From e07e10ae83bdf429f59c8c149173a8c4f29c481e Mon Sep 17 00:00:00 2001 From: Ketil Johnsen Date: Wed, 8 Oct 2025 12:51:11 +0200 Subject: drm/panthor: Ensure MCU is disabled on suspend Currently the Panthor driver needs the GPU to be powered down between suspend and resume. If this is not done, then the MCU_CONTROL register will be preserved as AUTO, which again will cause a premature FW boot on resume. The FW will go directly into fatal state in this case. This case needs to be handled as there is no guarantee that the GPU will be powered down after the suspend callback on all platforms. The fix is to call panthor_fw_stop() in "pre-reset" path to ensure the MCU_CONTROL register is cleared (set DISABLE). This matches well with the already existing call to panthor_fw_start() from the "post-reset" path. Signed-off-by: Ketil Johnsen Acked-by: Boris Brezillon Reviewed-by: Steven Price Fixes: 2718d91816ee ("drm/panthor: Add the FW logical block") Signed-off-by: Steven Price Link: https://lore.kernel.org/r/20251008105112.4077015-1-ketil.johnsen@arm.com --- drivers/gpu/drm/panthor/panthor_fw.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c index 36f1034839c2..44a995835188 100644 --- a/drivers/gpu/drm/panthor/panthor_fw.c +++ b/drivers/gpu/drm/panthor/panthor_fw.c @@ -1099,6 +1099,7 @@ void panthor_fw_pre_reset(struct panthor_device *ptdev, bool on_hang) } panthor_job_irq_suspend(&ptdev->fw->irq); + panthor_fw_stop(ptdev); } /** -- cgit v1.2.3 From db74b04edce1bc86b9a5acc724c7ca06f427ab60 Mon Sep 17 00:00:00 2001 From: Marek Vasut Date: Sat, 11 Oct 2025 12:59:53 +0200 Subject: drm/bridge: lt9211: Drop check for last nibble of version register There is now a new LT9211 rev. U5, which reports chip ID 0x18 0x01 0xe4 . The previous LT9211 reported chip ID 0x18 0x01 0xe3 , which is what the driver checks for right now. Since there is a possibility there will be yet another revision of the LT9211 in the future, drop the last version nibble check to allow all future revisions of the chip to work with this driver. This fix makes LT9211 rev. U5 work with this driver. Fixes: 8ce4129e3de4 ("drm/bridge: lt9211: Add Lontium LT9211 bridge driver") Signed-off-by: Marek Vasut Reviewed-by: Dmitry Baryshkov Link: https://lore.kernel.org/r/20251011110017.12521-1-marek.vasut@mailbox.org Signed-off-by: Dmitry Baryshkov --- drivers/gpu/drm/bridge/lontium-lt9211.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/bridge/lontium-lt9211.c b/drivers/gpu/drm/bridge/lontium-lt9211.c index 399fa7eebd49..03fc8fd10f20 100644 --- a/drivers/gpu/drm/bridge/lontium-lt9211.c +++ b/drivers/gpu/drm/bridge/lontium-lt9211.c @@ -121,8 +121,7 @@ static int lt9211_read_chipid(struct lt9211 *ctx) } /* Test for known Chip ID. */ - if (chipid[0] != REG_CHIPID0_VALUE || chipid[1] != REG_CHIPID1_VALUE || - chipid[2] != REG_CHIPID2_VALUE) { + if (chipid[0] != REG_CHIPID0_VALUE || chipid[1] != REG_CHIPID1_VALUE) { dev_err(ctx->dev, "Unknown Chip ID: 0x%02x 0x%02x 0x%02x\n", chipid[0], chipid[1], chipid[2]); return -EINVAL; -- cgit v1.2.3 From 1696b0cfcf004a3af34ffe4c57a14e837ef18144 Mon Sep 17 00:00:00 2001 From: Zhanjun Dong Date: Mon, 29 Sep 2025 11:29:04 -0400 Subject: drm/i915/guc: Skip communication warning on reset in progress GuC IRQ and tasklet handler receive just single G2H message, and let other messages to be received from next tasklet. During this chained tasklet process, if reset process started, communication will be disabled. Skip warning for this condition. Fixes: 65dd4ed0f4e1 ("drm/i915/guc: Don't receive all G2H messages in irq handler") Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/15018 Signed-off-by: Zhanjun Dong Reviewed-by: Vinay Belgaumkar Signed-off-by: Daniele Ceraolo Spurio Link: https://lore.kernel.org/r/20250929152904.269776-1-zhanjun.dong@intel.com (cherry picked from commit 604b5ee4a653a70979ce689dbd6a5d942eb016bf) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c index 3e7e5badcc2b..2c651ec024ef 100644 --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c @@ -1325,9 +1325,16 @@ static int ct_receive(struct intel_guc_ct *ct) static void ct_try_receive_message(struct intel_guc_ct *ct) { + struct intel_guc *guc = ct_to_guc(ct); int ret; - if (GEM_WARN_ON(!ct->enabled)) + if (!ct->enabled) { + GEM_WARN_ON(!guc_to_gt(guc)->uc.reset_in_progress); + return; + } + + /* When interrupt disabled, message handling is not expected */ + if (!guc->interrupts.enabled) return; ret = ct_receive(ct); -- cgit v1.2.3 From 760039c95c78490c5c66ef584fcd536797ed6a2f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ville=20Syrj=C3=A4l=C3=A4?= Date: Fri, 3 Oct 2025 17:57:30 +0300 Subject: drm/i915/frontbuffer: Move bo refcounting intel_frontbuffer_{get,release}() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Currently xe's intel_frontbuffer implementation forgets to hold a reference on the bo. This makes the entire thing extremely fragile as the cleanup order now depends on bo references held by other things (namely intel_fb_bo_framebuffer_fini()). Move the bo refcounting to intel_frontbuffer_{get,release}() so that both i915 and xe do this the same way. I first tried to fix this by having xe do the refcounting from its intel_bo_set_frontbuffer() implementation (which is what i915 does currently), but turns out xe's drm_gem_object_free() can sleep and thus drm_gem_object_put() isn't safe to call while we hold fb_tracking.lock. Fixes: 10690b8a49bc ("drm/i915/display: Add intel_fb_bo_framebuffer_fini") Signed-off-by: Ville Syrjälä Link: https://patchwork.freedesktop.org/patch/msgid/20251003145734.7634-2-ville.syrjala@linux.intel.com Reviewed-by: Jani Nikula (cherry picked from commit eb4d490729a5fd8dc5a76d334f8d01fec7c14bbe) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/display/intel_frontbuffer.c | 10 +++++++++- drivers/gpu/drm/i915/gem/i915_gem_object_frontbuffer.h | 2 -- 2 files changed, 9 insertions(+), 3 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/i915/display/intel_frontbuffer.c b/drivers/gpu/drm/i915/display/intel_frontbuffer.c index 43be5377ddc1..73ed28ac9573 100644 --- a/drivers/gpu/drm/i915/display/intel_frontbuffer.c +++ b/drivers/gpu/drm/i915/display/intel_frontbuffer.c @@ -270,6 +270,8 @@ static void frontbuffer_release(struct kref *ref) spin_unlock(&display->fb_tracking.lock); i915_active_fini(&front->write); + + drm_gem_object_put(obj); kfree_rcu(front, rcu); } @@ -287,6 +289,8 @@ intel_frontbuffer_get(struct drm_gem_object *obj) if (!front) return NULL; + drm_gem_object_get(obj); + front->obj = obj; kref_init(&front->ref); atomic_set(&front->bits, 0); @@ -299,8 +303,12 @@ intel_frontbuffer_get(struct drm_gem_object *obj) spin_lock(&display->fb_tracking.lock); cur = intel_bo_set_frontbuffer(obj, front); spin_unlock(&display->fb_tracking.lock); - if (cur != front) + + if (cur != front) { + drm_gem_object_put(obj); kfree(front); + } + return cur; } diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_frontbuffer.h b/drivers/gpu/drm/i915/gem/i915_gem_object_frontbuffer.h index b6dc3d1b9bb1..b682969e3a29 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_object_frontbuffer.h +++ b/drivers/gpu/drm/i915/gem/i915_gem_object_frontbuffer.h @@ -89,12 +89,10 @@ i915_gem_object_set_frontbuffer(struct drm_i915_gem_object *obj, if (!front) { RCU_INIT_POINTER(obj->frontbuffer, NULL); - drm_gem_object_put(intel_bo_to_drm_bo(obj)); } else if (rcu_access_pointer(obj->frontbuffer)) { cur = rcu_dereference_protected(obj->frontbuffer, true); kref_get(&cur->ref); } else { - drm_gem_object_get(intel_bo_to_drm_bo(obj)); rcu_assign_pointer(obj->frontbuffer, front); } -- cgit v1.2.3 From 86af6b90e0556fcefbc6e98eb78bdce90327ee76 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ville=20Syrj=C3=A4l=C3=A4?= Date: Fri, 3 Oct 2025 17:57:31 +0300 Subject: drm/i915/fb: Fix the set_tiling vs. addfb race, again MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit intel_frontbuffer_get() is what locks out subsequent set_tiling changes to the bo. Thus the fence vs. modifier check must be done after intel_frontbuffer_get(), or else a concurrent set_tiling ioctl might sneak in and change the fence after the check has been done. Close the race again. See commit dd689287b977 ("drm/i915: Prevent concurrent tiling/framebuffer modifications") for the previous instance. v2: Reorder intel_user_framebuffer_destroy() to match the unwind (Jani) Cc: Jouni Högander Reviewed-by: Jani Nikula Fixes: 10690b8a49bc ("drm/i915/display: Add intel_fb_bo_framebuffer_fini") Signed-off-by: Ville Syrjälä Link: https://patchwork.freedesktop.org/patch/msgid/20251003145734.7634-3-ville.syrjala@linux.intel.com (cherry picked from commit 1d1e4ded216017f8febd91332ee337f0e0e79285) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/display/intel_fb.c | 38 +++++++++++++++++---------------- 1 file changed, 20 insertions(+), 18 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/i915/display/intel_fb.c b/drivers/gpu/drm/i915/display/intel_fb.c index 22a4a1575d22..b817ff44c043 100644 --- a/drivers/gpu/drm/i915/display/intel_fb.c +++ b/drivers/gpu/drm/i915/display/intel_fb.c @@ -2113,10 +2113,10 @@ static void intel_user_framebuffer_destroy(struct drm_framebuffer *fb) if (intel_fb_uses_dpt(fb)) intel_dpt_destroy(intel_fb->dpt_vm); - intel_frontbuffer_put(intel_fb->frontbuffer); - intel_fb_bo_framebuffer_fini(intel_fb_bo(fb)); + intel_frontbuffer_put(intel_fb->frontbuffer); + kfree(intel_fb); } @@ -2218,15 +2218,17 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, int ret = -EINVAL; int i; + /* + * intel_frontbuffer_get() must be done before + * intel_fb_bo_framebuffer_init() to avoid set_tiling vs. addfb race. + */ + intel_fb->frontbuffer = intel_frontbuffer_get(obj); + if (!intel_fb->frontbuffer) + return -ENOMEM; + ret = intel_fb_bo_framebuffer_init(fb, obj, mode_cmd); if (ret) - return ret; - - intel_fb->frontbuffer = intel_frontbuffer_get(obj); - if (!intel_fb->frontbuffer) { - ret = -ENOMEM; - goto err; - } + goto err_frontbuffer_put; ret = -EINVAL; if (!drm_any_plane_has_format(display->drm, @@ -2235,7 +2237,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, drm_dbg_kms(display->drm, "unsupported pixel format %p4cc / modifier 0x%llx\n", &mode_cmd->pixel_format, mode_cmd->modifier[0]); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } max_stride = intel_fb_max_stride(display, mode_cmd->pixel_format, @@ -2246,7 +2248,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, mode_cmd->modifier[0] != DRM_FORMAT_MOD_LINEAR ? "tiled" : "linear", mode_cmd->pitches[0], max_stride); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } /* FIXME need to adjust LINOFF/TILEOFF accordingly. */ @@ -2254,7 +2256,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, drm_dbg_kms(display->drm, "plane 0 offset (0x%08x) must be 0\n", mode_cmd->offsets[0]); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } drm_helper_mode_fill_fb_struct(display->drm, fb, info, mode_cmd); @@ -2264,7 +2266,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, if (mode_cmd->handles[i] != mode_cmd->handles[0]) { drm_dbg_kms(display->drm, "bad plane %d handle\n", i); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } stride_alignment = intel_fb_stride_alignment(fb, i); @@ -2272,7 +2274,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, drm_dbg_kms(display->drm, "plane %d pitch (%d) must be at least %u byte aligned\n", i, fb->pitches[i], stride_alignment); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } if (intel_fb_is_gen12_ccs_aux_plane(fb, i)) { @@ -2282,7 +2284,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, drm_dbg_kms(display->drm, "ccs aux plane %d pitch (%d) must be %d\n", i, fb->pitches[i], ccs_aux_stride); - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; } } @@ -2291,7 +2293,7 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, ret = intel_fill_fb_info(display, intel_fb); if (ret) - goto err_frontbuffer_put; + goto err_bo_framebuffer_fini; if (intel_fb_uses_dpt(fb)) { struct i915_address_space *vm; @@ -2317,10 +2319,10 @@ int intel_framebuffer_init(struct intel_framebuffer *intel_fb, err_free_dpt: if (intel_fb_uses_dpt(fb)) intel_dpt_destroy(intel_fb->dpt_vm); +err_bo_framebuffer_fini: + intel_fb_bo_framebuffer_fini(obj); err_frontbuffer_put: intel_frontbuffer_put(intel_fb->frontbuffer); -err: - intel_fb_bo_framebuffer_fini(obj); return ret; } -- cgit v1.2.3 From 9858ea4c29c283f0a8a3cdbb42108d464ece90a8 Mon Sep 17 00:00:00 2001 From: Matthew Schwartz Date: Thu, 9 Oct 2025 14:19:00 +0200 Subject: Revert "drm/amd/display: Only restore backlight after amdgpu_dm_init or dm_resume" This fix regressed the original issue that commit 7875afafba84 ("drm/amd/display: Fix brightness level not retained over reboot") solved, so revert it until a different approach to solve the regression that it caused with AMD_PRIVATE_COLOR is found. Fixes: a490c8d77d50 ("drm/amd/display: Only restore backlight after amdgpu_dm_init or dm_resume") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4620 Cc: stable@vger.kernel.org Signed-off-by: Matthew Schwartz Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 12 ++++-------- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h | 7 ------- 2 files changed, 4 insertions(+), 15 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c index 0d03e324d5b9..6597475e245d 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c @@ -2085,8 +2085,6 @@ static int amdgpu_dm_init(struct amdgpu_device *adev) dc_hardware_init(adev->dm.dc); - adev->dm.restore_backlight = true; - adev->dm.hpd_rx_offload_wq = hpd_rx_irq_create_workqueue(adev); if (!adev->dm.hpd_rx_offload_wq) { drm_err(adev_to_drm(adev), "failed to create hpd rx offload workqueue.