diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 430 |
1 files changed, 273 insertions, 157 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 0e07f35e270c..7dcac3bfb771 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -46,15 +46,13 @@ #include <linux/dma-buf.h> static void i915_gem_flush_free_objects(struct drm_i915_private *i915); -static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj); -static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj); static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj) { - if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) + if (obj->cache_dirty) return false; - if (!i915_gem_object_is_coherent(obj)) + if (!obj->cache_coherent) return true; return obj->pin_display; @@ -145,9 +143,9 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, struct i915_ggtt *ggtt = &dev_priv->ggtt; struct drm_i915_gem_get_aperture *args = data; struct i915_vma *vma; - size_t pinned; + u64 pinned; - pinned = 0; + pinned = ggtt->base.reserved; mutex_lock(&dev->struct_mutex); list_for_each_entry(vma, &ggtt->base.active_list, vm_link) if (i915_vma_is_pinned(vma)) @@ -235,6 +233,14 @@ err_phys: return st; } +static void __start_cpu_write(struct drm_i915_gem_object *obj) +{ + obj->base.read_domains = I915_GEM_DOMAIN_CPU; + obj->base.write_domain = I915_GEM_DOMAIN_CPU; + if (cpu_write_needs_clflush(obj)) + obj->cache_dirty = true; +} + static void __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj, struct sg_table *pages, @@ -247,11 +253,10 @@ __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj, if (needs_clflush && (obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0 && - !i915_gem_object_is_coherent(obj)) + !obj->cache_coherent) drm_clflush_sg(pages); - obj->base.read_domains = I915_GEM_DOMAIN_CPU; - obj->base.write_domain = I915_GEM_DOMAIN_CPU; + __start_cpu_write(obj); } static void @@ -686,6 +691,12 @@ i915_gem_dumb_create(struct drm_file *file, args->size, &args->handle); } +static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj) +{ + return !(obj->cache_level == I915_CACHE_NONE || + obj->cache_level == I915_CACHE_WT); +} + /** * Creates a new mm object and returns a handle to it. * @dev: drm device pointer @@ -705,6 +716,66 @@ i915_gem_create_ioctl(struct drm_device *dev, void *data, args->size, &args->handle); } +static inline enum fb_op_origin +fb_write_origin(struct drm_i915_gem_object *obj, unsigned int domain) +{ + return (domain == I915_GEM_DOMAIN_GTT ? + obj->frontbuffer_ggtt_origin : ORIGIN_CPU); +} + +static void +flush_write_domain(struct drm_i915_gem_object *obj, unsigned int flush_domains) +{ + struct drm_i915_private *dev_priv = to_i915(obj->base.dev); + + if (!(obj->base.write_domain & flush_domains)) + return; + + /* No actual flushing is required for the GTT write domain. Writes + * to it "immediately" go to main memory as far as we know, so there's + * no chipset flush. It also doesn't land in render cache. + * + * However, we do have to enforce the order so that all writes through + * the GTT land before any writes to the device, such as updates to + * the GATT itself. + * + * We also have to wait a bit for the writes to land from the GTT. + * An uncached read (i.e. mmio) seems to be ideal for the round-trip + * timing. This issue has only been observed when switching quickly + * between GTT writes and CPU reads from inside the kernel on recent hw, + * and it appears to only affect discrete GTT blocks (i.e. on LLC + * system agents we cannot reproduce this behaviour). + */ + wmb(); + + switch (obj->base.write_domain) { + case I915_GEM_DOMAIN_GTT: + if (INTEL_GEN(dev_priv) >= 6 && !HAS_LLC(dev_priv)) { + if (intel_runtime_pm_get_if_in_use(dev_priv)) { + spin_lock_irq(&dev_priv->uncore.lock); + POSTING_READ_FW(RING_ACTHD(dev_priv->engine[RCS]->mmio_base)); + spin_unlock_irq(&dev_priv->uncore.lock); + intel_runtime_pm_put(dev_priv); + } + } + + intel_fb_obj_flush(obj, + fb_write_origin(obj, I915_GEM_DOMAIN_GTT)); + break; + + case I915_GEM_DOMAIN_CPU: + i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC); + break; + + case I915_GEM_DOMAIN_RENDER: + if (gpu_write_needs_clflush(obj)) + obj->cache_dirty = true; + break; + } + + obj->base.write_domain = 0; +} + static inline int __copy_to_user_swizzled(char __user *cpu_vaddr, const char *gpu_vaddr, int gpu_offset, @@ -785,8 +856,7 @@ int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, if (ret) return ret; - if (i915_gem_object_is_coherent(obj) || - !static_cpu_has(X86_FEATURE_CLFLUSH)) { + if (obj->cache_coherent || !static_cpu_has(X86_FEATURE_CLFLUSH)) { ret = i915_gem_object_set_to_cpu_domain(obj, false); if (ret) goto err_unpin; @@ -794,14 +864,15 @@ int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, goto out; } - i915_gem_object_flush_gtt_write_domain(obj); + flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); /* If we're not in the cpu read domain, set ourself into the gtt * read domain and manually flush cachelines (if required). This * optimizes for the case when the gpu will dirty the data * anyway again before the next pread happens. */ - if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) + if (!obj->cache_dirty && + !(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) *needs_clflush = CLFLUSH_BEFORE; out: @@ -837,8 +908,7 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, if (ret) return ret; - if (i915_gem_object_is_coherent(obj) || - !static_cpu_has(X86_FEATURE_CLFLUSH)) { + if (obj->cache_coherent || !static_cpu_has(X86_FEATURE_CLFLUSH)) { ret = i915_gem_object_set_to_cpu_domain(obj, true); if (ret) goto err_unpin; @@ -846,21 +916,23 @@ int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, goto out; } - i915_gem_object_flush_gtt_write_domain(obj); + flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); /* If we're not in the cpu write domain, set ourself into the * gtt write domain and manually flush cachelines (as required). * This optimizes for the case when the gpu will use the data * right away and we therefore have to clflush anyway. */ - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) + if (!obj->cache_dirty) { *needs_clflush |= CLFLUSH_AFTER; - /* Same trick applies to invalidate partially written cachelines read - * before writing. - */ - if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) - *needs_clflush |= CLFLUSH_BEFORE; + /* + * Same trick applies to invalidate partially written + * cachelines read before writing. + */ + if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) + *needs_clflush |= CLFLUSH_BEFORE; + } out: intel_fb_obj_invalidate(obj, ORIGIN_CPU); @@ -1501,13 +1573,6 @@ err: return ret; } -static inline enum fb_op_origin -write_origin(struct drm_i915_gem_object *obj, unsigned domain) -{ - return (domain == I915_GEM_DOMAIN_GTT ? - obj->frontbuffer_ggtt_origin : ORIGIN_CPU); -} - static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) { struct drm_i915_private *i915; @@ -1591,10 +1656,12 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, if (err) goto out_unpin; - if (read_domains & I915_GEM_DOMAIN_GTT) - err = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0); + if (read_domains & I915_GEM_DOMAIN_WC) + err = i915_gem_object_set_to_wc_domain(obj, write_domain); + else if (read_domains & I915_GEM_DOMAIN_GTT) + err = i915_gem_object_set_to_gtt_domain(obj, write_domain); else - err = i915_gem_object_set_to_cpu_domain(obj, write_domain != 0); + err = i915_gem_object_set_to_cpu_domain(obj, write_domain); /* And bump the LRU for this access */ i915_gem_object_bump_inactive_ggtt(obj); @@ -1602,7 +1669,8 