diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 3242 |
1 files changed, 2012 insertions, 1230 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index d33d02d13c9..d893e4da5dc 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -26,10 +26,12 @@ */ #include <drm/drmP.h> +#include <drm/drm_vma_manager.h> #include <drm/i915_drm.h> #include "i915_drv.h" #include "i915_trace.h" #include "intel_drv.h" +#include <linux/oom.h> #include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/swap.h> @@ -37,15 +39,13 @@ #include <linux/dma-buf.h> 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 __must_check int i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj, - unsigned alignment, - bool map_and_fenceable, - bool nonblocking); -static int i915_gem_phys_pwrite(struct drm_device *dev, - struct drm_i915_gem_object *obj, - struct drm_i915_gem_pwrite *args, - struct drm_file *file); +static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj, + bool force); +static __must_check int +i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, + bool readonly); +static void +i915_gem_object_retire(struct drm_i915_gem_object *obj); static void i915_gem_write_fence(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj); @@ -53,11 +53,29 @@ static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj, struct drm_i915_fence_reg *fence, bool enable); -static int i915_gem_inactive_shrink(struct shrinker *shrinker, - struct shrink_control *sc); -static long i915_gem_purge(struct drm_i915_private *dev_priv, long target); -static void i915_gem_shrink_all(struct drm_i915_private *dev_priv); -static void i915_gem_object_truncate(struct drm_i915_gem_object *obj); +static unsigned long i915_gem_shrinker_count(struct shrinker *shrinker, + struct shrink_control *sc); +static unsigned long i915_gem_shrinker_scan(struct shrinker *shrinker, + struct shrink_control *sc); +static int i915_gem_shrinker_oom(struct notifier_block *nb, + unsigned long event, + void *ptr); +static unsigned long i915_gem_purge(struct drm_i915_private *dev_priv, long target); +static unsigned long i915_gem_shrink_all(struct drm_i915_private *dev_priv); + +static bool cpu_cache_is_coherent(struct drm_device *dev, + enum i915_cache_level level) +{ + return HAS_LLC(dev) || level != I915_CACHE_NONE; +} + +static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj) +{ + if (!cpu_cache_is_coherent(obj->base.dev, obj->cache_level)) + return true; + + return obj->pin_display; +} static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj) { @@ -75,26 +93,29 @@ static inline void i915_gem_object_fence_lost(struct drm_i915_gem_object *obj) static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv, size_t size) { + spin_lock(&dev_priv->mm.object_stat_lock); dev_priv->mm.object_count++; dev_priv->mm.object_memory += size; + spin_unlock(&dev_priv->mm.object_stat_lock); } static void i915_gem_info_remove_obj(struct drm_i915_private *dev_priv, size_t size) { + spin_lock(&dev_priv->mm.object_stat_lock); dev_priv->mm.object_count--; dev_priv->mm.object_memory -= size; + spin_unlock(&dev_priv->mm.object_stat_lock); } static int -i915_gem_wait_for_error(struct drm_device *dev) +i915_gem_wait_for_error(struct i915_gpu_error *error) { - struct drm_i915_private *dev_priv = dev->dev_private; - struct completion *x = &dev_priv->error_completion; - unsigned long flags; int ret; - if (!atomic_read(&dev_priv->mm.wedged)) +#define EXIT_COND (!i915_reset_in_progress(error) || \ + i915_terminally_wedged(error)) + if (EXIT_COND) return 0; /* @@ -102,32 +123,26 @@ i915_gem_wait_for_error(struct drm_device *dev) * userspace. If it takes that long something really bad is going on and * we should simply try to bail out and fail as gracefully as possible. */ - ret = wait_for_completion_interruptible_timeout(x, 10*HZ); + ret = wait_event_interruptible_timeout(error->reset_queue, + EXIT_COND, + 10*HZ); if (ret == 0) { DRM_ERROR("Timed out waiting for the gpu reset to complete\n"); return -EIO; } else if (ret < 0) { return ret; } +#undef EXIT_COND - if (atomic_read(&dev_priv->mm.wedged)) { - /* GPU is hung, bump the completion count to account for - * the token we just consumed so that we never hit zero and - * end up waiting upon a subsequent completion event that - * will never happen. - */ - spin_lock_irqsave(&x->wait.lock, flags); - x->done++; - spin_unlock_irqrestore(&x->wait.lock, flags); - } return 0; } int i915_mutex_lock_interruptible(struct drm_device *dev) { + struct drm_i915_private *dev_priv = dev->dev_private; int ret; - ret = i915_gem_wait_for_error(dev); + ret = i915_gem_wait_for_error(&dev_priv->gpu_error); if (ret) return ret; @@ -142,13 +157,14 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) static inline bool i915_gem_object_is_inactive(struct drm_i915_gem_object *obj) { - return obj->gtt_space && !obj->active; + return i915_gem_obj_bound_any(obj) && !obj->active; } int i915_gem_init_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { + struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_gem_init *args = data; if (drm_core_check_feature(dev, DRIVER_MODESET)) @@ -163,8 +179,9 @@ i915_gem_init_ioctl(struct drm_device *dev, void *data, return -ENODEV; mutex_lock(&dev->struct_mutex); - i915_gem_init_global_gtt(dev, args->gtt_start, - args->gtt_end, args->gtt_end); + i915_gem_setup_global_gtt(dev, args->gtt_start, args->gtt_end, + args->gtt_end); + dev_priv->gtt.mappable_end = args->gtt_end; mutex_unlock(&dev->struct_mutex); return 0; @@ -181,17 +198,151 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, pinned = 0; mutex_lock(&dev->struct_mutex); - list_for_each_entry(obj, &dev_priv->mm.bound_list, gtt_list) - if (obj->pin_count) - pinned += obj->gtt_space->size; + list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) + if (i915_gem_obj_is_pinned(obj)) + pinned += i915_gem_obj_ggtt_size(obj); mutex_unlock(&dev->struct_mutex); - args->aper_size = dev_priv->mm.gtt_total; + args->aper_size = dev_priv->gtt.base.total; args->aper_available_size = args->aper_size - pinned; return 0; } +static void i915_gem_object_detach_phys(struct drm_i915_gem_object *obj) +{ + drm_dma_handle_t *phys = obj->phys_handle; + + if (!phys) + return; + + if (obj->madv == I915_MADV_WILLNEED) { + struct address_space *mapping = file_inode(obj->base.filp)->i_mapping; + char *vaddr = phys->vaddr; + int i; + + for (i = 0; i < obj->base.size / PAGE_SIZE; i++) { + struct page *page = shmem_read_mapping_page(mapping, i); + if (!IS_ERR(page)) { + char *dst = kmap_atomic(page); + memcpy(dst, vaddr, PAGE_SIZE); + drm_clflush_virt_range(dst, PAGE_SIZE); + kunmap_atomic(dst); + + set_page_dirty(page); + mark_page_accessed(page); + page_cache_release(page); + } + vaddr += PAGE_SIZE; + } + i915_gem_chipset_flush(obj->base.dev); + } + +#ifdef CONFIG_X86 + set_memory_wb((unsigned long)phys->vaddr, phys->size / PAGE_SIZE); +#endif + drm_pci_free(obj->base.dev, phys); + obj->phys_handle = NULL; +} + +int +i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, + int align) +{ + drm_dma_handle_t *phys; + struct address_space *mapping; + char *vaddr; + int i; + + if (obj->phys_handle) { + if ((unsigned long)obj->phys_handle->vaddr & (align -1)) + return -EBUSY; + + return 0; + } + + if (obj->madv != I915_MADV_WILLNEED) + return -EFAULT; + + if (obj->base.filp == NULL) + return -EINVAL; + + /* create a new object */ + phys = drm_pci_alloc(obj->base.dev, obj->base.size, align); + if (!phys) + return -ENOMEM; + + vaddr = phys->vaddr; +#ifdef CONFIG_X86 + set_memory_wc((unsigned long)vaddr, phys->size / PAGE_SIZE); +#endif + mapping = file_inode(obj->base.filp)->i_mapping; + for (i = 0; i < obj->base.size / PAGE_SIZE; i++) { + struct page *page; + char *src; + + page = shmem_read_mapping_page(mapping, i); + if (IS_ERR(page)) { +#ifdef CONFIG_X86 + set_memory_wb((unsigned long)phys->vaddr, phys->size / PAGE_SIZE); +#endif + drm_pci_free(obj->base.dev, phys); + return PTR_ERR(page); + } + + src = kmap_atomic(page); + memcpy(vaddr, src, PAGE_SIZE); + kunmap_atomic(src); + + mark_page_accessed(page); + page_cache_release(page); + + vaddr += PAGE_SIZE; + } + + obj->phys_handle = phys; + return 0; +} + +static int +i915_gem_phys_pwrite(struct drm_i915_gem_object *obj, + struct drm_i915_gem_pwrite *args, + struct drm_file *file_priv) +{ + struct drm_device *dev = obj->base.dev; + void *vaddr = obj->phys_handle->vaddr + args->offset; + char __user *user_data = to_user_ptr(args->data_ptr); + + if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) { + unsigned long unwritten; + + /* The physical object once assigned is fixed for the lifetime + * of the obj, so we can safely drop the lock and continue + * to access vaddr. + */ + mutex_unlock(&dev->struct_mutex); + unwritten = copy_from_user(vaddr, user_data, args->size); + mutex_lock(&dev->struct_mutex); + if (unwritten) + return -EFAULT; + } + + i915_gem_chipset_flush(dev); + return 0; +} + +void *i915_gem_object_alloc(struct drm_device *dev) +{ + struct drm_i915_private *dev_priv = dev->dev_private; + return kmem_cache_zalloc(dev_priv->slab, GFP_KERNEL); +} + +void i915_gem_object_free(struct drm_i915_gem_object *obj) +{ + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + kmem_cache_free(dev_priv->slab, obj); +} + static int i915_gem_create(struct drm_file *file, struct drm_device *dev, @@ -212,16 +363,10 @@ i915_gem_create(struct drm_file *file, return -ENOMEM; ret = drm_gem_handle_create(file, &obj->base, &handle); - if (ret) { - drm_gem_object_release(&obj->base); - i915_gem_info_remove_obj(dev->dev_private, obj->base.size); - kfree(obj); - return ret; - } - /* drop reference from allocate - handle holds it now */ - drm_gem_object_unreference(&obj->base); - trace_i915_gem_object_create(obj); + drm_gem_object_unreference_unlocked(&obj->base); + if (ret) + return ret; *handle_p = handle; return 0; @@ -233,19 +378,12 @@ i915_gem_dumb_create(struct drm_file *file, struct drm_mode_create_dumb *args) { /* have to work out size/pitch and return them */ - args->pitch = ALIGN(args->width * ((args->bpp + 7) / 8), 64); + args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 64); args->size = args->pitch * args->height; return i915_gem_create(file, dev, args->size, &args->handle); } -int i915_gem_dumb_destroy(struct drm_file *file, - struct drm_device *dev, - uint32_t handle) -{ - return drm_gem_handle_delete(file, handle); -} - /** * Creates a new mm object and returns a handle to it. */ @@ -259,14 +397,6 @@ i915_gem_create_ioctl(struct drm_device *dev, void *data, args->size, &args->handle); } -static int i915_gem_object_needs_bit17_swizzle(struct drm_i915_gem_object *obj) -{ - drm_i915_private_t *dev_priv = obj->base.dev->dev_private; - - return dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_9_10_17 && - obj->tiling_mode != I915_TILING_NONE; -} - static inline int __copy_to_user_swizzled(char __user *cpu_vaddr, const char *gpu_vaddr, int gpu_offset, @@ -319,6 +449,44 @@ __copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset, return 0; } +/* + * Pins the specified object's pages and synchronizes the object with + * GPU accesses. Sets needs_clflush to non-zero if the caller should + * flush the object from the CPU cache. + */ +int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, + int *needs_clflush) +{ + int ret; + + *needs_clflush = 0; + + if (!obj->base.filp) + return -EINVAL; + + if (!(obj->base.read_domains & 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. */ + *needs_clflush = !cpu_cache_is_coherent(obj->base.dev, + obj->cache_level); + ret = i915_gem_object_wait_rendering(obj, true); + if (ret) + return ret; + + i915_gem_object_retire(obj); + } + + ret = i915_gem_object_get_pages(obj); + if (ret) + return ret; + + i915_gem_object_pin_pages(obj); + + return ret; +} + /* Per-page copy function for the shmem pread fastpath. * Flushes invalid cachelines before reading the target if * needs_clflush is set. */ @@ -407,44 +575,24 @@ i915_gem_shmem_pread(struct drm_device *dev, loff_t offset; int shmem_page_offset, page_length, ret = 0; int obj_do_bit17_swizzling, page_do_bit17_swizzling; - int hit_slowpath = 0; int prefaulted = 0; int needs_clflush = 0; - struct scatterlist *sg; - int i; + struct sg_page_iter sg_iter; - user_data = (char __user *) (uintptr_t) args->data_ptr; + user_data = to_user_ptr(args->data_ptr); remain = args->size; obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); - if (!