\n"); @@ -3442,7 +3440,6 @@ static int dm_resume(struct amdgpu_ip_block *ip_block) dc_set_power_state(dm->dc, DC_ACPI_CM_POWER_STATE_D0); dc_resume(dm->dc); - adev->dm.restore_backlight = true; amdgpu_dm_irq_resume_early(adev); @@ -9969,6 +9966,7 @@ static void amdgpu_dm_commit_streams(struct drm_atomic_state *state, bool mode_set_reset_required = false; u32 i; struct dc_commit_streams_params params = {dc_state->streams, dc_state->stream_count}; + bool set_backlight_level = false; /* Disable writeback */ for_each_old_connector_in_state(state, connector, old_con_state, i) { @@ -10088,6 +10086,7 @@ static void amdgpu_dm_commit_streams(struct drm_atomic_state *state, acrtc->hw_mode = new_crtc_state->mode; crtc->hwmode = new_crtc_state->mode; mode_set_reset_required = true; + set_backlight_level = true; } else if (modereset_required(new_crtc_state)) { drm_dbg_atomic(dev, "Atomic commit: RESET. crtc id %d:[%p]\n", @@ -10144,16 +10143,13 @@ static void amdgpu_dm_commit_streams(struct drm_atomic_state *state, * to fix a flicker issue. * It will cause the dm->actual_brightness is not the current panel brightness * level. (the dm->brightness is the correct panel level) - * So we set the backlight level with dm->brightness value after initial - * set mode. Use restore_backlight flag to avoid setting backlight level - * for every subsequent mode set. + * So we set the backlight level with dm->brightness value after set mode */ - if (dm->restore_backlight) { + if (set_backlight_level) { for (i = 0; i < dm->num_of_edps; i++) { if (dm->backlight_dev[i]) amdgpu_dm_backlight_set_level(dm, i, dm->brightness[i]); } - dm->restore_backlight = false; } } diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h index 009f206226f0..db75e991ac7b 100644 --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.h @@ -630,13 +630,6 @@ struct amdgpu_display_manager { */ u32 actual_brightness[AMDGPU_DM_MAX_NUM_EDP]; - /** - * @restore_backlight: - * - * Flag to indicate whether to restore backlight after modeset. - */ - bool restore_backlight; - /** * @aux_hpd_discon_quirk: * -- cgit v1.2.3 From 5c05bcf6ae7732da1bd4dc1958d527b5f07f216a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Timur=20Krist=C3=B3f?= Date: Fri, 26 Sep 2025 20:26:12 +0200 Subject: drm/amd/pm: Disable MCLK switching on SI at high pixel clocks MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit On various SI GPUs, a flickering can be observed near the bottom edge of the screen when using a single 4K 60Hz monitor over DP. Disabling MCLK switching works around this problem. Reviewed-by: Alex Deucher Signed-off-by: Timur Kristóf Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c b/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c index cf9932e68055..3a9522c17fee 100644 --- a/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +++ b/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c @@ -3500,6 +3500,11 @@ static void si_apply_state_adjust_rules(struct amdgpu_device *adev, * for these GPUs to calculate bandwidth requirements. */ if (high_pixelclock_count) { + /* Work around flickering lines at the bottom edge + * of the screen when using a single 4K 60Hz monitor. + */ + disable_mclk_switching = true; + /* On Oland, we observe some flickering when two 4K 60Hz * displays are connected, possibly because voltage is too low. * Raise the voltage by requiring a higher SCLK. -- cgit v1.2.3 From 7bdd91abf0cb3ea78160e2e78fb58b12f6a38d55 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Timur=20Krist=C3=B3f?= Date: Fri, 26 Sep 2025 20:26:13 +0200 Subject: drm/amd: Disable ASPM on SI MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Enabling ASPM causes randoms hangs on Tahiti and Oland on Zen4. It's unclear if this is a platform-specific or GPU-specific issue. Disable ASPM on SI for the time being. Reviewed-by: Alex Deucher Signed-off-by: Timur Kristóf Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 7a899fb4de29..3d032c4e2dce 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1882,6 +1882,13 @@ static bool amdgpu_device_pcie_dynamic_switching_supported(struct amdgpu_device static bool amdgpu_device_aspm_support_quirk(struct amdgpu_device *adev) { + /* Enabling ASPM causes randoms hangs on Tahiti and Oland on Zen4. + * It's unclear if this is a platform-specific or GPU-specific issue. + * Disable ASPM on SI for the time being. + */ + if (adev->family == AMDGPU_FAMILY_SI) + return true; + #if IS_ENABLED(CONFIG_X86) struct cpuinfo_x86 *c = &cpu_data(0); -- cgit v1.2.3 From 5d55ed19d4190d2c210ac05ac7a53f800a8c6fe5 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= Date: Wed, 27 Aug 2025 14:47:23 +0200 Subject: drm/amdgpu: remove two invalid BUG_ON()s MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Those can be triggered trivially by userspace. Signed-off-by: Christian König Reviewed-by: Alex Deucher Acked-by: Timur Kristóf Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 2 -- drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 2 -- 2 files changed, 4 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index 66c47c466532..d61eb9f187c6 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -5862,8 +5862,6 @@ static void gfx_v11_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; - BUG_ON(ib->flags & AMDGPU_IB_FLAG_CE); - header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); control |= ib->length_dw | (vmid << 24); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c index 710ec9c34e43..93fde0f9af87 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c @@ -4419,8 +4419,6 @@ static void gfx_v12_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, unsigned vmid = AMDGPU_JOB_GET_VMID(job); u32 header, control = 0; - BUG_ON(ib->flags & AMDGPU_IB_FLAG_CE); - header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); control |= ib->length_dw | (vmid << 24); -- cgit v1.2.3 From 8f74c70be57527d7b79e2ecf6de1a154d148254d Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= Date: Mon, 22 Sep 2025 14:18:16 +0200 Subject: drm/amdgpu: block CE CS if not explicitely allowed by module option MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The Constant Engine found on gfx6-gfx10 HW has been a notorious source of problems. RADV never used it in the first place, radeonsi only used it for a few releases around 2017 for gfx6-gfx9 before dropping support for it as well. While investigating another problem I just recently found that submitting to the CE seems to be completely broken on gfx9 for quite a while. Since nobody complained about that problem it most likely means that nobody is using any of the affected radeonsi versions on current Linux kernels any more. So to potentially phase out the support for the CE and eliminate another source of problems block submitting CE IBs unless it is enabled again using a debug flag. Signed-off-by: Christian König Reviewed-by: Alex Deucher Acked-by: Timur Kristóf Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 ++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 +++++++- 3 files changed, 14 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 2a0df4cabb99..6f5b4a0e0a34 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1290,6 +1290,7 @@ struct amdgpu_device { bool debug_disable_gpu_ring_reset; bool debug_vm_userptr; bool debug_disable_ce_logs; + bool debug_enable_ce_cs; /* Protection for the following isolation structure */ struct mutex enforce_isolation_mutex; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 9cd7741d2254..ba9fb08db094 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -364,6 +364,12 @@ static int amdgpu_cs_p2_ib(struct amdgpu_cs_parser *p, if (p->uf_bo && ring->funcs->no_user_fence) return -EINVAL; + if (!p->adev->debug_enable_ce_cs && + chunk_ib->flags & AMDGPU_IB_FLAG_CE) { + dev_err_ratelimited(p->adev->dev, "CE CS is blocked, use debug=0x400 to override\n"); + return -EINVAL; + } + if (chunk_ib->ip_type == AMDGPU_HW_IP_GFX && chunk_ib->flags & AMDGPU_IB_FLAG_PREEMPT) { if (chunk_ib->flags & AMDGPU_IB_FLAG_CE) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index bff25ef3e2d0..61268aa82df4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -144,7 +144,8 @@ enum AMDGPU_DEBUG_MASK { AMDGPU_DEBUG_DISABLE_GPU_RING_RESET = BIT(6), AMDGPU_DEBUG_SMU_POOL = BIT(7), AMDGPU_DEBUG_VM_USERPTR = BIT(8), - AMDGPU_DEBUG_DISABLE_RAS_CE_LOG = BIT(9) + AMDGPU_DEBUG_DISABLE_RAS_CE_LOG = BIT(9), + AMDGPU_DEBUG_ENABLE_CE_CS = BIT(10) }; unsigned int amdgpu_vram_limit = UINT_MAX; @@ -2289,6 +2290,11 @@ static void amdgpu_init_debug_options(struct amdgpu_device *adev) pr_info("debug: disable kernel logs of correctable errors\n"); adev->debug_disable_ce_logs = true; } + + if (amdgpu_debug_mask & AMDGPU_DEBUG_ENABLE_CE_CS) { + pr_info("debug: allowing command submission to CE engine\n"); + adev->debug_enable_ce_cs = true; + } } static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev, unsigned long flags) -- cgit v1.2.3 From 357d90be2c7aaa526a840cddffd2b8d676fe75a6 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 26 Sep 2025 17:31:32 -0400 Subject: drm/amdgpu: fix handling of harvesting for ip_discovery firmware Chips which use the IP discovery firmware loaded by the driver reported incorrect harvesting information in the ip discovery table in sysfs because the driver only uses the ip discovery firmware for populating sysfs and not for direct parsing for the driver itself as such, the fields that are used to print the harvesting info in sysfs report incorrect data for some IPs. Populate the relevant fields for this case as well. Fixes: 514678da56da ("drm/amdgpu/discovery: fix fw based ip discovery") Acked-by: Tom St Denis Reviewed-by: Lijo Lazar Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c index 73401f0aeb34..dd7b2b796427 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c @@ -1033,7 +1033,9 @@ static uint8_t amdgpu_discovery_get_harvest_info(struct amdgpu_device *adev, /* Until a uniform way is figured, get mask based on hwid */ switch (hw_id) { case VCN_HWID: - harvest = ((1 << inst) & adev->vcn.inst_mask) == 0; + /* VCN vs UVD+VCE */ + if (!amdgpu_ip_version(adev, VCE_HWIP, 0)) + harvest = ((1 << inst) & adev->vcn.inst_mask) == 0; break; case DMU_HWID: if (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK) @@ -2565,7 +2567,9 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); vega10_reg_base_init(adev); adev->sdma.num_instances = 2; + adev->sdma.sdma_mask = 3; adev->gmc.num_umc = 4; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 0, 0); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 0, 0); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 0, 0); @@ -2592,7 +2596,9 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); vega10_reg_base_init(adev); adev->sdma.num_instances = 2; + adev->sdma.sdma_mask = 3; adev->gmc.num_umc = 4; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 3, 0); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 3, 0); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 0, 1); @@ -2619,8 +2625,10 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); vega10_reg_base_init(adev); adev->sdma.num_instances = 1; + adev->sdma.sdma_mask = 1; adev->vcn.num_vcn_inst = 1; adev->gmc.num_umc = 2; + adev->gfx.xcc_mask = 1; if (adev->apu_flags & AMD_APU_IS_RAVEN2) { adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 2, 0); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 2, 0); @@ -2665,7 +2673,9 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); vega20_reg_base_init(adev); adev->sdma.num_instances = 2; + adev->sdma.sdma_mask = 3; adev->gmc.num_umc = 8; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 0); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 0); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 2, 0); @@ -2693,8 +2703,10 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); arct_reg_base_init(adev); adev->sdma.num_instances = 8; + adev->sdma.sdma_mask = 0xff; adev->vcn.num_vcn_inst = 2; adev->gmc.num_umc = 8; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 1); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 1); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 2, 1); @@ -2726,8 +2738,10 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) amdgpu_discovery_init(adev); aldebaran_reg_base_init(adev); adev->sdma.num_instances = 5; + adev->sdma.sdma_mask = 0x1f; adev->vcn.num_vcn_inst = 2; adev->gmc.num_umc = 4; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(9, 4, 2); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(9, 4, 2); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(4, 4, 0); @@ -2762,6 +2776,8 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev) } else { cyan_skillfish_reg_base_init(adev); adev->sdma.num_instances = 2; + adev->sdma.sdma_mask = 3; + adev->gfx.xcc_mask = 1; adev->ip_versions[MMHUB_HWIP][0] = IP_VERSION(2, 0, 3); adev->ip_versions[ATHUB_HWIP][0] = IP_VERSION(2, 0, 3); adev->ip_versions[OSSSYS_HWIP][0] = IP_VERSION(5, 0, 1); -- cgit v1.2.3 From 1f22fcb88bfef26a966e9eb242c692c6bf253d47 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Mon, 15 Sep 2025 12:37:32 -0400 Subject: drm/amdgpu: handle wrap around in reemit handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Compare the sequence numbers directly. Fixes: 77cc0da39c7c ("drm/amdgpu: track ring state associated with a fence") Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index fd8cca241da6..e270df30c279 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -790,14 +790,19 @@ void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring, struct dma_fence *unprocessed; struct dma_fence __rcu **ptr; struct amdgpu_fence *fence; - u64 wptr, i, seqno; + u64 wptr; + u32 seq, last_seq; - seqno = amdgpu_fence_read(ring); + last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask; + seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask; wptr = ring->fence_drv.signalled_wptr; ring->ring_backup_entries_to_copy = 0; - for (i = seqno + 1; i <= ring->fence_drv.sync_seq; ++i) { - ptr = &ring->fence_drv.fences[i & ring->fence_drv.num_fences_mask]; + do { + last_seq++; + last_seq &= ring->fence_drv.num_fences_mask; + + ptr = &ring->fence_drv.fences[last_seq]; rcu_read_lock(); unprocessed = rcu_dereference(*ptr); @@ -813,7 +818,7 @@ void amdgpu_ring_backup_unprocessed_commands(struct amdgpu_ring *ring, wptr = fence->wptr; } rcu_read_unlock(); - } + } while (last_seq != seq); } /* -- cgit v1.2.3 From ff780f4f80323148d43198f2052c14160c8428d3 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 3 Sep 2025 13:48:23 -0400 Subject: drm/amdgpu: set an error on all fences from a bad context MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When we backup ring contents to reemit after a queue reset, we don't backup ring contents from the bad context. When we signal the fences, we should set an error on those fences as well. v2: misc cleanups v3: add locking for fence error, fix comment (Christian) v4: fix wrap around, locking (Christian) Fixes: 77cc0da39c7c ("drm/amdgpu: track ring state associated with a fence") Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 39 +++++++++++++++++++++++++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 2 +- 3 files changed, 37 insertions(+), 6 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index e270df30c279..18a7829122d2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -758,11 +758,42 @@ void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring) * @fence: fence of the ring to signal * */ -void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *fence) +void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *af) { - dma_fence_set_error(&fence->base, -ETIME); - amdgpu_fence_write(fence->ring, fence->seq); - amdgpu_fence_process(fence->ring); + struct dma_fence *unprocessed; + struct dma_fence __rcu **ptr; + struct amdgpu_fence *fence; + struct amdgpu_ring *ring = af->ring; + unsigned long flags; + u32 seq, last_seq; + + last_seq = amdgpu_fence_read(ring) & ring->fence_drv.num_fences_mask; + seq = ring->fence_drv.sync_seq & ring->fence_drv.num_fences_mask; + + /* mark all fences from the guilty context with an error */ + spin_lock_irqsave(&ring->fence_drv.lock, flags); + do { + last_seq++; + last_seq &= ring->fence_drv.num_fences_mask; + + ptr = &ring->fence_drv.fences[last_seq]; + rcu_read_lock(); + unprocessed = rcu_dereference(*ptr); + + if (unprocessed && !dma_fence_is_signaled_locked(unprocessed)) { + fence = container_of(unprocessed, struct amdgpu_fence, base); + + if (fence == af) + dma_fence_set_error(&fence->base, -ETIME); + else if (fence->context == af->context) + dma_fence_set_error(&fence->base, -ECANCELED); + } + rcu_read_unlock(); + } while (last_seq != seq); + spin_unlock_irqrestore(&ring->fence_drv.lock, flags); + /* signal the guilty fence */ + amdgpu_fence_write(ring, af->seq); + amdgpu_fence_process(ring); } void amdgpu_fence_save_wptr(struct dma_fence *fence) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index 8f6ce948c684..5ec5c3ff22bb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -811,7 +811,7 @@ int amdgpu_ring_reset_helper_end(struct amdgpu_ring *ring, if (r) return r; - /* signal the fence of the bad job */ + /* signal the guilty fence and set an error on all fences from the context */ if (guilty_fence) amdgpu_fence_driver_guilty_force_completion(guilty_fence); /* Re-emit the non-guilty commands */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h index b6b649179776..4b46e3c26ff3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h @@ -155,7 +155,7 @@ extern const struct drm_sched_backend_ops amdgpu_sched_ops; void amdgpu_fence_driver_clear_job_fences(struct amdgpu_ring *ring); void amdgpu_fence_driver_set_error(struct amdgpu_ring *ring, int error); void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring); -void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *fence); +void amdgpu_fence_driver_guilty_force_completion(struct amdgpu_fence *af); void amdgpu_fence_save_wptr(struct dma_fence *fence); int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring); -- cgit v1.2.3 From 6df8e84aa6b5b1812cc2cacd6b3f5ccbb18cda2b Mon Sep 17 00:00:00 2001 From: Gui-Dong Han Date: Wed, 8 Oct 2025 03:43:27 +0000 Subject: drm/amdgpu: use atomic functions with memory barriers for vm fault info The atomic variable vm_fault_info_updated is used to synchronize access to adev->gmc.vm_fault_info between the interrupt handler and get_vm_fault_info(). The default atomic functions like atomic_set() and atomic_read() do not provide memory barriers. This allows for CPU instruction reordering, meaning the memory accesses to vm_fault_info and the vm_fault_info_updated flag are not guaranteed to occur in the intended order. This creates a race condition that can lead to inconsistent or stale data being used. The previous implementation, which used an explicit mb(), was incomplete and inefficient. It failed to account for all potential CPU reorderings, such as the access of vm_fault_info being reordered before the atomic_read of the flag. This approach is also more verbose and less performant than using the proper atomic functions with acquire/release semantics. Fix this by switching to atomic_set_release() and atomic_read_acquire(). These functions provide the necessary acquire and release semantics, which act as memory barriers to ensure the correct order of operations. It is also more efficient and idiomatic than using explicit full memory barriers. Fixes: b97dfa27ef3a ("drm/amdgpu: save vm fault information for amdkfd") Cc: stable@vger.kernel.org Signed-off-by: Gui-Dong Han Signed-off-by: Felix Kuehling Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 5 ++--- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 7 +++---- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 7 +++---- 3 files changed, 8 insertions(+), 11 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 83020963dfde..a2ca9acf8c4e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -2329,10 +2329,9 @@ void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem) int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev, struct kfd_vm_fault_info *mem) { - if (atomic_read(&adev->gmc.vm_fault_info_updated) == 1) { + if (atomic_read_acquire(&adev->gmc.vm_fault_info_updated) == 1) { *mem = *adev->gmc.vm_fault_info; - mb(); /* make sure read happened */ - atomic_set(&adev->gmc.vm_fault_info_updated, 0); + atomic_set_release(&adev->gmc.vm_fault_info_updated, 0); } return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 93d7ccb7d013..0e5e54d0a9a5 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -1068,7 +1068,7 @@ static int gmc_v7_0_sw_init(struct amdgpu_ip_block *ip_block) GFP_KERNEL); if (!adev->gmc.vm_fault_info) return -ENOMEM; - atomic_set(&adev->gmc.vm_fault_info_updated, 0); + atomic_set_release(&adev->gmc.vm_fault_info_updated, 0); return 0; } @@ -1290,7 +1290,7 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev, vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID); if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid) - && !atomic_read(&adev->gmc.vm_fault_info_updated)) { + && !atomic_read_acquire(&adev->gmc.vm_fault_info_updated)) { struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info; u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, @@ -1306,8 +1306,7 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev, info->prot_read = protections & 0x8 ? true : false; info->prot_write = protections & 0x10 ? true : false; info->prot_exec = protections & 0x20 ? true : false; - mb(); - atomic_set(&adev->gmc.vm_fault_info_updated, 1); + atomic_set_release(&adev->gmc.vm_fault_info_updated, 1); } return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index c5e2a2c41e06..e1509480dfc2 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -1183,7 +1183,7 @@ static int gmc_v8_0_sw_init(struct amdgpu_ip_block *ip_block) GFP_KERNEL); if (!adev->gmc.vm_fault_info) return -ENOMEM; - atomic_set(&adev->gmc.vm_fault_info_updated, 0); + atomic_set_release(&adev->gmc.vm_fault_info_updated, 0); return 0; } @@ -1478,7 +1478,7 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev, vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID); if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid) - && !atomic_read(&adev->gmc.vm_fault_info_updated)) { + && !atomic_read_acquire(&adev->gmc.vm_fault_info_updated)) { struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info; u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, @@ -1494,8 +1494,7 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev, info->prot_read = protections & 0x8 ? true : false; info->prot_write = protections & 0x10 ? true : false; info->prot_exec = protections & 0x20 ? true : false; - mb(); - atomic_set(&adev->gmc.vm_fault_info_updated, 1); + atomic_set_release(&adev->gmc.vm_fault_info_updated, 1); } return 0; -- cgit v1.2.3 From 6917112af2ba36c5f19075eb9f2933ffd07e55bf Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Timur=20Krist=C3=B3f?