[PATCH] Revert "drm/amd/display: avoid large on-stack structures"

[Hamza Mahfooz]

On 6/4/24 13:45, Aurabindo Pillai wrote:
> This reverts commit 44069f0f9b1fe577c5d4f05fa9eb02db8c618adc since
> the code path is called from FPU context, and triggers error like:
> 
> [   26.924055] BUG: sleeping function called from invalid context at include/linux/sched/mm.h:306
> [   26.924060] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1022, name: modprobe
> [   26.924063] preempt_count: 2, expected: 0
> [   26.924064] RCU nest depth: 0, expected: 0
> [   26.924066] Preemption disabled at:
> [   26.924067] [<ffffffffc089e5e0>] dc_fpu_begin+0x30/0xd0 [amdgpu]
> [   26.924322] CPU: 9 PID: 1022 Comm: modprobe Not tainted 6.8.0+ #20
> [   26.924325] Hardware name: System manufacturer System Product Name/CROSSHAIR VI HERO, BIOS 6302 10/23/2018
> [   26.924326] Call Trace:
> [   26.924327]  <TASK>
> [   26.924329]  dump_stack_lvl+0x37/0x50
> [   26.924333]  ? dc_fpu_begin+0x30/0xd0 [amdgpu]
> [   26.924589]  dump_stack+0x10/0x20
> [   26.924592]  __might_resched+0x16a/0x1c0
> [   26.924596]  __might_sleep+0x42/0x70
> [   26.924598]  __kmalloc_node_track_caller+0x2ad/0x4b0
> [   26.924601]  ? dm_helpers_allocate_gpu_mem+0x12/0x20 [amdgpu]
> [   26.924855]  ? dcn401_update_bw_bounding_box+0x2a/0xf0 [amdgpu]
> [   26.925122]  kmemdup+0x20/0x50
> [   26.925124]  ? kernel_fpu_begin_mask+0x6b/0xe0
> [   26.925127]  ? kmemdup+0x20/0x50
> [   26.925129]  dcn401_update_bw_bounding_box+0x2a/0xf0 [amdgpu]
> [   26.925393]  dc_create+0x311/0x670 [amdgpu]
> [   26.925649]  amdgpu_dm_init+0x2aa/0x1fa0 [amdgpu]
> [   26.925903]  ? irq_work_queue+0x38/0x50
> [   26.925907]  ? vprintk_emit+0x1e7/0x270
> [   26.925910]  ? dev_printk_emit+0x83/0xb0
> [   26.925914]  ? amdgpu_device_rreg+0x17/0x20 [amdgpu]
> [   26.926133]  dm_hw_init+0x14/0x30 [amdgpu]
> ---
>   drivers/gpu/drm/amd/display/dc/core/dc_state.c   | 16 +++++-----------
>   .../display/dc/resource/dcn401/dcn401_resource.c | 16 +++++-----------
>   2 files changed, 10 insertions(+), 22 deletions(-)

You probably want something like
https://patchwork.freedesktop.org/patch/597044/ instead.

> 
> diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_state.c b/drivers/gpu/drm/amd/display/dc/core/dc_state.c
> index 8ea9391c60b7..70928223b642 100644
> --- a/drivers/gpu/drm/amd/display/dc/core/dc_state.c
> +++ b/drivers/gpu/drm/amd/display/dc/core/dc_state.c
> @@ -193,11 +193,7 @@ static void init_state(struct dc *dc, struct dc_state *state)
>   struct dc_state *dc_state_create(struct dc *dc, struct dc_state_create_params *params)
>   {
>   #ifdef CONFIG_DRM_AMD_DC_FP
> -	struct dml2_configuration_options *dml2_opt;
> -
> -	dml2_opt = kmemdup(&dc->dml2_options, sizeof(*dml2_opt), GFP_KERNEL);
> -	if (!dml2_opt)
> -		return NULL;
> +	struct dml2_configuration_options dml2_opt = dc->dml2_options;
>   #endif
>   	struct dc_state *state = kvzalloc(sizeof(struct dc_state),
>   			GFP_KERNEL);
> @@ -211,14 +207,12 @@ struct dc_state *dc_state_create(struct dc *dc, struct dc_state_create_params *p
>   
>   #ifdef CONFIG_DRM_AMD_DC_FP
>   	if (dc->debug.using_dml2) {
> -		dml2_opt->use_clock_dc_limits = false;
> -		dml2_create(dc, dml2_opt, &state->bw_ctx.dml2);
> +		dml2_opt.use_clock_dc_limits = false;
> +		dml2_create(dc, &dml2_opt, &state->bw_ctx.dml2);
>   
> -		dml2_opt->use_clock_dc_limits = true;
> -		dml2_create(dc, dml2_opt, &state->bw_ctx.dml2_dc_power_source);
> +		dml2_opt.use_clock_dc_limits = true;
> +		dml2_create(dc, &dml2_opt, &state->bw_ctx.dml2_dc_power_source);
>   	}
> -
> -	kfree(dml2_opt);
>   #endif
>   
>   	kref_init(&state->refcount);
> diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn401/dcn401_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn401/dcn401_resource.c
> index 8dfb0a3d21cb..247bac177d1b 100644
> --- a/drivers/gpu/drm/amd/display/dc/resource/dcn401/dcn401_resource.c
> +++ b/drivers/gpu/drm/amd/display/dc/resource/dcn401/dcn401_resource.c
> @@ -1581,27 +1581,21 @@ static struct dc_cap_funcs cap_funcs = {
>   
>   static void dcn401_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_params)
>   {
> -	struct dml2_configuration_options *dml2_opt;
> -
> -	dml2_opt = kmemdup(&dc->dml2_options, sizeof(*dml2_opt), GFP_KERNEL);
> -	if (!dml2_opt)
> -		return;
> +	struct dml2_configuration_options dml2_opt = dc->dml2_options;
>   
>   	DC_FP_START();
>   
>   	dcn401_update_bw_bounding_box_fpu(dc, bw_params);
>   
> -	dml2_opt->use_clock_dc_limits = false;
> +	dml2_opt.use_clock_dc_limits = false;
>   	if (dc->debug.using_dml2 && dc->current_state && dc->current_state->bw_ctx.dml2)
> -		dml2_reinit(dc, dml2_opt, &dc->current_state->bw_ctx.dml2);
> +		dml2_reinit(dc, &dml2_opt, &dc->current_state->bw_ctx.dml2);
>   
> -	dml2_opt->use_clock_dc_limits = true;
> +	dml2_opt.use_clock_dc_limits = true;
>   	if (dc->debug.using_dml2 && dc->current_state && dc->current_state->bw_ctx.dml2_dc_power_source)
> -		dml2_reinit(dc, dml2_opt, &dc->current_state->bw_ctx.dml2_dc_power_source);
> +		dml2_reinit(dc, &dml2_opt, &dc->current_state->bw_ctx.dml2_dc_power_source);
>   
>   	DC_FP_END();
> -
> -	kfree(dml2_opt);
>   }
>   
>   enum dc_status dcn401_patch_unknown_plane_state(struct dc_plane_state *plane_state)
-- 
Hamza