On 2024/12/11 0:20, Mike Snitzer wrote:
On Mon, Nov 25, 2024 at 03:49:01PM +0800, LongPing Wei wrote:
On 2024/11/25 14:11, Christoph Hellwig wrote:
On Mon, Nov 25, 2024 at 03:07:24PM +0900, Damien Le Moal wrote:
100% agree. But in this case, dm-crypt sees a single BIO which gets split when
submitted to the underlying device. But the split issues the requests in
reverse order, which breaks write ordering.
I'd like to see verification of this. People who send these kinds of
odd old kernel reports tend to have weird patches applied to their tree
as well.
Without an upstream reproducer that actually shows issues I think
this can be safely ignored.
Thanks for your suggestions.
I will try to make another patch of fault injection to reproduce it on qemu.
If/when you do find it is relevant for latest upstream and/or
6.6-only: please add a "Fixes:" tag and also be explicit about which
kernel(s) your fix is applicable to.
Mike
Hi, Mike
We met this issue only on 6.6 now, as we work on android devices with
linux-6.6 and zoned storage device.
We will got android devices with linux-6.12 and zoned storage device
after half a year and then we can try to reproduce it with scheduler
being set to none.
I have tried to reproduce it with QEMU and nulblk, but failed as our
test suite is deeply coupled with our android devices.
Thanks
LongPing
