drm/amd/display: Fix unsafe uses of kernel mode FPU

The point of isolating code that uses kernel mode FPU in separate
compilation units is to ensure that even implicit uses of, e.g., SIMD
registers for spilling occur only in a context where this is permitted,
i.e., from inside a kernel_fpu_begin/end block.

This is important on arm64, which uses -mgeneral-regs-only to build all
kernel code, with the exception of such compilation units where FP or
SIMD registers are expected to be used. Given that the compiler may
invent uses of FP/SIMD anywhere in such a unit, none of its code may be
accessible from outside a kernel_fpu_begin/end block.

This means that all callers into such compilation units must use the
DC_FP start/end macros, which must not occur there themselves. For
robustness, all functions with external linkage that reside there should
call dc_assert_fp_enabled() to assert that the FPU context was set up
correctly.

Fix this for the DCN35, DCN351 and DCN36 implementations.

Cc: Austin Zheng <[email protected]>
Cc: Jun Lei <[email protected]>
Cc: Harry Wentland <[email protected]>
Cc: Leo Li <[email protected]>
Cc: Rodrigo Siqueira <[email protected]>
Cc: Alex Deucher <[email protected]>
Cc: "Christian König" <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Ard Biesheuvel <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Cc: [email protected]
Signed-off-by: Alex Deucher <[email protected]>

Author: ardbiesheuvel

drivers/gpu/drm/amd/display/dc/dml/dcn31/dcn31_fpu.c

@@ -808,13 +808,17 @@ void dcn316_update_bw_bounding_box(struct dc *dc, struct clk_bw_params *bw_param
 
 int dcn_get_max_non_odm_pix_rate_100hz(struct _vcs_dpi_soc_bounding_box_st *soc)
 {
+	dc_assert_fp_enabled();
+
 	return soc->clock_limits[0].dispclk_mhz * 10000.0 / (1.0 + soc->dcn_downspread_percent / 100.0);
 }
 
 int dcn_get_approx_det_segs_required_for_pstate(
 		struct _vcs_dpi_soc_bounding_box_st *soc,
 		int pix_clk_100hz, int bpp, int seg_size_kb)
 {
+	dc_assert_fp_enabled();
+
 	/* Roughly calculate required crb to hide latency. In practice there is slightly
 	 * more buffer available for latency hiding
 	 */

drivers/gpu/drm/amd/display/dc/dml/dcn35/dcn35_fpu.c

@@ -445,6 +445,8 @@ int dcn35_populate_dml_pipes_from_context_fpu(struct dc *dc,
 	bool upscaled = false;
 	const unsigned int max_allowed_vblank_nom = 1023;
 
+	dc_assert_fp_enabled();
+
 	dcn31_populate_dml_pipes_from_context(dc, context, pipes,
 					      validate_mode);
 
@@ -498,9 +500,7 @@ int dcn35_populate_dml_pipes_from_context_fpu(struct dc *dc,
 
 		pipes[pipe_cnt].pipe.src.unbounded_req_mode = false;
 
-		DC_FP_START();
 		dcn31_zero_pipe_dcc_fraction(pipes, pipe_cnt);
-		DC_FP_END();
 
 		pipes[pipe_cnt].pipe.dest.vfront_porch = timing->v_front_porch;
 		pipes[pipe_cnt].pipe.src.dcc_rate = 3;
@@ -581,6 +581,8 @@ void dcn35_decide_zstate_support(struct dc *dc, struct dc_state *context)
 	unsigned int i, plane_count = 0;
 	DC_LOGGER_INIT(dc->ctx->logger);
 
+	dc_assert_fp_enabled();
+
 	for (i = 0; i < dc->res_pool->pipe_count; i++) {
 		if (context->res_ctx.pipe_ctx[i].plane_state)
 			plane_count++;

drivers/gpu/drm/amd/display/dc/dml/dcn351/dcn351_fpu.c

@@ -478,6 +478,8 @@ int dcn351_populate_dml_pipes_from_context_fpu(struct dc *dc,
 	bool upscaled = false;
 	const unsigned int max_allowed_vblank_nom = 1023;
 
+	dc_assert_fp_enabled();
+
 	dcn31_populate_dml_pipes_from_context(dc, context, pipes,
 					      validate_mode);
 
@@ -531,9 +533,7 @@ int dcn351_populate_dml_pipes_from_context_fpu(struct dc *dc,
 
 		pipes[pipe_cnt].pipe.src.unbounded_req_mode = false;
 
-		DC_FP_START();
 		dcn31_zero_pipe_dcc_fraction(pipes, pipe_cnt);
-		DC_FP_END();
 
 		pipes[pipe_cnt].pipe.dest.vfront_porch = timing->v_front_porch;
 		pipes[pipe_cnt].pipe.src.dcc_rate = 3;

drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c

@@ -1760,6 +1760,20 @@ enum dc_status dcn35_patch_unknown_plane_state(struct dc_plane_state *plane_stat
 }
 
