Re: [PATCH] drm/amd/display: fix usage of DC_FPU_{BEGIN,END} with PREEMPT_RT

[Bert Karwatzki <spasswolf@web.de>]

Am Dienstag, dem 28.07.2026 um 05:51 +0500 schrieb mikhail.v.gavrilov@gmail.com:
> On Mon, 2026-07-27 at 12:50 +0200, Bert Karwatzki wrote:
> > On PREEMPT_RT kernels kvzalloc_obj() can sleep because spin_lock is
> > converted to rt_mutex. dc_create_plane_state() can be called while
> > inside an FPU-guarded region, resuling in "scheduling while atomic"
> > errors on PREEMPT_RT kernels.
> >  Fix this by calling kvzalloc_obj() with
> > DC_RUN_WITH_PREEMPTION_ENABLED().
> > Also fix the error path in dc_create_stream_for_sink().
> > 
> > Fixes: 3539437f354b ("drm/amd/display: Move FPU Guards From DML To DC
> > - Part 1")
> > Link:
> > https://lore.kernel.org/lkml/20260723123449.6494-1-spasswolf@web.de/
> > 
> > Signed-off-by: Bert Karwatzki <spasswolf@web.de>
> > ---
> >  drivers/gpu/drm/amd/display/dc/core/dc_stream.c  | 5 +++--
> >  drivers/gpu/drm/amd/display/dc/core/dc_surface.c | 4 ++--
> >  2 files changed, 5 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
> > b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
> > index dbc12640b01c..4ac835777b58 100644
> > --- a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
> > +++ b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c
> > @@ -233,8 +233,9 @@ struct dc_stream_state
> > *dc_create_stream_for_sink(
> >  
> >  fail:
> >  	if (stream) {
> > -		kfree(stream->update_scratch);
> > -		kfree(stream);
> > +		if (stream->update_scratch)
> > +			DC_RUN_WITH_PREEMPTION_ENABLED(kfree(stream-
> > > update_scratch));
> > +		DC_RUN_WITH_PREEMPTION_ENABLED(kfree(stream));
> >  	}
> >  
> >  	return NULL;
> > diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_surface.c
> > b/drivers/gpu/drm/amd/display/dc/core/dc_surface.c
> > index 88e825a6582c..d5c6427796b6 100644
> > --- a/drivers/gpu/drm/amd/display/dc/core/dc_surface.c
> > +++ b/drivers/gpu/drm/amd/display/dc/core/dc_surface.c
> > @@ -85,8 +85,8 @@ uint8_t  dc_plane_get_pipe_mask(struct dc_state
> > *dc_state, const struct dc_plane
> >  
> > *********************************************************************
> > *********/
> >  struct dc_plane_state *dc_create_plane_state(const struct dc *dc)
> >  {
> > -	struct dc_plane_state *plane_state =
> > kvzalloc_obj(*plane_state,
> > -							 
> > GFP_ATOMIC);
> > +	struct dc_plane_state *plane_state;
> > +	DC_RUN_WITH_PREEMPTION_ENABLED(plane_state =
> > kvzalloc_obj(*plane_state, GFP_ATOMIC));
> >  
> >  	if (NULL == plane_state)
> >  		return NULL;
> 
> Hi Bert,
> 
> You may not be aware that the same allocation is already wrapped once,
> at the dcn32 call site:
> 
>   183182235f6d ("drm/amd/display: Wrap DCN32 phantom-plane allocation
> in DC_RUN_WITH_PREEMPTION_ENABLED")
> 
> That one only covers the dcn32 DML1 path, while your trace goes through
> dcn401_validate_bandwidth() and dml21 - so wrapping the call site could
> never have caught your case.  Which is a good argument for guarding the
> allocation in the callee, as you do: dc_create_plane_state() is reached
> from every DCN and both DML generations.

I did try a different approach, using DC_RUN_WITH_PREEMPTION_ENABLED further
upward in the calltrace (This would have the slight benefit of more code
running with preemption enabled). This does not work because dml21_utils.c is compiled
as a _LINUX_FPU_COMPILATION_UNIT, so DC_RUN_WITH_PREEMPTION_ENABLED does nothing.
(see drivers/gpu/drm/amd/display/amdgpu_dm/dc_fpu.h)

diff --git a/drivers/gpu/drm/amd/display/dc/dml2_0/dml21/dml21_utils.c b/drivers/gpu/drm/amd/display/dc/dml2_0/dml21/dml21_utils.c
index 835fece1d46a..8c92deded8cd 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2_0/dml21/dml21_utils.c
+++ b/drivers/gpu/drm/amd/display/dc/dml2_0/dml21/dml21_utils.c
@@ -286,7 +286,8 @@ static struct dc_plane_state *dml21_add_phantom_plane(struct dml2_context *dml_c
        (void)plane_programming;
        struct dc_plane_state *phantom_plane;
 
-       phantom_plane = dml_ctx->config.svp_pstate.callbacks.create_phantom_plane(dc, context, main_plane);
+       DC_RUN_WITH_PREEMPTION_ENABLED(phantom_plane
+                       = dml_ctx->config.svp_pstate.callbacks.create_phantom_plane(dc, context, main_plane));
        if (!phantom_plane)
                return NULL;
 
diff --git a/drivers/gpu/drm/amd/display/dc/dml2_0/dml2_mall_phantom.c b/drivers/gpu/drm/amd/display/dc/dml2_0/dml2_mall_phantom.c
index 4543a60a0683..464bab9fc13b 100644
--- a/drivers/gpu/drm/amd/display/dc/dml2_0/dml2_mall_phantom.c
+++ b/drivers/gpu/drm/amd/display/dc/dml2_0/dml2_mall_phantom.c
@@ -769,10 +769,11 @@ static void enable_phantom_plane(struct dml2_context *ctx,
                if (curr_pipe->top_pipe && curr_pipe->top_pipe->plane_state == curr_pipe->plane_state) {
                        phantom_plane = prev_phantom_plane;
                } else {
-                       phantom_plane = ctx->config.svp_pstate.callbacks.create_phantom_plane(
+                       DC_RUN_WITH_PREEMPTION_ENABLED(phantom_plane =
+                                       ctx->config.svp_pstate.callbacks.create_phantom_plane(
                                        ctx->config.svp_pstate.callbacks.dc,
                                        state,
-                                       curr_pipe->plane_state);
+                                       curr_pipe->plane_state));
                        if (!phantom_plane)
                                return;
                }




Bert Karwatzki