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, mutex_unlock(&dev->struct_mutex); if (write_domain != 0) - intel_fb_obj_invalidate(obj, write_origin(obj, write_domain)); + intel_fb_obj_invalidate(obj, + fb_write_origin(obj, write_domain)); out_unpin: i915_gem_object_unpin_pages(obj); @@ -1737,6 +1805,9 @@ static unsigned int tile_row_pages(struct drm_i915_gem_object *obj) * into userspace. (This view is aligned and sized appropriately for * fenced access.) * + * 2 - Recognise WC as a separate cache domain so that we can flush the + * delayed writes via GTT before performing direct access via WC. + * * Restrictions: * * * snoopable objects cannot be accessed via the GTT. It can cause machine @@ -1764,7 +1835,7 @@ static unsigned int tile_row_pages(struct drm_i915_gem_object *obj) */ int i915_gem_mmap_gtt_version(void) { - return 1; + return 2; } static inline struct i915_ggtt_view @@ -2228,7 +2299,7 @@ void __i915_gem_object_put_pages(struct drm_i915_gem_object *obj, if (obj->mm.mapping) { void *ptr; - ptr = ptr_mask_bits(obj->mm.mapping); + ptr = page_mask_bits(obj->mm.mapping); if (is_vmalloc_addr(ptr)) vunmap(ptr); else @@ -2285,8 +2356,8 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) struct page *page; unsigned long last_pfn = 0; /* suppress gcc warning */ unsigned int max_segment; + gfp_t noreclaim; int ret; - gfp_t gfp; /* Assert that the object is not currently in any GPU domain. As it * wasn't in the GTT, there shouldn't be any way it could have been in @@ -2315,22 +2386,30 @@ rebuild_st: * Fail silently without starting the shrinker */ mapping = obj->base.filp->f_mapping; - gfp = mapping_gfp_constraint(mapping, ~(__GFP_IO | __GFP_RECLAIM)); - gfp |= __GFP_NORETRY | __GFP_NOWARN; + noreclaim = mapping_gfp_constraint(mapping, ~__GFP_RECLAIM); + noreclaim |= __GFP_NORETRY | __GFP_NOWARN; + sg = st->sgl; st->nents = 0; for (i = 0; i < page_count; i++) { - page = shmem_read_mapping_page_gfp(mapping, i, gfp); - if (unlikely(IS_ERR(page))) { - i915_gem_shrink(dev_priv, - page_count, - I915_SHRINK_BOUND | - I915_SHRINK_UNBOUND | - I915_SHRINK_PURGEABLE); + const unsigned int shrink[] = { + I915_SHRINK_BOUND | I915_SHRINK_UNBOUND | I915_SHRINK_PURGEABLE, + 0, + }, *s = shrink; + gfp_t gfp = noreclaim; + + do { page = shmem_read_mapping_page_gfp(mapping, i, gfp); - } - if (unlikely(IS_ERR(page))) { - gfp_t reclaim; + if (likely(!IS_ERR(page))) + break; + + if (!*s) { + ret = PTR_ERR(page); + goto err_sg; + } + + i915_gem_shrink(dev_priv, 2 * page_count, *s++); + cond_resched(); /* We've tried hard to allocate the memory by reaping * our own buffer, now let the real VM do its job and @@ -2340,15 +2419,26 @@ rebuild_st: * defer the oom here by reporting the ENOMEM back * to userspace. */ - reclaim = mapping_gfp_mask(mapping); - reclaim |= __GFP_NORETRY; /* reclaim, but no oom */ - - page = shmem_read_mapping_page_gfp(mapping, i, reclaim); - if (IS_ERR(page)) { - ret = PTR_ERR(page); - goto err_sg; + if (!*s) { + /* reclaim and warn, but no oom */ + gfp = mapping_gfp_mask(mapping); + + /* Our bo are always dirty and so we require + * kswapd to reclaim our pages (direct reclaim + * does not effectively begin pageout of our + * buffers on its own). However, direct reclaim + * only waits for kswapd when under allocation + * congestion. So as a result __GFP_RECLAIM is + * unreliable and fails to actually reclaim our + * dirty pages -- unless you try over and over + * again with !