[PATCH 3/3] drm/amdkfd: Skip update vmid in while update queue

[Joshi, Mukul]

[AMD Official Use Only - AMD Internal Distribution Only]

> -----Original Message-----
> From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Deng,
> Emily
> Sent: Monday, February 24, 2025 8:05 PM
> To: Deng, Emily <Emily.Deng at amd.com>; Kuehling, Felix
> <Felix.Kuehling at amd.com>
> Cc: amd-gfx at lists.freedesktop.org
> Subject: RE: [PATCH 3/3] drm/amdkfd: Skip update vmid in while update
> queue
>
> [AMD Official Use Only - AMD Internal Distribution Only]
>
> [AMD Official Use Only - AMD Internal Distribution Only]
>
> Ping......
>
> >-----Original Message-----
> >From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of
> >Deng, Emily
> >Sent: Monday, February 24, 2025 9:53 AM
> >To: Kuehling, Felix <Felix.Kuehling at amd.com>
> >Cc: amd-gfx at lists.freedesktop.org
> >Subject: RE: [PATCH 3/3] drm/amdkfd: Skip update vmid in while update
> >queue
> >
> >[AMD Official Use Only - AMD Internal Distribution Only]
> >
> >[AMD Official Use Only - AMD Internal Distribution Only]
> >
> >Hi Felix,
> >    Could you help review this? Thanks.
> >
> >Emily Deng
> >Best Wishes
> >
> >
> >
> >>-----Original Message-----
> >>From: Deng, Emily <Emily.Deng at amd.com>
> >>Sent: Friday, February 21, 2025 9:44 AM
> >>To: Deng, Emily <Emily.Deng at amd.com>; amd-gfx at lists.freedesktop.org
> >>Subject: RE: [PATCH 3/3] drm/amdkfd: Skip update vmid in while update
> >>queue
> >>
> >>[AMD Official Use Only - AMD Internal Distribution Only]
> >>
> >>Ping......
> >>
> >>Emily Deng
> >>Best Wishes
> >>
> >>
> >>
> >>>-----Original Message-----
> >>>From: Emily Deng <Emily.Deng at amd.com>
> >>>Sent: Thursday, February 20, 2025 2:25 PM
> >>>To: amd-gfx at lists.freedesktop.org
> >>>Cc: Deng, Emily <Emily.Deng at amd.com>
> >>>Subject: [PATCH 3/3] drm/amdkfd: Skip update vmid in while update
> >>>queue
> >>>
> >>>Avoid updating the vmid to 0 during the queue update process, as this
> >>>may trigger a wptr poll address page fault when a ring doorbell is
> >>>activated  in
> >>doorbell_mode=1.

Have you observed this page fault? If you have it, can you please paste the page fault backtrace.

> >>>
> >>>Signed-off-by: Emily Deng <Emily.Deng at amd.com>
> >>>---
> >>> drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 4 ++--
> >>> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>>diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
> >>>b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
> >>>index 6b38967d5631..3028c16264b2 100644
> >>>--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
> >>>+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
> >>>@@ -219,6 +219,8 @@ static void init_mqd(struct mqd_manager *mm,
> void
> >**mqd,
> >>>               m->cp_hqd_wg_state_offset = q->ctl_stack_size;
> >>>       }
> >>>
> >>>+      m->cp_hqd_vmid = q->vmid;
> >>>+

q->vmid would be 0 at the time of calling init_mqd when using HW scheduler as it's the
HWS which assigns the VMID.
Driver only assigns VMID when there is no HWS, which is not a production use-case.

> >>>       *mqd = m;
> >>>       if (gart_addr)
> >>>               *gart_addr = addr;
> >>>@@ -288,8 +290,6 @@ static void update_mqd(struct mqd_manager
> *mm,
> >>>void *mqd,
> >>>
> >>>       m->cp_hqd_iq_timer = 0;
> >>>
> >>>-      m->cp_hqd_vmid = q->vmid;

Maybe we can just remove his vmid assignment if this is indeed causing a page fault.
But I haven't seen a page fault because of this before.

Regards,
Mukul

> >>>-
> >>>       if (q->format == KFD_QUEUE_FORMAT_AQL) {
> >>>               m->cp_hqd_pq_control |=
> >>>CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
> >>>                               2 <<
> >>>CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT |
> >>>--
> >>>2.36.1
> >>