From 3a47280b768748992ee34bd52c394c60b2845af3 Mon Sep 17 00:00:00 2001 From: Ankit Nautiyal Date: Mon, 14 Apr 2025 08:12:56 +0530 Subject: [PATCH 01/16] drm/i915/dp: Check for HAS_DSC_3ENGINES while configuring DSC slices DSC 12 slices configuration is used for some specific cases with Ultrajoiner. This can be supported only when each of the 4 joined pipes have 3 DSC engines each. Add the missing check for 3 DSC engines support before using 3 DSC slices per pipe. Fixes: be7f5fcdf4a0 ("drm/i915/dp: Enable 3 DSC engines for 12 slices") Cc: Ankit Nautiyal Cc: Suraj Kandpal Cc: # v6.14+ Signed-off-by: Ankit Nautiyal Reviewed-by: Suraj Kandpal Link: https://lore.kernel.org/r/20250414024256.2782702-3-ankit.k.nautiyal@intel.com (cherry picked from commit da9b1c61e7f7b327dd70c5f073ba04d419a55ef8) Signed-off-by: Jani Nikula --- drivers/gpu/drm/i915/display/intel_dp.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c index 9476aaa91900..392c3653d0d7 100644 --- a/drivers/gpu/drm/i915/display/intel_dp.c +++ b/drivers/gpu/drm/i915/display/intel_dp.c @@ -1050,10 +1050,11 @@ u8 intel_dp_dsc_get_slice_count(const struct intel_connector *connector, u8 test_slice_count = valid_dsc_slicecount[i] * num_joined_pipes; /* - * 3 DSC Slices per pipe need 3 DSC engines, - * which is supported only with Ultrajoiner. + * 3 DSC Slices per pipe need 3 DSC engines, which is supported only + * with Ultrajoiner only for some platforms. */ - if (valid_dsc_slicecount[i] == 3 && num_joined_pipes != 4) + if (valid_dsc_slicecount[i] == 3 && + (!HAS_DSC_3ENGINES(display) || num_joined_pipes != 4)) continue; if (test_slice_count > -- 2.51.0 From 447fab30955cf7dba7dd563f42b67c02284860c8 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Christian=20K=C3=B6nig?= Date: Fri, 28 Mar 2025 18:58:17 +0100 Subject: [PATCH 02/16] drm/amdgpu: use a dummy owner for sysfs triggered cleaner shaders v4 MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Otherwise triggering sysfs multiple times without other submissions in between only runs the shader once. v2: add some comment v3: re-add missing cast v4: squash in semicolon fix Signed-off-by: Christian König Reviewed-by: Srinivasan Shanmugam Signed-off-by: Alex Deucher (cherry picked from commit 8b2ae7d492675e8af8902f103364bef59382b935) --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 72af5e5a894a..cf2df7790077 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -1438,9 +1438,11 @@ static int amdgpu_gfx_run_cleaner_shader_job(struct amdgpu_ring *ring) struct amdgpu_device *adev = ring->adev; struct drm_gpu_scheduler *sched = &ring->sched; struct drm_sched_entity entity; + static atomic_t counter; struct dma_fence *f; struct amdgpu_job *job; struct amdgpu_ib *ib; + void *owner; int i, r; /* Initialize the scheduler entity */ @@ -1451,9 +1453,15 @@ static int amdgpu_gfx_run_cleaner_shader_job(struct amdgpu_ring *ring) goto err; } - r = amdgpu_job_alloc_with_ib(ring->adev, &entity, NULL, - 64, 0, - &job); + /* + * Use some unique dummy value as the owner to make sure we execute + * the cleaner shader on each submission. The value just need to change + * for each submission and is otherwise meaningless. + */ + owner = (void *)(unsigned long)atomic_inc_return(&counter); + + r = amdgpu_job_alloc_with_ib(ring->adev, &entity, owner, + 64, 0, &job); if (r) goto err; -- 2.51.0 From 1657793def101dac7c9d3b2250391f6a3dd934ba Mon Sep 17 00:00:00 2001 From: Mario Limonciello Date: Tue, 8 Apr 2025 13:09:57 -0500 Subject: [PATCH 03/16] drm/amd: Forbid suspending into non-default suspend states On systems that default to 'deep' some userspace software likes to try to suspend in 'deep' first. If there is a failure for any reason (such as -ENOMEM) the failure is ignored and then it will try to use 's2idle' as a fallback. This fails, but more importantly it leads to graphical problems. Forbid this behavior and only allow suspending in the last state supported by the system. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4093 Acked-by: Alex Deucher Link: https://lore.kernel.org/r/20250408180957.4027643-1-superm1@kernel.org Signed-off-by: Mario Limonciello Signed-off-by: Alex Deucher (cherry picked from commit 2aabd44aa8a3c08da3d43264c168370f6da5e81d) --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 14 +++++++++++++- 2 files changed, 14 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 2c04ae133848..ef6e78224fdf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1123,6 +1123,7 @@ struct amdgpu_device { bool in_s3; bool in_s4; bool in_s0ix; + suspend_state_t last_suspend_state; enum pp_mp1_state mp1_state; struct amdgpu_doorbell_index doorbell_index; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 26bf896f1444..24ee4710f807 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -2548,8 +2548,20 @@ static int amdgpu_pmops_suspend(struct device *dev) adev->in_s0ix = true; else if (amdgpu_acpi_is_s3_active(adev)) adev->in_s3 = true; - if (!adev->in_s0ix && !adev->in_s3) + if (!adev->in_s0ix && !adev->in_s3) { + /* don't allow going deep first time followed by s2idle the next time */ + if (adev->last_suspend_state != PM_SUSPEND_ON && + adev->last_suspend_state != pm_suspend_target_state) { + drm_err_once(drm_dev, "Unsupported suspend state %d\n", + pm_suspend_target_state); + return -EINVAL; + } return 0; + } + + /* cache the state last used for suspend */ + adev->last_suspend_state = pm_suspend_target_state; + return amdgpu_device_suspend(drm_dev, true); } -- 2.51.0 From e7afa85a0d0eba5bf2c0a446ff622ebdbc9812d6 Mon Sep 17 00:00:00 2001 From: ZhenGuo Yin Date: Tue, 8 Apr 2025 16:18:28 +0800 Subject: [PATCH 04/16] drm/amdgpu: fix warning of drm_mm_clean MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Kernel doorbell BOs needs to be freed before ttm_fini. Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4145 Fixes: 54c30d2a8def ("drm/amdgpu: create kernel doorbell pages") Acked-by: Alex Deucher Reviewed-by: Christian König Signed-off-by: ZhenGuo Yin Signed-off-by: Alex Deucher (cherry picked from commit 39938a8ed979e398faa3791a47e282c82bcc6f04) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index b34b915203f2..7f354cd532dc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -3510,6 +3510,7 @@ static int amdgpu_device_ip_fini(struct amdgpu_device *adev) amdgpu_device_mem_scratch_fini(adev); amdgpu_ib_pool_fini(adev); amdgpu_seq64_fini(adev); + amdgpu_doorbell_fini(adev); } if (adev->ip_blocks[i].version->funcs->sw_fini) { r = adev->ip_blocks[i].version->funcs->sw_fini(&adev->ip_blocks[i]); @@ -4858,7 +4859,6 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev) iounmap(adev->rmmio); adev->rmmio = NULL; - amdgpu_doorbell_fini(adev); drm_dev_exit(idx); } -- 2.51.0 From 2036be31741b00f030530381643a8b35a5a42b5c Mon Sep 17 00:00:00 2001 From: David Rosca Date: Mon, 7 Apr 2025 13:12:11 +0200 Subject: [PATCH 05/16] drm/amdgpu: Add back JPEG to video caps for carrizo and newer JPEG is not supported on Vega only. Fixes: 0a6e7b06bdbe ("drm/amdgpu: Remove JPEG from vega and carrizo video caps") Signed-off-by: David Rosca Reviewed-by: Leo Liu Signed-off-by: Alex Deucher (cherry picked from commit 0f4dfe86fe922c37bcec99dce80a15b4d5d4726d) Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/vi.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index 86d8bc10d90a..9b3510e53112 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -239,6 +239,13 @@ static const struct amdgpu_video_codec_info cz_video_codecs_decode_array[] = .max_pixels_per_frame = 4096 * 4096, .max_level = 186, }, + { + .codec_type = AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, + .max_width = 4096, + .max_height = 4096, + .max_pixels_per_frame = 4096 * 4096, + .max_level = 0, + }, }; static const struct amdgpu_video_codecs cz_video_codecs_decode = -- 2.51.0 From cd9e6d6fdd2de60bfb4672387c17d4ee7157cf8e Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Tue, 8 Apr 2025 21:27:15 -0400 Subject: [PATCH 06/16] drm/amd/display/dml2: use vzalloc rather than kzalloc The structures are large and they do not require contiguous memory so use vzalloc. Fixes: 70839da63605 ("drm/amd/display: Add new DCN401 sources") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4126 Cc: Aurabindo Pillai Reviewed-by: Aurabindo Pillai Signed-off-by: Alex Deucher (cherry picked from commit 20c50a9a793300a1fc82f3ddd0e3c68f8213fbef) Cc: stable@vger.kernel.org --- .../gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c | 11 ++++++----- drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c | 6 ++++-- 2 files changed, 10 insertions(+), 7 deletions(-) diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c index 94e99e540691..5d16f36ec95c 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_wrapper.c @@ -2,6 +2,7 @@ // // Copyright 2024 Advanced Micro Devices, Inc. +#include #include "dml2_internal_types.h" #include "dml_top.h" @@ -13,11 +14,11 @@ static bool dml21_allocate_memory(struct dml2_context **dml_ctx) { - *dml_ctx = kzalloc(sizeof(struct dml2_context), GFP_KERNEL); + *dml_ctx = vzalloc(sizeof(struct dml2_context)); if (!(*dml_ctx)) return false; - (*dml_ctx)->v21.dml_init.dml2_instance = kzalloc(sizeof(struct dml2_instance), GFP_KERNEL); + (*dml_ctx)->v21.dml_init.dml2_instance = vzalloc(sizeof(struct dml2_instance)); if (!((*dml_ctx)->v21.dml_init.dml2_instance)) return false; @@ -27,7 +28,7 @@ static bool dml21_allocate_memory(struct dml2_context **dml_ctx) (*dml_ctx)->v21.mode_support.display_config = &(*dml_ctx)->v21.display_config; (*dml_ctx)->v21.mode_programming.display_config = (*dml_ctx)->v21.mode_support.display_config; - (*dml_ctx)->v21.mode_programming.programming = kzalloc(sizeof(struct dml2_display_cfg_programming), GFP_KERNEL); + (*dml_ctx)->v21.mode_programming.programming = vzalloc(sizeof(struct dml2_display_cfg_programming)); if (!