Hi Andi,
On Friday, 17 October 2025 10:45:23 CEST Andi Shyti wrote:
> Hi Janusz,
>
> > diff --git a/drivers/gpu/drm/i915/gt/selftest_migrate.c b/drivers/gpu/drm/
i915/gt/selftest_migrate.c
> > index 54bc447efce0b..cde755751a0ba 100644
> > --- a/drivers/gpu/drm/i915/gt/selftest_migrate.c
> > +++ b/drivers/gpu/drm/i915/gt/selftest_migrate.c
> > @@ -710,7 +710,8 @@ static int threaded_migrate(struct intel_migrate
*migrate,
> > thread[i].tsk = tsk;
> > }
> >
> > - msleep(10 * n_cpus); /* start all threads before we kthread_stop()
*/
> > + /* start all threads before we kthread_stop() */
> > + msleep((intel_vm_no_concurrent_access_wa(migrate->context->vm->i915)
? 100 : 10) * n_cpus);
>
> First question, why 100? Second question why not 100 for
> everyone?
100ms because 10ms occurred too short on CHV / BXT+VTD with GGTT workarounds,
and I've empirically determined a sufficient delay of 100ms, as mentioned in
commit description. Do you want me to add that information to the inline
comment as well?
Not for everyone because 10ms has proven to be sufficient for platforms with
no CHV / BXT+VTD specific workarounds. Do you want me to use a single value?
Thanks,
Janusz
>
> Andi
>
> > for (i = 0; i < n_cpus; ++i) {
> > struct task_struct *tsk = thread[i].tsk;
>
