diff options
Diffstat (limited to 'drivers/gpu/drm')
96 files changed, 2098 insertions, 1477 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 306f75700bf8..048cfe073dae 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -496,6 +496,7 @@ struct amdgpu_bo_va_mapping { /* bo virtual addresses in a specific vm */ struct amdgpu_bo_va { + struct mutex mutex; /* protected by bo being reserved */ struct list_head bo_list; struct fence *last_pt_update; @@ -538,6 +539,7 @@ struct amdgpu_bo { /* Constant after initialization */ struct amdgpu_device *adev; struct drm_gem_object gem_base; + struct amdgpu_bo *parent; struct ttm_bo_kmap_obj dma_buf_vmap; pid_t pid; @@ -928,8 +930,6 @@ struct amdgpu_vm_id { }; struct amdgpu_vm { - struct mutex mutex; - struct rb_root va; /* protecting invalidated */ @@ -956,6 +956,8 @@ struct amdgpu_vm { struct amdgpu_vm_id ids[AMDGPU_MAX_RINGS]; /* for interval tree */ spinlock_t it_lock; + /* protecting freed */ + spinlock_t freed_lock; }; struct amdgpu_vm_manager { @@ -1262,7 +1264,8 @@ struct amdgpu_cs_parser { struct ww_acquire_ctx ticket; /* user fence */ - struct amdgpu_user_fence uf; + struct amdgpu_user_fence uf; + struct amdgpu_bo_list_entry uf_entry; }; struct amdgpu_job { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 3afcf0237c25..25a3e2485cc2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -127,6 +127,37 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, return 0; } +static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p, + struct drm_amdgpu_cs_chunk_fence *fence_data) +{ + struct drm_gem_object *gobj; + uint32_t handle; + + handle = fence_data->handle; + gobj = drm_gem_object_lookup(p->adev->ddev, p->filp, + fence_data->handle); + if (gobj == NULL) + return -EINVAL; + + p->uf.bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); + p->uf.offset = fence_data->offset; + + if (amdgpu_ttm_tt_has_userptr(p->uf.bo->tbo.ttm)) { + drm_gem_object_unreference_unlocked(gobj); + return -EINVAL; + } + + p->uf_entry.robj = amdgpu_bo_ref(p->uf.bo); + p->uf_entry.prefered_domains = AMDGPU_GEM_DOMAIN_GTT; + p->uf_entry.allowed_domains = AMDGPU_GEM_DOMAIN_GTT; + p->uf_entry.priority = 0; + p->uf_entry.tv.bo = &p->uf_entry.robj->tbo; + p->uf_entry.tv.shared = true; + + drm_gem_object_unreference_unlocked(gobj); + return 0; +} + int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) { union drm_amdgpu_cs *cs = data; @@ -207,26 +238,15 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) case AMDGPU_CHUNK_ID_FENCE: size = sizeof(struct drm_amdgpu_cs_chunk_fence); - if (p->chunks[i].length_dw * sizeof(uint32_t) >= size) { - uint32_t handle; - struct drm_gem_object *gobj; - struct drm_amdgpu_cs_chunk_fence *fence_data; - - fence_data = (void *)p->chunks[i].kdata; - handle = fence_data->handle; - gobj = drm_gem_object_lookup(p->adev->ddev, - p->filp, handle); - if (gobj == NULL) { - ret = -EINVAL; - goto free_partial_kdata; - } - - p->uf.bo = gem_to_amdgpu_bo(gobj); - p->uf.offset = fence_data->offset; - } else { + if (p->chunks[i].length_dw * sizeof(uint32_t) < size) { ret = -EINVAL; goto free_partial_kdata; } + + ret = amdgpu_cs_user_fence_chunk(p, (void *)p->chunks[i].kdata); + if (ret) + goto free_partial_kdata; + break; case AMDGPU_CHUNK_ID_DEPENDENCIES: @@ -389,6 +409,9 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p) p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm, &p->validated); + if (p->uf.bo) + list_add(&p->uf_entry.tv.head, &p->validated); + if (need_mmap_lock) down_read(¤t->mm->mmap_sem); @@ -486,8 +509,8 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bo for (i = 0; i < parser->num_ibs; i++) amdgpu_ib_free(parser->adev, &parser->ibs[i]); kfree(parser->ibs); - if (parser->uf.bo) - drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); + amdgpu_bo_unref(&parser->uf.bo); + amdgpu_bo_unref(&parser->uf_entry.robj); } static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, @@ -776,7 +799,7 @@ static int amdgpu_cs_free_job(struct amdgpu_job *job) amdgpu_ib_free(job->adev, &job->ibs[i]); kfree(job->ibs); if (job->uf.bo) - drm_gem_object_unreference_unlocked(&job->uf.bo->gem_base); + amdgpu_bo_unref(&job->uf.bo); return 0; } @@ -784,8 +807,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { struct amdgpu_device *adev = dev->dev_private; union drm_amdgpu_cs *cs = data; - struct amdgpu_fpriv *fpriv = filp->driver_priv; - struct amdgpu_vm *vm = &fpriv->vm; struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; int i, r; @@ -803,7 +824,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) r = amdgpu_cs_handle_lockup(adev, r); return r; } - mutex_lock(&vm->mutex); r = amdgpu_cs_parser_relocs(&parser); if (r == -ENOMEM) DRM_ERROR("Not enough memory for command submission!