From 471db2c2d4f80ee94225a1ef246e4f5011733e50 Mon Sep 17 00:00:00 2001 From: "Lin.Cao" Date: Thu, 15 May 2025 10:07:13 +0800 Subject: drm/scheduler: signal scheduled fence when kill job MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit When an entity from application B is killed, drm_sched_entity_kill() removes all jobs belonging to that entity through drm_sched_entity_kill_jobs_work(). If application A's job depends on a scheduled fence from application B's job, and that fence is not properly signaled during the killing process, application A's dependency cannot be cleared. This leads to application A hanging indefinitely while waiting for a dependency that will never be resolved. Fix this issue by ensuring that scheduled fences are properly signaled when an entity is killed, allowing dependent applications to continue execution. Signed-off-by: Lin.Cao Reviewed-by: Philipp Stanner Signed-off-by: Christian König Link: https://lore.kernel.org/r/20250515020713.1110476-1-lincao12@amd.com --- drivers/gpu/drm/scheduler/sched_entity.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c index bd39db7bb240..e671aa241720 100644 --- a/drivers/gpu/drm/scheduler/sched_entity.c +++ b/drivers/gpu/drm/scheduler/sched_entity.c @@ -176,6 +176,7 @@ static void drm_sched_entity_kill_jobs_work(struct work_struct *wrk) { struct drm_sched_job *job = container_of(wrk, typeof(*job), work); + drm_sched_fence_scheduled(job->s_fence, NULL); drm_sched_fence_finished(job->s_fence, -ESRCH); WARN_ON(job->s_fence->parent); job->sched->ops->free_job(job); -- cgit v1.2.3 From 03bcbbb3995ba5df43af9aba45334e35f2dfe27b Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Tue, 20 May 2025 09:14:00 +0200 Subject: dummycon: Trigger redraw when switching consoles with deferred takeover Signal vt subsystem to redraw console when switching to dummycon with deferred takeover enabled. Makes the console switch to fbcon and displays the available output. With deferred takeover enabled, dummycon acts as the placeholder until the first output to the console happens. At that point, fbcon takes over. If the output happens while dummycon is not active, it cannot inform fbcon. This is the case if the vt subsystem runs in graphics mode. A typical graphical boot starts plymouth, a display manager and a compositor; all while leaving out dummycon. Switching to a text-mode console leaves the console with dummycon even if a getty terminal has been started. Returning true from dummycon's con_switch helper signals the vt subsystem to redraw the screen. If there's output available dummycon's con_putc{s} helpers trigger deferred takeover of fbcon, which sets a display mode and displays the output. If no output is available, dummycon remains active. v2: - make the comment slightly more verbose (Javier) Signed-off-by: Thomas Zimmermann Reported-by: Andrei Borzenkov Closes: https://bugzilla.suse.com/show_bug.cgi?id=1242191 Tested-by: Andrei Borzenkov Acked-by: Javier Martinez Canillas Fixes: 83d83bebf401 ("console/fbcon: Add support for deferred console takeover") Cc: Hans de Goede Cc: linux-fbdev@vger.kernel.org Cc: dri-devel@lists.freedesktop.org Cc: # v4.19+ Link: https://lore.kernel.org/r/20250520071418.8462-1-tzimmermann@suse.de --- drivers/video/console/dummycon.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-) diff --git a/drivers/video/console/dummycon.c b/drivers/video/console/dummycon.c index 139049368fdc..7d02470f19b9 100644 --- a/drivers/video/console/dummycon.c +++ b/drivers/video/console/dummycon.c @@ -85,6 +85,15 @@ static bool dummycon_blank(struct vc_data *vc, enum vesa_blank_mode blank, /* Redraw, so that we get putc(s) for output done while blanked */ return true; } + +static bool dummycon_switch(struct vc_data *vc) +{ + /* + * Redraw, so that we get putc(s) for output done while switched + * away. Informs deferred consoles to take over the display. + */ + return true; +} #else static void dummycon_putc(struct vc_data *vc, u16 c, unsigned int y, unsigned int x) { } @@ -95,6 +104,10 @@ static bool dummycon_blank(struct vc_data *vc, enum vesa_blank_mode blank, { return false; } +static bool dummycon_switch(struct vc_data *vc) +{ + return false; +} #endif static const char *dummycon_startup(void) @@ -124,11 +137,6 @@ static bool dummycon_scroll(struct vc_data *vc, unsigned int top, return false; } -static bool dummycon_switch(struct vc_data *vc) -{ - return false; -} - /* * The console `switch' structure for the dummy console * -- cgit v1.2.3 From a01e93ee44f7ed76f872d0ede82f8d31bf0a048a Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Tue, 6 May 2025 11:13:03 +0200 Subject: accel/ivpu: Improve buffer object logging - Fix missing alloc log when drm_gem_handle_create() fails in drm_vma_node_allow() and open callback is not called - Add ivpu_bo->ctx_id that enables to log the actual context id instead of using 0 as default - Add couple WARNs and errors so we can catch more memory corruption issues Fixes: 37dee2a2f433 ("accel/ivpu: Improve buffer object debug logs") Cc: stable@vger.kernel.org # v6.8+ Reviewed-by: Jeff Hugo Reviewed-by: Lizhi Hou Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250506091303.262034-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_gem.c | 25 +++++++++++++++++-------- drivers/accel/ivpu/ivpu_gem.h | 1 + 2 files changed, 18 insertions(+), 8 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c index 8741c73b92ce..c193a80241f5 100644 --- a/drivers/accel/ivpu/ivpu_gem.c +++ b/drivers/accel/ivpu/ivpu_gem.c @@ -28,7 +28,7 @@ static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, con { ivpu_dbg(vdev, BO, "%6s: bo %8p vpu_addr %9llx size %8zu ctx %d has_pages %d dma_mapped %d mmu_mapped %d wc %d imported %d\n", - action, bo, bo->vpu_addr, ivpu_bo_size(bo), bo->ctx ? bo->ctx->id : 0, + action, bo, bo->vpu_addr, ivpu_bo_size(bo), bo->ctx_id, (bool)bo->base.pages, (bool)bo->base.sgt, bo->mmu_mapped, bo->base.map_wc, (bool)bo->base.base.import_attach); } @@ -94,8 +94,6 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx, ivpu_err(vdev, "Failed to add BO to context %u: %d\n", ctx->id, ret); } - ivpu_dbg_bo(vdev, bo, "alloc"); - mutex_unlock(&bo->lock); drm_dev_exit(idx); @@ -215,7 +213,7 @@ fail_detach: return ERR_PTR(ret); } -static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags) +static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, u32 ctx_id) { struct drm_gem_shmem_object *shmem; struct ivpu_bo *bo; @@ -233,6 +231,7 @@ static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 fla return ERR_CAST(shmem); bo = to_ivpu_bo(&shmem->base); + bo->ctx_id = ctx_id; bo->base.map_wc = flags & DRM_IVPU_BO_WC; bo->flags = flags; @@ -240,6 +239,8 @@ static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 fla list_add_tail(&bo->bo_list_node, &vdev->bo_list); mutex_unlock(&vdev->bo_list_lock); + ivpu_dbg_bo(vdev, bo, "alloc"); + return bo; } @@ -278,8 +279,13 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj) mutex_unlock(&vdev->bo_list_lock); drm_WARN_ON(&vdev->drm, !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ)); + drm_WARN_ON(&vdev->drm, ivpu_bo_size(bo) == 0); + drm_WARN_ON(&vdev->drm, bo->base.vaddr); ivpu_bo_unbind_locked(bo); + drm_WARN_ON(&vdev->drm, bo->mmu_mapped); + drm_WARN_ON(&vdev->drm, bo->ctx); + mutex_destroy(&bo->lock); drm_WARN_ON(obj->dev, bo->base.pages_use_count > 1); @@ -314,7 +320,7 @@ int ivpu_bo_create_ioctl(struct drm_device *dev, void *data, struct drm_file *fi if (size == 0) return -EINVAL; - bo = ivpu_bo_alloc(vdev, size, args->flags); + bo = ivpu_bo_alloc(vdev, size, args->flags, file_priv->ctx.id); if (IS_ERR(bo)) { ivpu_err(vdev, "Failed to allocate BO: %pe (ctx %u size %llu flags 0x%x)", bo, file_priv->ctx.id, args->size, args->flags); @@ -322,7 +328,10 @@ int ivpu_bo_create_ioctl(struct drm_device *dev, void *data, struct drm_file *fi } ret = drm_gem_handle_create(file, &bo->base.base, &args->handle); - if (!ret) + if (ret) + ivpu_err(vdev, "Failed to create handle for BO: %pe (ctx %u size %llu flags 0x%x)", + bo, file_priv->ctx.id, args->size, args->flags); + else args->vpu_addr = bo->vpu_addr; drm_gem_object_put(&bo->base.base); @@ -345,7 +354,7 @@ ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(range->end)); drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(size)); - bo = ivpu_bo_alloc(vdev, size, flags); + bo = ivpu_bo_alloc(vdev, size, flags, IVPU_GLOBAL_CONTEXT_MMU_SSID); if (IS_ERR(bo)) { ivpu_err(vdev, "Failed to