Re: [PATCH 00/12 v5] Multiqueue for MMC/SD

[Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>]

[-- Attachment #1: Type: text/plain, Size: 3059 bytes --]


Hi Linus,

On Friday, November 10, 2017 11:01:31 AM Linus Walleij wrote:
> This is the fifth iteration of this patch set.
> 
> I *HOPE* that we can scrap this patch set and merge Adrian's
> patches instead, because they also bring CQE support which is
> nice. I had some review comments on his series, mainly that
> it needs to kill off the legacy block layer code path that
> noone likes anyway.
> 
> So this is mainly an academic and inspirational exercise.
> Whatever remains of this refactoring, if anything, I can
> certainly do on top of Adrian's patches as well.
> 
> What changed since v4 is the error path, since Adrian pointed
> out that the error handling seems to be fragile. It was indeed
> fragile... To make sure things work properly I have run long
> test rounds with fault injection, essentially:
> 
> Enable FAULT_INJECTION, FAULT_INJECTION_DEBUG_FS,
>        FAIL_MMC_REQUEST
> cd /debug/mmc3/fail_mmc_request/
> echo 1 > probability
> echo -1 > times
> 
> Then running a dd to the card, also increased the error rate
> to 10% and completed tests successfully, but at this error
> rate the MMC stack sometimes exceeds the retry limit and the
> dd command fails (as is appropriate).
> 
> Removing a card during I/O does not work well however :/
> So I guess I would need to work on that if this series should
> continue. (Hopefully unlikely.)
> 
> 
> Linus Walleij (12):
>   mmc: core: move the asynchronous post-processing
>   mmc: core: add a workqueue for completing requests
>   mmc: core: replace waitqueue with worker
>   mmc: core: do away with is_done_rcv
>   mmc: core: do away with is_new_req
>   mmc: core: kill off the context info
>   mmc: queue: simplify queue logic
>   mmc: block: shuffle retry and error handling
>   mmc: queue: stop flushing the pipeline with NULL
>   mmc: queue/block: pass around struct mmc_queue_req*s
>   mmc: block: issue requests in massive parallel
>   mmc: switch MMC/SD to use blk-mq multiqueueing v5
> 
>  drivers/mmc/core/block.c    | 557 +++++++++++++++++++++++---------------------
>  drivers/mmc/core/block.h    |   5 +-
>  drivers/mmc/core/bus.c      |   1 -
>  drivers/mmc/core/core.c     | 217 ++++++++++-------
>  drivers/mmc/core/core.h     |  11 +-
>  drivers/mmc/core/host.c     |   1 -
>  drivers/mmc/core/mmc_test.c |  31 +--
>  drivers/mmc/core/queue.c    | 252 ++++++++------------
>  drivers/mmc/core/queue.h    |  16 +-
>  include/linux/mmc/core.h    |   3 +-
>  include/linux/mmc/host.h    |  31 +--
>  11 files changed, 557 insertions(+), 568 deletions(-)

This works much better than initial version and a simple dd read
test shows more consistent results than with vanilla kernel.

However there are still some issues:

1. 30 seconds delay on "Waiting for /dev to be fully populated..."
   during boot

2. reboot command no longer works (there is a livelock after
   "The system is going down for reboot NOW!" message)

Full log (together with SysRq-l & SysRq-t outputs) attached.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

[-- Attachment #2: log.txt.gz --]
[-- Type: application/gzip, Size: 53535 bytes --]