Re: [PATCH blktests v5 2/3] bcache: add bcache/002

On Mar 06, 2026 / 14:14, Daniel Wagner wrote:
> On Fri, Mar 06, 2026 at 12:27:51PM +0000, Shinichiro Kawasaki wrote:
> > On Mar 05, 2026 / 16:01, Daniel Wagner wrote:
> > > Add test case from Stephen Zhang [1][2], which is
> > >=20
> > > 1. Wait for fio to complete (wait $fio_pid)
> > > 2. Wait a few seconds for I/O to drain
> > > 3. Then check the utilization
> > >=20
> > > If utilization is still > 0% after I/O completes, then there's a real
> > > accounting leak.
> > >=20
> > > [1] https://lore.kernel.org/linux-bcache/CANubcdX7eNbH_bo4-f94DUbdiEb=
[email protected]/
> > > [2] https://lore.kernel.org/all/CANubcdU92Uv7wNdZH357LLTnfwPouekXNTPS=
[email protected]
> >=20
> > Thanks for the updates. It's good that we do not need iostat :)
> >=20
> > I ran this test case with v7.0-rc2 kernel, and it passes. Good.
> >=20
> > I also ran this test case with older kernel without the kernel fix [1].=
 I
> > expected failure, but it passed. I ran the Stephen's original test scri=
pt and it
> > did not fail either. So I guess my test environment does not fulfill th=
e
> > conditions to recreate the failure. Daniel, did you observe this test c=
ase
> > failed with any older kernel?
>=20
> Same here, it doesn't fail on older kernels either. Maybe we should just
> not merge this test then? Maybe Stephen can have a look and figure out
> what needs to be done to trigger the condition more reliable.

I see. Before merging this patch, I would like to make sure that this test =
case
is valuable, and it does not just waste time and power of CI systems.

Stephen,

May I ask you to share what kind of devices you used to recreate the failur=
e?
Also, could you share the kernel .config to recreate the failure? I would l=
ike
use them to recreate the failure on my test system.

>=20
> > Let me leave nit comments in line. If this patch does not respin, I can=
 fold in
> > the changes for the nit comments.
>=20
> Sure, just fix it up if you decide to apply it.

Got it, thanks.=