allocate BO: %pe (vpu_addr 0x%llx size %llu flags 0x%x)", bo, range->start, size, flags); @@ -452,7 +461,7 @@ static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p) mutex_lock(&bo->lock); drm_printf(p, "%-9p %-3u 0x%-12llx %-10lu 0x%-8x %-4u", - bo, bo->ctx ? bo->ctx->id : 0, bo->vpu_addr, bo->base.base.size, + bo, bo->ctx_id, bo->vpu_addr, bo->base.base.size, bo->flags, kref_read(&bo->base.base.refcount)); if (bo->base.pages) diff --git a/drivers/accel/ivpu/ivpu_gem.h b/drivers/accel/ivpu/ivpu_gem.h index a222a9ec9d61..0c93118c85bd 100644 --- a/drivers/accel/ivpu/ivpu_gem.h +++ b/drivers/accel/ivpu/ivpu_gem.h @@ -21,6 +21,7 @@ struct ivpu_bo { u64 vpu_addr; u32 flags; u32 job_status; /* Valid only for command buffer */ + u32 ctx_id; bool mmu_mapped; }; -- cgit v1.2.3 From 1c2c0e29f24360b3130c005a3c261cb8c7b363c6 Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Tue, 6 May 2025 11:20:30 +0200 Subject: accel/ivpu: Use firmware names from upstream repo Use FW names from linux-firmware repo instead of deprecated ones. The vpu_37xx.bin style names were never released and were only used for internal testing, so it is safe to remove them. Fixes: c140244f0cfb ("accel/ivpu: Add initial Panther Lake support") Cc: stable@vger.kernel.org # v6.13+ Reviewed-by: Lizhi Hou Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250506092030.280276-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_fw.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c index ccaaf6c100c0..9db741695401 100644 --- a/drivers/accel/ivpu/ivpu_fw.c +++ b/drivers/accel/ivpu/ivpu_fw.c @@ -55,18 +55,18 @@ static struct { int gen; const char *name; } fw_names[] = { - { IVPU_HW_IP_37XX, "vpu_37xx.bin" }, + { IVPU_HW_IP_37XX, "intel/vpu/vpu_37xx_v1.bin" }, { IVPU_HW_IP_37XX, "intel/vpu/vpu_37xx_v0.0.bin" }, - { IVPU_HW_IP_40XX, "vpu_40xx.bin" }, + { IVPU_HW_IP_40XX, "intel/vpu/vpu_40xx_v1.bin" }, { IVPU_HW_IP_40XX, "intel/vpu/vpu_40xx_v0.0.bin" }, - { IVPU_HW_IP_50XX, "vpu_50xx.bin" }, + { IVPU_HW_IP_50XX, "intel/vpu/vpu_50xx_v1.bin" }, { IVPU_HW_IP_50XX, "intel/vpu/vpu_50xx_v0.0.bin" }, }; /* Production fw_names from the table above */ -MODULE_FIRMWARE("intel/vpu/vpu_37xx_v0.0.bin"); -MODULE_FIRMWARE("intel/vpu/vpu_40xx_v0.0.bin"); -MODULE_FIRMWARE("intel/vpu/vpu_50xx_v0.0.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_37xx_v1.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_40xx_v1.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_50xx_v1.bin"); static int ivpu_fw_request(struct ivpu_device *vdev) { -- cgit v1.2.3 From 4557cc834712eca4eae7adbd9f0a06bdd8f79c99 Mon Sep 17 00:00:00 2001 From: Karol Wachowski Date: Thu, 15 May 2025 11:41:24 +0200 Subject: accel/ivpu: Reorder Doorbell Unregister and Command Queue Destruction Refactor ivpu_cmdq_unregister() to ensure the doorbell is unregistered before destroying the command queue. The NPU firmware requires doorbells to be unregistered prior to command queue destruction. If doorbell remains registered when command queue destroy command is sent firmware will automatically unregister the doorbell, making subsequent unregister attempts no-operations (NOPs). Ensure compliance with firmware expectations by moving the doorbell unregister call ahead of the command queue destruction logic, thus preventing unnecessary NOP operation. Fixes: 465a3914b254 ("accel/ivpu: Add API for command queue create/destroy/submit") Signed-off-by: Karol Wachowski Reviewed-by: Jeff Hugo Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250515094124.255141-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_job.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c index b28da35c30b6..1c8e283ad985 100644 --- a/drivers/accel/ivpu/ivpu_job.c +++ b/drivers/accel/ivpu/ivpu_job.c @@ -247,6 +247,10 @@ static int ivpu_cmdq_unregister(struct ivpu_file_priv *file_priv, struct ivpu_cm if (!cmdq->db_id) return 0; + ret = ivpu_jsm_unregister_db(vdev, cmdq->db_id); + if (!ret) + ivpu_dbg(vdev, JOB, "DB %d unregistered\n", cmdq->db_id); + if (vdev->fw->sched_mode == VPU_SCHEDULING_MODE_HW) { ret = ivpu_jsm_hws_destroy_cmdq(vdev, file_priv->ctx.id, cmdq->id); if (!ret) @@ -254,10 +258,6 @@ static int ivpu_cmdq_unregister(struct ivpu_file_priv *file_priv, struct