[PATCH 6.4 213/292] drm/ttm: never consider pinned BOs for eviction&swap

[PATCH 6.4 213/292] drm/ttm: never consider pinned BOs for eviction&swap

[Greg Kroah-Hartman]

From: Christian König <christian.koenig@amd.com>

commit a2848d08742c8e8494675892c02c0d22acbe3cf8 upstream.

There is a small window where we have already incremented the pin count
but not yet moved the bo from the lru to the pinned list.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reported-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
Tested-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Link: https://patchwork.freedesktop.org/patch/msgid/20230707120826.3701-1-christian.koenig@amd.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 drivers/gpu/drm/ttm/ttm_bo.c |    6 ++++++
 1 file changed, 6 insertions(+)

--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -517,6 +517,12 @@ static bool ttm_bo_evict_swapout_allowab
 {
 	bool ret = false;
 
+	if (bo->pin_count) {
+		*locked = false;
+		*busy = false;
+		return false;
+	}
+
 	if (bo->base.resv == ctx->resv) {
 		dma_resv_assert_held(bo->base.resv);
 		if (ctx->allow_res_evict)

Re: [PATCH 6.4 213/292] drm/ttm: never consider pinned BOs for eviction&swap

[Thomas Backlund]

Den 2023-07-21 kl. 19:05, skrev Greg Kroah-Hartman:
> From: Christian König <christian.koenig@amd.com>
> 
> commit a2848d08742c8e8494675892c02c0d22acbe3cf8 upstream.
> 
> There is a small window where we have already incremented the pin count
> but not yet moved the bo from the lru to the pinned list.
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> Reported-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
> Tested-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
> Acked-by: Alex Deucher <alexander.deucher@amd.com>
> Cc: stable@vger.kernel.org
> Link: https://patchwork.freedesktop.org/patch/msgid/20230707120826.3701-1-christian.koenig@amd.com
> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c |    6 ++++++
>   1 file changed, 6 insertions(+)
> 
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -517,6 +517,12 @@ static bool ttm_bo_evict_swapout_allowab
>   {
>   	bool ret = false;
>   
> +	if (bo->pin_count) {
> +		*locked = false;
> +		*busy = false;
> +		return false;
> +	}
> +
>   	if (bo->base.resv == ctx->resv) {
>   		dma_resv_assert_held(bo->base.resv);
>   		if (ctx->allow_res_evict)
> 


This one will trigger GPF and needs a follow-up fix that is not upstream 
yet:
https://patchwork.freedesktop.org/patch/547897/

as reported on LKML in thread:
[bug/bisected] commit a2848d08742c8e8494675892c02c0d22acbe3cf8 cause 
general protection fault, probably for non-canonical address 
0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN NOPTI

--
Thomas

Re: [PATCH 6.4 213/292] drm/ttm: never consider pinned BOs for eviction&swap

[Greg Kroah-Hartman]

On Fri, Jul 21, 2023 at 04:51:01PM +0000, Thomas Backlund wrote:
> Den 2023-07-21 kl. 19:05, skrev Greg Kroah-Hartman:
> > From: Christian König <christian.koenig@amd.com>
> > 
> > commit a2848d08742c8e8494675892c02c0d22acbe3cf8 upstream.
> > 
> > There is a small window where we have already incremented the pin count
> > but not yet moved the bo from the lru to the pinned list.
> > 
> > Signed-off-by: Christian König <christian.koenig@amd.com>
> > Reported-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
> > Tested-by: Pelloux-Prayer, Pierre-Eric <Pierre-eric.Pelloux-prayer@amd.com>
> > Acked-by: Alex Deucher <alexander.deucher@amd.com>
> > Cc: stable@vger.kernel.org
> > Link: https://patchwork.freedesktop.org/patch/msgid/20230707120826.3701-1-christian.koenig@amd.com
> > Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > ---
> >   drivers/gpu/drm/ttm/ttm_bo.c |    6 ++++++
> >   1 file changed, 6 insertions(+)
> > 
> > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > @@ -517,6 +517,12 @@ static bool ttm_bo_evict_swapout_allowab
> >   {
> >   	bool ret = false;
> >   
> > +	if (bo->pin_count) {
> > +		*locked = false;
> > +		*busy = false;
> > +		return false;
> > +	}
> > +
> >   	if (bo->base.resv == ctx->resv) {
> >   		dma_resv_assert_held(bo->base.resv);
> >   		if (ctx->allow_res_evict)
> > 
> 
> 
> This one will trigger GPF and needs a follow-up fix that is not upstream 
> yet:
> https://patchwork.freedesktop.org/patch/547897/
> 
> as reported on LKML in thread:
> [bug/bisected] commit a2848d08742c8e8494675892c02c0d22acbe3cf8 cause 
> general protection fault, probably for non-canonical address 
> 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN NOPTI

Thanks for letting me know, I've dropped this from the queues now.
Please let us know when we should add it back.

greg k-h