__GFP_NORETRY. However, we still + * want to fail this allocation rather than + * trigger the out-of-memory killer and for + * this we want the future __GFP_MAYFAIL. + */ } - } + } while (1); + if (!i || sg->length >= max_segment || page_to_pfn(page) != last_pfn + 1) { @@ -2560,7 +2650,7 @@ void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj, } GEM_BUG_ON(!obj->mm.pages); - ptr = ptr_unpack_bits(obj->mm.mapping, has_type); + ptr = page_unpack_bits(obj->mm.mapping, &has_type); if (ptr && has_type != type) { if (pinned) { ret = -EBUSY; @@ -2582,7 +2672,7 @@ void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj, goto err_unpin; } - obj->mm.mapping = ptr_pack_bits(ptr, type); + obj->mm.mapping = page_pack_bits(ptr, type); } out_unlock: @@ -2967,12 +3057,14 @@ static void engine_set_wedged(struct intel_engine_cs *engine) */ if (i915.enable_execlists) { + struct execlist_port *port = engine->execlist_port; unsigned long flags; + unsigned int n; spin_lock_irqsave(&engine->timeline->lock, flags); - i915_gem_request_put(engine->execlist_port[0].request); - i915_gem_request_put(engine->execlist_port[1].request); + for (n = 0; n < ARRAY_SIZE(engine->execlist_port); n++) + i915_gem_request_put(port_request(&port[n])); memset(engine->execlist_port, 0, sizeof(engine->execlist_port)); engine->execlist_queue = RB_ROOT; engine->execlist_first = NULL; @@ -3101,8 +3193,6 @@ i915_gem_idle_work_handler(struct work_struct *work) struct drm_i915_private *dev_priv = container_of(work, typeof(*dev_priv), gt.idle_work.work); struct drm_device *dev = &dev_priv->drm; - struct intel_engine_cs *engine; - enum intel_engine_id id; bool rearm_hangcheck; if (!READ_ONCE(dev_priv->gt.awake)) @@ -3140,10 +3230,8 @@ i915_gem_idle_work_handler(struct work_struct *work) if (wait_for(intel_engines_are_idle(dev_priv), 10)) DRM_ERROR("Timeout waiting for engines to idle\n"); - for_each_engine(engine, dev_priv, id) { - intel_engine_disarm_breadcrumbs(engine); - i915_gem_batch_pool_fini(&engine->batch_pool); - } + intel_engines_mark_idle(dev_priv); + i915_gem_timelines_mark_idle(dev_priv); GEM_BUG_ON(!dev_priv->gt.awake); dev_priv->gt.awake = false; @@ -3173,6 +3261,10 @@ void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file) if (vma->vm->file == fpriv) i915_vma_close(vma); + vma = obj->vma_hashed; + if (vma && vma->ctx->file_priv == fpriv) + i915_vma_unlink_ctx(vma); + if (i915_gem_object_is_active(obj) && !i915_gem_object_has_active_reference(obj)) { i915_gem_object_set_active_reference(obj); @@ -3298,6 +3390,10 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915, unsigned int flags) { int ret; + /* If the device is asleep, we have no requests outstanding */ + if (!READ_ONCE(i915->gt.awake)) + return 0; + if (flags & I915_WAIT_LOCKED) { struct i915_gem_timeline *tl; @@ -3320,73 +3416,89 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915, unsigned int flags) return ret; } -/** Flushes the GTT write domain for the object if it's dirty. */ -static void -i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj) +static void __i915_gem_object_flush_for_display(struct drm_i915_gem_object *obj) { - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); - - if (obj->base.write_domain != I915_GEM_DOMAIN_GTT) - return; - - /* No actual flushing is required for the GTT write domain. Writes - * to it "immediately" go to main memory as far as we know, so there's - * no chipset flush. It also doesn't land in render cache. - * - * However, we do have to enforce the order so that all writes through - * the GTT land before any writes to the device, such as updates to - * the GATT itself. - * - * We also have to wait a bit for the writes to land from the GTT. - * An uncached read (i.e. mmio) seems to be ideal for the round-trip - * timing. This issue has only been observed when switching quickly - * between GTT writes and CPU reads from inside the kernel on recent hw, - * and it appears to only affect discrete GTT blocks (i.e. on LLC - * system agents we cannot reproduce this behaviour). + /* + * We manually flush the CPU domain so that we can override and + * force the flush for the display, and perform it asyncrhonously. */ - wmb(); - if (INTEL_GEN(dev_priv) >= 6 && !HAS_LLC(dev_priv)) { - if (intel_runtime_pm_get_if_in_use(dev_priv)) { - spin_lock_irq(&dev_priv->uncore.lock); - POSTING_READ_FW(RING_ACTHD(dev_priv->engine[RCS]->mmio_base)); - spin_unlock_irq(&dev_priv->uncore.lock); - intel_runtime_pm_put(dev_priv); - } - } - - intel_fb_obj_flush(obj, write_origin(obj, I915_GEM_DOMAIN_GTT)); - + flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); + if (obj->cache_dirty) + i915_gem_clflush_object(obj, I915_CLFLUSH_FORCE); obj->base.write_domain = 0; } -/** Flushes the CPU write domain for the object if it's dirty. */ -static void -i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj) +void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj) { - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) + if (!READ_ONCE(obj->pin_display)) return; - i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC); - obj->base.write_domain = 0; + mutex_lock(&obj->base.dev->struct_mutex); + __i915_gem_object_flush_for_display(obj); + mutex_unlock(&obj->base.dev->struct_mutex); } -static void __i915_gem_object_flush_for_display(struct drm_i915_gem_object *obj) +/** + * Moves a single object to the WC read, and possibly write domain. + * @obj: object to act on + * @write: ask for write access or read only + * + * This function returns when the move is complete, including waiting on + * flushes to occur. + */ +int +i915_gem_object_set_to_wc_domain(struct drm_i915_gem_object *obj, bool write) { - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU && !obj->cache_dirty) - return; + int ret; - i915_gem_clflush_object(obj, I915_CLFLUSH_FORCE); - obj->base.write_domain = 0; -} + lockdep_assert_held(&obj->base.dev->struct_mutex); -void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj) -{ - if (!READ_ONCE(obj->pin_display)) - return; + ret = i915_gem_object_wait(obj, + I915_WAIT_INTERRUPTIBLE | + I915_WAIT_LOCKED | + (write ? I915_WAIT_ALL : 0), + MAX_SCHEDULE_TIMEOUT, + NULL); + if (ret) + return ret; - mutex_lock(&obj->base.dev->struct_mutex); - __i915_gem_object_flush_for_display(obj); - mutex_unlock(&obj->base.dev->struct_mutex); + if (obj->base.write_domain == I915_GEM_DOMAIN_WC) + return 0; + + /* Flush and acquire obj->pages so that we are coherent through + * direct access in memory with previous cached writes through + * shmemfs and that our cache domain tracking remains valid. + * For example, if the obj->filp was moved to swap without us + * being notified and releasing the pages, we would mistakenly + * continue to assume that the obj remained out of the CPU cached + * domain. + */ + ret = i915_gem_object_pin_pages(obj); + if (ret) + return ret; + + flush_write_domain(obj, ~I915_GEM_DOMAIN_WC); + + /* Serialise direct access to this object with the barriers for + * coherent writes from the GPU, by effectively invalidating the + * WC domain upon first access. + */ + if ((obj->base.read_domains & I915_GEM_DOMAIN_WC) == 0) + mb(); + + /* It should now be out of any other write domains, and we can update + * the domain values for our changes. + */ + GEM_BUG_ON((obj->base.write_domain & ~I915_GEM_DOMAIN_WC) != 0); + obj->base.read_domains |= I915_GEM_DOMAIN_WC; + if (write) { + obj->base.read_domains = I915_GEM_DOMAIN_WC; + obj->base.write_domain = I915_GEM_DOMAIN_WC; + obj->mm.dirty = true; + } + + i915_gem_object_unpin_pages(obj); + return 0; } /** @@ -3428,7 +3540,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) if (ret) return ret; - i915_gem_object_flush_cpu_write_domain(obj); + flush_write_domain(obj, ~I915_GEM_DOMAIN_GTT); /* Serialise direct access to this object with the barriers for * coherent writes from the GPU, by effectively invalidating the @@ -3571,13 +3683,11 @@ restart: } } - if (obj->base.write_domain == I915_GEM_DOMAIN_CPU && - i915_gem_object_is_coherent(obj)) - obj->cache_dirty = true; - list_for_each_entry(vma, &obj->vma_list, obj_link) vma->node.color = cache_level; obj->cache_level = cache_level; + obj->cache_coherent = i915_gem_object_is_coherent(obj); + obj->cache_dirty = true; /* Always invalidate stale cachelines */ return 0; } @@ -3799,10 +3909,7 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) if (ret) return ret; - if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) - return 0; - - i915_gem_object_flush_gtt_write_domain(obj); + flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); /* Flush the CPU cache if it's still invalid. */ if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) { @@ -3813,15 +3920,13 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) /* It should now be out of any other write domains, and we can update * the domain values for our changes. */ - GEM_BUG_ON((obj->base.write_domain & ~I915_GEM_DOMAIN_CPU) != 0); + GEM_BUG_ON(obj->base.write_domain & ~I915_GEM_DOMAIN_CPU); /* If we're writing through the CPU, then the GPU read domains will * need to be invalidated at next use. */ - if (write) { - obj->base.read_domains = I915_GEM_DOMAIN_CPU; - obj->base.write_domain = I915_GEM_DOMAIN_CPU; - } + if (write) + __start_cpu_write(obj); return 0; } @@ -3996,7 +4101,7 @@ __busy_set_if_active(const struct dma_fence *fence, if (i915_gem_request_completed(rq)) return 0; - return flag(rq->engine->exec_id); + return flag(rq->engine->uabi_id); } static __always_inline unsigned int @@ -4153,7 +4258,6 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj, INIT_LIST_HEAD(&obj->global_link); INIT_LIST_HEAD(&obj->userfault_link); - INIT_LIST_HEAD(&obj->obj_exec_link); INIT_LIST_HEAD(&obj->vma_list); INIT_LIST_HEAD(&obj->batch_pool_link); @@ -4195,7 +4299,7 @@ i915_gem_object_create(struct drm_i915_private *dev_priv, u64 size) * catch if we ever need to fix it. In the meantime, if you do spot * such a local variable, please consider fixing! */ - if (WARN_ON(size >> PAGE_SHIFT > INT_MAX)) + if (size >> PAGE_SHIFT > INT_MAX) return ERR_PTR(-E2BIG); if (overflows_type(size, obj->base.size)) @@ -4218,6 +4322,7 @@ i915_gem_object_create(struct drm_i915_private *dev_priv, u64 size) mapping = obj->base.filp->f_mapping; mapping_set_gfp_mask(mapping, mask); + GEM_BUG_ON(!(mapping_gfp_mask(mapping) & __GFP_RECLAIM)); i915_gem_object_init(obj, &i915_gem_object_ops); @@ -4241,6 +4346,9 @@ i915_gem_object_create(struct drm_i915_private *dev_priv, u64 size) } else obj->cache_level = I915_CACHE_NONE; + obj->cache_coherent = i915_gem_object_is_coherent(obj); + obj->cache_dirty = !obj->cache_coherent; + trace_i915_gem_object_create(obj); return obj; @@ -4289,7 +4397,6 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915, GEM_BUG_ON(i915_gem_object_is_active(obj)); list_for_each_entry_safe(vma, vn, &obj->vma_list, obj_link) { - GEM_BUG_ON(!i915_vma_is_ggtt(vma)); GEM_BUG_ON(i915_vma_is_active(vma)); vma->flags &= ~I915_VMA_PIN_MASK; i915_vma_close(vma); @@ -4302,6 +4409,8 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915, intel_runtime_pm_put(i915); mutex_unlock(&i915->drm.struct_mutex); + cond_resched(); + llist_for_each_entry_safe(obj, on, freed, freed) { GEM_BUG_ON(obj->bind_count); GEM_BUG_ON(atomic_read(&obj->frontbuffer_bits)); @@ -4349,8 +4458,11 @@ static void __i915_gem_free_work(struct work_struct *work) * unbound now. */ - while ((freed = llist_del_all(&i915->mm.free_list))) + while ((freed = llist_del_all(&i915->mm.free_list))) { __i915_gem_free_objects(i915, freed); + if (need_resched()) + break; + } } static void __i915_gem_free_object_rcu(struct rcu_head *head) @@ -4415,10 +4527,9 @@ void i915_gem_sanitize(struct drm_i915_private *i915) * try to take over. The only way to remove the earlier state * is by resetting. However, resetting on earlier gen is tricky as * it may impact