ivpu_cm cmdq->id, file_priv->ctx.id); } - ret = ivpu_jsm_unregister_db(vdev, cmdq->db_id); - if (!ret) - ivpu_dbg(vdev, JOB, "DB %d unregistered\n", cmdq->db_id); - xa_erase(&file_priv->vdev->db_xa, cmdq->db_id); cmdq->db_id = 0; -- cgit v1.2.3 From 5dc1ea903588a73fb03b3a3e5a041a7c63a4bccd Mon Sep 17 00:00:00 2001 From: Michael Walle Date: Tue, 20 May 2025 09:41:10 +0200 Subject: drm/panel-simple: fix the warnings for the Evervision VGG644804 The panel lacked the connector type which causes a warning. Adding the connector type reveals wrong bus_flags and bits per pixel. Fix all of it. Fixes: 1319f2178bdf ("drm/panel-simple: add Evervision VGG644804 panel entry") Signed-off-by: Michael Walle Reviewed-by: Neil Armstrong Signed-off-by: Neil Armstrong Link: https://lore.kernel.org/r/20250520074110.655114-1-mwalle@kernel.org --- drivers/gpu/drm/panel/panel-simple.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/panel/panel-simple.c b/drivers/gpu/drm/panel/panel-simple.c index 33a37539de57..3aaac96c0bfb 100644 --- a/drivers/gpu/drm/panel/panel-simple.c +++ b/drivers/gpu/drm/panel/panel-simple.c @@ -2199,13 +2199,14 @@ static const struct display_timing evervision_vgg644804_timing = { static const struct panel_desc evervision_vgg644804 = { .timings = &evervision_vgg644804_timing, .num_timings = 1, - .bpc = 8, + .bpc = 6, .size = { .width = 115, .height = 86, }, .bus_format = MEDIA_BUS_FMT_RGB666_1X7X3_SPWG, - .bus_flags = DRM_BUS_FLAG_DE_HIGH | DRM_BUS_FLAG_PIXDATA_SAMPLE_NEGEDGE, + .bus_flags = DRM_BUS_FLAG_DE_HIGH, + .connector_type = DRM_MODE_CONNECTOR_LVDS, }; static const struct display_timing evervision_vgg804821_timing = { -- cgit v1.2.3 From 98d3f772ca7d6822bdfc8c960f5f909574db97c9 Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Wed, 28 May 2025 17:43:25 +0200 Subject: accel/ivpu: Use dma_resv_lock() instead of a custom mutex This fixes a potential race conditions in: - ivpu_bo_unbind_locked() where we modified the shmem->sgt without holding the dma_resv_lock(). - ivpu_bo_print_info() where we read the shmem->pages without holding the dma_resv_lock(). Using dma_resv_lock() also protects against future syncronisation issues that may arise when accessing drm_gem_shmem_object or drm_gem_object members. Fixes: 42328003ecb6 ("accel/ivpu: Refactor BO creation functions") Cc: stable@vger.kernel.org # v6.9+ Reviewed-by: Lizhi Hou Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250528154325.500684-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_gem.c | 63 +++++++++++++++++++++++-------------------- drivers/accel/ivpu/ivpu_gem.h | 1 - 2 files changed, 34 insertions(+), 30 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c index c193a80241f5..5908268ca45e 100644 --- a/drivers/accel/ivpu/ivpu_gem.c +++ b/drivers/accel/ivpu/ivpu_gem.c @@ -33,6 +33,16 @@ static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, con (bool)bo->base.base.import_attach); } +static inline int ivpu_bo_lock(struct ivpu_bo *bo) +{ + return dma_resv_lock(bo->base.base.resv, NULL); +} + +static inline void ivpu_bo_unlock(struct ivpu_bo *bo) +{ + dma_resv_unlock(bo->base.base.resv); +} + /* * ivpu_bo_pin() - pin the backing physical pages and map them to VPU. * @@ -43,22 +53,22 @@ static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, con int __must_check ivpu_bo_pin(struct ivpu_bo *bo) { struct ivpu_device *vdev = ivpu_bo_to_vdev(bo); + struct sg_table *sgt; int ret = 0; - mutex_lock(&bo->lock); - ivpu_dbg_bo(vdev, bo, "pin"); - drm_WARN_ON(&vdev->drm, !bo->ctx); - if (!bo->mmu_mapped) { - struct sg_table *sgt = drm_gem_shmem_get_pages_sgt(&bo->base); + sgt = drm_gem_shmem_get_pages_sgt(&bo->base); + if (IS_ERR(sgt)) { + ret = PTR_ERR(sgt); + ivpu_err(vdev, "Failed to map BO in IOMMU: %d\n", ret); + return ret; + } - if (IS_ERR(sgt)) { - ret = PTR_ERR(sgt); - ivpu_err(vdev, "Failed to map BO in IOMMU: %d\n", ret); - goto unlock; - } + ivpu_bo_lock(bo); + if (!bo->mmu_mapped) { + drm_WARN_ON(&vdev->drm, !bo->ctx); ret = ivpu_mmu_context_map_sgt(vdev, bo->ctx, bo->vpu_addr, sgt, ivpu_bo_is_snooped(bo)); if (ret) { @@ -69,7 +79,7 @@ int __must_check ivpu_bo_pin(struct ivpu_bo *bo) } unlock: - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); return ret; } @@ -84,7 +94,7 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx, if (!drm_dev_enter(&vdev->drm, &idx)) return -ENODEV; - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); ret = ivpu_mmu_context_insert_node(ctx, range, ivpu_bo_size(bo), &bo->mm_node); if (!ret) { @@ -94,7 +104,7 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx, ivpu_err(vdev, "Failed to add BO to context %u: %d\n", ctx->id, ret); } - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); drm_dev_exit(idx); @@ -105,7 +115,7 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo) { struct ivpu_device *vdev = ivpu_bo_to_vdev(bo); - lockdep_assert(lockdep_is_held(&bo->lock) || !kref_read(&bo->base.base.refcount)); + lockdep_assert(dma_resv_held(bo->base.base.resv) || !kref_read(&bo->base.base.refcount)); if (bo->mmu_mapped) { drm_WARN_ON(&vdev->drm, !bo->ctx); @@ -123,14 +133,12 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo) if (bo->base.base.import_attach) return; - dma_resv_lock(bo->base.base.resv, NULL); if (bo->base.sgt) { dma_unmap_sgtable(vdev->drm.dev, bo->base.sgt, DMA_BIDIRECTIONAL, 0); sg_free_table(bo->base.sgt); kfree(bo->base.sgt); bo->base.sgt = NULL; } - dma_resv_unlock(bo->base.base.resv); } void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx) @@ -142,12 +150,12 @@ void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_m mutex_lock(&vdev->bo_list_lock); list_for_each_entry(bo, &vdev->bo_list, bo_list_node) { - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); if (bo->ctx == ctx) { ivpu_dbg_bo(vdev, bo, "unbind"); ivpu_bo_unbind_locked(bo); } - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); } mutex_unlock(&vdev->bo_list_lock); } @@ -167,7 +175,6 @@ struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t siz bo->base.pages_mark_dirty_on_put = true; /* VPU can dirty a BO anytime */ INIT_LIST_HEAD(&bo->bo_list_node); - mutex_init(&bo->lock); return &bo->base.base; } @@ -286,8 +293,6 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj) drm_WARN_ON(&vdev->drm, bo->mmu_mapped); drm_WARN_ON(&vdev->drm, bo->ctx); - mutex_destroy(&bo->lock); - drm_WARN_ON(obj->dev, bo->base.pages_use_count > 1); drm_gem_shmem_free(&bo->base); } @@ -370,9 +375,9 @@ ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, goto err_put; if (flags & DRM_IVPU_BO_MAPPABLE) { - dma_resv_lock(bo->base.base.resv, NULL); + ivpu_bo_lock(bo); ret = drm_gem_shmem_vmap(&bo->base, &map); - dma_resv_unlock(bo->base.base.resv); + ivpu_bo_unlock(bo); if (ret) goto err_put; @@ -395,9 +400,9 @@ void ivpu_bo_free(struct ivpu_bo *bo) struct iosys_map map = IOSYS_MAP_INIT_VADDR(bo->base.vaddr); if (bo->flags & DRM_IVPU_BO_MAPPABLE) { - dma_resv_lock(bo->base.base.resv, NULL); + ivpu_bo_lock(bo); drm_gem_shmem_vunmap(&bo->base, &map); - dma_resv_unlock(bo->base.base.resv); + ivpu_bo_unlock(bo); } drm_gem_object_put(&bo->base.base); @@ -416,12 +421,12 @@ int ivpu_bo_info_ioctl(struct drm_device *dev, void *data, struct drm_file *file bo = to_ivpu_bo(obj); - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); args->flags = bo->flags; args->mmap_offset = drm_vma_node_offset_addr(&obj->vma_node); args->vpu_addr = bo->vpu_addr; args->size = obj->size; - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); drm_gem_object_put(obj); return ret; @@ -458,7 +463,7 @@ int ivpu_bo_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p) { - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); drm_printf(p, "%-9p %-3u 0x%-12llx %-10lu 0x%-8x %-4u", bo, bo->ctx_id, bo->vpu_addr, bo->base.base.size, @@ -475,7 +480,7 @@ static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p) drm_printf(p, "\n"); - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); } void ivpu_bo_list(struct drm_device *dev, struct drm_printer *p) diff --git a/drivers/accel/ivpu/ivpu_gem.h b/drivers/accel/ivpu/ivpu_gem.h index 0c93118c85bd..aa8ff14f7aae 100644 --- a/drivers/accel/ivpu/ivpu_gem.h +++ b/drivers/accel/ivpu/ivpu_gem.h @@ -17,7 +17,6 @@ struct ivpu_bo { struct list_head bo_list_node; struct drm_mm_node mm_node; - struct mutex lock; /* Protects: ctx, mmu_mapped, vpu_addr */ u64 vpu_addr; u32 flags; u32 job_status; /* Valid only for command buffer */ -- cgit v1.2.3 From a47e36dc5d90dc664cac87304c17d50f1595d634 