\n"); @@ -888,7 +908,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) out: amdgpu_cs_parser_fini(&parser, r, reserved_buffers); - mutex_unlock(&vm->mutex); r = amdgpu_cs_handle_lockup(adev, r); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c index e173a5a02f0d..5580d3420c3a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c @@ -73,6 +73,8 @@ static void amdgpu_flip_work_func(struct work_struct *__work) struct drm_crtc *crtc = &amdgpuCrtc->base; unsigned long flags; unsigned i; + int vpos, hpos, stat, min_udelay; + struct drm_vblank_crtc *vblank = &crtc->dev->vblank[work->crtc_id]; amdgpu_flip_wait_fence(adev, &work->excl); for (i = 0; i < work->shared_count; ++i) @@ -81,6 +83,41 @@ static void amdgpu_flip_work_func(struct work_struct *__work) /* We borrow the event spin lock for protecting flip_status */ spin_lock_irqsave(&crtc->dev->event_lock, flags); + /* If this happens to execute within the "virtually extended" vblank + * interval before the start of the real vblank interval then it needs + * to delay programming the mmio flip until the real vblank is entered. + * This prevents completing a flip too early due to the way we fudge + * our vblank counter and vblank timestamps in order to work around the + * problem that the hw fires vblank interrupts before actual start of + * vblank (when line buffer refilling is done for a frame). It + * complements the fudging logic in amdgpu_get_crtc_scanoutpos() for + * timestamping and amdgpu_get_vblank_counter_kms() for vblank counts. + * + * In practice this won't execute very often unless on very fast + * machines because the time window for this to happen is very small. + */ + for (;;) { + /* GET_DISTANCE_TO_VBLANKSTART returns distance to real vblank + * start in hpos, and to the "fudged earlier" vblank start in + * vpos. + */ + stat = amdgpu_get_crtc_scanoutpos(adev->ddev, work->crtc_id, + GET_DISTANCE_TO_VBLANKSTART, + &vpos, &hpos, NULL, NULL, + &crtc->hwmode); + + if ((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) != + (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE) || + !(vpos >= 0 && hpos <= 0)) + break; + + /* Sleep at least until estimated real start of hw vblank */ + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); + min_udelay = (-hpos + 1) * max(vblank->linedur_ns / 1000, 5); + usleep_range(min_udelay, 2 * min_udelay); + spin_lock_irqsave(&crtc->dev->event_lock, flags); + }; + /* do the flip (mmio) */ adev->mode_info.funcs->page_flip(adev, work->crtc_id, work->base); /* set the flip status */ @@ -109,7 +146,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work) } else DRM_ERROR("failed to reserve buffer after flip\n"); - drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base); + amdgpu_bo_unref(&work->old_rbo); kfree(work->shared); kfree(work); } @@ -148,8 +185,8 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc, obj = old_amdgpu_fb->obj; /* take a reference to the old object */ - drm_gem_object_reference(obj); work->old_rbo = gem_to_amdgpu_bo(obj); + amdgpu_bo_ref(work->old_rbo); new_amdgpu_fb = to_amdgpu_framebuffer(fb); obj = new_amdgpu_fb->obj; @@ -222,7 +259,7 @@ pflip_cleanup: amdgpu_bo_unreserve(new_rbo); cleanup: - drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base); + amdgpu_bo_unref(&work->old_rbo); fence_put(work->excl); for (i = 0; i < work->shared_count; ++i) fence_put(work->shared[i]); @@ -712,6 +749,15 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc, * \param dev Device to query. * \param pipe Crtc to query. * \param flags Flags from caller (DRM_CALLED_FROM_VBLIRQ or 0). + * For driver internal use only also supports these flags: + * + * USE_REAL_VBLANKSTART to use the real start of vblank instead + * of a fudged earlier start of vblank. + * + * GET_DISTANCE_TO_VBLANKSTART to return distance to the + * fudged earlier start of vblank in *vpos and the distance + * to true start of vblank in *hpos. + * * \param *vpos Location where vertical scanout position should be stored. * \param *hpos Location where horizontal scanout position should go. * \param *stime Target location for timestamp taken immediately before @@ -776,10 +822,40 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe, vbl_end = 0; } + /* Called from driver internal vblank counter query code? */ + if (flags & GET_DISTANCE_TO_VBLANKSTART) { + /* Caller wants distance from real vbl_start in *hpos */ + *hpos = *vpos - vbl_start; + } + + /* Fudge vblank to start a few scanlines earlier to handle the + * problem that vblank irqs fire a few scanlines before start + * of vblank. Some driver internal callers need the true vblank + * start to be used and signal this via the USE_REAL_VBLANKSTART flag. + * + * The cause of the "early" vblank irq is that the irq is triggered + * by the line buffer logic when the line buffe |