Re: [PATCH RFC 0/3] md/raid1: data corruption with serialization

On Thu, Feb 5, 2026 at 9:48 AM Yu Kuai <[email protected]> wrote:
>
> Hi,
>
> 在 2026/2/5 9:06, Xiao Ni 写道:
> > And let me change to this:
> > Setup: RAID1 with one fast disk and one slow disk. Two requests R1 and
> > R2, each split into LO and HI parts
> >
> > Step 1-10: Request R1 Processing
> >    - R1-LO: writes to fast disk → completes
> >    - R1-LO: writes to slow disk → starts (slow)
> >    - R1-HI: incorrectly waits for R1-LO (wrong sector range check)
> >    - R1-HI: R1-LO finishes and wakes up serial_io_wait. R1-HI finally
> > writes to fast disk → completes
>
> R1-hi should not wait for r1-lo, as they're not overlap. And as I said
> in the first reply, if here are still problem with WriteMostly case, please
> rebase and keep serialization for rdev.

Yes, it's a bug and patch03 fixes. I'll re-send patches based on
rdev-serialization.

>
> --
> Thansk,
> Kuai
>