= Date: Mon, 13 Oct 2025 08:06:42 +0200 Subject: drm/amd/powerplay: Fix CIK shutdown temperature MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Remove extra multiplication. CIK GPUs such as Hawaii appear to use PP_TABLE_V0 in which case the shutdown temperature is hardcoded in smu7_init_dpm_defaults and is already multiplied by 1000. The value was mistakenly multiplied another time by smu7_get_thermal_temperature_range. Fixes: 4ba082572a42 ("drm/amd/powerplay: export the thermal ranges of VI asics (V2)") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/1676 Reviewed-by: Alex Deucher Signed-off-by: Timur Kristóf Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c b/drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c index 8da882c51856..9b28c0728269 100644 --- a/drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c +++ b/drivers/gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c @@ -5444,8 +5444,7 @@ static int smu7_get_thermal_temperature_range(struct pp_hwmgr *hwmgr, thermal_data->max = table_info->cac_dtp_table->usSoftwareShutdownTemp * PP_TEMPERATURE_UNITS_PER_CENTIGRADES; else if (hwmgr->pp_table_version == PP_TABLE_V0) - thermal_data->max = data->thermal_temp_setting.temperature_shutdown * - PP_TEMPERATURE_UNITS_PER_CENTIGRADES; + thermal_data->max = data->thermal_temp_setting.temperature_shutdown; thermal_data->sw_ctf_threshold = thermal_data->max; -- cgit v1.2.3 From 74de0eaa00eac2e0cbad1dda6dcf8f44ab27629e Mon Sep 17 00:00:00 2001 From: Sathishkumar S Date: Fri, 10 Oct 2025 23:32:40 +0530 Subject: drm/amdgpu: fix bit shift logic BIT_ULL(n) sets nth bit, remove explicit shift and set the position Fixes: a7a411e24626 ("drm/amdgpu: fix shift-out-of-bounds in amdgpu_debugfs_jpeg_sched_mask_set") Signed-off-by: Sathishkumar S Reviewed-by: Leo Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c index 6b7d66b6d4cc..63ee6ba6a931 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.c @@ -371,7 +371,7 @@ static int amdgpu_debugfs_jpeg_sched_mask_set(void *data, u64 val) for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) { ring = &adev->jpeg.inst[i].ring_dec[j]; - if (val & (BIT_ULL(1) << ((i * adev->jpeg.num_jpeg_rings) + j))) + if (val & (BIT_ULL((i * adev->jpeg.num_jpeg_rings) + j))) ring->sched.ready = true; else ring->sched.ready = false; -- cgit v1.2.3 From 33cc891b56b93cad1a83263eaf2e417436f70c82 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= Date: Tue, 7 Oct 2025 10:10:52 +0200 Subject: drm/amdgpu: hide VRAM sysfs attributes on GPUs without VRAM MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Otherwise accessing them can cause a crash. Signed-off-by: Christian König Tested-by: Mangesh Gadre Acked-by: Alex Deucher Reviewed-by: Arunpravin Paneer Selvam Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c index a5adb2ed9b3c..9d934c07fa6b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c @@ -234,6 +234,9 @@ static umode_t amdgpu_vram_attrs_is_visible(struct kobject *kobj, !adev->gmc.vram_vendor) return 0; + if (!ttm_resource_manager_used(&adev->mman.vram_mgr.manager)) + return 0; + return attr->mode; } -- cgit v1.2.3 From 883f309add55060233bf11c1ea6947140372920f Mon Sep 17 00:00:00 2001 From: "Jesse.Zhang" Date: Mon, 13 Oct 2025 13:46:12 +0800 Subject: drm/amdgpu: Fix NULL pointer dereference in VRAM logic for APU devices MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Previously, APU platforms (and other scenarios with uninitialized VRAM managers) triggered a NULL pointer dereference in `ttm_resource_manager_usage()`. The root cause is not that the `struct ttm_resource_manager *man` pointer itself is NULL, but that `man->bdev` (the backing device pointer within the manager) remains uninitialized (NULL) on APUs—since APUs lack dedicated VRAM and do not fully set up VRAM manager structures. When `ttm_resource_manager_usage()` attempts to acquire `man->bdev->lru_lock`, it dereferences the NULL `man->bdev`, leading to a kernel OOPS. 1. **amdgpu_cs.c**: Extend the existing bandwidth control check in `amdgpu_cs_get_threshold_for_moves()` to include a check for `ttm_resource_manager_used()`. If the manager is not used (uninitialized `bdev`), return 0 for migration thresholds immediately—skipping VRAM-specific logic that would trigger the NULL dereference. 2. **amdgpu_kms.c**: Update the `AMDGPU_INFO_VRAM_USAGE` ioctl and memory info reporting to use a conditional: if the manager is used, return the real VRAM usage; otherwise, return 0. This avoids accessing `man->bdev` when it is NULL. 3. **amdgpu_virt.c**: Modify the vf2pf (virtual function to physical function) data write path. Use `ttm_resource_manager_used()` to check validity: if the manager is usable, calculate `fb_usage` from VRAM usage; otherwise, set `fb_usage` to 0 (APUs have no discrete framebuffer to report). This approach is more robust than APU-specific checks because it: - Works for all scenarios where the VRAM manager is uninitialized (not just APUs), - Aligns with TTM's design by using its native helper function, - Preserves correct behavior for discrete GPUs (which have fully initialized `man->bdev` and pass the `ttm_resource_manager_used()` check). v4: use ttm_resource_manager_used(&adev->mman.vram_mgr.manager) instead of checking the adev->gmc.is_app_apu flag (Christian) Reviewed-by: Christian König Suggested-by: Lijo Lazar Signed-off-by: Jesse Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 7 ++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 4 ++-- 3 files changed, 7 insertions(+), 6 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index ba9fb08db094..2f6a96af7fb1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -708,7 +708,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev, */ const s64 us_upper_bound = 200000; - if (!adev->mm_stats.log2_max_MBps) { + if ((!adev->mm_stats.log2_max_MBps) || !ttm_resource_manager_used(&adev->mman.vram_mgr.manager)) { *max_bytes = 0; *max_vis_bytes = 0; return; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index a9327472c651..b3e6b3fcdf2c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -758,7 +758,8 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) ui64 = atomic64_read(&adev->num_vram_cpu_page_faults); return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0; case AMDGPU_INFO_VRAM_USAGE: - ui64 = ttm_resource_manager_usage(&adev->mman.vram_mgr.manager); + ui64 = ttm_resource_manager_used(&adev->mman.vram_mgr.manager) ? + ttm_resource_manager_usage(&adev->mman.vram_mgr.manager) : 0; return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0; case AMDGPU_INFO_VIS_VRAM_USAGE: ui64 = amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr); @@ -804,8 +805,8 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) mem.vram.usable_heap_size = adev->gmc.real_vram_size - atomic64_read(&adev->vram_pin_size) - AMDGPU_VM_RESERVED_VRAM; - mem.vram.heap_usage = - ttm_resource_manager_usage(vram_man); + mem.vram.heap_usage = ttm_resource_manager_used(&adev->mman.vram_mgr.manager) ? + ttm_resource_manager_usage(vram_man) : 0; mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4; mem.cpu_accessible_vram.total_heap_size = diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c index 3328ab63376b..f96beb96c75c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c @@ -598,8 +598,8 @@ static int amdgpu_virt_write_vf2pf_data(struct amdgpu_device *adev) vf2pf_info->driver_cert = 0; vf2pf_info->os_info.all = 0; - vf2pf_info->fb_usage = - ttm_resource_manager_usage(&adev->mman.vram_mgr.manager) >> 20; + vf2pf_info->fb_usage = ttm_resource_manager_used(&adev->mman.vram_mgr.manager) ? + ttm_resource_manager_usage(&adev->mman.vram_mgr.manager) >> 20 : 0; vf2pf_info->fb_vis_usage = amdgpu_vram_mgr_vis_usage(&adev->mman.vram_mgr) >> 20; vf2pf_info->fb_size = adev->gmc.real_vram_size >> 20; -- cgit v1.2.3 From d0de79f66a80eeb849033fae34bd07a69ce72235 Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Thu, 9 Oct 2025 10:45:42 -0400 Subject: drm/amdgpu: fix gfx12 mes packet status return check GFX12 MES uses low 32 bits of status return for success (1 or 0) and high bits for debug information if low bits are 0. GFX11 MES doesn't do this so checking full 64-bit status return for 1 or 0 is still valid. Signed-off-by: Jonathan Kim Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c index aff06f06aeee..e3149196143e 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c @@ -228,7 +228,12 @@ static int mes_v12_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes, pipe, x_pkt->header.opcode); r = amdgpu_fence_wait_polling(ring, seq, timeout); - if (r < 1 || !*status_ptr) { + + /* + * status_ptr[31:0] == 0 (fail) or status_ptr[63:0] == 1 (success). + * If status_ptr[31:0] == 0 then status_ptr[63:32] will have debug error information. + */ + if (r < 1 || !(lower_32_bits(*status_ptr))) { if (misc_op_str) dev_err(adev->dev, "MES(%d) failed to respond to msg=%s (%s)\n", -- cgit v1.2.3 From 8745ca5efb2aad0b6591d9b8cd48573ea49c929d Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Thu, 9 Oct 2025 10:48:09 -0400 Subject: drm/amdgpu: fix initialization of doorbell array for detect and hang Initialized doorbells should be set to invalid rather than 0 to prevent driver from over counting hung doorbells since it checks against the invalid value to begin with. Signed-off-by: Jonathan Kim Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c index 5bf9be073cdd..30e1fb510600 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c @@ -409,7 +409,7 @@ int amdgpu_mes_detect_and_reset_hung_queues(struct amdgpu_device *adev, return -EINVAL; /* Clear the doorbell array before detection */ - memset(adev->mes.hung_queue_db_array_cpu_addr, 0, + memset(adev->mes.hung_queue_db_array_cpu_addr, AMDGPU_MES_INVALID_DB_OFFSET, adev->mes.hung_queue_db_array_size * sizeof(u32)); input.queue_type = queue_type; input.detect_only = detect_only; -- cgit v1.2.3 From 0ef930e1faca6418316e5b9a3b4d1f6ae9e5b240 Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Thu, 9 Oct 2025 11:28:19 -0400 Subject: drm/amdgpu: fix hung reset queue array memory allocation By design the MES will return an array result that is twice the number of hung doorbells it can report. i.e. if up k reported doorbells are supported, then the second half of the array, also of length k, holds the HQD information (type/queue/pipe) where queue 1 corresponds to index 0 and k, queue 2 corresponds to index 1 and k + 1 etc ... The driver will use the HDQ info to target queue/pipe reset for hardware scheduled user compute queues. Signed-off-by: Jonathan Kim Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 7 ++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 1 + drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 6 +++--- drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 8 +++++--- drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 8 +++++--- 5 files changed, 20 insertions(+), 10 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c index 30e1fb510600..94973018f761 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c @@ -420,12 +420,17 @@ int amdgpu_mes_detect_and_reset_hung_queues(struct amdgpu_device *adev, dev_err(adev->dev, "failed to detect and reset\n"); } else { *hung_db_num = 0; - for (i = 0; i < adev->mes.hung_queue_db_array_size; i++) { + for (i = 0; i < adev->mes.hung_queue_hqd_info_offset; i++) { if (db_array[i] != AMDGPU_MES_INVALID_DB_OFFSET) { hung_db_array[i] = db_array[i]; *hung_db_num += 1; } } + + /* + * TODO: return HQD info for MES scheduled user compute queue reset cases + * stored in hung_db_array hqd info offset to full array size + */ } return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h index 6b506fc72f58..97c137c90f97 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h @@ -149,6 +149,7 @@ struct amdgpu_mes { void *resource_1_addr[AMDGPU_MAX_MES_PIPES]; int hung_queue_db_array_size; + int hung_queue_hqd_info_offset; struct amdgpu_bo *hung_queue_db_array_gpu_obj; uint64_t hung_queue_db_array_gpu_addr; void *hung_queue_db_array_cpu_addr; diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c index 2db9b2c63693..1cd9eaeef38f 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c @@ -208,10 +208,10 @@ static int mes_userq_detect_and_reset(struct amdgpu_device *adev, struct amdgpu_userq_mgr *uqm, *tmp; unsigned int hung_db_num = 