On Wed, May 13, 2026 at 1:29 AM Boris Brezillon
<[email protected]> wrote:
>
> On Tue, 12 May 2026 14:04:43 -0700
> Chia-I Wu <[email protected]> wrote:
>
> > On Tue, May 12, 2026 at 5:14 AM Boris Brezillon
> > <[email protected]> wrote:
> > >
> > > Add a specific spinlock for events processing, and force processing
> > > of events in the panthor_sched_report_fw_events() path rather than
> > > deferring it to a work item. We also fast-track fence signalling by
> > > making the job completion logic IRQ-safe.
> > >
> > > Note that it requires changing a couple spin_lock() into
> > > spin_lock_irqsave() when those are taken inside a events_lock section.
> > >
> > > Signed-off-by: Boris Brezillon <[email protected]>
> > > ---
> > > drivers/gpu/drm/panthor/panthor_sched.c | 332
> > > +++++++++++++++-----------------
> > > 1 file changed, 155 insertions(+), 177 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/panthor/panthor_sched.c
> > > b/drivers/gpu/drm/panthor/panthor_sched.c
> > > index 5b34032deff8..fbf76b59b7ef 100644
> > > --- a/drivers/gpu/drm/panthor/panthor_sched.c
> > > +++ b/drivers/gpu/drm/panthor/panthor_sched.c
> > > @@ -177,18 +177,6 @@ struct panthor_scheduler {
> > > */
> > > struct work_struct sync_upd_work;
> > >
> > > - /**
> > > - * @fw_events_work: Work used to process FW events outside the
> > > interrupt path.
> > > - *
> > > - * Even if the interrupt is threaded, we need any event processing
> > > - * that require taking the panthor_scheduler::lock to be processed
> > > - * outside the interrupt path so we don't block the tick logic
> > > when
> > > - * it calls panthor_fw_{csg,wait}_wait_acks(). Since most of the
> > > - * event processing requires taking this lock, we just delegate
> > > all
> > > - * FW event processing to the scheduler workqueue.
> > > - */
> > > - struct work_struct fw_events_work;
> > > -
> > > /**
> > > * @fw_events: Bitmask encoding pending FW events.
> > > */
> > If we process all fw events in the irq context, we can remove
> > fw_events as well. More on this below.
>
> Oops, forgot to remove this field, indeed.
>
> > > @@ -254,6 +242,15 @@ struct panthor_scheduler {
> > > struct list_head waiting;
> > > } groups;
> > >
> > > + /**
> > > + * @events_lock: Lock taken when processing events.
> > > + *
> > > + * This also needs to be taken when csg_slots are updated, to
> > > make sure
> > > + * the event processing logic doesn't touch groups that have left
> > > the CSG
> > > + * slot.
> > > + */
> > > + spinlock_t events_lock;
> > > +
> > > /**
> > > * @csg_slots: FW command stream group slots.
> > It looks like read access can use either lock (process context) or
> > events_lock (irq context), while write access must use events_lock
> > (process context). Can we put that into the comment, or if makes
> > sense, enforce that with accessor functions?
>
> You're right. I'll mention that updates to csg_slots[] must be done
> with both the ::lock and ::events_lock held, while reads can be done
> with any of them held.
>
> >
> >
> > > */
> > > @@ -676,9 +673,6 @@ struct panthor_group {
> > > */
> > > struct panthor_kernel_bo *protm_suspend_buf;
> > >
> > > - /** @sync_upd_work: Work used to check/signal job fences. */
> > > - struct work_struct sync_upd_work;
> > > -
> > Can we make this a preparatory commit, where group_sync_upd_work is
> > replaced by group_check_job_completion?
>
> I'll try to split that up.
>
> >
> > Multiple things happen in this commit. I try to identify things that
> > can be separate commits. If this does not make sense, feel free to
> > ignore.
> >
> > > /** @tiler_oom_work: Work used to process tiler OOM events
> > > happening on this group. */
> > > struct work_struct tiler_oom_work;
> > >
>
> [...]
>
> > > /**
> > > * panthor_sched_report_fw_events() - Report FW events to the scheduler.
> > > * @ptdev: Device.
> > > @@ -1902,8 +1953,19 @@ void panthor_sched_report_fw_events(struct
> > > panthor_device *ptdev, u32 events)
> > This can be renamed to panthor_sched_handle_fw_events.
>
> It's not quite handling events though. For most of them, it's really
> just deferring the processing to work items, SYNC_UPDATE is the
> exception.
panthor_sched_report_fw_events no longer just queues
process_fw_events_work. It processes fw events immediately. If
"handle" is not the right verb, perhaps we can go with "process".
>
> >
> > > if (!ptdev->scheduler)
> > > return;
> > >
> > > - atomic_or(events, &ptdev->scheduler->fw_events);
> > > - sched_queue_work(ptdev->scheduler, fw_events);
> > > + guard(spinlock_irqsave)(&ptdev->scheduler->events_lock);
> > > +
> > > + if (events & JOB_INT_GLOBAL_IF) {
> > > + sched_process_global_irq_locked(ptdev);
> > > + events &= ~JOB_INT_GLOBAL_IF;
> > > + }
> > > +
> > > + while (events) {
> > > + u32 csg_id = ffs(events) - 1;
> > > +
> > > + sched_process_csg_irq_locked(ptdev, csg_id);
> > > + events &= ~BIT(csg_id);
> > > + }
> > This handles all fw events in the irq context. Are there concerns that
> > it may take too long? I might be wrong, but it seems possible to
> > handle only CSG_SYNC_UPDATE and defer the rest as before.
>
> I started with just the SYNC_UPDATE processing done in the hard-irq
> context, but after auditing the other stuff done in the handler, I
> realized it's basically just deferring all actual processing to work
> items. Yes, there's the overhead of demuxing the events from the
> ack/req regs, but part of this is already done to get to SYNC_UPDATE
> anyway, so at this point we're probably better off demuxing everything
> and scheduling works for all kind of events.
>
> I also compared the perfs between the two approaches (though I didn't
> do as much testing as I did with the new version, so I might have
> missed something), and it didn't seem to matter at all, because the
> interrupts we receive the most are SYNC_UPDATE and IDLE events, and
> those are at the same level.
Looking at ftrace irq events, when there is one active csg,
panthor-job takes 6us (median) / 17us (95%) / 27us (slowest).
I don't have a good sense if that's considered normal in hardirq. But
if that is ever an issue, and if the majority of the time is spent in
CSG_SYNC_UPDATE anyway, we can always revert the last patch to move
processing to threaded handler.