(obj->base.read_domains & 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->cache_level == I915_CACHE_NONE) - needs_clflush = 1; - if (obj->gtt_space) { - ret = i915_gem_object_set_to_gtt_domain(obj, false); - if (ret) - return ret; - } - } - - ret = i915_gem_object_get_pages(obj); + ret = i915_gem_obj_prepare_shmem_read(obj, &needs_clflush); if (ret) return ret; - i915_gem_object_pin_pages(obj); - offset = args->offset; - for_each_sg(obj->pages->sgl, sg, obj->pages->nents, i) { - struct page *page; - - if (i < offset >> PAGE_SHIFT) - continue; + for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, + offset >> PAGE_SHIFT) { + struct page *page = sg_page_iter_page(&sg_iter); if (remain <= 0) break; @@ -459,7 +607,6 @@ i915_gem_shmem_pread(struct drm_device *dev, if ((shmem_page_offset + page_length) > PAGE_SIZE) page_length = PAGE_SIZE - shmem_page_offset; - page = sg_page(sg); page_do_bit17_swizzling = obj_do_bit17_swizzling && (page_to_phys(page) & (1 << 17)) != 0; @@ -469,10 +616,9 @@ i915_gem_shmem_pread(struct drm_device *dev, if (ret == 0) goto next_page; - hit_slowpath = 1; mutex_unlock(&dev->struct_mutex); - if (!prefaulted) { + if (likely(!i915.prefault_disable) && !prefaulted) { ret = fault_in_multipages_writeable(user_data, remain); /* Userspace is tricking us, but we've already clobbered * its pages with the prefault and promised to write the @@ -488,12 +634,10 @@ i915_gem_shmem_pread(struct drm_device *dev, mutex_lock(&dev->struct_mutex); -next_page: - mark_page_accessed(page); - if (ret) goto out; +next_page: remain -= page_length; user_data += page_length; offset += page_length; @@ -502,12 +646,6 @@ next_page: out: i915_gem_object_unpin_pages(obj); - if (hit_slowpath) { - /* Fixup: Kill any reinstated backing storage pages */ - if (obj->madv == __I915_MADV_PURGED) - i915_gem_object_truncate(obj); - } - return ret; } @@ -528,7 +666,7 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data, return 0; if (!access_ok(VERIFY_WRITE, - (char __user *)(uintptr_t)args->data_ptr, + to_user_ptr(args->data_ptr), args->size)) return -EFAULT; @@ -601,13 +739,13 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev, struct drm_i915_gem_pwrite *args, struct drm_file *file) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; ssize_t remain; loff_t offset, page_base; char __user *user_data; int page_offset, page_length, ret; - ret = i915_gem_object_pin(obj, 0, true, true); + ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE | PIN_NONBLOCK); if (ret) goto out; @@ -619,10 +757,10 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev, if (ret) goto out_unpin; - user_data = (char __user *) (uintptr_t) args->data_ptr; + user_data = to_user_ptr(args->data_ptr); remain = args->size; - offset = obj->gtt_offset + args->offset; + offset = i915_gem_obj_ggtt_offset(obj) + args->offset; while (remain > 0) { /* Operation in this page @@ -641,7 +779,7 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev, * source page isn't available. Return the error and we'll * retry in the slow path. */ - if (fast_user_write(dev_priv->mm.gtt_mapping, page_base, + if (fast_user_write(dev_priv->gtt.mappable, page_base, page_offset, user_data, page_length)) { ret = -EFAULT; goto out_unpin; @@ -653,7 +791,7 @@ i915_gem_gtt_pwrite_fast(struct drm_device *dev, } out_unpin: - i915_gem_object_unpin(obj); + i915_gem_object_ggtt_unpin(obj); out: return ret; } @@ -679,9 +817,8 @@ shmem_pwrite_fast(struct page *page, int shmem_page_offset, int page_length, if (needs_clflush_before) drm_clflush_virt_range(vaddr + shmem_page_offset, page_length); - ret = __copy_from_user_inatomic_nocache(vaddr + shmem_page_offset, - user_data, - page_length); + ret = __copy_from_user_inatomic(vaddr + shmem_page_offset, + user_data, page_length); if (needs_clflush_after) drm_clflush_virt_range(vaddr + shmem_page_offset, page_length); @@ -738,10 +875,9 @@ i915_gem_shmem_pwrite(struct drm_device *dev, int hit_slowpath = 0; int needs_clflush_after = 0; int needs_clflush_before = 0; - int i; - struct scatterlist *sg; + struct sg_page_iter sg_iter; - user_data = (char __user *) (uintptr_t) args->data_ptr; + user_data = to_user_ptr(args->data_ptr); remain = args->size; obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); @@ -751,19 +887,18 @@ i915_gem_shmem_pwrite(struct drm_device *dev, * write domain and manually flush cachelines (if required). This * optimizes for the case when the gpu will use the data * right away and we therefore have to clflush anyway. */ - if (obj->cache_level == I915_CACHE_NONE) - needs_clflush_after = 1; - if (obj->gtt_space) { - ret = i915_gem_object_set_to_gtt_domain(obj, true); - if (ret) - return ret; - } + needs_clflush_after = cpu_write_needs_clflush(obj); + ret = i915_gem_object_wait_rendering(obj, false); + if (ret) + return ret; + + i915_gem_object_retire(obj); } - /* Same trick applies for invalidate partially written cachelines before - * writing. */ - if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU) - && obj->cache_level == I915_CACHE_NONE) - needs_clflush_before = 1; + /* Same trick applies to invalidate partially written cachelines read + * before writing. */ + if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) + needs_clflush_before = + !cpu_cache_is_coherent(dev, obj->cache_level); ret = i915_gem_object_get_pages(obj); if (ret) @@ -774,13 +909,11 @@ i915_gem_shmem_pwrite(struct drm_device *dev, offset = args->offset; obj->dirty = 1; - for_each_sg(obj->pages->sgl, sg, obj->pages->nents, i) { - struct page *page; + for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, + offset >> PAGE_SHIFT) { + struct page *page = sg_page_iter_page(&sg_iter); int partial_cacheline_write; - if (i < offset >> PAGE_SHIFT) - continue; - if (remain <= 0) break; @@ -802,7 +935,6 @@ i915_gem_shmem_pwrite(struct drm_device *dev, ((shmem_page_offset | page_length) & (boot_cpu_data.x86_clflush_size - 1)); - page = sg_page(sg); page_do_bit17_swizzling = obj_do_bit17_swizzling && (page_to_phys(page) & (1 << 17)) != 0; @@ -822,13 +954,10 @@ i915_gem_shmem_pwrite(struct