0; int queue_id, r, i; - u32 db_array[4]; + u32 db_array[8]; - if (db_array_size > 4) { - dev_err(adev->dev, "DB array size (%d vs 4) too small\n", + if (db_array_size > 8) { + dev_err(adev->dev, "DB array size (%d vs 8) too small\n", db_array_size); return -EINVAL; } diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c index e82188431f79..da575bb1377f 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c @@ -66,7 +66,8 @@ static int mes_v11_0_kiq_hw_fini(struct amdgpu_device *adev); #define GFX_MES_DRAM_SIZE 0x80000 #define MES11_HW_RESOURCE_1_SIZE (128 * AMDGPU_GPU_PAGE_SIZE) -#define MES11_HUNG_DB_OFFSET_ARRAY_SIZE 4 +#define MES11_HUNG_DB_OFFSET_ARRAY_SIZE 8 /* [0:3] = db offset, [4:7] = hqd info */ +#define MES11_HUNG_HQD_INFO_OFFSET 4 static void mes_v11_0_ring_set_wptr(struct amdgpu_ring *ring) { @@ -1720,8 +1721,9 @@ static int mes_v11_0_early_init(struct amdgpu_ip_block *ip_block) struct amdgpu_device *adev = ip_block->adev; int pipe, r; - adev->mes.hung_queue_db_array_size = - MES11_HUNG_DB_OFFSET_ARRAY_SIZE; + adev->mes.hung_queue_db_array_size = MES11_HUNG_DB_OFFSET_ARRAY_SIZE; + adev->mes.hung_queue_hqd_info_offset = MES11_HUNG_HQD_INFO_OFFSET; + for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) { if (!adev->enable_mes_kiq && pipe == AMDGPU_MES_KIQ_PIPE) continue; diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c index e3149196143e..7f3512d9de07 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c @@ -47,7 +47,8 @@ static int mes_v12_0_kiq_hw_fini(struct amdgpu_device *adev); #define MES_EOP_SIZE 2048 -#define MES12_HUNG_DB_OFFSET_ARRAY_SIZE 4 +#define MES12_HUNG_DB_OFFSET_ARRAY_SIZE 8 /* [0:3] = db offset [4:7] hqd info */ +#define MES12_HUNG_HQD_INFO_OFFSET 4 static void mes_v12_0_ring_set_wptr(struct amdgpu_ring *ring) { @@ -1904,8 +1905,9 @@ static int mes_v12_0_early_init(struct amdgpu_ip_block *ip_block) struct amdgpu_device *adev = ip_block->adev; int pipe, r; - adev->mes.hung_queue_db_array_size = - MES12_HUNG_DB_OFFSET_ARRAY_SIZE; + adev->mes.hung_queue_db_array_size = MES12_HUNG_DB_OFFSET_ARRAY_SIZE; + adev->mes.hung_queue_hqd_info_offset = MES12_HUNG_HQD_INFO_OFFSET; + for (pipe = 0; pipe < AMDGPU_MAX_MES_PIPES; pipe++) { r = amdgpu_mes_init_microcode(adev, pipe); if (r) -- cgit v1.2.3 From 277bb0f83e98261018ddd82b7ab8154bb9b93237 Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Thu, 5 Jun 2025 10:18:37 -0400 Subject: drm/amdgpu: enable suspend/resume all for gfx 12 Suspend/resume all gangs has been available for GFX12 for a while now so enable it. Signed-off-by: Jonathan Kim Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c index 94973018f761..4883adcfbb4b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c @@ -691,14 +691,11 @@ out: bool amdgpu_mes_suspend_resume_all_supported(struct amdgpu_device *adev) { uint32_t mes_rev = adev->mes.sched_version & AMDGPU_MES_VERSION_MASK; - bool is_supported = false; - if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0) && - amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(12, 0, 0) && - mes_rev >= 0x63) - is_supported = true; - - return is_supported; + return ((amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0) && + amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(12, 0, 0) && + mes_rev >= 0x63) || + amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(12, 0, 0)); } /* Fix me -- node_id is used to identify the correct MES instances in the future */ -- cgit v1.2.3 From 079ae5118e1f0dcf5b1ab68ffdb5760b06ed79a2 Mon Sep 17 00:00:00 2001 From: Jonathan Kim Date: Wed, 18 Jun 2025 10:31:15 -0400 Subject: drm/amdkfd: fix suspend/resume all calls in mes based eviction path Suspend/resume all gangs should be done with the device lock is held. Signed-off-by: Jonathan Kim Acked-by: Alex Deucher Reviewed-by: Harish Kasiviswanathan Signed-off-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 73 +++++++--------------- 1 file changed, 21 insertions(+), 52 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 6c5c7c1bf5ed..6e7bc983fc0b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1209,6 +1209,15 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, pr_debug_ratelimited("Evicting process pid %d queues\n", pdd->process->lead_thread->pid); + if (dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); + retval = suspend_all_queues_mes(dqm); + if (retval) { + dev_err(dev, "Suspending all queues failed"); + goto out; + } + } + /* Mark all queues as evicted. Deactivate all active queues on * the qpd. */ @@ -1221,23 +1230,27 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, decrement_queue_count(dqm, qpd, q); if (dqm->dev->kfd->shared_resources.enable_mes) { - int err; - - err = remove_queue_mes(dqm, q, qpd); - if (err) { + retval = remove_queue_mes(dqm, q, qpd); + if (retval) { dev_err(dev, "Failed to evict queue %d\n", q->properties.queue_id); - retval = err; + goto out; } } } - pdd->last_evict_timestamp = get_jiffies_64(); - if (!dqm->dev->kfd->shared_resources.enable_mes) + + if (!dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); retval = execute_queues_cpsch(dqm, qpd->is_debug ? KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES : KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0, USE_DEFAULT_GRACE_PERIOD); + } else { + retval = resume_all_queues_mes(dqm); + if (retval) + dev_err(dev, "Resuming all queues failed"); + } out: dqm_unlock(dqm); @@ -3098,61 +3111,17 @@ out: return ret; } -static int kfd_dqm_evict_pasid_mes(struct device_queue_manager *dqm, - struct qcm_process_device *qpd) -{ - struct device *dev = dqm->dev->adev->dev; - int ret = 0; - - /* Check if process is already evicted */ - dqm_lock(dqm); - if (qpd->evicted) { - /* Increment the evicted count to make sure the - * process stays evicted before its terminated. - */ - qpd->evicted++; - dqm_unlock(dqm); - goto out; - } - dqm_unlock(dqm); - - ret = suspend_all_queues_mes(dqm); - if (ret) { - dev_err(dev, "Suspending all queues failed"); - goto out; - } - - ret = dqm->ops.evict_process_queues(dqm, qpd); - if (ret) { - dev_err(dev, "Evicting process queues failed"); - goto out; - } - - ret = resume_all_queues_mes(dqm); - if (ret) - dev_err(dev, "Resuming all queues failed"); - -out: - return ret; -} - int kfd_evict_process_device(struct kfd_process_device *pdd) { struct device_queue_manager *dqm; struct kfd_process *p; - int ret = 0; p = pdd->process; dqm = pdd->dev->dqm; WARN(debug_evictions, "Evicting pid %d", p->lead_thread->pid); - if (dqm->dev->kfd->shared_resources.enable_mes) - ret = kfd_dqm_evict_pasid_mes(dqm, &pdd->qpd); - else - ret = dqm->ops.evict_process_queues(dqm, &pdd->qpd); - - return ret; + return dqm->ops.evict_process_queues(dqm, &pdd->qpd); } int reserve_debug_trap_vmid(struct device_queue_manager *dqm, -- cgit v1.2.3 From e5ae8d1eb08a3e27fff4ae264af4c8056d908639 Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Fri, 12 Sep 2025 15:31:45 -0700 Subject: drm/xe: Increase global invalidation timeout to 1000us The previous timeout of 500us seems to be too small; panning the map in the Roll20 VTT in Firefox on a KDE/Wayland desktop reliably triggered timeouts within a few seconds of usage, causing the monitor to freeze and the following to be printed to dmesg: [Jul30 13:44] xe 0000:03:00.0: [drm] *ERROR* GT0: Global invalidation timeout [Jul30 13:48] xe 0000:03:00.0: [drm] *ERROR* [CRTC:82:pipe A] flip_done timed out I haven't hit a single timeout since increasing it to 1000us even after several multi-hour testing sessions. Fixes: 0dd2dd0182bc ("drm/xe: Move DSB l2 flush to a more sensible place") Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/5710 Signed-off-by: Kenneth Graunke Cc: stable@vger.kernel.org Cc: Maarten Lankhorst Reviewed-by: Shuicheng Lin Link: https://lore.kernel.org/r/20250912223254.147940-1-kenneth@whitecape.org Signed-off-by: Lucas De Marchi (cherry picked from commit 146046907b56578263434107f5a7d5051847c459) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_device.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c index 2883b39c9b37..34d33965eac2 100644 --- a/drivers/gpu/drm/xe/xe_device.c +++ b/drivers/gpu/drm/xe/xe_device.c @@ -1070,7 +1070,7 @@ void xe_device_l2_flush(struct xe_device *xe) spin_lock(>->global_invl_lock); xe_mmio_write32(>->mmio, XE2_GLOBAL_INVAL, 0x1); - if (xe_mmio_wait32(>->mmio, XE2_GLOBAL_INVAL, 0x1, 0x0, 500, NULL, true)) + if (xe_mmio_wait32(>->mmio, XE2_GLOBAL_INVAL, 0x1, 0x0, 1000, NULL, true)) xe_gt_err_once(gt, "Global invalidation timeout\n"); spin_unlock(>->global_invl_lock); -- cgit v1.2.3 From 7ac74613e5f2ef3450f44fd2127198662c2563a9 Mon Sep 17 00:00:00 2001 From: Matthew Brost Date: Thu, 9 Oct 2025 04:06:18 -0700 Subject: drm/xe: Don't allow evicting of BOs in same VM in array of VM binds MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit An array of VM binds can potentially evict other buffer objects (BOs) within the same VM under certain conditions, which may lead to NULL pointer dereferences later in the bind pipeline. To prevent this, clear the allow_res_evict flag in the xe_bo_validate call. v2: - Invert polarity of no_res_evict (Thomas) - Add comment in code explaining issue (Thomas) Cc: stable@vger.kernel.org Reported-by: Paulo Zanoni Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/6268 Fixes: 774b5fa509a9 ("drm/xe: Avoid evicting object of the same vm in none fault mode") Fixes: 77f2ef3f16f5 ("drm/xe: Lock all gpuva ops during VM bind IOCTL") Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs") Signed-off-by: Matthew Brost Tested-by: Paulo Zanoni Reviewed-by: Thomas Hellström Link: https://lore.kernel.org/r/20251009110618.3481870-1-matthew.brost@intel.com (cherry picked from commit 8b9ba8d6d95fe75fed6b0480bb03da4b321bea08) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_vm.c | 32 +++++++++++++++++++++++--------- drivers/gpu/drm/xe/xe_vm_types.h | 1 + 2 files changed, 24 insertions(+), 9 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 027e6ce648c5..f602b874e054 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -2832,7 +2832,7 @@ static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm, } static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma, - bool validate) + bool res_evict, bool validate) { struct xe_bo *bo = xe_vma_bo(vma); struct xe_vm *vm = xe_vma_vm(vma); @@ -2843,7 +2843,8 @@ static int vma_lock_and_validate(struct drm_exec *exec, struct xe_vma *vma, err = drm_exec_lock_obj(exec, &bo->ttm.base); if (!err && validate) err = xe_bo_validate(bo, vm, - !xe_vm_in_preempt_fence_mode(vm), exec); + !xe_vm_in_preempt_fence_mode(vm) && + res_evict, exec); } return err; @@ -2913,14 +2914,23 @@ static int prefetch_ranges(struct xe_vm *vm, struct xe_vma_op *op) } static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm, - struct xe_vma_op *op) + struct xe_vma_ops *vops, struct xe_vma_op *op) { int err = 0; + bool res_evict; + + /* + * We only allow evicting a BO within the VM if it is not part of an + * array of binds, as an array of binds can evict another BO within the + * bind. + */ + res_evict = !