Mon Sep 17 00:00:00 2001 From: Karol Wachowski Date: Wed, 28 May 2025 17:42:53 +0200 Subject: accel/ivpu: Trigger device recovery on engine reset/resume failure Trigger full device recovery when the driver fails to restore device state via engine reset and resume operations. This is necessary because, even if submissions from a faulty context are blocked, the NPU may still process previously submitted faulty jobs if the engine reset fails to abort them. Such jobs can continue to generate faults and occupy device resources. When engine reset is ineffective, the only way to recover is to perform a full device recovery. Fixes: dad945c27a42 ("accel/ivpu: Add handling of VPU_JSM_STATUS_MVNCI_CONTEXT_VIOLATION_HW") Cc: stable@vger.kernel.org # v6.15+ Signed-off-by: Karol Wachowski Reviewed-by: Lizhi Hou Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250528154253.500556-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_job.c | 6 ++++-- drivers/accel/ivpu/ivpu_jsm_msg.c | 9 +++++++-- 2 files changed, 11 insertions(+), 4 deletions(-) diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c index 1c8e283ad985..fae8351aa330 100644 --- a/drivers/accel/ivpu/ivpu_job.c +++ b/drivers/accel/ivpu/ivpu_job.c @@ -986,7 +986,8 @@ void ivpu_context_abort_work_fn(struct work_struct *work) return; if (vdev->fw->sched_mode == VPU_SCHEDULING_MODE_HW) - ivpu_jsm_reset_engine(vdev, 0); + if (ivpu_jsm_reset_engine(vdev, 0)) + return; mutex_lock(&vdev->context_list_lock); xa_for_each(&vdev->context_xa, ctx_id, file_priv) { @@ -1009,7 +1010,8 @@ void ivpu_context_abort_work_fn(struct work_struct *work) if (vdev->fw->sched_mode != VPU_SCHEDULING_MODE_HW) goto runtime_put; - ivpu_jsm_hws_resume_engine(vdev, 0); + if (ivpu_jsm_hws_resume_engine(vdev, 0)) + return; /* * In hardware scheduling mode NPU already has stopped processing jobs * and won't send us any further notifications, thus we have to free job related resources diff --git a/drivers/accel/ivpu/ivpu_jsm_msg.c b/drivers/accel/ivpu/ivpu_jsm_msg.c index 219ab8afefab..0256b2dfefc1 100644 --- a/drivers/accel/ivpu/ivpu_jsm_msg.c +++ b/drivers/accel/ivpu/ivpu_jsm_msg.c @@ -7,6 +7,7 @@ #include "ivpu_hw.h" #include "ivpu_ipc.h" #include "ivpu_jsm_msg.h" +#include "ivpu_pm.h" #include "vpu_jsm_api.h" const char *ivpu_jsm_msg_type_to_str(enum vpu_ipc_msg_type type) @@ -163,8 +164,10 @@ int ivpu_jsm_reset_engine(struct ivpu_device *vdev, u32 engine) ret = ivpu_ipc_send_receive(vdev, &req, VPU_JSM_MSG_ENGINE_RESET_DONE, &resp, VPU_IPC_CHAN_ASYNC_CMD, vdev->timeout.jsm); - if (ret) + if (ret) { ivpu_err_ratelimited(vdev, "Failed to reset engine %d: %d\n", engine, ret); + ivpu_pm_trigger_recovery(vdev, "Engine reset failed"); + } return ret; } @@ -354,8 +357,10 @@ int ivpu_jsm_hws_resume_engine(struct ivpu_device *vdev, u32 engine) ret = ivpu_ipc_send_receive(vdev, &req, VPU_JSM_MSG_HWS_RESUME_ENGINE_DONE, &resp, VPU_IPC_CHAN_ASYNC_CMD, vdev->timeout.jsm); - if (ret) + if (ret) { ivpu_err_ratelimited(vdev, "Failed to resume engine %d: %d\n", engine, ret); + ivpu_pm_trigger_recovery(vdev, "Engine resume failed"); + } return ret; } -- cgit v1.2.3 From 91274fd4ed9ba110b02c53d71d2778b7d13b49ac Mon Sep 17 00:00:00 2001 From: Jacek Lawrynowicz Date: Wed, 28 May 2025 19:12:20 +0200 Subject: accel/ivpu: Fix warning in ivpu_gem_bo_free() Don't WARN if imported buffers are in use in ivpu_gem_bo_free() as they can be indeed used in the original context/driver. Fixes: 647371a6609d ("accel/ivpu: Add GEM buffer object management") Cc: stable@vger.kernel.org # v6.3 Reviewed-by: Jeff Hugo Reviewed-by: Lizhi Hou Signed-off-by: Jacek Lawrynowicz Link: https://lore.kernel.org/r/20250528171220.513225-1-jacek.lawrynowicz@linux.intel.com --- drivers/accel/ivpu/ivpu_gem.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c index 5908268ca45e..248bfebeaa22 100644 --- a/drivers/accel/ivpu/ivpu_gem.c +++ b/drivers/accel/ivpu/ivpu_gem.c @@ -285,7 +285,8 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj) list_del(&bo->bo_list_node); mutex_unlock(&vdev->bo_list_lock); - drm_WARN_ON(&vdev->drm, !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ)); + drm_WARN_ON(&vdev->drm, !drm_gem_is_imported(&bo->base.base) && + !