drm_device *dev, mutex_lock(&dev->struct_mutex); -next_page: - set_page_dirty(page); - mark_page_accessed(page); - if (ret) goto out; +next_page: remain -= page_length; user_data += page_length; offset += page_length; @@ -838,19 +967,20 @@ out: i915_gem_object_unpin_pages(obj); if (hit_slowpath) { - /* Fixup: Kill any reinstated backing storage pages */ - if (obj->madv == __I915_MADV_PURGED) - i915_gem_object_truncate(obj); - /* and flush dirty cachelines in case the object isn't in the cpu write - * domain anymore. */ - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) { - i915_gem_clflush_object(obj); - intel_gtt_chipset_flush(); + /* + * Fixup: Flush cpu caches in case we didn't flush the dirty + * cachelines in-line while writing and the object moved + * out of the cpu write domain while we've dropped the lock. + */ + if (!needs_clflush_after && + obj->base.write_domain != I915_GEM_DOMAIN_CPU) { + if (i915_gem_clflush_object(obj, obj->pin_display)) + i915_gem_chipset_flush(dev); } } if (needs_clflush_after) - intel_gtt_chipset_flush(); + i915_gem_chipset_flush(dev); return ret; } @@ -872,14 +1002,16 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, return 0; if (!access_ok(VERIFY_READ, - (char __user *)(uintptr_t)args->data_ptr, + to_user_ptr(args->data_ptr), args->size)) return -EFAULT; - ret = fault_in_multipages_readable((char __user *)(uintptr_t)args->data_ptr, - args->size); - if (ret) - return -EFAULT; + if (likely(!i915.prefault_disable)) { + ret = fault_in_multipages_readable(to_user_ptr(args->data_ptr), + args->size); + if (ret) + return -EFAULT; + } ret = i915_mutex_lock_interruptible(dev); if (ret) @@ -915,14 +1047,14 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, * pread/pwrite currently are reading and writing from the CPU * perspective, requiring manual detiling by the client. */ - if (obj->phys_obj) { - ret = i915_gem_phys_pwrite(dev, obj, args, file); + if (obj->phys_handle) { + ret = i915_gem_phys_pwrite(obj, args, file); goto out; } - if (obj->cache_level == I915_CACHE_NONE && - obj->tiling_mode == I915_TILING_NONE && - obj->base.write_domain != I915_GEM_DOMAIN_CPU) { + if (obj->tiling_mode == I915_TILING_NONE && + obj->base.write_domain != I915_GEM_DOMAIN_CPU && + cpu_write_needs_clflush(obj)) { ret = i915_gem_gtt_pwrite_fast(dev, obj, args, file); /* Note that the gtt paths might fail with non-page-backed user * pointers (e.g. gtt mappings when moving data between @@ -940,26 +1072,17 @@ unlock: } int -i915_gem_check_wedge(struct drm_i915_private *dev_priv, +i915_gem_check_wedge(struct i915_gpu_error *error, bool interruptible) { - if (atomic_read(&dev_priv->mm.wedged)) { - struct completion *x = &dev_priv->error_completion; - bool recovery_complete; - unsigned long flags; - - /* Give the error handler a chance to run. */ - spin_lock_irqsave(&x->wait.lock, flags); - recovery_complete = x->done > 0; - spin_unlock_irqrestore(&x->wait.lock, flags); - + if (i915_reset_in_progress(error)) { /* Non-interruptible callers can't handle -EAGAIN, hence return * -EIO unconditionally for these. */ if (!interruptible) return -EIO; - /* Recovery complete, but still wedged means reset failure. */ - if (recovery_complete) + /* Recovery complete, but the reset failed ... */ + if (i915_terminally_wedged(error)) return -EIO; return -EAGAIN; @@ -973,98 +1096,156 @@ i915_gem_check_wedge(struct drm_i915_private *dev_priv, * equal. */ static int -i915_gem_check_olr(struct intel_ring_buffer *ring, u32 seqno) +i915_gem_check_olr(struct intel_engine_cs *ring, u32 seqno) { int ret; BUG_ON(!mutex_is_locked(&ring->dev->struct_mutex)); ret = 0; - if (seqno == ring->outstanding_lazy_request) - ret = i915_add_request(ring, NULL, NULL); + if (seqno == ring->outstanding_lazy_seqno) + ret = i915_add_request(ring, NULL); return ret; } +static void fake_irq(unsigned long data) +{ + wake_up_process((struct task_struct *)data); +} + +static bool missed_irq(struct drm_i915_private *dev_priv, + struct intel_engine_cs *ring) +{ + return test_bit(ring->id, &dev_priv->gpu_error.missed_irq_rings); +} + +static bool can_wait_boost(struct drm_i915_file_private *file_priv) +{ + if (file_priv == NULL) + return true; + + return !atomic_xchg(&file_priv->rps_wait_boost, true); +} + /** * __wait_seqno - wait until execution of seqno has finished * @ring: the ring expected to report seqno * @seqno: duh! + * @reset_counter: reset sequence associated with the given seqno * @interruptible: do an interruptible wait (normally yes) * @timeout: in - how long to wait (NULL forever); out - how much time remaining * + * Note: It is of utmost importance that the passed in seqno and reset_counter + * values have been read by the caller in an smp safe manner. Where read-side + * locks are involved, it is sufficient to read the reset_counter before + * unlocking the lock that protects the seqno. For lockless tricks, the + * reset_counter _must_ be read before, and an appropriate smp_rmb must be + * inserted. + * * Returns 0 if the seqno was found within the alloted time. Else returns the * errno with remaining time filled in timeout argument. */ -static int __wait_seqno(struct intel_ring_buffer *ring, u32 seqno, - bool interruptible, struct timespec *timeout) -{ - drm_i915_private_t *dev_priv = ring->dev->dev_private; - struct timespec before, now, wait_time={1,0}; - unsigned long timeout_jiffies; - long end; - bool wait_forever = true; +static int __wait_seqno(struct intel_engine_cs *ring, u32 seqno, + unsigned reset_counter, + bool interruptible, + struct timespec *timeout, + struct drm_i915_file_private *file_priv) +{ + struct drm_device *dev = ring->dev; + struct drm_i915_private *dev_priv = dev->dev_private; + const bool irq_test_in_progress = + ACCESS_ONCE(dev_priv->gpu_error.test_irq_rings) & intel_ring_flag(ring); + struct timespec before, now; + DEFINE_WAIT(wait); + unsigned long timeout_expire; int ret; + WARN(dev_priv->pm.irqs_disabled, "IRQs disabled\n"); + if (i915_seqno_passed(ring->get_seqno(ring, true), seqno)) return 0; - trace_i915_gem_request_wait_begin(ring, seqno); + timeout_expire = timeout ? jiffies + timespec_to_jiffies_timeout(timeout) : 0; - if (timeout != NULL) { - wait_time = *timeout; - wait_forever = false; + if (INTEL_INFO(dev)->gen >= 6 && can_wait_boost(file_priv)) { + gen6_rps_boost(dev_priv); + if (file_priv) + mod_delayed_work(dev_priv->wq, + &file_priv->mm.idle_work, + msecs_to_jiffies(100)); } - timeout_jiffies = timespec_to_jiffies(&wait_time); - - if (WARN_ON(!ring->irq_get(ring))) + if (!irq_test_in_progress && WARN_ON(!ring->irq_get(ring))) return -ENODEV; - /* Record current time in case interrupted by signal, or wedged * */ + /* Record current time in case interrupted by signal, or wedged */ + trace_i915_gem_request_wait_begin(ring, seqno); getrawmonotonic(&before); + for (;;) { + struct timer_list timer; + + prepare_to_wait(&ring->irq_queue, &wait, + interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE); + + /* We need to check whether any gpu reset happened in between + * the caller grabbing the seqno and now ... */ + if (reset_counter != atomic_read(&dev_priv->gpu_error.reset_counter)) { + /* ... but upgrade the -EAGAIN to an -EIO if the gpu + * is truely gone. */ + ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible); + if (ret == 0) + ret = -EAGAIN; + break; + } -#define EXIT_COND \ - (i915_seqno_passed(ring->get_seqno(ring, false), seqno) || \ - atomic_read(&dev_priv->mm.wedged)) - do { - if (interruptible) - end = wait_event_interruptible_timeout(ring->irq_queue, - EXIT_COND, - timeout_jiffies); - else - end = wait_event_timeout(ring->irq_queue, EXIT_COND, - timeout_jiffies); + if (i915_seqno_passed(ring->get_seqno(ring, false), seqno)) { + ret = 0; + break; + } - ret = i915_gem_check_wedge(dev_priv, interruptible); - if (ret) - end = ret; - } while (end == 0 && wait_forever); + if (interruptible && signal_pending(current)) { + ret = -ERESTARTSYS; + break; + } - getrawmonotonic(&now); + if (timeout && time_after_eq(jiffies, timeout_expire)) { + ret = -ETIME; + break; + } + + timer.function = NULL; + if (timeout || missed_irq(dev_priv, ring)) { + unsigned long expire; - ring->irq_put(ring); + setup_timer_on_stack(&timer, fake_irq, (unsigned long)current); + expire = missed_irq(dev_priv, ring) ? jiffies + 1 : timeout_expire; + mod_timer(&timer, expire); + } + + io_schedule(); + + if (timer.function) { + del_singleshot_timer_sync(&timer); + destroy_timer_on_stack(&timer); + } + } + getrawmonotonic(&now); trace_i915_gem_request_wait_end(ring, seqno); -#undef EXIT_COND + + if (!irq_test_in_progress) + ring->irq_put(ring); + + finish_wait(&ring->irq_queue, &wait); if (timeout) { struct timespec sleep_time = timespec_sub(now, before); *timeout = timespec_sub(*timeout, sleep_time); - } - - switch (end) { - case -EIO: - case -EAGAIN: /* Wedged */ - case -ERESTARTSYS: /* Signal */ - return (int)end; - case 0: /* Timeout */ - if (timeout) + if (!timespec_valid(timeout)) /* i.e. negative time remains */ set_normalized_timespec(timeout, 0, 0); - return -ETIME; - default: /* Completed */ - WARN_ON(end < 0); /* We're not aware of other errors */ - return 0; } + + return ret; } /** @@ -1072,7 +1253,7 @@ static int __wait_seqno(struct intel_ring_buffer *ring, u32 seqno, * request and object lists appropriately for that event. */ int -i915_wait_seqno(struct intel_ring_buffer *ring, uint32_t seqno) +i915_wait_seqno(struct intel_engine_cs *ring, uint32_t seqno) { struct drm_device *dev = ring->dev; struct drm_i915_private *dev_priv = dev->dev_private; @@ -1082,7 +1263,7 @@ i915_wait_seqno(struct intel_ring_buffer *ring, uint32_t seqno) BUG_ON(!mutex_is_locked(&dev->struct_mutex)); BUG_ON(seqno == 0); - ret = i915_gem_check_wedge(dev_priv, interruptible); + ret = i915_gem_check_wedge(&dev_priv->gpu_error, interruptible); if (ret) return ret; @@ -1090,7 +1271,28 @@ i915_wait_seqno(struct intel_ring_buffer *ring, uint32_t seqno) if (ret) return ret; - return __wait_seqno(ring, seqno, interruptible, NULL); + return __wait_seqno(ring, seqno, + atomic_read(&dev_priv->gpu_error.reset_counter), + interruptible, NULL, NULL); +} + +static int +i915_gem_object_wait_rendering__tail(struct drm_i915_gem_object *obj, + struct intel_engine_cs *ring) +{ + if (!obj->active) + return 0; + + /* Manually manage the write flush as we may have not yet + * retired the buffer. + * + * Note that the last_write_seqno is always the earlier of + * the two (read/write) seqno, so if we haved successfully waited, + * we know we have passed the last write. + */ + obj->last_write_seqno = 0; + + return 0; } /** @@ -1101,7 +1303,7 @@ static __must_check int i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, bool readonly) { - struct intel_ring_buffer *ring = obj->ring; + struct intel_engine_cs *ring = obj->ring; u32 seqno; int ret; @@ -1113,18 +1315,7 @@ i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, if (ret) return ret; - i915_gem_retire_requests_ring(ring); - - /* Manually manage the write flush as we may have not yet - * retired the buffer. - */ - if (obj->last_write_seqno && - i915_seqno_passed(seqno, obj->last_write_seqno)) { - obj->last_write_seqno = 0; - obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS; - } - - return 0; + return i915_gem_object_wait_rendering__tail(obj, ring); } /* A nonblocking variant of the above wait. This is a highly dangerous routine @@ -1132,11 +1323,13 @@ i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, */ static __must_check int i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj, + struct drm_i915_file_private *file_priv, bool readonly) { struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; - struct intel_ring_buffer *ring = obj->ring; + struct intel_engine_cs *ring = obj->ring; + unsigned reset_counter; u32 seqno; int ret; @@ -1147,7 +1340,7 @@ i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj, if (seqno == 0) return 0; - ret = i915_gem_check_wedge(dev_priv, true); + ret = i915_gem_check_wedge(&dev_priv->gpu_error, true); if (ret) return ret; @@ -1155,22 +1348,14 @@ i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj, if (ret) return ret; + reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter); mutex_unlock(&dev->struct_mutex); - ret = __wait_seqno(ring, seqno, true, NULL); + ret = __wait_seqno(ring, seqno, reset_counter, true, NULL, file_priv); mutex_lock(&dev->struct_mutex); + if (ret) + return ret; - i915_gem_retire_requests_ring(ring); - - /* Manually manage the write flush as we may have not yet - * retired the buffer. - */ - if (obj->last_write_seqno && - i915_seqno_passed(seqno, obj->last_write_seqno)) { - obj->last_write_seqno = 0; - obj->base.write_domain &= ~I915_GEM_GPU_DOMAINS; - } - - return ret; + return i915_gem_object_wait_rendering__tail(obj, ring); } /** @@ -1214,7 +1399,9 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, * We will repeat the flush holding the lock in the normal manner * to catch cases where we are gazumped. */ - ret = i915_gem_object_wait_rendering__nonblocking(obj, !write_domain); + ret = i915_gem_object_wait_rendering__nonblocking(obj, + file->driver_priv, + !write_domain); if (ret) goto unref; @@ -1260,8 +1447,8 @@ i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, } /* Pinned buffers may be scanout, so flush the cache */ - if (obj->pin_count) - i915_gem_object_flush_cpu_write_domain(obj); + if (obj->pin_display) + i915_gem_object_flush_cpu_write_domain(obj, true); drm_gem_object_unreference(&obj->base); unlock: @@ -1328,12 +1515,14 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) { struct drm_i915_gem_object *obj = to_intel_bo(vma->vm_private_data); struct drm_device *dev = obj->base.dev; - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; pgoff_t page_offset; unsigned long pfn; int ret = 0; bool write = !!(vmf->flags & FAULT_FLAG_WRITE); + intel_runtime_pm_get(dev_priv); + /* We don't use vmf->pgoff since that has the fake offset */ page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >> PAGE_SHIFT; @@ -1344,39 +1533,44 @@ int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) trace_i915_gem_object_fault(obj, page_offset, true, write); - /* Now bind it into the GTT if needed */ - if (!obj->map_and_fenceable) { - ret = i915_gem_object_unbind(obj); - if (ret) - goto unlock; - } - if (!obj->gtt_space) { - ret = i915_gem_object_bind_to_gtt(obj, 0, true, false); - if (ret) - goto unlock; + /* Try to flush the object off the GPU first without holding the lock. + * Upon reacquiring the lock, we will perform our sanity checks and then + * repeat the flush holding the lock in the normal manner to catch cases + * where we are gazumped. + */ + ret = i915_gem_object_wait_rendering__nonblocking(obj, NULL, !write); + if (ret) + goto unlock; - ret = i915_gem_object_set_to_gtt_domain(obj, write); - if (ret) - goto unlock; + /* Access to snoopable pages through the GTT is incoherent. */ + if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(dev)) { + ret = -EFAULT; + goto unlock; } - if (!obj->has_global_gtt_mapping) - i915_gem_gtt_bind_object(obj, obj->cache_level); - - ret = i915_gem_object_get_fence(obj); + /* Now bind it into the GTT if needed */ + ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE); if (ret) goto unlock; - if (i915_gem_object_is_inactive(obj)) - list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); + ret = i915_gem_object_set_to_gtt_domain(obj, write); + if (ret) + goto unpin; + + ret = i915_gem_object_get_fence(obj); + if (ret) + goto unpin; obj->fault_mappable = true; - pfn = ((dev_priv->mm.gtt_base_addr + obj->gtt_offset) >> PAGE_SHIFT) + - page_offset; + pfn = dev_priv->gtt.mappable_base + i915_gem_obj_ggtt_offset(obj); + pfn >>= PAGE_SHIFT; + pfn += page_offset; /* Finally, remap it using the new GTT offset */ ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn); +unpin: + i915_gem_object_ggtt_unpin(obj); unlock: mutex_unlock(&dev->struct_mutex); out: @@ -1385,17 +1579,16 @@ out: /* If this -EIO is due to a gpu hang, give the reset code a * chance to clean up the mess. Otherwise return the proper * SIGBUS. */ - if (!atomic_read(&dev_priv->mm.wedged)) - return VM_FAULT_SIGBUS; + if (i915_terminally_wedged(&dev_priv->gpu_error)) { + ret = VM_FAULT_SIGBUS; + break; + } case -EAGAIN: - /* Give the error handler a chance to run and move the - * objects off the GPU active list. Next time we service the - * fault, we should be able to transition the page into the - * GTT without touching the GPU (and so avoid further - * EIO/EGAIN). If the GPU is wedged, then there is no issue - * with coherency, just lost writes. + /* + * EAGAIN means the gpu is hung and we'll wait for the error + * handler to reset everything when re-faulting in + * i915_mutex_lock_interruptible. */ - set_need_resched(); case 0: case -ERESTARTSYS: case -EINTR: @@ -1404,13 +1597,23 @@ out: * EBUSY is ok: this just means that another thread * already did the job. */ - return VM_FAULT_NOPAGE; + ret = VM_FAULT_NOPAGE; + break; case -ENOMEM: - return VM_FAULT_OOM; + ret = VM_FAULT_OOM; + break; + case -ENOSPC: + case -EFAULT: + ret = VM_FAULT_SIGBUS; + break; default: - WARN_ON_ONCE(ret); - return VM_FAULT_SIGBUS; + WARN_ONCE(ret, "unhandled error in i915_gem_fault: %i\n", ret); + ret = VM_FAULT_SIGBUS; + break; } + + intel_runtime_pm_put(dev_priv); + return ret; } /** @@ -1433,15 +1636,21 @@ i915_gem_release_mmap(struct drm_i915_gem_object *obj) if (!obj->fault_mappable) return; - if (obj->base.dev->dev_mapping) - unmap_mapping_range(obj->base.dev->dev_mapping, - (loff_t)obj->base.map_list.hash.key<<PAGE_SHIFT, - obj->base.size, 1); - + drm_vma_node_unmap(&obj->base.vma_node, + obj->base.dev->anon_inode->i_mapping); obj->fault_mappable = false; } -static uint32_t +void +i915_gem_release_all_mmaps(struct drm_i915_private *dev_priv) +{ + struct drm_i915_gem_object *obj; + + list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) + i915_gem_release_mmap(obj); +} + +uint32_t i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode) { uint32_t gtt_size; @@ -1469,16 +1678,15 @@ i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode) * Return the required GTT alignment for an object, taking into account * potential fence register mapping. */ -static uint32_t -i915_gem_get_gtt_alignment(struct drm_device *dev, - uint32_t size, - int tiling_mode) +uint32_t +i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size, + int tiling_mode, bool fenced) { /* * Minimum alignment is 4k (GTT page size), but might be greater * if