones <redacted>; 
linux-mmc@vger.kernel.org; Linux Kernel Mailing List 
[off-list ref]
Subject: Re: [PATCH] mmc: core: Apply trim broken quirk to R1J57L

+ Fabio, Haibo

On Thu, 21 Jan 2021 at 10:54, Wang, Xiaolei 
[off-list ref]
wrote:

quoted

Hi

quoted

Are you sure this is an eMMC problem and not a mmc host driver issue?

quoted

Can you elaborate more what happens?

When I use the mkfs.f2fs tool to format the eMMC file system on the 
imx8qxp board, because mkfs.f2fs will use secdiscard first, when 
entering mmc_blk_issue_secdiscard_rq erase, once the parameters are

passed into MMC_SECURE_TRIM1_ARG, this function will take a long time 
to return .

quoted

The program has not ended, has been in TASK_UNINTERRUPTIBLE state.

I compared the mkfs.ext4 tool to format the file system. Because it 
directly uses mmc_blk_issue_discard_rq, this is a normal formatting

phenomenon.

quoted

mmc_blk_issue_secdiscard_rq and mmc_blk_issue_discard_rq are just

different commands and parameters sent by the host as a bus.

quoted

 I did not see the description of trim in the data sheet. Could the 
host driver

cause this problem?

Yes, it can - and we have had issues like these before. So before 
adding a card quirk, I think we need to make sure this isn't the case.

When using MMC_SECURE_TRIM1_ARG, it's very likely that the request 
takes longer to complete.

The mmc host is responsible for either dealing with busy detection 
with the help of its HW/controller - or if that can't be supported, 
the mmc core falls back to polling the card for busy with a CMD13.

See mmc_do_erase() in /drivers/mmc/core/core.c

quoted

Note:
The host driver I use is sdhci-esdhc-imx.c

Alright, I have looped in Fabio and Haibo that knows this driver, 
let's see if they can help.

Hi Ulf and Xiaolei,

I use the mkfs.f2fs tool format a partition on one Sandisk eMMC chip on imx8 platform, do not meet issue.
I add some log, confirm mkfs.f2fs call the mmc_blk_issue_secdiscard_rq(), and the arg is MMC_SECURE_TRIM1_ARG.
For the busy detection, imx usdhc has one register bit represent the dat0 signal level. So for your mmc device, it should be the mmc device trim feature problem.

Hi UIf and haibo

I did the verification again and found that the kernel version I used before was v5.4. The latest kernel does not have this problem.
 This patch is not accurate. I will confirm later whether it is caused by the missing patch of v5.4.

Thanks
Xiaolei

Best Regards
Haibo

quoted

Thanks
Xiaolei

Kind regards
Uffe

quoted

-----Original Message-----
From: Ulf Hansson <redacted>
Sent: Wednesday, January 20, 2021 9:41 PM
To: Wang, Xiaolei <redacted>
Cc: Pali Rohár <pali@kernel.org>; Lee Jones <redacted>; 
linux-mmc@vger.kernel.org; Linux Kernel Mailing List 
[off-list ref]
Subject: Re: [PATCH] mmc: core: Apply trim broken quirk to R1J57L

[Please note this e-mail is from an EXTERNAL e-mail address]

On Mon, 18 Jan 2021 at 05:27, Xiaolei Wang 
[off-list ref]

wrote:

quoted

quoted

R1J57L mmc chip hw capibility indicates that it supports trim 
function, but this function does not work properly, the SDIO bus 
does not respond, and the IO has been waiting so set quirks to 
skip trim

Are you sure this is an eMMC problem and not a mmc host driver issue?

Can you elaborate more what happens?

Kind regards
Uffe

quoted

Signed-off-by: Xiaolei Wang <redacted>
---
 drivers/mmc/core/quirks.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/mmc/core/quirks.h b/drivers/mmc/core/quirks.h 
index d68e6e513a4f..63e02391c133 100644
--- a/drivers/mmc/core/quirks.h
+++ b/drivers/mmc/core/quirks.h

@@ -89,6 +89,8 @@ static const struct mmc_fixup __maybe_unused

mmc_blk_fixups[] = {

quoted

quoted

                  MMC_QUIRK_SEC_ERASE_TRIM_BROKEN),
        MMC_FIXUP("VZL00M", CID_MANFID_SAMSUNG,

CID_OEMID_ANY, add_quirk_mmc,

quoted

quoted

                  MMC_QUIRK_SEC_ERASE_TRIM_BROKEN),
+       MMC_FIXUP("R1J57L", CID_MANFID_MICRON,

CID_OEMID_ANY, add_quirk_mmc,

quoted

quoted

+                 MMC_QUIRK_SEC_ERASE_TRIM_BROKEN),

        /*
         *  On Some Kingston eMMCs, performing trim can result in

@@

quoted

quoted

-98,6 +100,8 @@ static const struct mmc_fixup __maybe_unused

mmc_blk_fixups[] = {

quoted

quoted

                  MMC_QUIRK_TRIM_BROKEN),
        MMC_FIXUP("V10016", CID_MANFID_KINGSTON,

CID_OEMID_ANY, add_quirk_mmc,

quoted

quoted

                  MMC_QUIRK_TRIM_BROKEN),
+       MMC_FIXUP("R1J57L", CID_MANFID_MICRON,

CID_OEMID_ANY, add_quirk_mmc,

quoted

quoted

+                 MMC_QUIRK_TRIM_BROKEN),

        END_FIXUP
 };
--
2.25.1