On Mon, 14 Apr 2025, Benjamin Marzinski wrote:
> > cond_resched() shouldn't be removed because fsleep may fall back to
> > udelay.
>
> Again, your version is fine, but I'm not sure that cond_resched() was
> ever necessary, since there already is one in flush_delayed_bios().
> Also, at least the way it's currently coded, fsleep() will only resort
> to busy-waiting when the delay is 10 us or less, and the shortest it can
> be with this code is 62 us, so I don't think this cond_resched() will
> ever do anything.
Yes, but this is implementation detail that may change. Someone may change
fsleep to spin for larger timeout without knowing that dm-delay depends on
fsleep not spinning.
> > The patch should increase target version.
> >
> > I fixed the patch so that it applies on the top Linus' tree and applied
> > it to the linux-dm tree.
> >
> > BTW. do we need to backport this to the stable kernels? I think not, but
> > if you have some reason why should we backport it, explain it.
>
> dm-delay is basically a testing target, so I agree that it seems
> unnecessary to backport this.
>
> -Ben
OK, so I removed the "Fixes:" tag.
Mikulas