((*dml_ctx)->v21.mode_programming.programming)) return false; @@ -115,8 +116,8 @@ bool dml21_create(const struct dc *in_dc, struct dml2_context **dml_ctx, const s void dml21_destroy(struct dml2_context *dml2) { - kfree(dml2->v21.dml_init.dml2_instance); - kfree(dml2->v21.mode_programming.programming); + vfree(dml2->v21.dml_init.dml2_instance); + vfree(dml2->v21.mode_programming.programming); } static void dml21_calculate_rq_and_dlg_params(const struct dc *dc, struct dc_state *context, struct resource_context *out_new_hw_state, diff --git a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c index f549a778f6f1..e89571874185 100644 --- a/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c +++ b/drivers/gpu/drm/amd/display/dc/dml2/dml2_wrapper.c @@ -24,6 +24,8 @@ * */ +#include + #include "display_mode_core.h" #include "dml2_internal_types.h" #include "dml2_utils.h" @@ -747,7 +749,7 @@ bool dml2_validate(const struct dc *in_dc, struct dc_state *context, struct dml2 static inline struct dml2_context *dml2_allocate_memory(void) { - return (struct dml2_context *) kzalloc(sizeof(struct dml2_context), GFP_KERNEL); + return (struct dml2_context *) vzalloc(sizeof(struct dml2_context)); } static void dml2_init(const struct dc *in_dc, const struct dml2_configuration_options *config, struct dml2_context **dml2) @@ -821,7 +823,7 @@ void dml2_destroy(struct dml2_context *dml2) if (dml2->architecture == dml2_architecture_21) dml21_destroy(dml2); - kfree(dml2); + vfree(dml2); } void dml2_extract_dram_and_fclk_change_support(struct dml2_context *dml2, -- 2.51.0 From c235a7132258ac30bd43d228222986022d21f5de Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Fri, 11 Apr 2025 17:40:26 +0530 Subject: [PATCH 07/16] drm/amdgpu: Use the right function for hdp flush There are a few prechecks made before HDP flush like a flush is not required on APU bare metal. Using hdp callback directly bypasses those checks. Use amdgpu_device_flush_hdp which takes care of prechecks. Signed-off-by: Lijo Lazar Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher (cherry picked from commit 1d9bff4cf8c53d33ee2ff1b11574e5da739ce61c) --- drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 8 ++++---- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 12 ++++++------ drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 6 +++--- drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/psp_v11_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/psp_v13_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/psp_v14_0.c | 2 +- 10 files changed, 23 insertions(+), 23 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c index a63ce747863f..23e6a05359c2 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c @@ -6114,7 +6114,7 @@ static int gfx_v10_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0); @@ -6192,7 +6192,7 @@ static int gfx_v10_0_cp_gfx_load_ce_microcode(struct amdgpu_device *adev) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_CE_IC_BASE_CNTL, VMID, 0); @@ -6269,7 +6269,7 @@ static int gfx_v10_0_cp_gfx_load_me_microcode(struct amdgpu_device *adev) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0); @@ -6644,7 +6644,7 @@ static int gfx_v10_0_cp_compute_load_microcode(struct amdgpu_device *adev) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index d57db42f9536..2a5c2a1ae3c7 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -2428,7 +2428,7 @@ static int gfx_v11_0_config_me_cache(struct amdgpu_device *adev, uint64_t addr) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0); @@ -2472,7 +2472,7 @@ static int gfx_v11_0_config_pfp_cache(struct amdgpu_device *adev, uint64_t addr) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0); @@ -2517,7 +2517,7 @@ static int gfx_v11_0_config_mec_cache(struct amdgpu_device *adev, uint64_t addr) } if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL); tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0); @@ -3153,7 +3153,7 @@ static int gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(struct amdgpu_device *adev) amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_data_obj); if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO, lower_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr)); @@ -3371,7 +3371,7 @@ static int gfx_v11_0_cp_gfx_load_me_microcode_rs64(struct amdgpu_device *adev) amdgpu_bo_unreserve(adev->gfx.me.me_fw_data_obj); if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO, lower_32_bits(adev->gfx.me.me_fw_gpu_addr)); @@ -4541,7 +4541,7 @@ static int gfx_v11_0_gfxhub_enable(struct amdgpu_device *adev) if (r) return r; - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ? false : true; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c index e7b58e470292..62a257a4a3e9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c @@ -2324,7 +2324,7 @@ static int gfx_v12_0_cp_gfx_load_pfp_microcode_rs64(struct amdgpu_device *adev) amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_data_obj); if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO, lower_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr)); @@ -2468,7 +2468,7 @@ static int gfx_v12_0_cp_gfx_load_me_microcode_rs64(struct amdgpu_device *adev) amdgpu_bo_unreserve(adev->gfx.me.me_fw_data_obj); if (amdgpu_emu_mode == 1) - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO, lower_32_bits(adev->gfx.me.me_fw_gpu_addr)); @@ -3426,7 +3426,7 @@ static int gfx_v12_0_gfxhub_enable(struct amdgpu_device *adev) if (r) return r; - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ? false : true; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c index 95d894a231fc..809b3a882d0d 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c @@ -268,7 +268,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, ack = hub->vm_inv_eng0_ack + hub->eng_distance * eng; /* flush hdp cache */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); /* This is necessary for SRIOV as well as for GFXOFF to function * properly under bare metal @@ -969,7 +969,7 @@ static int gmc_v10_0_gart_enable(struct amdgpu_device *adev) adev->hdp.funcs->init_registers(adev); /* Flush HDP after it is initialized */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ? false : true; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c index ad099f136f84..e74e26b6a4f2 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c @@ -229,7 +229,7 @@ static void gmc_v11_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, ack = hub->vm_inv_eng0_ack + hub->eng_distance * eng; /* flush hdp cache */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); /* This is necessary for SRIOV as well as for GFXOFF to function * properly under bare metal @@ -899,7 +899,7 @@ static int gmc_v11_0_gart_enable(struct amdgpu_device *adev) return r; /* Flush HDP after it is initialized */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ? false : true; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c index 05c026d0b0d9..c6f290704d47 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c @@ -297,7 +297,7 @@ static void gmc_v12_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, return; /* flush hdp cache */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); /* This is necessary for SRIOV as well as for GFXOFF to function * properly under bare metal @@ -881,7 +881,7 @@ static int gmc_v12_0_gart_enable(struct amdgpu_device *adev) return r; /* Flush HDP after it is initialized */ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ? false : true; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index 783e0c3b86b4..5effe8327d29 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -2435,7 +2435,7 @@ static int gmc_v9_0_hw_init(struct amdgpu_ip_block *ip_block) adev->hdp.funcs->init_registers(adev); /* After HDP is initialized, flush HDP.*/ - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) value = false; diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c index bb5dfc410a66..215543575f47 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v11_0.c @@ -533,7 +533,7 @@ static int psp_v11_0_memory_training(struct psp_context *psp, uint32_t ops) } memcpy_toio(adev->mman.aper_base_kaddr, buf, sz); - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); vfree(buf); drm_dev_exit(idx); } else { diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v13_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v13_0.c index cc621064610f..afdf8ce3b4c5 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v13_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v13_0.c @@ -610,7 +610,7 @@ static int psp_v13_0_memory_training(struct psp_context *psp, uint32_t ops) } memcpy_toio(adev->mman.aper_base_kaddr, buf, sz); - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); vfree(buf); drm_dev_exit(idx); } else { diff --git a/drivers/gpu/drm/amd/amdgpu/psp_v14_0.c b/drivers/gpu/drm/amd/amdgpu/psp_v14_0.c index 7c49c3f3c388..256288c6cd78 100644 --- a/drivers/gpu/drm/amd/amdgpu/psp_v14_0.c +++ b/drivers/gpu/drm/amd/amdgpu/psp_v14_0.c @@ -498,7 +498,7 @@ static int psp_v14_0_memory_training(struct psp_context *psp, uint32_t ops) } memcpy_toio(adev->mman.aper_base_kaddr, buf, sz); - adev->hdp.funcs->flush_hdp(adev, NULL); + amdgpu_device_flush_hdp(adev, NULL); vfree(buf); drm_dev_exit(idx); } else { -- 2.51.0 From 8260731ccad0451207b45844bb66eb161a209218 Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Wed, 16 Apr 2025 08:57:45 +0200 Subject: [PATCH 08/16] drm/gem: Internally test import_attach for imported objects MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Test struct drm_gem_object.import_attach to detect imported objects. During object clenanup, the dma_buf field might be NULL. Testing it in an object's free callback then incorrectly does a cleanup as for native objects. Happens for calls to drm_mode_destroy_dumb_ioctl() that clears the dma_buf field in drm_gem_object_exported_dma_buf_free(). v3: - only test for import_attach (Boris) v2: - use import_attach.dmabuf instead of dma_buf (Christian) Signed-off-by: Thomas Zimmermann Fixes: b57aa47d39e9 ("drm/gem: Test for imported GEM buffers with helper") Reported-by: Andy Yan Closes: https://lore.kernel.org/dri-devel/38d09d34.4354.196379aa560.Coremail.andyshrk@163.com/ Tested-by: Andy Yan Cc: Thomas Zimmermann Cc: Anusha Srivatsa Cc: Christian König Cc: Maarten Lankhorst Cc: Maxime Ripard Cc: David Airlie Cc: Simona Vetter Cc: Sumit Semwal Cc: "Christian König" Cc: dri-devel@lists.freedesktop.org Cc: linux-media@vger.kernel.org Cc: linaro-mm-sig@lists.linaro.org Reviewed-by: Boris Brezillon Reviewed-by: Simona Vetter Link: https://lore.kernel.org/r/20250416065820.26076-1-tzimmermann@suse.de --- include/drm/drm_gem.