Re: [PATCHv2 1/2] mmc: block: Remove error check of hw_reset on reset

[Ulf Hansson]

On Mon, 10 Oct 2022 at 14:36, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>
> On 10/10/22 11:08, Christian Löhle wrote:
> >
> >>
> >> -----Original Message-----
> >> From: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> >> Sent: Samstag, 8. Oktober 2022 10:38
> >> To: Christian Löhle <CLoehle@xxxxxxxxxxxxxx>; Ulf Hansson <ulf.hansson@xxxxxxxxxx>; Linux MMC List <linux-mmc@xxxxxxxxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx
> >> Subject: Re: [PATCHv2 1/2] mmc: block: Remove error check of hw_reset on reset
> >>
> >> On 7/10/22 18:42, Christian Löhle wrote:
> >>> Before switching back to the right partition in mmc_blk_reset there
> >>> used to be a check if hw_reset was even supported.
> >>> This return value was removed, so there is no reason to check.
> >>>
> >>> Fixes: fefdd3c91e0a ("mmc: core: Drop superfluous validations in
> >>> mmc_hw|sw_reset()")
> >>> Cc: stable@xxxxxxxxxxxxxxx
> >>>
> >>> Signed-off-by: Christian Loehle <cloehle@xxxxxxxxxxxxxx>
> >>> ---
> >>> -v2: Do not attempt to switch partitions if reset failed
> >>>
> >>>  drivers/mmc/core/block.c | 28 +++++++++++++---------------
> >>>  1 file changed, 13 insertions(+), 15 deletions(-)
> >>>
> >>> diff --git a/drivers/mmc/core/block.c b/drivers/mmc/core/block.c index
> >>> ce89611a136e..8db72cba2bbe 100644
> >>> --- a/drivers/mmc/core/block.c
> >>> +++ b/drivers/mmc/core/block.c
> >>> @@ -991,29 +991,27 @@ static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host,
> >>>                      int type)
> >>>  {
> >>>     int err;
> >>> +   struct mmc_blk_data *main_md = dev_get_drvdata(&host->card->dev);
> >>> +   int part_err;
> >>>
> >>>     if (md->reset_done & type)
> >>>             return -EEXIST;
> >>>
> >>>     md->reset_done |= type;
> >>>     err = mmc_hw_reset(host->card);
> >>> +   if (err)
> >>> +           return err;
> >>
> >> This could be a potential source of data corruption.
> >>
> >> There is no guarantee that a subsequent I/O will fail just because the reset failed.  Reading / writing the wrong partition would be disastrous, so we should always try to get back to the correct partition.
> >>
> >> I haven't looked at the possibility of just flagging the partition as invalid - need to be sure any subsequent I/O attempts still go through a path that switches the partition.
> >
> > I can see where youre coming from, but similarly a failing mmc_blk_part_switch doesn't imply all subsequent IO will fail.
> > Flagging the partition as invalid can be seen as rendering the system to a potentially useless state, which is a bit overboard for e.g. one CRC7 failure on the switch.
>
> I wasn't clear sorry.  I meant setting main_md->part_curr to a value
> that doesn't match any partition, thereby forcing the next I/O to
> switch partition first.

If I understand correctly, you are suggesting to ignore the return
code from mmc_hw_reset() and then always try to switch to the correct
partition. If we end up failing to switch the partition, then we
should set an invalid value in main_md->part_curr and return an error
code?

>
> > Not sure yet what the ideal behavior is, but either way I would go with v1 1/2 and v2 2/2 for now? That already fixes imo the most relevant potential data corruptions. (successful reset -> no switch)
> > Then we can come up with a a good handling for mmc_blk_reset or even around mmc_blk_part_switch.
> > Or what do you suggest?
>
> What about what I just described above.
>

It seems reasonable to me (assuming I have understood correctly).

One additional thing though. I would appreciate some comments in the
code, so it becomes clear of what goes on.

Kind regards
Uffe