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ)); drm_WARN_ON(&vdev->drm, ivpu_bo_size(bo) == 0); drm_WARN_ON(&vdev->drm, bo->base.vaddr); -- cgit v1.2.3 From 2f29b5c231011b94007d2c8a6d793992f2275db1 Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Wed, 28 May 2025 10:02:08 +0200 Subject: video: screen_info: Relocate framebuffers behind PCI bridges Apply PCI host-bridge window offsets to screen_info framebuffers. Fixes invalid access to I/O memory. Resources behind a PCI host bridge can be relocated by a certain offset in the kernel's CPU address range used for I/O. The framebuffer memory range stored in screen_info refers to the CPU addresses as seen during boot (where the offset is 0). During boot up, firmware may assign a different memory offset to the PCI host bridge and thereby relocating the framebuffer address of the PCI graphics device as seen by the kernel. The information in screen_info must be updated as well. The helper pcibios_bus_to_resource() performs the relocation of the screen_info's framebuffer resource (given in PCI bus addresses). The result matches the I/O-memory resource of the PCI graphics device (given in CPU addresses). As before, we store away the information necessary to later update the information in screen_info itself. Commit 78aa89d1dfba ("firmware/sysfb: Update screen_info for relocated EFI framebuffers") added the code for updating screen_info. It is based on similar functionality that pre-existed in efifb. Efifb uses a pointer to the PCI resource, while the newer code does a memcpy of the region. Hence efifb sees any updates to the PCI resource and avoids the issue. v3: - Only use struct pci_bus_region for PCI bus addresses (Bjorn) - Clarify address semantics in commit messages and comments (Bjorn) v2: - Fixed tags (Takashi, Ivan) - Updated information on efifb Signed-off-by: Thomas Zimmermann Reviewed-by: Javier Martinez Canillas Reported-by: "Ivan T. Ivanov" Closes: https://bugzilla.suse.com/show_bug.cgi?id=1240696 Tested-by: "Ivan T. Ivanov" Fixes: 78aa89d1dfba ("firmware/sysfb: Update screen_info for relocated EFI framebuffers") Cc: dri-devel@lists.freedesktop.org Cc: # v6.9+ Link: https://lore.kernel.org/r/20250528080234.7380-1-tzimmermann@suse.de --- drivers/video/screen_info_pci.c | 79 ++++++++++++++++++++++++++--------------- 1 file changed, 50 insertions(+), 29 deletions(-) diff --git a/drivers/video/screen_info_pci.c b/drivers/video/screen_info_pci.c index 6c5833517141..66bfc1d0a6dc 100644 --- a/drivers/video/screen_info_pci.c +++ b/drivers/video/screen_info_pci.c @@ -7,8 +7,8 @@ static struct pci_dev *screen_info_lfb_pdev; static size_t screen_info_lfb_bar; -static resource_size_t screen_info_lfb_offset; -static struct resource screen_info_lfb_res = DEFINE_RES_MEM(0, 0); +static resource_size_t screen_info_lfb_res_start; // original start of resource +static resource_size_t screen_info_lfb_offset; // framebuffer offset within resource static bool __screen_info_relocation_is_valid(const struct screen_info *si, struct resource *pr) { @@ -31,7 +31,7 @@ void screen_info_apply_fixups(void) if (screen_info_lfb_pdev) { struct resource *pr = &screen_info_lfb_pdev->resource[screen_info_lfb_bar]; - if (pr->start != screen_info_lfb_res.start) { + if (pr->start != screen_info_lfb_res_start) { if (__screen_info_relocation_is_valid(si, pr)) { /* * Only update base if we have an actual @@ -47,46 +47,67 @@ void screen_info_apply_fixups(void) } } +static int __screen_info_lfb_pci_bus_region(const struct screen_info *si, unsigned int type, + struct pci_bus_region *r) +{ + u64 base, size; + + base = __screen_info_lfb_base(si); + if (!base) + return -EINVAL; + + size = __screen_info_lfb_size(si, type); + if (!size) + return -EINVAL; + + r->start = base; + r->end = base + size - 1; + + return 0; +} + static void screen_info_fixup_lfb(struct pci_dev *pdev) { unsigned int type; - struct resource res[SCREEN_INFO_MAX_RESOURCES]; - size_t i, numres; + struct pci_bus_region bus_region; int ret; + struct resource r = { + .flags = IORESOURCE_MEM, + }; + const struct resource *pr; const struct screen_info *si = &screen_info; if (screen_info_lfb_pdev) return; // already found type = screen_info_video_type(si); - if (type != VIDEO_TYPE_EFI) - return; // only applies to EFI + if (!