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h index 2bf893eabb4b..bcd54020d6ba 100644 --- a/include/drm/drm_gem.h +++ b/include/drm/drm_gem.h @@ -585,8 +585,7 @@ static inline bool drm_gem_object_is_shared_for_memory_stats(struct drm_gem_obje */ static inline bool drm_gem_is_imported(const struct drm_gem_object *obj) { - /* The dma-buf's priv field points to the original GEM object. */ - return obj->dma_buf && (obj->dma_buf->priv != obj); + return !!obj->import_attach; } #ifdef CONFIG_LOCKDEP -- 2.51.0 From 76c332d119f9048c6e16b52359f401510f18b2ff Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Wed, 16 Apr 2025 10:38:05 +0200 Subject: [PATCH 09/16] drm/mgag200: Fix value in register MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Fix an off-by-one error when setting the vblanking start in . Commit d6460bd52c27 ("drm/mgag200: Add dedicated variables for blanking fields") switched the value from crtc_vdisplay to crtc_vblank_start, which DRM helpers copy from the former. The commit missed to subtract one though. Reported-by: Wakko Warner Closes: https://lore.kernel.org/dri-devel/CAMwc25rKPKooaSp85zDq2eh-9q4UPZD=RqSDBRp1fAagDnmRmA@mail.gmail.com/ Reported-by: Сергей Closes: https://lore.kernel.org/all/5b193b75-40b1-4342-a16a-ae9fc62f245a@gmail.com/ Closes: https://bbs.archlinux.org/viewtopic.php?id=303819 Signed-off-by: Thomas Zimmermann Fixes: d6460bd52c27 ("drm/mgag200: Add dedicated variables for blanking fields") Cc: Thomas Zimmermann Cc: Jocelyn Falempe Cc: Dave Airlie Cc: dri-devel@lists.freedesktop.org Cc: # v6.12+ Reviewed-by: Jocelyn Falempe Tested-by: Wakko Warner Link: https://lore.kernel.org/r/20250416083847.51764-1-tzimmermann@suse.de --- drivers/gpu/drm/mgag200/mgag200_mode.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/gpu/drm/mgag200/mgag200_mode.c b/drivers/gpu/drm/mgag200/mgag200_mode.c index fb71658c3117..6067d08aeee3 100644 --- a/drivers/gpu/drm/mgag200/mgag200_mode.c +++ b/drivers/gpu/drm/mgag200/mgag200_mode.c @@ -223,7 +223,7 @@ void mgag200_set_mode_regs(struct mga_device *mdev, const struct drm_display_mod vsyncstr = mode->crtc_vsync_start - 1; vsyncend = mode->crtc_vsync_end - 1; vtotal = mode->crtc_vtotal - 2; - vblkstr = mode->crtc_vblank_start; + vblkstr = mode->crtc_vblank_start - 1; vblkend = vtotal + 1; linecomp = vdispend; -- 2.51.0 From 6405f5b70b1c240ffddef01c7a140498f47d4fe7 Mon Sep 17 00:00:00 2001 From: Lucas De Marchi Date: Wed, 9 Apr 2025 21:59:34 -0700 Subject: [PATCH 10/16] drm/xe: Set LRC addresses before guc load The metadata saved in the ADS is read by GuC when it's initialized. Saving the addresses to the LRCs when they are populated is too late as GuC will keep using the old ones. This was causing GuC to use the RCS LRC for any engine class. It's not a big problem on a Linux-only scenario since the they are used by GuC only on media engines when the watchdog is triggered. However, in a virtualization scenario with Windows as the VF, it causes the wrong LRCs to be loaded as the watchdog is used for all engines. Fix it by letting guc_golden_lrc_init() initialize the metadata, like other *_init() functions, and later guc_golden_lrc_populate() to copy the LRCs to the right places. The former is called before the second GuC load, while the latter is called after LRCs have been recorded. Cc: Chee Yin Wong Cc: John Harrison Cc: Matt Roper Cc: Matthew Brost Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs") Cc: # v6.11+ Reviewed-by: Matthew Brost Tested-by: Chee Yin Wong Link: https://lore.kernel.org/r/20250409-fix-guc-ads-v1-1-494135f7a5d0@intel.com Signed-off-by: Lucas De Marchi (cherry picked from commit c31a0b6402d15b530514eee9925adfcb8cfbb1c9) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_guc_ads.c | 75 ++++++++++++++++++++------------- 1 file changed, 45 insertions(+), 30 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c index e7c9e095a19f..7031542a70ce 100644 --- a/drivers/gpu/drm/xe/xe_guc_ads.c +++ b/drivers/gpu/drm/xe/xe_guc_ads.c @@ -490,24 +490,52 @@ static void fill_engine_enable_masks(struct xe_gt *gt, engine_enable_mask(gt, XE_ENGINE_CLASS_OTHER)); } -static void guc_prep_golden_lrc_null(struct xe_guc_ads *ads) +/* + * Write the offsets corresponding to the golden LRCs. The actual data is + * populated later by guc_golden_lrc_populate() + */ +static void guc_golden_lrc_init(struct xe_guc_ads *ads) { struct xe_device *xe = ads_to_xe(ads); + struct xe_gt *gt = ads_to_gt(ads); struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads), offsetof(struct __guc_ads_blob, system_info)); - u8 guc_class; + size_t alloc_size, real_size; + u32 addr_ggtt, offset; + int class; + + offset = guc_ads_golden_lrc_offset(ads); + addr_ggtt = xe_bo_ggtt_addr(ads->bo) + offset; + + for (class = 0; class < XE_ENGINE_CLASS_MAX; ++class) { + u8 guc_class; + + guc_class = xe_engine_class_to_guc_class(class); - for (guc_class = 0; guc_class <= GUC_MAX_ENGINE_CLASSES; ++guc_class) { if (!info_map_read(xe, &info_map, engine_enabled_masks[guc_class])) continue; + real_size = xe_gt_lrc_size(gt, class); + alloc_size = PAGE_ALIGN(real_size); + + /* + * This