Christoph Hellwig <hch@xxxxxx> writes:
> On Fri, Sep 04, 2009 at 12:29:04AM +0200, Zdenek Kabelac wrote:
>> Ok - another bisect game played - and unexpected winner is:
>>
>> (fat: add ->sync_fs)
>>
>> f83d6d46e7adf241a064a4a425e5cd8a8fd8925f
>>
>> Reverting this commit with current -rc8 kernel makes the system happy
>> during the suspend/resume cycle. Obviously it has it price :) so just
>> plain revert is probably not a good solution so the problem looks
>> 'more serious' (fat is not the only fs with this patch) thus adding
>> original author to this thread.
>From it, I suspect the possible reason seems to read mmc after remove
event. I.e. the following sequence or something
sync fs process
[...]
removed mmc event
[...]
fat_sync_fs() <- sync again?
fat_clusters_flush()
sb_bread() <- read block on removed mmc
Can you add dump_stack() to the top of fat_sync_fs()? I hope it tells
why fat_sync_fs() is called (it is called from device unplug event?).
Well, that commit seems a bit strange. It calls fat_clusters_flush()
unconditionally without checking sb->s_dirt. However, if my guess is
right, "sync after removed event" itself sounds like the issue in
suspend process.
Thanks.
> Note that when you rever this patch on a current kernel you do actually
> get different behvaviour than when going back to before this commit.
>
> In 2.6.30 we called ->write_super in the various sync functions and
> then ->sync_fs, in 2.6.31-rc8 you would not call any syncing at all
> anymore. I think this patch might just be a symptom for a situation
> where the suspend code causes a sync and the mmc driver can't handle
> it anymore.
--
OGAWA Hirofumi <hirofumi@xxxxxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