 
+static int populate_dml_pipes_from_context_fpu(struct dc *dc,
+					       struct dc_state *context,
+					       display_e2e_pipe_params_st *pipes,
+					       enum dc_validate_mode validate_mode)
+{
+	int ret;
+
+	DC_FP_START();
+	ret = dcn35_populate_dml_pipes_from_context_fpu(dc, context, pipes, validate_mode);
+	DC_FP_END();
+
+	return ret;
+}
+
 static struct resource_funcs dcn35_res_pool_funcs = {
 	.destroy = dcn35_destroy_resource_pool,
 	.link_enc_create = dcn35_link_encoder_create,
@@ -1770,7 +1784,7 @@ static struct resource_funcs dcn35_res_pool_funcs = {
 	.validate_bandwidth = dcn35_validate_bandwidth,
 	.calculate_wm_and_dlg = NULL,
 	.update_soc_for_wm_a = dcn31_update_soc_for_wm_a,
-	.populate_dml_pipes = dcn35_populate_dml_pipes_from_context_fpu,
+	.populate_dml_pipes = populate_dml_pipes_from_context_fpu,
 	.acquire_free_pipe_as_secondary_dpp_pipe = dcn20_acquire_free_pipe_for_layer,
 	.release_pipe = dcn20_release_pipe,
 	.add_stream_to_ctx = dcn30_add_stream_to_ctx,

drivers/gpu/drm/amd/display/dc/resource/dcn351/dcn351_resource.c

@@ -1732,6 +1732,21 @@ static enum dc_status dcn351_validate_bandwidth(struct dc *dc,
 	return out ? DC_OK : DC_FAIL_BANDWIDTH_VALIDATE;
 }
 
+static int populate_dml_pipes_from_context_fpu(struct dc *dc,
+					       struct dc_state *context,
+					       display_e2e_pipe_params_st *pipes,
+					       enum dc_validate_mode validate_mode)
+{
+	int ret;
+
+	DC_FP_START();
+	ret = dcn351_populate_dml_pipes_from_context_fpu(dc, context, pipes, validate_mode);
+	DC_FP_END();
+
+	return ret;
+
+}
+
 static struct resource_funcs dcn351_res_pool_funcs = {
 	.destroy = dcn351_destroy_resource_pool,
 	.link_enc_create = dcn35_link_encoder_create,
@@ -1742,7 +1757,7 @@ static struct resource_funcs dcn351_res_pool_funcs = {
 	.validate_bandwidth = dcn351_validate_bandwidth,
 	.calculate_wm_and_dlg = NULL,
 	.update_soc_for_wm_a = dcn31_update_soc_for_wm_a,
-	.populate_dml_pipes = dcn351_populate_dml_pipes_from_context_fpu,
+	.populate_dml_pipes = populate_dml_pipes_from_context_fpu,
 	.acquire_free_pipe_as_secondary_dpp_pipe = dcn20_acquire_free_pipe_for_layer,
 	.release_pipe = dcn20_release_pipe,
 	.add_stream_to_ctx = dcn30_add_stream_to_ctx,

drivers/gpu/drm/amd/display/dc/resource/dcn36/dcn36_resource.c

@@ -1734,6 +1734,20 @@ static enum dc_status dcn35_validate_bandwidth(struct dc *dc,
 }
 
 
+static int populate_dml_pipes_from_context_fpu(struct dc *dc,
+					       struct dc_state *context,
+					       display_e2e_pipe_params_st *pipes,
+					       enum dc_validate_mode validate_mode)
+{
+	int ret;
+
+	DC_FP_START();
+	ret = dcn35_populate_dml_pipes_from_context_fpu(dc, context, pipes, validate_mode);
+	DC_FP_END();
+
+	return ret;
+}
+
 static struct resource_funcs dcn36_res_pool_funcs = {
 	.destroy = dcn36_destroy_resource_pool,
 	.link_enc_create = dcn35_link_encoder_create,
@@ -1744,7 +1758,7 @@ static struct resource_funcs dcn36_res_pool_funcs = {
 	.validate_bandwidth = dcn35_validate_bandwidth,
 	.calculate_wm_and_dlg = NULL,
 	.update_soc_for_wm_a = dcn31_update_soc_for_wm_a,
-	.populate_dml_pipes = dcn35_populate_dml_pipes_from_context_fpu,
+	.populate_dml_pipes = populate_dml_pipes_from_context_fpu,
 	.acquire_free_pipe_as_secondary_dpp_pipe = dcn20_acquire_free_pipe_for_layer,
 	.release_pipe = dcn20_release_pipe,
 	.add_stream_to_ctx = dcn30_add_stream_to_ctx,