interface is slightly confusing. We need to pass the + * base address of the full golden context and the size of just + * the engine state, which is the section of the context image + * that starts after the execlists LRC registers. This is + * required to allow the GuC to restore just the engine state + * when a watchdog reset occurs. + * We calculate the engine state size by removing the size of + * what comes before it in the context image (which is identical + * on all engines). + */ ads_blob_write(ads, ads.eng_state_size[guc_class], - guc_ads_golden_lrc_size(ads) - - xe_lrc_skip_size(xe)); + real_size - xe_lrc_skip_size(xe)); ads_blob_write(ads, ads.golden_context_lrca[guc_class], - xe_bo_ggtt_addr(ads->bo) + - guc_ads_golden_lrc_offset(ads)); + addr_ggtt); + + addr_ggtt += alloc_size; } } @@ -857,7 +885,7 @@ void xe_guc_ads_populate_minimal(struct xe_guc_ads *ads) xe_map_memset(ads_to_xe(ads), ads_to_map(ads), 0, 0, ads->bo->size); guc_policies_init(ads); - guc_prep_golden_lrc_null(ads); + guc_golden_lrc_init(ads); guc_mapping_table_init_invalid(gt, &info_map); guc_doorbell_init(ads); @@ -883,7 +911,7 @@ void xe_guc_ads_populate(struct xe_guc_ads *ads) guc_policies_init(ads); fill_engine_enable_masks(gt, &info_map); guc_mmio_reg_state_init(ads); - guc_prep_golden_lrc_null(ads); + guc_golden_lrc_init(ads); guc_mapping_table_init(gt, &info_map); guc_capture_prep_lists(ads); guc_doorbell_init(ads); @@ -903,18 +931,22 @@ void xe_guc_ads_populate(struct xe_guc_ads *ads) guc_ads_private_data_offset(ads)); } -static void guc_populate_golden_lrc(struct xe_guc_ads *ads) +/* + * After the golden LRC's are recorded for each engine class by the first + * submission, copy them to the ADS, as initialized earlier by + * guc_golden_lrc_init(). + */ +static void guc_golden_lrc_populate(struct xe_guc_ads *ads) { struct xe_device *xe = ads_to_xe(ads); struct xe_gt *gt = ads_to_gt(ads); struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads), offsetof(struct __guc_ads_blob, system_info)); size_t total_size = 0, alloc_size, real_size; - u32 addr_ggtt, offset; + u32 offset; int class; offset = guc_ads_golden_lrc_offset(ads); - addr_ggtt = xe_bo_ggtt_addr(ads->bo) + offset; for (class = 0; class < XE_ENGINE_CLASS_MAX; ++class) { u8 guc_class; @@ -931,26 +963,9 @@ static void guc_populate_golden_lrc(struct xe_guc_ads *ads) alloc_size = PAGE_ALIGN(real_size); total_size += alloc_size; - /* - * This interface is slightly confusing. We need to pass the - * base address of the full golden context and the size of just - * the engine state, which is the section of the context image - * that starts after the execlists LRC registers. This is - * required to allow the GuC to restore just the engine state - * when a watchdog reset occurs. - * We calculate the engine state size by removing the size of - * what comes before it in the context image (which is identical - * on all engines). - */ - ads_blob_write(ads, ads.eng_state_size[guc_class], - real_size - xe_lrc_skip_size(xe)); - ads_blob_write(ads, ads.golden_context_lrca[guc_class], - addr_ggtt); - xe_map_memcpy_to(xe, ads_to_map(ads), offset, gt->default_lrc[class], real_size); - addr_ggtt += alloc_size; offset += alloc_size; } @@ -959,7 +974,7 @@ static void guc_populate_golden_lrc(struct xe_guc_ads *ads) void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads) { - guc_populate_golden_lrc(ads); + guc_golden_lrc_populate(ads); } static int guc_ads_action_update_policies(struct xe_guc_ads *ads, u32 policy_offset) -- 2.51.0 From 2577b202458cddff85cc154b1fe7f313e0d1f418 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Mon, 14 Apr 2025 14:25:40 +0100 Subject: [PATCH 11/16] drm/xe/userptr: fix notifier vs folio deadlock MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit User is reporting what smells like notifier vs folio deadlock, where migrate_pages_batch() on core kernel side is holding folio lock(s) and then interacting with the mappings of it, however those mappings are tied to some userptr, which means calling into the notifier callback and grabbing the notifier lock. With perfect timing it looks possible that the pages we pulled from the hmm fault can get sniped by migrate_pages_batch() at the same time that we are holding the notifier lock to mark the pages as accessed/dirty, but at this point we also want to grab the folio locks(s) to mark them as dirty, but if they are contended from notifier/migrate_pages_batch side then we deadlock since folio lock won't be dropped until we drop the notifier lock. Fortunately the mark_page_accessed/dirty is not really needed in the first place it seems and should have already been done by hmm fault, so just remove it. Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4765 Fixes: 0a98219bcc96 ("drm/xe/hmm: Don't dereference struct page pointers without notifier lock") Signed-off-by: Matthew Auld Cc: Thomas Hellström Cc: Matthew Brost Cc: # v6.10+ Reviewed-by: Thomas Hellström Reviewed-by: Matthew Brost Link: https://lore.kernel.org/r/20250414132539.26654-2-matthew.auld@intel.com (cherry picked from commit bd7c0cb695e87c0e43247be8196b4919edbe0e85) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_hmm.c | 24 ------------------------ 1 file changed, 24 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_hmm.c b/drivers/gpu/drm/xe/xe_hmm.c index c3cc0fa105e8..57b71956ddf4 100644 --- a/drivers/gpu/drm/xe/xe_hmm.c +++ b/drivers/gpu/drm/xe/xe_hmm.c @@ -19,29 +19,6 @@ static u64 xe_npages_in_range(unsigned long start, unsigned long end) return (end - start) >> PAGE_SHIFT; } -/** - * xe_mark_range_accessed() - mark a range is accessed, so core mm - * have such information for memory eviction or write back to - * hard disk - * @range: the range to mark - * @write: if write to this range, we mark pages in this range - * as dirty - */ -static void xe_mark_range_accessed(struct hmm_range *range, bool write) -{ - struct page *page; - u64 i, npages; - - npages = xe_npages_in_range(range->start, range->end); - for (i = 0; i < npages; i++) { - page = hmm_pfn_to_page(range->hmm_pfns[i]); - if (write) - set_page_dirty_lock(page); - - mark_page_accessed(page); - } -} - static int xe_alloc_sg(struct xe_device *xe, struct sg_table *st, struct hmm_range *range, struct rw_semaphore *notifier_sem) { @@ -331,7 +308,6 @@ int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma, if (ret) goto out_unlock; - xe_mark_range_accessed(&hmm_range, write); userptr->sg = &userptr->sgt; xe_hmm_userptr_set_mapped(uvma); userptr->notifier_seq = hmm_range.notifier_seq; -- 2.51.0 From 25583ad42d091819157832e894179200ba8b54ee Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Thu, 10 Apr 2025 17:27:17 +0100 Subject: [PATCH 12/16] drm/xe/dma_buf: stop relying on placement in unmap MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit The is_vram() is checking the current placement, however if we consider exported VRAM with dynamic dma-buf, it looks possible for the xe driver to async evict the memory, notifying the importer, however importer does not have to call unmap_attachment() immediately, but rather just as "soon as possible", like when the dma-resv idles. Following from this we would then pipeline the move, attaching the fence to the manager, and then update the current placement. But when the unmap_attachment() runs at some later point we might see that is_vram() is now false, and take the complete wrong path when dma-unmapping the sg, leading to explosions. To fix this check if the sgl was mapping a struct page. v2: - The attachment can be mapped multiple times it seems, so we can't really rely on encoding something in the attachment->priv. Instead see if the page_link has an encoded struct page. For vram we expect this to be NULL. Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4563 Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs") Signed-off-by: Matthew Auld Cc: Thomas Hellström Cc: Matthew Brost Cc: # v6.8+ Acked-by: Christian König Link: https://lore.kernel.org/r/20250410162716.159403-2-matthew.auld@intel.com (cherry picked from commit d755887f8e5a2a18e15e6632a5193e5feea18499) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_dma_buf.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_dma_buf.c b/drivers/gpu/drm/xe/xe_dma_buf.c index f67803e15a0e..f7a20264ea33 100644 --- a/drivers/gpu/drm/xe/xe_dma_buf.c +++ b/drivers/gpu/drm/xe/xe_dma_buf.c @@ -145,10 +145,7 @@ static void xe_dma_buf_unmap(struct dma_buf_attachment *attach, struct sg_table *sgt, enum dma_data_direction dir) { - struct dma_buf *dma_buf = attach->dmabuf; - struct xe_bo *bo = gem_to_xe_bo(dma_buf->priv); - - if (!xe_bo_is_vram(bo)) { + if (sg_page(sgt->sgl)) { dma_unmap_sgtable(attach->dev, sgt, dir, 0); sg_free_table(sgt); kfree(sgt); -- 2.51.0 From 78600df8f593407a3df2d6c48c35d0ad203d7fb4 Mon Sep 17 00:00:00 2001 From: Daniele Ceraolo Spurio Date: Wed, 16 Apr 2025 13:16:22 -0700 Subject: [PATCH 13/16] drm/xe/pxp: do not queue unneeded terminations from debugfs The PXP terminate debugfs currently unconditionally simulates a termination, no matter what the HW status is. This is unneeded if PXP is not in use and can cause errors if the HW init hasn't completed yet. To solve these issues, we can simply limit the terminations to the cases where PXP is fully initialized and in use. v2: s/pxp_status/ready/ to avoid confusion with pxp->status (John) Fixes: 385a8015b214 ("drm/xe/pxp: Add PXP debugfs support") Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4749 Signed-off-by: Daniele Ceraolo Spurio Cc: John Harrison Reviewed-by: John Harrison Link: https://lore.kernel.org/r/20250416201622.1295369-1-daniele.ceraolospurio@intel.com (cherry picked from commit ba1f62a0cac84757ca35f4217e3cd3a2654233ae) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_pxp_debugfs.