the display and we are uncertain about the stability - * of the reset, so we only reset recent machines with logical - * context support (that must be reset to remove any stray contexts). + * of the reset, so this could be applied to even earlier gen. */ - if (HAS_HW_CONTEXTS(i915)) { + if (INTEL_GEN(i915) >= 5) { int reset = intel_gpu_reset(i915, ALL_ENGINES); WARN_ON(reset && reset != -ENODEV); } @@ -4661,11 +4772,9 @@ bool intel_sanitize_semaphores(struct drm_i915_private *dev_priv, int value) if (value >= 0) return value; -#ifdef CONFIG_INTEL_IOMMU /* Enable semaphores on SNB when IO remapping is off */ - if (INTEL_INFO(dev_priv)->gen == 6 && intel_iommu_gfx_mapped) + if (IS_GEN6(dev_priv) && intel_vtd_active()) return false; -#endif return true; } @@ -4676,7 +4785,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv) mutex_lock(&dev_priv->drm.struct_mutex); - i915_gem_clflush_init(dev_priv); + dev_priv->mm.unordered_timeline = dma_fence_context_alloc(1); if (!i915.enable_execlists) { dev_priv->gt.resume = intel_legacy_submission_resume; @@ -4694,7 +4803,9 @@ int i915_gem_init(struct drm_i915_private *dev_priv) */ intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); - i915_gem_init_userptr(dev_priv); + ret = i915_gem_init_userptr(dev_priv); + if (ret) + goto out_unlock; ret = i915_gem_init_ggtt(dev_priv); if (ret) @@ -4799,12 +4910,16 @@ i915_gem_load_init(struct drm_i915_private *dev_priv) if (!dev_priv->dependencies) goto err_requests; + dev_priv->priorities = KMEM_CACHE(i915_priolist, SLAB_HWCACHE_ALIGN); + if (!dev_priv->priorities) + goto err_dependencies; + mutex_lock(&dev_priv->drm.struct_mutex); INIT_LIST_HEAD(&dev_priv->gt.timelines); err = i915_gem_timeline_init__global(dev_priv); mutex_unlock(&dev_priv->drm.struct_mutex); if (err) - goto err_dependencies; + goto err_priorities; INIT_LIST_HEAD(&dev_priv->context_list); INIT_WORK(&dev_priv->mm.free_work, __i915_gem_free_work); @@ -4822,14 +4937,14 @@ i915_gem_load_init(struct drm_i915_private *dev_priv) init_waitqueue_head(&dev_priv->pending_flip_queue); - dev_priv->mm.interruptible = true; - atomic_set(&dev_priv->mm.bsd_engine_dispatch_index, 0); spin_lock_init(&dev_priv->fb_tracking.lock); return 0; +err_priorities: + kmem_cache_destroy(dev_priv->priorities); err_dependencies: kmem_cache_destroy(dev_priv->dependencies); err_requests: @@ -4853,6 +4968,7 @@ void i915_gem_load_cleanup(struct drm_i915_private *dev_priv) WARN_ON(!list_empty(&dev_priv->gt.timelines)); mutex_unlock(&dev_priv->drm.struct_mutex); + kmem_cache_destroy(dev_priv->priorities); kmem_cache_destroy(dev_priv->dependencies); kmem_cache_destroy(dev_priv->requests); kmem_cache_destroy(dev_priv->vmas); @@ -4864,9 +4980,10 @@ void i915_gem_load_cleanup(struct drm_i915_private *dev_priv) int i915_gem_freeze(struct drm_i915_private *dev_priv) { - mutex_lock(&dev_priv->drm.struct_mutex); + /* Discard all purgeable objects, let userspace recover those as + * required after resuming. + */ i915_gem_shrink_all(dev_priv); - mutex_unlock(&dev_priv->drm.struct_mutex); return 0; } @@ -4891,17 +5008,16 @@ int i915_gem_freeze_late(struct drm_i915_private *dev_priv) * we update that state just before writing out the image. * * To try and reduce the hibernation image, we manually shrink - * the objects as well. + * the objects as well, see i915_gem_freeze() */ - mutex_lock(&dev_priv->drm.struct_mutex); i915_gem_shrink(dev_priv, -1UL, I915_SHRINK_UNBOUND); + i915_gem_drain_freed_objects(dev_priv); + mutex_lock(&dev_priv->drm.struct_mutex); for (p = phases; *p; p++) { - list_for_each_entry(obj, *p, global_link) { - obj->base.read_domains = I915_GEM_DOMAIN_CPU; - obj->base.write_domain = I915_GEM_DOMAIN_CPU; - } + list_for_each_entry(obj, *p, global_link) + __start_cpu_write(obj); } mutex_unlock(&dev_priv->drm.struct_mutex); |