[PATCH 0/1] drm/amdgpu/display: fix context switch issues in dcn20_validate_bandwidth

[PATCH 0/1] drm/amdgpu/display: fix context switch issues in dcn20_validate_bandwidth

[Daniel Kolesa]

This is likely the real fix for issues such as
https://gitlab.freedesktop.org/drm/amd/-/issues/1215, or perhaps
https://gitlab.freedesktop.org/drm/amd/-/issues/1191, and should
be probably backported into existing stable trees.

Daniel Kolesa (1):
  drm/amdgpu/display: use GFP_ATOMIC in
    dcn20_validate_bandwidth_internal

 drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

-- 
2.28.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdgpu/display: use GFP_ATOMIC in dcn20_validate_bandwidth_internal

[Daniel Kolesa]

GFP_KERNEL may and will sleep, and this is being executed in
a non-preemptible context; this will mess things up since it's
called inbetween DC_FP_START/END, and rescheduling will result
in the DC_FP_END later being called in a different context (or
just crashing if any floating point/vector registers/instructions
are used after the call is resumed in a different context).

Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
---
 drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
index 991eddd10952..c31d1f30e505 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
@@ -3141,7 +3141,7 @@ static bool dcn20_validate_bandwidth_internal(struct dc *dc, struct dc_state *co
 	int vlevel = 0;
 	int pipe_split_from[MAX_PIPES];
 	int pipe_cnt = 0;
-	display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_KERNEL);
+	display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_ATOMIC);
 	DC_LOGGER_INIT(dc->ctx->logger);
 
 	BW_VAL_TRACE_COUNT();
-- 
2.28.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/1] drm/amdgpu/display: use GFP_ATOMIC in dcn20_validate_bandwidth_internal

[Alex Deucher]

On Sat, Aug 8, 2020 at 4:51 PM Daniel Kolesa <daniel@octaforge.org> wrote:
>
> GFP_KERNEL may and will sleep, and this is being executed in
> a non-preemptible context; this will mess things up since it's
> called inbetween DC_FP_START/END, and rescheduling will result
> in the DC_FP_END later being called in a different context (or
> just crashing if any floating point/vector registers/instructions
> are used after the call is resumed in a different context).
>
> Signed-off-by: Daniel Kolesa <daniel@octaforge.org>

We should probably find a way to pre-allocate this, but in the
meantime, I'll apply the patch.

Thanks!

Alex


> ---
>  drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> index 991eddd10952..c31d1f30e505 100644
> --- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> +++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> @@ -3141,7 +3141,7 @@ static bool dcn20_validate_bandwidth_internal(struct dc *dc, struct dc_state *co
>         int vlevel = 0;
>         int pipe_split_from[MAX_PIPES];
>         int pipe_cnt = 0;
> -       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_KERNEL);
> +       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_ATOMIC);
>         DC_LOGGER_INIT(dc->ctx->logger);
>
>         BW_VAL_TRACE_COUNT();
> --
> 2.28.0
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/1] drm/amdgpu/display: use GFP_ATOMIC in dcn20_validate_bandwidth_internal

[Daniel Kolesa]

On Mon, Aug 10, 2020, at 17:38, Alex Deucher wrote:
> On Sat, Aug 8, 2020 at 4:51 PM Daniel Kolesa <daniel@octaforge.org> wrote:
> >
> > GFP_KERNEL may and will sleep, and this is being executed in
> > a non-preemptible context; this will mess things up since it's
> > called inbetween DC_FP_START/END, and rescheduling will result
> > in the DC_FP_END later being called in a different context (or
> > just crashing if any floating point/vector registers/instructions
> > are used after the call is resumed in a different context).
> >
> > Signed-off-by: Daniel Kolesa <daniel@octaforge.org>
> 
> We should probably find a way to pre-allocate this, but in the
> meantime, I'll apply the patch.

Indeed. But this should work as an immediate solution for people experiencing issues. For me it completely stopped dmesg noise on ppc64le and aarch64, and seemingly even fixed the screen flicker problem on POWER that I had to formerly work around by forcing the clocks to high (been running it for days and it's worked perfectly fine with adaptive during the time).

Daniel

> 
> Thanks!
> 
> Alex
> 
> 
> > ---
> >  drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> > index 991eddd10952..c31d1f30e505 100644
> > --- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> > +++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
> > @@ -3141,7 +3141,7 @@ static bool dcn20_validate_bandwidth_internal(struct dc *dc, struct dc_state *co
> >         int vlevel = 0;
> >         int pipe_split_from[MAX_PIPES];
> >         int pipe_cnt = 0;
> > -       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_KERNEL);
> > +       display_e2e_pipe_params_st *pipes = kzalloc(dc->res_pool->pipe_count * sizeof(display_e2e_pipe_params_st), GFP_ATOMIC);
> >         DC_LOGGER_INIT(dc->ctx->logger);
> >
> >         BW_VAL_TRACE_COUNT();
> > --
> > 2.28.0
> >
> > _______________________________________________
> > amd-gfx mailing list
> > amd-gfx@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx