Re: [PATCH 0/4] Extend MQRQ_XFER_SINGLE_BLOCK to Support Multi-Block Read Failures

[Ulf Hansson]

On Mon, 30 Mar 2026 at 05:28, Shawn Lin <shawn.lin@xxxxxxxxxxxxxx> wrote:
>
>
> Background and patchset description:
> Currently, the MMC driver handles multi-block read/write failures through
> two separate mechanisms:
> 1.  For read failures, the native retry mechanism is triggered internally
>     within the MMC core.
> 2.  For write failures, the corresponding request is re-queued with the
>     MQRQ_XFER_SINGLE_BLOCK flag for single-block recovery. This separation
>     creates inconsistency in error handling.
>
> To improve code uniformity and reliability, this patch series consolidates
> both read and write failure paths into a unified recovery model by enabling
> the use of MQRQ_XFER_SINGLE_BLOCK for handling multi-block read failures as
> well. Patch 1-3 are small improvement while reading the code. Patch 4
> implements the conversion of multi-block read failures to MQRQ_XFER_SINGLE_BLOCK
> mode.
>
> Testing Details:
> 1. Platforms: Tested on RK3576 with both CQE and non-CQE modes.
> 2. Methodology:
> - Background dd test continuously reading from eMMC.
> - Physically interrupting the eMMC data line during tests to simulate CRC errors.
>
> Testing Result:
> On CEQ mode, CRC error triggered on slot 4 (blkaddr 734208). It successfully
> transitioned to single-block retries using tags 25, 27, 24, 25, 26, etc.
> (see logs below):
> mmc0: starting CQE transfer for tag 4 blkaddr 734208
> mmc0: CQE transfer done tag 9
> mmc0: blksz 512 blocks 1024 flags 00000200 tsac 150 ms nsac 1000
> mmc0: 524288 bytes transferred: 0
> mmc0: starting CQE transfer for tag 5 blkaddr 735232
> mmc0: blksz 512 blocks 1024 flags 00000200 tsac 150 ms nsac 1000
> mmc0: running CQE recovery
> mmc0: starting CMD12 arg 00000000 flags 00000019
> mmc0: req failed (CMD12): -110, retrying...
> mmc0: req failed (CMD12): -110, retrying...
> mmc0: req failed (CMD12): -110, retrying...
> mmc0: req done (CMD12): -110: 00000000 00000000 00000000 00000000
> mmc0: starting CMD13 arg 00010000 flags 00000195
> mmc0: req done (CMD13): 0: 00400900 00000000 00000000 00000000
> mmc0: starting CMD48 arg 00000001 flags 00000019
> mmc0: req done (CMD48): 0: 00000900 00000000 00000000 00000000
> mmc0: CQE transfer done tag 4
> mmc0: 0 bytes transferred: -84
> mmc0: CQE transfer done tag 5
> mmc0: 0 bytes transferred: 0
> mmc0: starting CQE transfer for tag 25 blkaddr 734208
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: starting CQE transfer for tag 26 blkaddr 735232
> mmc0: CQE transfer done tag 25
> mmc0: 512 bytes transferred: 0
> mmc0: blksz 512 blocks 1024 flags 00000200 tsac 150 ms nsac 1000
> mmc0: starting CQE transfer for tag 27 blkaddr 734209
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CQE transfer done tag 26
> mmc0: 524288 bytes transferred: 0
> mmc0: CQE transfer done tag 27
> mmc0: 512 bytes transferred: 0
> mmc0: starting CQE transfer for tag 24 blkaddr 734210
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CQE transfer done tag 24
> mmc0: 512 bytes transferred: 0
> mmc0: starting CQE transfer for tag 25 blkaddr 734211
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CQE transfer done tag 25
> mmc0: 512 bytes transferred: 0
> mmc0: starting CQE transfer for tag 26 blkaddr 734212
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CQE transfer done tag 26
> mmc0: 512 bytes transferred: 0
>
> On non-CQE mode, a CRC error occurs during a multi-block read
> operation (CMD18) targeting sector 0x0002d000. Subsequently,
> mmc core initiates a single-block retry using CMD17, starting
> from the same sector 0x0002d000(see logs below):
> mmc0: starting CMD18 arg 0002d000 flags 000000b5
> mmc0: blksz 512 blocks 1024 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CMD12 arg 00000000 flags 00000095
> mmc0: req done : 0: 00000000 00000000 00000000 00000000
> mmc0: req done (CMD18): 0: 00000900 00000000 00000000 00000000
> mmc0: 0 bytes transferred: -84
> mmc0: (CMD12): 0: 00000b00 00000000 00000000 00000000
> mmc0: starting CMD13 arg 00010000 flags 00000195
> mmc0: req done (CMD13): 0: 00000900 00000000 00000000 00000000
> mmc0: <starting CMD23 arg 00000400 flags 00000015>
> mmc0: starting CMD18 arg 0002d400 flags 000000b5
> mmc0: blksz 512 blocks 1024 flags 00000200 tsac 150 ms nsac 1000
> mmc0: CMD12 arg 00000000 flags 00000095
> mmc0: req done : 0: 00000000 00000000 00000000 00000000
> mmc0: req done (CMD18): 0: 00000900 00000000 00000000 00000000
> mmc0: 524288 bytes transferred: 0
> mmc0: (CMD12): 0: 00000000 00000000 00000000 00000000
> mmc0: starting CMD17 arg 0002d000 flags 000000b5
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: req done (CMD17): 0: 00000900 00000000 00000000 00000000
> mmc0: 512 bytes transferred: 0
> mmc0: starting CMD17 arg 0002d001 flags 000000b5
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: req done (CMD17): 0: 00000900 00000000 00000000 00000000
> mmc0: 512 bytes transferred: 0
> mmc0: starting CMD17 arg 0002d002 flags 000000b5
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: req done (CMD17): 0: 00000900 00000000 00000000 00000000
> mmc0: 512 bytes transferred: 0
> mmc0: starting CMD17 arg 0002d003 flags 000000b5
> mmc0: blksz 512 blocks 1 flags 00000200 tsac 150 ms nsac 1000
> mmc0: req done (CMD17): 0: 00000900 00000000 00000000 00000000
> mmc0: 512 bytes transferred: 0
>
>
>
> Shawn Lin (4):
>   mmc: core: Replace the hard-coded shift value 9 with SECTOR_SHIFT
>   mmc: block: Convert to use DEFINE_SIMPLE_DEV_PM_OPS()
>   mmc: mmc_test: Replace hard-coded values with macros and consolidate
>     test parameters
>   mmc: block: Use MQRQ_XFER_SINGLE_BLOCK for both read and write
>     recovery
>
>  drivers/mmc/core/block.c    | 82 ++++----------------------------------------
>  drivers/mmc/core/core.c     |  3 +-
>  drivers/mmc/core/mmc_test.c | 83 ++++++++++++++++++++-------------------------
>  3 files changed, 45 insertions(+), 123 deletions(-)

Nice!

>
> --
> 2.7.4
>

It looks like you could have split this series into separate
standalone changes, but let's proceed this time.

The series applied for next, thanks!

Kind regards
Uffe