(vops->flags & XE_VMA_OPS_ARRAY_OF_BINDS); switch (op->base.op) { case DRM_GPUVA_OP_MAP: if (!op->map.invalidate_on_bind) err = vma_lock_and_validate(exec, op->map.vma, + res_evict, !xe_vm_in_fault_mode(vm) || op->map.immediate); break; @@ -2931,11 +2941,13 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm, err = vma_lock_and_validate(exec, gpuva_to_vma(op->base.remap.unmap->va), - false); + res_evict, false); if (!err && op->remap.prev) - err = vma_lock_and_validate(exec, op->remap.prev, true); + err = vma_lock_and_validate(exec, op->remap.prev, + res_evict, true); if (!err && op->remap.next) - err = vma_lock_and_validate(exec, op->remap.next, true); + err = vma_lock_and_validate(exec, op->remap.next, + res_evict, true); break; case DRM_GPUVA_OP_UNMAP: err = check_ufence(gpuva_to_vma(op->base.unmap.va)); @@ -2944,7 +2956,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm, err = vma_lock_and_validate(exec, gpuva_to_vma(op->base.unmap.va), - false); + res_evict, false); break; case DRM_GPUVA_OP_PREFETCH: { @@ -2959,7 +2971,7 @@ static int op_lock_and_prep(struct drm_exec *exec, struct xe_vm *vm, err = vma_lock_and_validate(exec, gpuva_to_vma(op->base.prefetch.va), - false); + res_evict, false); if (!err && !xe_vma_has_no_bo(vma)) err = xe_bo_migrate(xe_vma_bo(vma), region_to_mem_type[region], @@ -3005,7 +3017,7 @@ static int vm_bind_ioctl_ops_lock_and_prep(struct drm_exec *exec, return err; list_for_each_entry(op, &vops->list, link) { - err = op_lock_and_prep(exec, vm, op); + err = op_lock_and_prep(exec, vm, vops, op); if (err) return err; } @@ -3638,6 +3650,8 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } xe_vma_ops_init(&vops, vm, q, syncs, num_syncs); + if (args->num_binds > 1) + vops.flags |= XE_VMA_OPS_ARRAY_OF_BINDS; for (i = 0; i < args->num_binds; ++i) { u64 range = bind_ops[i].range; u64 addr = bind_ops[i].addr; diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h index da39940501d8..413353e1c225 100644 --- a/drivers/gpu/drm/xe/xe_vm_types.h +++ b/drivers/gpu/drm/xe/xe_vm_types.h @@ -476,6 +476,7 @@ struct xe_vma_ops { /** @flag: signify the properties within xe_vma_ops*/ #define XE_VMA_OPS_FLAG_HAS_SVM_PREFETCH BIT(0) #define XE_VMA_OPS_FLAG_MADVISE BIT(1) +#define XE_VMA_OPS_ARRAY_OF_BINDS BIT(2) u32 flags; #ifdef TEST_VM_OPS_ERROR /** @inject_error: inject error to test error handling */ -- cgit v1.2.3 From d30203739be798d3de5c84db3060e96f00c54e82 Mon Sep 17 00:00:00 2001 From: Lucas De Marchi Date: Thu, 18 Sep 2025 13:58:57 -0700 Subject: drm/xe: Move rebar to be done earlier MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit There may be cases in which the BAR0 also needs to move to accommodate the bigger BAR2. However if it's not released, the BAR2 resize fails. During the vram probe it can't be released as it's already in use by xe_mmio for early register access. Add a new function in xe_vram and let xe_pci call it directly before even early device probe. This allows the BAR2 to resize in cases BAR0 also needs to move, assuming there aren't other reasons to hold that move: [] xe 0000:03:00.0: vgaarb: deactivate vga console [] xe 0000:03:00.0: [drm] Attempting to resize bar from 8192MiB -> 16384MiB [] xe 0000:03:00.0: BAR 0 [mem 0x83000000-0x83ffffff 64bit]: releasing [] xe 0000:03:00.0: BAR 2 [mem 0x4000000000-0x41ffffffff 64bit pref]: releasing [] pcieport 0000:02:01.0: bridge window [mem 0x4000000000-0x41ffffffff 64bit pref]: releasing [] pcieport 0000:01:00.0: bridge window [mem 0x4000000000-0x41ffffffff 64bit pref]: releasing [] pcieport 0000:01:00.0: bridge window [mem 0x4000000000-0x43ffffffff 64bit pref]: assigned [] pcieport 0000:02:01.0: bridge window [mem 0x4000000000-0x43ffffffff 64bit pref]: assigned [] xe 0000:03:00.0: BAR 2 [mem 0x4000000000-0x43ffffffff 64bit pref]: assigned [] xe 0000:03:00.0: BAR 0 [mem 0x83000000-0x83ffffff 64bit]: assigned [] pcieport 0000:00:01.0: PCI bridge to [bus 01-04] [] pcieport 0000:00:01.0: bridge window [mem 0x83000000-0x840fffff] [] pcieport 0000:00:01.0: bridge window [mem 0x4000000000-0x44007fffff 64bit pref] [] pcieport 0000:01:00.0: PCI bridge to [bus 02-04] [] pcieport 0000:01:00.0: bridge window [mem 0x83000000-0x840fffff] [] pcieport 0000:01:00.0: bridge window [mem 0x4000000000-0x43ffffffff 64bit pref] [] pcieport 0000:02:01.0: PCI bridge to [bus 03] [] pcieport 0000:02:01.0: bridge window [mem 0x83000000-0x83ffffff] [] pcieport 0000:02:01.0: bridge window [mem 0x4000000000-0x43ffffffff 64bit pref] [] xe 0000:03:00.0: [drm] BAR2 resized to 16384M [] xe 0000:03:00.0: [drm:xe_pci_probe [xe]] BATTLEMAGE e221:0000 dgfx:1 gfx:Xe2_HPG (20.02) ... For BMG there are additional fix needed in the PCI side, but this helps getting it to a working resize. All the rebar logic is more pci-specific than xe-specific and can be done very early in the probe sequence. In future it would be good to move it out of xe_vram.c, but this refactor is left for later. Cc: Ilpo Järvinen Cc: stable@vger.kernel.org # 6.12+ Link: https://lore.kernel.org/intel-xe/fafda2a3-fc63-ce97-d22b-803f771a4d19@linux.intel.com Reviewed-by: Ilpo Järvinen Link: https://lore.kernel.org/r/20250918-xe-pci-rebar-2-v1-2-6c094702a074@intel.com Signed-off-by: Lucas De Marchi (cherry picked from commit 45e33f220fd625492c11e15733d8e9b4f9db82a4) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_pci.c | 2 ++ drivers/gpu/drm/xe/xe_vram.c | 34 ++++++++++++++++++++++++++-------- drivers/gpu/drm/xe/xe_vram.h | 1 + 3 files changed, 29 insertions(+), 8 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c index be91343829dd..9a6df79fc5b6 100644 --- a/drivers/gpu/drm/xe/xe_pci.c +++ b/drivers/gpu/drm/xe/xe_pci.c @@ -867,6 +867,8 @@ static int xe_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent) if (err) return err; + xe_vram_resize_bar(xe); + err = xe_device_probe_early(xe); /* * In Boot Survivability mode, no drm card is exposed and driver diff --git a/drivers/gpu/drm/xe/xe_vram.c b/drivers/gpu/drm/xe/xe_vram.c index b44ebf50fedb..652df7a5f4f6 100644 --- a/drivers/gpu/drm/xe/xe_vram.c +++ b/drivers/gpu/drm/xe/xe_vram.c @@ -26,15 +26,35 @@ #define BAR_SIZE_SHIFT 20 -static void -_resize_bar(struct xe_device *xe, int resno, resource_size_t size) +/* + * Release all the BARs that could influence/block LMEMBAR resizing, i.e. + * assigned IORESOURCE_MEM_64 BARs + */ +static void release_bars(struct pci_dev *pdev) +{ + struct resource *res; + int i; + + pci_dev_for_each_resource(pdev, res, i) { + /* Resource already un-assigned, do not reset it */ + if (!res->parent) + continue; + + /* No need to release unrelated BARs */ + if (!(res->flags & IORESOURCE_MEM_64)) + continue; + + pci_release_resource(pdev, i); + } +} + +static void resize_bar(struct xe_device *xe, int resno, resource_size_t size) { struct pci_dev *pdev = to_pci_dev(xe->drm.dev); int bar_size = pci_rebar_bytes_to_size(size); int ret; - if (pci_resource_len(pdev, resno)) - pci_release_resource(pdev, resno); + release_bars(pdev); ret = pci_resize_resource(pdev, resno, bar_size); if (ret) { @@ -50,7 +70,7 @@ _resize_bar(struct xe_device *xe, int resno, resource_size_t size) * if force_vram_bar_size is set, attempt to set to the requested size * else set to maximum possible size */ -static void resize_vram_bar(struct xe_device *xe) +void xe_vram_resize_bar(struct xe_device *xe) { int force_vram_bar_size = xe_modparam.force_vram_bar_size; struct pci_dev *pdev = to_pci_dev(xe->drm.dev); @@ -119,7 +139,7 @@ static void resize_vram_bar(struct xe_device *xe) pci_read_config_dword(pdev, PCI_COMMAND, &pci_cmd); pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd & ~PCI_COMMAND_MEMORY); - _resize_bar(xe, LMEM_BAR, rebar_size); + resize_bar(xe, LMEM_BAR, rebar_size); pci_assign_unassigned_bus_resources(pdev->bus); pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd); @@ -148,8 +168,6 @@ static int determine_lmem_bar_size(struct xe_device *xe, struct xe_vram_region * return -ENXIO; } - resize_vram_bar(xe); - lmem_bar->io_start = pci_resource_start(pdev, LMEM_BAR); lmem_bar->io_size = pci_resource_len(pdev, LMEM_BAR); if (!lmem_bar->io_size) diff --git a/drivers/gpu/drm/xe/xe_vram.h b/drivers/gpu/drm/xe/xe_vram.h index 72860f714fc6..13505cfb184d 100644 --- a/drivers/gpu/drm/xe/xe_vram.h +++ b/drivers/gpu/drm/xe/xe_vram.h @@ -11,6 +11,7 @@ struct xe_device; struct xe_vram_region; +void xe_vram_resize_bar(struct xe_device *xe); int xe_vram_probe(struct xe_device *xe); struct xe_vram_region *xe_vram_region_alloc(struct xe_device *xe, u8 id, u32 placement); -- cgit v1.2.3 From 1117e7d1e8e66bf7e40291178b829a8513f83a7a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= Date: Wed, 10 Sep 2025 18:09:39 +0200 Subject: drm/xe/migrate: Fix an error path MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The exhaustive eviction accidently changed an error path goto to a return. Fix this. Fixes: 59eabff2a352 ("drm/xe: Convert xe_bo_create_pin_map() for exhaustive eviction") Cc: Matthew Brost Signed-off-by: Thomas Hellström Reviewed-by: Francois Dugast Link: https://lore.kernel.org/r/20250910160939.103473-1-thomas.hellstrom@linux.intel.com (cherry picked from commit 381f1ed15159c4b3f00dd37cc70924dedebeb111) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_migrate.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c index 1d667fa36cf3..569869a2b339 100644 --- a/drivers/gpu/drm/xe/xe_migrate.c +++ b/drivers/gpu/drm/xe/xe_migrate.c @@ -434,7 +434,7 @@ int xe_migrate_init(struct xe_migrate *m) err = xe_migrate_lock_prepare_vm(tile, m, vm); if (err) - return err; + goto err_out; if (xe->info.has_usm) { struct xe_hw_engine *hwe = xe_gt_hw_engine(primary_gt, -- cgit v1.2.3 From 7413e9f2be6b2b0caff9c517efa123d988914bba Mon Sep 17 00:00:00 2001 From: Matthew Brost Date: Thu, 9 Oct 2025 06:06:29 -0700 Subject: drm/xe: Handle mixed mappings and existing VRAM on atomic faults Moving to VRAM will fail if mixed mappings are present or if the page is already located in VRAM. Atomic faults that require a move to VRAM currently retry without attempting to evict mixed mappings or locate existing VRAM mappings. This patch fixes the issue by attempting to evict mixed mappings or find existing VRAM pages when a move to VRAM fails during atomic fault handling. Fixes: a9ac0fa455b0 ("drm/xe: Strict migration policy for atomic SVM faults") Signed-off-by: Matthew Brost Reviewed-by: Himal Prasad Ghimiray Link: https://lore.kernel.org/r/20251009130629.3531962-1-matthew.brost@intel.com (cherry picked from commit 75188605c56d10c1bd3b1cd94f4872f349c3a9c8) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_svm.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c index 7e2db71ff34e..b268ee0d2271 100644 --- a/drivers/gpu/drm/xe/xe_svm.c +++ b/drivers/gpu/drm/xe/xe_svm.c @@ -1073,7 +1073,17 @@ retry: drm_dbg(&vm->xe->drm, "VRAM allocation failed, falling back to retrying fault, asid=%u, errno=%pe\n", vm->usm.asid, ERR_PTR(err)); - goto retry; + + /* + * In the devmem-only case, mixed mappings may + * be found. The get_pages function will fix + * these up to a single location, allowing the + * page fault handler to make forward progress. + */ + if (ctx.devmem_only) + goto get_pages; + else + goto retry; } else { drm_err(&vm->xe->drm, "VRAM allocation failed, retry count exceeded, asid=%u, errno=%pe\n", @@ -1083,6 +1093,7 @@ retry: } } +get_pages: get_pages_start = xe_svm_stats_ktime_get(); range_debug(range, "GET PAGES"); -- cgit v1.2.3 From 1852d27aa998272696680607b65a2ceac966104e Mon Sep 17 00:00:00 2001 From: Vinay Belgaumkar Date: Thu, 9 Oct 2025 18:10:47 -0700 Subject: drm/xe: Enable media sampler power gating Where applicable, enable media sampler power gating. Also, add it to the powergate_info debugfs. v2: Remove the sampler powergate status since it is cleared quickly anyway. v3: Use vcs mask (Rodrigo) and fix the version check for media v4: Remove extra spaces v5: Media samplers are independent of vcs mask, use Media version 1255 (Matt Roper) Fixes: 38e8c4184ea0 ("drm/xe: Enable Coarse Power Gating") Cc: Rodrigo Vivi Cc: Matt Roper Reviewed-by: Rodrigo Vivi Signed-off-by: Vinay Belgaumkar Link: https://lore.kernel.org/r/20251010011047.2047584-1-vinay.belgaumkar@intel.com Signed-off-by: Rodrigo Vivi (cherry picked from commit 4cbc08649a54c3d533df9832342d52d409dfbbf0) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/regs/xe_gt_regs.h | 1 + drivers/gpu/drm/xe/xe_gt_idle.c | 8 ++++++++ 2 files changed, 9 insertions(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/regs/xe_gt_regs.h b/drivers/gpu/drm/xe/regs/xe_gt_regs.h index 06cb6b02ec64..51f2a03847f9 100644 --- a/drivers/gpu/drm/xe/regs/xe_gt_regs.h +++ b/drivers/gpu/drm/xe/regs/xe_gt_regs.h @@ -342,6 +342,7 @@ #define POWERGATE_ENABLE XE_REG(0xa210) #define RENDER_POWERGATE_ENABLE REG_BIT(0) #define MEDIA_POWERGATE_ENABLE REG_BIT(1) +#define MEDIA_SAMPLERS_POWERGATE_ENABLE REG_BIT(2) #define VDN_HCP_POWERGATE_ENABLE(n) REG_BIT(3 + 2 * (n)) #define VDN_MFXVDENC_POWERGATE_ENABLE(n) REG_BIT(4 + 2 * (n)) diff --git a/drivers/gpu/drm/xe/xe_gt_idle.c b/drivers/gpu/drm/xe/xe_gt_idle.c index f8950a52d0a4..bdc9d9877ec4 100644 --- a/drivers/gpu/drm/xe/xe_gt_idle.c +++ b/drivers/gpu/drm/xe/xe_gt_idle.c @@ -124,6 +124,9 @@ void xe_gt_idle_enable_pg(struct xe_gt *gt) if (xe_gt_is_main_type(gt)) gtidle->powergate_enable |= RENDER_POWERGATE_ENABLE; + if (MEDIA_VERx100(xe) >= 1100 && MEDIA_VERx100(xe) < 1255) + gtidle->powergate_enable |= MEDIA_SAMPLERS_POWERGATE_ENABLE; + if (xe->info.platform != XE_DG1) { for (i = XE_HW_ENGINE_VCS0, j = 0; i <= XE_HW_ENGINE_VCS7; ++i, ++j) { if ((gt->info.engine_mask & BIT(i))) @@ -246,6 +249,11 @@ int xe_gt_idle_pg_print(struct xe_gt *gt, struct drm_printer *p) drm_printf(p, "Media Slice%d Power Gate Status: %s\n", n, str_up_down(pg_status & media_slices[n].status_bit)); } + + if (MEDIA_VERx100(xe) >= 1100 && MEDIA_VERx100(xe) < 1255) + drm_printf(p, "Media Samplers Power Gating Enabled: %s\n", + str_yes_no(pg_enabled & MEDIA_SAMPLERS_POWERGATE_ENABLE)); + return 0; } -- cgit v1.2.3 From 9f64b3cd051b825de0a2a9f145c8e003200cedd5 Mon Sep 17 00:00:00 2001 From: Shuicheng Lin Date: Fri, 10 Oct 2025 17:25:29 +0000 Subject: drm/xe/guc: Check GuC running state before deregistering exec queue In normal operation, a registered exec queue is disabled and deregistered through the GuC, and freed only after the GuC confirms completion. However, if the driver is forced to unbind while the exec queue is still running, the user may call exec_destroy() after the GuC has already been stopped and CT communication disabled. In this case, the driver cannot receive a response from the GuC, preventing proper cleanup of exec queue resources. Fix this by directly releasing the resources when GuC is not running. Here is the failure dmesg log: " [ 468.089581] ---[ end trace 0000000000000000 ]--- [ 468.089608] pci 0000:03:00.0: [drm] *ERROR* GT0: GUC ID manager unclean (1/65535) [ 468.090558] pci 0000:03:00.0: [drm] GT0: total 65535 [ 468.090562] pci 0000:03:00.0: [drm] GT0: used 1 [ 468.090564] pci 0000:03:00.0: [drm] GT0: range 1..1 (1) [ 468.092716] ------------[ cut here ]------------ [ 468.092719] WARNING: CPU: 14 PID: 4775 at drivers/gpu/drm/xe/xe_ttm_vram_mgr.c:298 ttm_vram_mgr_fini+0xf8/0x130 [xe] " v2: use xe_uc_fw_is_running() instead of xe_guc_ct_enabled(). As CT may go down and come back during VF migration. Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs") Cc: stable@vger.kernel.org Cc: Matthew Brost Signed-off-by: Shuicheng Lin Reviewed-by: Matthew Brost Signed-off-by: Matthew Brost Link: https://lore.kernel.org/r/20251010172529.2967639-2-shuicheng.lin@intel.com (cherry picked from commit 9b42321a02c50a12b2beb6ae9469606257fbecea) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_guc_submit.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c index 53024eb5670b..94ed8159496f 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.c +++ b/drivers/gpu/drm/xe/xe_guc_submit.c @@ -44,6 +44,7 @@ #include "xe_ring_ops_types.h" #include "xe_sched_job.h" #include "xe_trace.h" +#include "xe_uc_fw.h" #include "xe_vm.h" static struct xe_guc * @@ -1489,7 +1490,17 @@ static void __guc_exec_queue_process_msg_cleanup(struct xe_sched_msg *msg) xe_gt_assert(guc_to_gt(guc), !(q->flags & EXEC_QUEUE_FLAG_PERMANENT)); trace_xe_exec_queue_cleanup_entity(q); - if (exec_queue_registered(q)) + /* + * Expected state transitions for cleanup: + * - If the exec queue is registered and GuC firmware is running, we must first + * disable scheduling and deregister the queue to ensure proper teardown and + * resource release in the GuC, then destroy the exec queue on driver side. + * - If the GuC is already stopped (e.g., during driver unload or GPU reset), + * we cannot expect a response for the deregister request. In this case, + * it is safe to directly destroy the exec queue on driver side, as the GuC + * will not process further requests and all resources must be cleaned up locally. + */ + if (exec_queue_registered(q) && xe_uc_fw_is_running(&guc->fw)) disable_scheduling_deregister(guc, q); else __guc_exec_queue_destroy(guc, q); -- cgit v1.2.3 From 095232711f23179053ca26bcf046ca121a91a465 Mon Sep 17 00:00:00 2001 From: Francesco Valla Date: Fri, 3 Oct 2025 12:33:03 +0200 Subject: drm/draw: fix color truncation in drm_draw_fill24 The color parameter passed to drm_draw_fill24() was truncated to 16 bits, leading to an incorrect color drawn to the target iosys_map. Fix this behavior, widening the parameter to 32 bits. Fixes: 31fa2c1ca0b2 ("drm/panic: Move drawing functions to drm_draw") Signed-off-by: Francesco Valla Reviewed-by: Jocelyn Falempe Link: https://lore.kernel.org/r/20251003-drm_draw_fill24_fix-v1-1-8fb7c1c2a893@valla.it Signed-off-by: Jocelyn Falempe --- drivers/gpu/drm/drm_draw.c | 2 +- drivers/gpu/drm/drm_draw_internal.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/drm_draw.c b/drivers/gpu/drm/drm_draw.c index 9dc0408fbbea..5b956229c82f 100644 --- a/drivers/gpu/drm/drm_draw.c +++ b/drivers/gpu/drm/drm_draw.c @@ -127,7 +127,7 @@ EXPORT_SYMBOL(drm_draw_fill16); void drm_draw_fill24(struct iosys_map *dmap, unsigned int dpitch, unsigned int height, unsigned int width, - u16 color) + u32 color) { unsigned int y, x; diff --git a/drivers/gpu/drm/drm_draw_internal.h b/drivers/gpu/drm/drm_draw_internal.h index f121ee7339dc..20cb404e23ea 100644 --- a/drivers/gpu/drm/drm_draw_internal.h +++ b/drivers/gpu/drm/drm_draw_internal.h @@ -47,7 +47,7 @@ void drm_draw_fill16(struct iosys_map *dmap, unsigned int dpitch, void drm_draw_fill24(struct iosys_map *dmap, unsigned int dpitch, unsigned int height, unsigned int width, - u16 color); + u32 color); void drm_draw_fill32(struct iosys_map *dmap, unsigned int dpitch, unsigned int height, unsigned int width, -- cgit v1.2.3 From 7f38a1487555604bc4e210fa7cc9b1bce981c40e Mon Sep 17 00:00:00 2001 From: Alok Tiwari Date: Sun, 12 Oct 2025 07:20:01 -0700 Subject: drm/rockchip: vop2: use correct destination rectangle height check The vop2_plane_atomic_check() function incorrectly checks drm_rect_width(dest) twice instead of verifying both width and height. Fix the second condition to use drm_rect_height(dest) so that invalid destination rectangles with height < 4 are correctly rejected. Fixes: 604be85547ce ("drm/rockchip: Add VOP2 driver") Signed-off-by: Alok Tiwari Reviewed-by: Andy Yan Signed-off-by: Heiko Stuebner Link: https://lore.kernel.org/r/20251012142005.660727-1-alok.a.tiwari@oracle.com --- drivers/gpu/drm/rockchip/rockchip_drm_vop2.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_vop2.c b/drivers/gpu/drm/rockchip/rockchip_drm_vop2.c index b50927a824b4..7ec7bea5e38e 100644 --- a/drivers/gpu/drm/rockchip/rockchip_drm_vop2.c +++ b/drivers/gpu/drm/rockchip/rockchip_drm_vop2.c @@ -1031,7 +1031,7 @@ static int vop2_plane_atomic_check(struct drm_plane *plane, return format; if (drm_rect_width(src) >> 16 < 4 || drm_rect_height(src) >> 16 < 4 || - drm_rect_width(dest) < 4 || drm_rect_width(dest) < 4) { + drm_rect_width(dest) < 4 || drm_rect_height(dest) < 4) { drm_err(vop2->drm, "Invalid size: %dx%d->%dx%d, min size is 4x4\n", drm_rect_width(src) >> 16, drm_rect_height(src) >> 16, drm_rect_width(dest), drm_rect_height(dest)); -- cgit v1.2.3 From 6f719373b943a955fee6fc2012aed207b65e2854 Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Tue, 14 Oct 2025 10:46:34 +0200 Subject: drm/ast: Blank with VGACR17 sync enable, always clear VGACRB6 sync off Blank the display by disabling sync pulses with VGACR17<7>. Unblank by reenabling them. This VGA setting should be supported by all Aspeed hardware. Ast currently blanks via sync-off bits in VGACRB6. Not all BMCs handle VGACRB6 correctly. After disabling sync during a reboot, some BMCs do not reenable it after the soft reset. The display output remains dark. When the display is off during boot, some BMCs set the sync-off bits in VGACRB6, so the display remains dark. Observed with Blackbird AST2500 BMCs. Clearing the sync-off bits unconditionally fixes these issues. Also do not modify VGASR1's SD bit for blanking, as it only disables GPU access to video memory. v2: - init vgacrb6 correctly (Jocelyn) Signed-off-by: Thomas Zimmermann Fixes: ce3d99c83495 ("drm: Call drm_atomic_helper_shutdown() at shutdown time for misc drivers") Tested-by: Nick Bowler Reported-by: Nick Bowler Closes: https://lore.kernel.org/dri-devel/wpwd7rit6t4mnu6kdqbtsnk5bhftgslio6e2jgkz6kgw6cuvvr@xbfswsczfqsi/ Cc: Douglas Anderson Cc: Dave Airlie Cc: Thomas Zimmermann Cc: Jocelyn Falempe Cc: dri-devel@lists.freedesktop.org Cc: # v6.7+ Reviewed-by: Jocelyn Falempe Link: https://lore.kernel.org/r/20251014084743.18242-1-tzimmermann@suse.de --- drivers/gpu/drm/ast/ast_mode.c | 18 ++++++++++-------- drivers/gpu/drm/ast/ast_reg.h | 1 + 2 files changed, 11 insertions(+), 8 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/ast/ast_mode.c b/drivers/gpu/drm/ast/ast_mode.c index b4e8edc7c767..30b011ed0a05 100644 --- a/drivers/gpu/drm/ast/ast_mode.c +++ b/drivers/gpu/drm/ast/ast_mode.c @@ -836,22 +836,24 @@ ast_crtc_helper_atomic_flush(struct drm_crtc *crtc, static void ast_crtc_helper_atomic_enable(struct drm_crtc *crtc, struct drm_atomic_state *state) { struct ast_device *ast = to_ast_device(crtc->dev); + u8 vgacr17 = 0x00; + u8 vgacrb6 = 0xff; - ast_set_index_reg_mask(ast, AST_IO_VGACRI, 0xb6, 0xfc, 0x00); - ast_set_index_reg_mask(ast, AST_IO_VGASRI, 0x01, 0xdf, 0x00); + vgacr17 |= AST_IO_VGACR17_SYNC_ENABLE; + vgacrb6 &= ~(AST_IO_VGACRB6_VSYNC_OFF | AST_IO_VGACRB6_HSYNC_OFF); + + ast_set_index_reg_mask(ast, AST_IO_VGACRI, 0x17, 0x7f, vgacr17); + ast_set_index_reg_mask(ast, AST_IO_VGACRI, 0xb6, 0xfc, vgacrb6); } static void ast_crtc_helper_atomic_disable(struct drm_crtc *crtc, struct drm_atomic_state *state) { struct drm_crtc_state *old_crtc_state = drm_atomic_get_old_crtc_state(state, crtc); struct ast_device *ast = to_ast_device(crtc->dev); - u8 vgacrb6; + u8 vgacr17 = 0xff; - ast_set_index_reg_mask(ast, AST_IO_VGASRI, 0x01, 0xdf, AST_IO_VGASR1_SD); - - vgacrb6 = AST_IO_VGACRB6_VSYNC_OFF | - AST_IO_VGACRB6_HSYNC_OFF; - ast_set_index_reg_mask(ast, AST_IO_VGACRI, 0xb6, 0xfc, vgacrb6); + vgacr17 &= ~AST_IO_VGACR17_SYNC_ENABLE; + ast_set_index_reg_mask(ast, AST_IO_VGACRI, 0x17, 0x7f, vgacr17); /* * HW cursors require the underlying primary plane and CRTC to diff --git a/drivers/gpu/drm/ast/ast_reg.h b/drivers/gpu/drm/ast/ast_reg.h index e15adaf3a80e..30578e3b07e4 100644 --- a/drivers/gpu/drm/ast/ast_reg.h +++ b/drivers/gpu/drm/ast/ast_reg.h @@ -29,6 +29,7 @@ #define AST_IO_VGAGRI (0x4E) #define AST_IO_VGACRI (0x54) +#define AST_IO_VGACR17_SYNC_ENABLE BIT(7) /* called "Hardware reset" in docs */ #define AST_IO_VGACR80_PASSWORD (0xa8) #define AST_IO_VGACR99_VGAMEM_RSRV_MASK GENMASK(1, 0) #define AST_IO_VGACRA1_VGAIO_DISABLED BIT(1) -- cgit v1.2.3 From 95355766e5871e9cdc574be5a3b115392ad33aea Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jouni=20H=C3=B6gander?= Date: Mon, 22 Sep 2025 13:27:25 +0300 Subject: drm/i915/psr: Deactivate PSR only on LNL and when selective fetch enabled MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Using intel_psr_exit in frontbuffer flush on older platforms seems to be causing problems. Sending single full frame update using intel_psr_force_update is anyways more optimal compared to psr deactivate/activate -> move back to this approach on PSR1, PSR HW tracking and Panel Replay full frame update and use deactivate/activate only on LunarLake and only when selective fetch is enabled. Tested-by: Lemen Tested-by: Koos Vriezen Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/14946 Signed-off-by: Jouni Högander Reviewed-by: Mika Kahola Link: https://lore.kernel.org/r/20250922102725.2752742-1-jouni.hogander@intel.com (cherry picked from commit 924adb0bbdd8fef25fd229c76e3f602c3e8752ee) Signed-off-by: Rodrigo Vivi --- drivers/gpu/drm/i915/display/intel_psr.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/i915/display/intel_psr.c b/drivers/gpu/drm/i915/display/intel_psr.c index 01bf304c705f..10eb93a34cf2 100644 --- a/drivers/gpu/drm/i915/display/intel_psr.c +++ b/drivers/gpu/drm/i915/display/intel_psr.c @@ -3402,6 +3402,7 @@ static void _psr_flush_handle(struct intel_dp *intel_dp) struct intel_display *display = to_intel_display(intel_dp); if (DISPLAY_VER(display) < 20 && intel_dp->psr.psr2_sel_fetch_enabled) { + /* Selective fetch prior LNL */ if (intel_dp->psr.psr2_sel_fetch_cff_enabled) { /* can we turn CFF off? */ if (intel_dp->psr.busy_frontbuffer_bits == 0) @@ -3420,12 +3421,19 @@ static void _psr_flush_handle(struct intel_dp *intel_dp) intel_psr_configure_full_frame_update(intel_dp); intel_psr_force_update(intel_dp); + } else if (!intel_dp->psr.psr2_sel_fetch_enabled) { + /* + * PSR1 on all platforms + * PSR2 HW tracking + * Panel Replay Full frame update + */ + intel_psr_force_update(intel_dp); } else { + /* Selective update LNL onwards */ intel_psr_exit(intel_dp); } - if ((!intel_dp->psr.psr2_sel_fetch_enabled || DISPLAY_VER(display) >= 20) && - !intel_dp->psr.busy_frontbuffer_bits) + if (!intel_dp->psr.active && !intel_dp->psr.busy_frontbuffer_bits) queue_work(display->wq.unordered, &intel_dp->psr.work); } -- cgit v1.2.3 From 7987b93e3a11a7a95ddf2b21563d3286661b999c Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Thomas=20Hellstr=C3=B6m?