__screen_info_has_lfb(type)) + return; // only applies to EFI; maybe VESA - ret = screen_info_resources(si, res, ARRAY_SIZE(res)); + ret = __screen_info_lfb_pci_bus_region(si, type, &bus_region); if (ret < 0) return; - numres = ret; - for (i = 0; i < numres; ++i) { - struct resource *r = &res[i]; - const struct resource *pr; - - if (!(r->flags & IORESOURCE_MEM)) - continue; - pr = pci_find_resource(pdev, r); - if (!pr) - continue; - - /* - * We've found a PCI device with the framebuffer - * resource. Store away the parameters to track - * relocation of the framebuffer aperture. - */ - screen_info_lfb_pdev = pdev; - screen_info_lfb_bar = pr - pdev->resource; - screen_info_lfb_offset = r->start - pr->start; - memcpy(&screen_info_lfb_res, r, sizeof(screen_info_lfb_res)); - } + /* + * Translate the PCI bus address to resource. Account + * for an offset if the framebuffer is behind a PCI host + * bridge. + */ + pcibios_bus_to_resource(pdev->bus, &r, &bus_region); + + pr = pci_find_resource(pdev, &r); + if (!pr) + return; + + /* + * We've found a PCI device with the framebuffer + * resource. Store away the parameters to track + * relocation of the framebuffer aperture. + */ + screen_info_lfb_pdev = pdev; + screen_info_lfb_bar = pr - pdev->resource; + screen_info_lfb_offset = r.start - pr->start; + screen_info_lfb_res_start = bus_region.start; } DECLARE_PCI_FIXUP_CLASS_HEADER(PCI_ANY_ID, PCI_ANY_ID, PCI_BASE_CLASS_DISPLAY, 16, screen_info_fixup_lfb); -- cgit v1.2.3 From f670b50ef5e4a69bf4d2ec5ac6a9228d93b13a7a Mon Sep 17 00:00:00 2001 From: Thomas Zimmermann Date: Tue, 3 Jun 2025 17:48:20 +0200 Subject: sysfb: Fix screen_info type check for VGA MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Use the helper screen_info_video_type() to get the framebuffer type from struct screen_info. Handle supported values in sorted switch statement. Reading orig_video_isVGA is unreliable. On most systems it is a VIDEO_TYPE_ constant. On some systems with VGA it is simply set to 1 to signal the presence of a VGA output. See vga_probe() for an example. Retrieving the screen_info type with the helper screen_info_video_type() detects these cases and returns the appropriate VIDEO_TYPE_ constant. For VGA, sysfb creates a device named "vga-framebuffer". The sysfb code has been taken from vga16fb, where it likely didn't work correctly either. With this bugfix applied, vga16fb loads for compatible vga-framebuffer devices. Fixes: 0db5b61e0dc0 ("fbdev/vga16fb: Create EGA/VGA devices in sysfb code") Cc: Thomas Zimmermann Cc: Javier Martinez Canillas Cc: Alex Deucher Cc: Tzung-Bi Shih Cc: Helge Deller Cc: "Uwe Kleine-König" Cc: Zsolt Kajtar Cc: # v6.1+ Signed-off-by: Thomas Zimmermann Reviewed-by: Tzung-Bi Shih Reviewed-by: Javier Martinez Canillas Link: https://lore.kernel.org/r/20250603154838.401882-1-tzimmermann@suse.de --- drivers/firmware/sysfb.c | 26 ++++++++++++++++++-------- 1 file changed, 18 insertions(+), 8 deletions(-) diff --git a/drivers/firmware/sysfb.c b/drivers/firmware/sysfb.c index 7c5c03f274b9..889e5b05c739 100644 --- a/drivers/firmware/sysfb.c +++ b/drivers/firmware/sysfb.c @@ -143,6 +143,7 @@ static __init int sysfb_init(void) { struct screen_info *si = &screen_info; struct device *parent; + unsigned int type; struct simplefb_platform_data mode; const char *name; bool compatible; @@ -170,17 +171,26 @@ static __init int sysfb_init(void) goto put_device; } + type = screen_info_video_type(si); + /* if the FB is incompatible, create a legacy framebuffer device */ - if (si->orig_video_isVGA == VIDEO_TYPE_EFI) - name = "efi-framebuffer"; - else if (si->orig_video_isVGA == VIDEO_TYPE_VLFB) - name = "vesa-framebuffer"; - else if (si->orig_video_isVGA == VIDEO_TYPE_VGAC) - name = "vga-framebuffer"; - else if (si->orig_video_isVGA == VIDEO_TYPE_EGAC) + switch (type) { + case VIDEO_TYPE_EGAC: name = "ega-framebuffer"; - else + break; + case VIDEO_TYPE_VGAC: + name = "vga-framebuffer"; + break; + case VIDEO_TYPE_VLFB: + name = "vesa-framebuffer"; + break; + case VIDEO_TYPE_EFI: + name = "efi-framebuffer"; + break; + default: name = "platform-framebuffer"; + break; + } pd = platform_device_alloc(name, 0); if (!pd) { -- cgit v1.2.3