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_pxp_debugfs.c b/drivers/gpu/drm/xe/xe_pxp_debugfs.c index ccfbacf08efc..525a2f6bb076 100644 --- a/drivers/gpu/drm/xe/xe_pxp_debugfs.c +++ b/drivers/gpu/drm/xe/xe_pxp_debugfs.c @@ -66,9 +66,18 @@ static int pxp_terminate(struct seq_file *m, void *data) { struct xe_pxp *pxp = node_to_pxp(m->private); struct drm_printer p = drm_seq_file_printer(m); + int ready = xe_pxp_get_readiness_status(pxp); - if (!xe_pxp_is_enabled(pxp)) - return -ENODEV; + if (ready < 0) + return ready; /* disabled or error occurred */ + else if (!ready) + return -EBUSY; /* init still in progress */ + + /* no need for a termination if PXP is not active */ + if (pxp->status != XE_PXP_ACTIVE) { + drm_printf(&p, "PXP not active\n"); + return 0; + } /* simulate a termination interrupt */ spin_lock_irq(&pxp->xe->irq.lock); -- 2.51.0 From 9d78f02503227d3554d26cf8ca73276105c98f3e Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Mon, 17 Mar 2025 08:00:06 -0700 Subject: [PATCH 14/16] drm/msm/a6xx+: Don't let IB_SIZE overflow IB_SIZE is only b0..b19. Starting with a6xx gen3, additional fields were added above the IB_SIZE. Accidentially setting them can cause badness. Fix this by properly defining the CP_INDIRECT_BUFFER packet and using the generated builder macro to ensure unintended bits are not set. v2: add missing type attribute for IB_BASE v3: fix offset attribute in xml Reported-by: Connor Abbott Fixes: a83366ef19ea ("drm/msm/a6xx: add A640/A650 to gpulist") Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/643396/ --- drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 8 ++++---- drivers/gpu/drm/msm/registers/adreno/adreno_pm4.xml | 7 +++++++ 2 files changed, 11 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c index 1820c167fcee..28c659c72493 100644 --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c @@ -242,10 +242,10 @@ static void a6xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) break; fallthrough; case MSM_SUBMIT_CMD_BUF: - OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3); + OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3); OUT_RING(ring, lower_32_bits(submit->cmd[i].iova)); OUT_RING(ring, upper_32_bits(submit->cmd[i].iova)); - OUT_RING(ring, submit->cmd[i].size); + OUT_RING(ring, A5XX_CP_INDIRECT_BUFFER_2_IB_SIZE(submit->cmd[i].size)); ibs++; break; } @@ -377,10 +377,10 @@ static void a7xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) break; fallthrough; case MSM_SUBMIT_CMD_BUF: - OUT_PKT7(ring, CP_INDIRECT_BUFFER_PFE, 3); + OUT_PKT7(ring, CP_INDIRECT_BUFFER, 3); OUT_RING(ring, lower_32_bits(submit->cmd[i].iova)); OUT_RING(ring, upper_32_bits(submit->cmd[i].iova)); - OUT_RING(ring, submit->cmd[i].size); + OUT_RING(ring, A5XX_CP_INDIRECT_BUFFER_2_IB_SIZE(submit->cmd[i].size)); ibs++; break; } diff --git a/drivers/gpu/drm/msm/registers/adreno/adreno_pm4.xml b/drivers/gpu/drm/msm/registers/adreno/adreno_pm4.xml index 55a35182858c..5a6ae9fc3194 100644 --- a/drivers/gpu/drm/msm/registers/adreno/adreno_pm4.xml +++ b/drivers/gpu/drm/msm/registers/adreno/adreno_pm4.xml @@ -2259,5 +2259,12 @@ opcode: CP_LOAD_STATE4 (30) (4 dwords) + + + + + + + -- 2.51.0 From cfa00a625f1c730e93f96b5b4ba7c1b4dc286c79 Mon Sep 17 00:00:00 2001 From: Guoqing Jiang Date: Fri, 6 Dec 2024 19:45:31 +0800 Subject: [PATCH 15/16] drm/exynos: Remove unnecessary checking It is not needed since drm_atomic_helper_shutdown checks it. Signed-off-by: Guoqing Jiang Signed-off-by: Inki Dae --- drivers/gpu/drm/exynos/exynos_drm_drv.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/gpu/drm/exynos/exynos_drm_drv.c b/drivers/gpu/drm/exynos/exynos_drm_drv.c index f313ae7bc3a3..6cc7bf77bcac 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_drv.c +++ b/drivers/gpu/drm/exynos/exynos_drm_drv.c @@ -355,8 +355,7 @@ static void exynos_drm_platform_shutdown(struct platform_device *pdev) { struct drm_device *drm = platform_get_drvdata(pdev); - if (drm) - drm_atomic_helper_shutdown(drm); + drm_atomic_helper_shutdown(drm); } static struct platform_driver exynos_drm_platform_driver = { -- 2.51.0 From 0253dadc772e83aaa67aea8bf24a71e7ffe13cb0 Mon Sep 17 00:00:00 2001 From: Wentao Liang Date: Thu, 6 Mar 2025 12:27:20 +0800 Subject: [PATCH 16/16] drm/exynos/vidi: Remove redundant error handling in vidi_get_modes() In the vidi_get_modes() function, if either drm_edid_dup() or drm_edid_alloc() fails, the function will immediately return 0, indicating that no display modes can be retrieved. However, in the event of failure in these two functions, it is still necessary to call the subsequent drm_edid_connector_update() function with a NULL drm_edid as an argument. This ensures that operations such as connector settings are performed in its callee function, _drm_edid_connector_property_update. To maintain the integrity of the operation, redundant error handling needs to be removed. Signed-off-by: Wentao Liang Signed-off-by: Inki Dae --- drivers/gpu/drm/exynos/exynos_drm_vidi.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/drivers/gpu/drm/exynos/exynos_drm_vidi.c b/drivers/gpu/drm/exynos/exynos_drm_vidi.c index 08cf79a62025..e644e2382d77 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_vidi.c +++ b/drivers/gpu/drm/exynos/exynos_drm_vidi.c @@ -312,9 +312,6 @@ static int vidi_get_modes(struct drm_connector *connector) else drm_edid = drm_edid_alloc(fake_edid_info, sizeof(fake_edid_info)); - if (!drm_edid) - return 0; - drm_edid_connector_update(connector, drm_edid); count = drm_edid_connector_add_modes(connector); -- 2.51.0