= Date: Fri, 10 Oct 2025 12:41:48 +0200 Subject: drm/xe/svm: Ensure data will be migrated to system if indicated by madvise. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit If the location madvise() is set to DRM_XE_PREFERRED_LOC_DEFAULT_SYSTEM, the drm_pagemap in the SVM gpu fault handler will be set to NULL. However there is nothing that explicitly migrates the data to system if it is already present in device memory. In that case, set the device memory owner to NULL to ensure data gets properly migrated to system on page-fault. v2: - Remove redundant dpagemap assignment (Himal Prasad Ghimiray) Signed-off-by: Thomas Hellström Reviewed-by: Matthew Brost #v1 Reviewed-by: Himal Prasad Ghimiray Link: https://lore.kernel.org/r/20251010104149.72783-2-thomas.hellstrom@linux.intel.com Fixes: 10aa5c806030 ("drm/gpusvm, drm/xe: Fix userptr to not allow device private pages") (cherry picked from commit 2cfcea7a745794f9b8e265a309717ca6ba335fc4) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_svm.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c index b268ee0d2271..da2a412f80c0 100644 --- a/drivers/gpu/drm/xe/xe_svm.c +++ b/drivers/gpu/drm/xe/xe_svm.c @@ -1034,6 +1034,9 @@ retry: if (err) return err; + dpagemap = xe_vma_resolve_pagemap(vma, tile); + if (!dpagemap && !ctx.devmem_only) + ctx.device_private_page_owner = NULL; range = xe_svm_range_find_or_insert(vm, fault_addr, vma, &ctx); if (IS_ERR(range)) @@ -1054,7 +1057,6 @@ retry: range_debug(range, "PAGE FAULT"); - dpagemap = xe_vma_resolve_pagemap(vma, tile); if (--migrate_try_count >= 0 && xe_svm_range_needs_migrate_to_vram(range, vma, !!dpagemap || ctx.devmem_only)) { ktime_t migrate_start = xe_svm_stats_ktime_get(); -- cgit v1.2.3 From 6d36f65ba551d28710c3e1aaceecacf19df0cd8f Mon Sep 17 00:00:00 2001 From: Matt Roper Date: Mon, 13 Oct 2025 08:30:15 -0700 Subject: drm/xe/kunit: Fix kerneldoc for parameterized tests Kunit's generate_params() was recently updated to take an additional test context parameter. Xe's IP and platform parameter generators were updated accordingly at the same time, but the new parameter was not added to the functions' kerneldoc, resulting in the following warnings: Warning: drivers/gpu/drm/xe/tests/xe_pci.c:78 function parameter 'test' not described in 'xe_pci_fake_data_gen_params' Warning: drivers/gpu/drm/xe/tests/xe_pci.c:254 function parameter 'test' not described in 'xe_pci_graphics_ip_gen_param' Warning: drivers/gpu/drm/xe/tests/xe_pci.c:278 function parameter 'test' not described in 'xe_pci_media_ip_gen_param' Warning: drivers/gpu/drm/xe/tests/xe_pci.c:302 function parameter 'test' not described in 'xe_pci_id_gen_param' Warning: drivers/gpu/drm/xe/tests/xe_pci.c:390 function parameter 'test' not described in 'xe_pci_live_device_gen_param' 5 warnings as errors Document the new parameter to eliminate the warnings and make CI happy. Fixes: b9a214b5f6aa ("kunit: Pass parameterized test context to generate_params()") Reviewed-by: Shuicheng Lin Link: https://lore.kernel.org/r/20251013153014.2362879-2-matthew.d.roper@intel.com Signed-off-by: Matt Roper (cherry picked from commit 89e347f8a70165d1e8d88a93d875da7742c902ce) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/tests/xe_pci.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/tests/xe_pci.c b/drivers/gpu/drm/xe/tests/xe_pci.c index 69e2840c7ef0..663a79ec960d 100644 --- a/drivers/gpu/drm/xe/tests/xe_pci.c +++ b/drivers/gpu/drm/xe/tests/xe_pci.c @@ -66,6 +66,7 @@ KUNIT_ARRAY_PARAM(platform, cases, xe_pci_fake_data_desc); /** * xe_pci_fake_data_gen_params - Generate struct xe_pci_fake_data parameters + * @test: test context object * @prev: the pointer to the previous parameter to iterate from or NULL * @desc: output buffer with minimum size of KUNIT_PARAM_DESC_SIZE * @@ -242,6 +243,7 @@ KUNIT_ARRAY_PARAM(pci_id, pciidlist, xe_pci_id_kunit_desc); /** * xe_pci_graphics_ip_gen_param - Generate graphics struct xe_ip parameters + * @test: test context object * @prev: the pointer to the previous parameter to iterate from or NULL * @desc: output buffer with minimum size of KUNIT_PARAM_DESC_SIZE * @@ -266,6 +268,7 @@ EXPORT_SYMBOL_IF_KUNIT(xe_pci_graphics_ip_gen_param); /** * xe_pci_media_ip_gen_param - Generate media struct xe_ip parameters + * @test: test context object * @prev: the pointer to the previous parameter to iterate from or NULL * @desc: output buffer with minimum size of KUNIT_PARAM_DESC_SIZE * @@ -290,6 +293,7 @@ EXPORT_SYMBOL_IF_KUNIT(xe_pci_media_ip_gen_param); /** * xe_pci_id_gen_param - Generate struct pci_device_id parameters + * @test: test context object * @prev: the pointer to the previous parameter to iterate from or NULL * @desc: output buffer with minimum size of KUNIT_PARAM_DESC_SIZE * @@ -376,6 +380,7 @@ EXPORT_SYMBOL_IF_KUNIT(xe_pci_fake_device_init); /** * xe_pci_live_device_gen_param - Helper to iterate Xe devices as KUnit parameters + * @test: test context object * @prev: the previously returned value, or NULL for the first iteration * @desc: the buffer for a parameter name * -- cgit v1.2.3 From 6a91af25cdbce2086d85cc4994cf791bda3a2c90 Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Fri, 10 Oct 2025 17:20:21 +0100 Subject: drm/xe/migrate: don't misalign current bytes If current bytes exceeds the max copy size, ensure the clamped size still accounts for the XE_CACHELINE_BYTES alignment, otherwise we trigger the assert in xe_migrate_vram with the size now being out of alignment. Fixes: 8c2d61e0e916 ("drm/xe/migrate: don't overflow max copy size") Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/6212 Signed-off-by: Matthew Auld Cc: Stuart Summers Cc: Matthew Brost Reviewed-by: Matthew Brost Link: https://lore.kernel.org/r/20251010162020.190962-2-matthew.auld@intel.com (cherry picked from commit 641bcf8731d21b56760e3646a39a65f471e9efd1) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_migrate.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c index 569869a2b339..a36ce7dce8cc 100644 --- a/drivers/gpu/drm/xe/xe_migrate.c +++ b/drivers/gpu/drm/xe/xe_migrate.c @@ -2113,7 +2113,9 @@ int xe_migrate_access_memory(struct xe_migrate *m, struct xe_bo *bo, if (current_bytes & ~PAGE_MASK) { int pitch = 4; - current_bytes = min_t(int, current_bytes, S16_MAX * pitch); + current_bytes = min_t(int, current_bytes, + round_down(S16_MAX * pitch, + XE_CACHELINE_BYTES)); } __fence = xe_migrate_vram(m, current_bytes, -- cgit v1.2.3 From 225bc03d85427e7e3821d6f99f4f2d4a09350dda Mon Sep 17 00:00:00 2001 From: Matthew Auld Date: Fri, 10 Oct 2025 16:24:58 +0100 Subject: drm/xe/evict: drop bogus assert MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This assert can trigger here with non pin_map users that select LATE_RESTORE, since the vmap is allowed to be NULL given that save/restore can now use the blitter instead. The check here doesn't seem to have much value anymore given that we no longer move pinned memory, so any existing vmap is left well alone, and doesn't need to be recreated upon restore, so just drop the assert here. Fixes: 86f69c26113c ("drm/xe: use backup object for pinned save/restore") Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/6213 Signed-off-by: Matthew Auld Cc: Thomas Hellström Cc: Matthew Brost Reviewed-by: Thomas Hellström Link: https://lore.kernel.org/r/20251010152457.177884-2-matthew.auld@intel.com (cherry picked from commit a10b4a69c7f8f596d2c5218fbe84430734fab3b2) Signed-off-by: Lucas De Marchi --- drivers/gpu/drm/xe/xe_bo_evict.c | 8 -------- 1 file changed, 8 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/xe/xe_bo_evict.c b/drivers/gpu/drm/xe/xe_bo_evict.c index d5dbc51e8612..bc5b4c5fab81 100644 --- a/drivers/gpu/drm/xe/xe_bo_evict.c +++ b/drivers/gpu/drm/xe/xe_bo_evict.c @@ -182,7 +182,6 @@ int xe_bo_evict_all(struct xe_device *xe) static int xe_bo_restore_and_map_ggtt(struct xe_bo *bo) { - struct xe_device *xe = xe_bo_device(bo); int ret; ret = xe_bo_restore_pinned(bo); @@ -201,13 +200,6 @@ static int xe_bo_restore_and_map_ggtt(struct xe_bo *bo) } } - /* - * We expect validate to trigger a move VRAM and our move code - * should setup the iosys map. - */ - xe_assert(xe, !(bo->flags & XE_BO_FLAG_PINNED_LATE_RESTORE) || - !iosys_map_is_null(&bo->vmap)); - return 0; } -- cgit v1.2.3 From 5801e65206b065b0b2af032f7f1eef222aa2fd83 Mon Sep 17 00:00:00 2001 From: Tvrtko Ursulin Date: Wed, 15 Oct 2025 09:40:15 +0100 Subject: drm/sched: Fix potential double free in drm_sched_job_add_resv_dependencies MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When adding dependencies with drm_sched_job_add_dependency(), that function consumes the fence reference both on success and failure, so in the latter case the dma_fence_put() on the error path (xarray failed to expand) is a double free. Interestingly this bug appears to have been present ever since commit ebd5f74255b9 ("drm/sched: Add dependency tracking"), since the code back then looked like this: drm_sched_job_add_implicit_dependencies(): ... for (i = 0; i < fence_count; i++) { ret = drm_sched_job_add_dependency(job, fences[i]); if (ret) break; } for (; i < fence_count; i++) dma_fence_put(fences[i]); Which means for the failing 'i' the dma_fence_put was already a double free. Possibly there were no users at that time, or the test cases were insufficient to hit it. The bug was then only noticed and fixed after commit 9c2ba265352a ("drm/scheduler: use new iterator in drm_sched_job_add_implicit_dependencies v2") landed, with its fixup of commit 4eaf02d6076c ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies"). At that point it was a slightly different flavour of a double free, which commit 963d0b356935 ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies harder") noticed and attempted to fix. But it only moved the double free from happening inside the drm_sched_job_add_dependency(), when releasing the reference not yet obtained, to the caller, when releasing the reference already released by the former in the failure case. As such it is not easy to identify the right target for the fixes tag so lets keep it simple and just continue the chain. While fixing we also improve the comment and explain the reason for taking the reference and not dropping it. Signed-off-by: Tvrtko Ursulin Fixes: 963d0b356935 ("drm/scheduler: fix drm_sched_job_add_implicit_dependencies harder") Reported-by: Dan Carpenter Closes: https://lore.kernel.org/dri-devel/aNFbXq8OeYl3QSdm@stanley.mountain/ Cc: Christian König Cc: Rob Clark Cc: Daniel Vetter Cc: Matthew Brost Cc: Danilo Krummrich Cc: Philipp Stanner Cc: Christian König Cc: dri-devel@lists.freedesktop.org Cc: stable@vger.kernel.org # v5.16+ Signed-off-by: Philipp Stanner Link: https://lore.kernel.org/r/20251015084015.6273-1-tvrtko.ursulin@igalia.com --- drivers/gpu/drm/scheduler/sched_main.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) (limited to 'drivers/gpu') diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index 46119aacb809..c39f0245e3a9 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -965,13 +965,14 @@ int drm_sched_job_add_resv_dependencies(struct drm_sched_job *job, dma_resv_assert_held(resv); dma_resv_for_each_fence(&cursor, resv, usage, fence) { - /* Make sure to grab an additional ref on the added fence */ - dma_fence_get(fence); - ret = drm_sched_job_add_dependency(job, fence); - if (ret) { - dma_fence_put(fence); + /* + * As drm_sched_job_add_dependency always consumes the fence + * reference (even when it fails), and dma_resv_for_each_fence + * is not obtaining one, we need to grab one before calling. + */ + ret = drm_sched_job_add_dependency(job, dma_fence_get(fence)); + if (ret) return ret; - } } return 0; } -- cgit v1.2.3