a fence register is needed for the object. */ - if (INTEL_INFO(dev)->gen >= 4 || + if (INTEL_INFO(dev)->gen >= 4 || (!fenced && IS_G33(dev)) || tiling_mode == I915_TILING_NONE) return 4096; @@ -1489,46 +1697,19 @@ i915_gem_get_gtt_alignment(struct drm_device *dev, return i915_gem_get_gtt_size(dev, size, tiling_mode); } -/** - * i915_gem_get_unfenced_gtt_alignment - return required GTT alignment for an - * unfenced object - * @dev: the device - * @size: size of the object - * @tiling_mode: tiling mode of the object - * - * Return the required GTT alignment for an object, only taking into account - * unfenced tiled surface requirements. - */ -uint32_t -i915_gem_get_unfenced_gtt_alignment(struct drm_device *dev, - uint32_t size, - int tiling_mode) -{ - /* - * Minimum alignment is 4k (GTT page size) for sane hw. - */ - if (INTEL_INFO(dev)->gen >= 4 || IS_G33(dev) || - tiling_mode == I915_TILING_NONE) - return 4096; - - /* Previous hardware however needs to be aligned to a power-of-two - * tile height. The simplest method for determining this is to reuse - * the power-of-tile object size. - */ - return i915_gem_get_gtt_size(dev, size, tiling_mode); -} - static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj) { struct drm_i915_private *dev_priv = obj->base.dev->dev_private; int ret; - if (obj->base.map_list.map) + if (drm_vma_node_has_offset(&obj->base.vma_node)) return 0; + dev_priv->mm.shrinker_no_lock_stealing = true; + ret = drm_gem_create_mmap_offset(&obj->base); if (ret != -ENOSPC) - return ret; + goto out; /* Badly fragmented mmap space? The only way we can recover * space is by destroying unwanted objects. We can't randomly release @@ -1540,17 +1721,18 @@ static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj) i915_gem_purge(dev_priv, obj->base.size >> PAGE_SHIFT); ret = drm_gem_create_mmap_offset(&obj->base); if (ret != -ENOSPC) - return ret; + goto out; i915_gem_shrink_all(dev_priv); - return drm_gem_create_mmap_offset(&obj->base); + ret = drm_gem_create_mmap_offset(&obj->base); +out: + dev_priv->mm.shrinker_no_lock_stealing = false; + + return ret; } static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) { - if (!obj->base.map_list.map) - return; - drm_gem_free_mmap_offset(&obj->base); } @@ -1574,14 +1756,14 @@ i915_gem_mmap_gtt(struct drm_file *file, goto unlock; } - if (obj->base.size > dev_priv->mm.gtt_mappable_end) { + if (obj->base.size > dev_priv->gtt.mappable_end) { ret = -E2BIG; goto out; } if (obj->madv != I915_MADV_WILLNEED) { - DRM_ERROR("Attempting to mmap a purgeable buffer\n"); - ret = -EINVAL; + DRM_DEBUG("Attempting to mmap a purgeable buffer\n"); + ret = -EFAULT; goto out; } @@ -1589,7 +1771,7 @@ i915_gem_mmap_gtt(struct drm_file *file, if (ret) goto out; - *offset = (u64)obj->base.map_list.hash.key << PAGE_SHIFT; + *offset = drm_vma_node_offset_addr(&obj->base.vma_node); out: drm_gem_object_unreference(&obj->base); @@ -1622,12 +1804,16 @@ i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data, return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset); } +static inline int +i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj) +{ + return obj->madv == I915_MADV_DONTNEED; +} + /* Immediately discard the backing storage */ static void i915_gem_object_truncate(struct drm_i915_gem_object *obj) { - struct inode *inode; - i915_gem_object_free_mmap_offset(obj); if (obj->base.filp == NULL) @@ -1638,24 +1824,35 @@ i915_gem_object_truncate(struct drm_i915_gem_object *obj) * To do this we must instruct the shmfs to drop all of its * backing pages, *now*. */ - inode = obj->base.filp->f_path.dentry->d_inode; - shmem_truncate_range(inode, 0, (loff_t)-1); - + shmem_truncate_range(file_inode(obj->base.filp), 0, (loff_t)-1); obj->madv = __I915_MADV_PURGED; } -static inline int -i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj) +/* Try to discard unwanted pages */ +static void +i915_gem_object_invalidate(struct drm_i915_gem_object *obj) { - return obj->madv == I915_MADV_DONTNEED; + struct address_space *mapping; + + switch (obj->madv) { + case I915_MADV_DONTNEED: + i915_gem_object_truncate(obj); + case __I915_MADV_PURGED: + return; + } + + if (obj->base.filp == NULL) + return; + + mapping = file_inode(obj->base.filp)->i_mapping, + invalidate_mapping_pages(mapping, 0, (loff_t)-1); } static void i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj) { - int page_count = obj->base.size / PAGE_SIZE; - struct scatterlist *sg; - int ret, i; + struct sg_page_iter sg_iter; + int ret; BUG_ON(obj->madv == __I915_MADV_PURGED); @@ -1665,7 +1862,7 @@ i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj) * hope for the best. */ WARN_ON(ret != -EIO); - i915_gem_clflush_object(obj); + i915_gem_clflush_object(obj, true); obj->base.read_domains = obj->base.write_domain = I915_GEM_DOMAIN_CPU; } @@ -1675,8 +1872,8 @@ i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj) if (obj->madv == I915_MADV_DONTNEED) obj->dirty = 0; - for_each_sg(obj->pages->sgl, sg, page_count, i) { - struct page *page = sg_page(sg); + for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, 0) { + struct page *page = sg_page_iter_page(&sg_iter); if (obj->dirty) set_page_dirty(page); @@ -1692,7 +1889,7 @@ i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj) kfree(obj->pages); } -static int +int i915_gem_object_put_pages(struct drm_i915_gem_object *obj) { const struct drm_i915_gem_object_ops *ops = obj->ops; @@ -1700,62 +1897,107 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj) if (obj->pages == NULL) return 0; - BUG_ON(obj->gtt_space); - if (obj->pages_pin_count) return -EBUSY; + BUG_ON(i915_gem_obj_bound_any(obj)); + + /* ->put_pages might need to allocate memory for the bit17 swizzle + * array, hence protect them from being reaped by removing them from gtt + * lists early. */ + list_del(&obj->global_list); + ops->put_pages(obj); obj->pages = NULL; - list_del(&obj->gtt_list); - if (i915_gem_object_is_purgeable(obj)) - i915_gem_object_truncate(obj); + i915_gem_object_invalidate(obj); return 0; } -static long -i915_gem_purge(struct drm_i915_private *dev_priv, long target) +static unsigned long +__i915_gem_shrink(struct drm_i915_private *dev_priv, long target, + bool purgeable_only) { - struct drm_i915_gem_object *obj, *next; - long count = 0; + struct list_head still_in_list; + struct drm_i915_gem_object *obj; + unsigned long count = 0; - list_for_each_entry_safe(obj, next, - &dev_priv->mm.unbound_list, - gtt_list) { - if (i915_gem_object_is_purgeable(obj) && - i915_gem_object_put_pages(obj) == 0) { + /* + * As we may completely rewrite the (un)bound list whilst unbinding + * (due to retiring requests) we have to strictly process only + * one element of the list at the time, and recheck the list + * on every iteration. + * + * In particular, we must hold a reference whilst removing the + * object as we may end up waiting for and/or retiring the objects. + * This might release the final reference (held by the active list) + * and result in the object being freed from under us. This is + * similar to the precautions the eviction code must take whilst + * removing objects. + * + * Also note that although these lists do not hold a reference to + * the object we can safely grab one here: The final object + * unreferencing and the bound_list are both protected by the + * dev->struct_mutex and so we won't ever be able to observe an + * object on the bound_list with a reference count equals 0. + */ + INIT_LIST_HEAD(&still_in_list); + while (count < target && !list_empty(&dev_priv->mm.unbound_list)) { + obj = list_first_entry(&dev_priv->mm.unbound_list, + typeof(*obj), global_list); + list_move_tail(&obj->global_list, &still_in_list); + + if (!i915_gem_object_is_purgeable(obj) && purgeable_only) + continue; + + drm_gem_object_reference(&obj->base); + + if (i915_gem_object_put_pages(obj) == 0) count += obj->base.size >> PAGE_SHIFT; - if (count >= target) - return count; - } + + drm_gem_object_unreference(&obj->base); } + list_splice(&still_in_list, &dev_priv->mm.unbound_list); + + INIT_LIST_HEAD(&still_in_list); + while (count < target && !list_empty(&dev_priv->mm.bound_list)) { + struct i915_vma *vma, *v; - list_for_each_entry_safe(obj, next, - &dev_priv->mm.inactive_list, - mm_list) { - if (i915_gem_object_is_purgeable(obj) && - i915_gem_object_unbind(obj) == 0 && - i915_gem_object_put_pages(obj) == 0) { + obj = list_first_entry(&dev_priv->mm.bound_list, + typeof(*obj), global_list); + list_move_tail(&obj->global_list, &still_in_list); + + if (!i915_gem_object_is_purgeable(obj) && purgeable_only) + continue; + + drm_gem_object_reference(&obj->base); + + list_for_each_entry_safe(vma, v, &obj->vma_list, vma_link) + if (i915_vma_unbind(vma)) + break; + + if (i915_gem_object_put_pages(obj) == 0) count += obj->base.size >> PAGE_SHIFT; - if (count >= target) - return count; - } + + drm_gem_object_unreference(&obj->base); } + list_splice(&still_in_list, &dev_priv->mm.bound_list); return count; } -static void -i915_gem_shrink_all(struct drm_i915_private *dev_priv) +static unsigned long +i915_gem_purge(struct drm_i915_private *dev_priv, long target) { - struct drm_i915_gem_object *obj, *next; + return __i915_gem_shrink(dev_priv, target, true); +} +static unsigned long +i915_gem_shrink_all(struct drm_i915_private *dev_priv) +{ i915_gem_evict_everything(dev_priv->dev); - - list_for_each_entry_safe(obj, next, &dev_priv->mm.unbound_list, gtt_list) - i915_gem_object_put_pages(obj); + return __i915_gem_shrink(dev_priv, LONG_MAX, false); } static int @@ -1766,7 +2008,9 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) struct address_space *mapping; struct sg_table *st; struct scatterlist *sg; + struct sg_page_iter sg_iter; struct page *page; + unsigned long last_pfn = 0; /* suppress gcc warning */ gfp_t gfp; /* Assert that the object is not currently in any GPU domain. As it @@ -1782,7 +2026,6 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) page_count = obj->base.size / PAGE_SIZE; if (sg_alloc_table(st, page_count, GFP_KERNEL)) { - sg_free_table(st); kfree(st); return -ENOMEM; } @@ -1792,11 +2035,13 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) * * Fail silently without starting the shrinker */ - mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping; + mapping = file_inode(obj->base.filp)->i_mapping; gfp = mapping_gfp_mask(mapping); - gfp |= __GFP_NORETRY | __GFP_NOWARN; + gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD; gfp &= ~(__GFP_IO | __GFP_WAIT); - for_each_sg(st->sgl, sg, page_count, i) { + sg = st->sgl; + st->nents = 0; + for (i = 0; i < page_count; i++) { page = shmem_read_mapping_page_gfp(mapping, i, gfp); if (IS_ERR(page)) { i915_gem_purge(dev_priv, page_count); @@ -1807,7 +2052,7 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) * our own buffer, now let the real VM do its job and * go down in flames if truly OOM. */ - gfp &= ~(__GFP_NORETRY | __GFP_NOWARN); + gfp &= ~(__GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD); gfp |= __GFP_IO | __GFP_WAIT; i915_gem_shrink_all(dev_priv); @@ -1815,25 +2060,60 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) if (IS_ERR(page)) goto err_pages; - gfp |= __GFP_NORETRY | __GFP_NOWARN; + gfp |= __GFP_NORETRY | __GFP_NOWARN | __GFP_NO_KSWAPD; gfp &= ~(__GFP_IO | __GFP_WAIT); } +#ifdef CONFIG_SWIOTLB + if (swiotlb_nr_tbl()) { + st->nents++; + sg_set_page(sg, page, PAGE_SIZE, 0); + sg = sg_next(sg); + continue; + } +#endif + if (!i || page_to_pfn(page) != last_pfn + 1) { + if (i) + sg = sg_next(sg); + st->nents++; + sg_set_page(sg, page, PAGE_SIZE, 0); + } else { + sg->length += PAGE_SIZE; + } + last_pfn = page_to_pfn(page); - sg_set_page(sg, page, PAGE_SIZE, 0); + /* Check that the i965g/gm workaround works. */ + WARN_ON((gfp & __GFP_DMA32) && (last_pfn >= 0x00100000UL)); } +#ifdef CONFIG_SWIOTLB + if (!swiotlb_nr_tbl()) +#endif + sg_mark_end(sg); + obj->pages = st; if (i915_gem_object_needs_bit17_swizzle(obj)) i915_gem_object_do_bit_17_swizzle(obj); - obj->pages = st; return 0; err_pages: - for_each_sg(st->sgl, sg, i, page_count) - page_cache_release(sg_page(sg)); + sg_mark_end(sg); + for_each_sg_page(st->sgl, &sg_iter, st->nents, 0) + page_cache_release(sg_page_iter_page(&sg_iter)); sg_free_table(st); kfree(st); - return PTR_ERR(page); + + /* shmemfs first checks if there is enough memory to allocate the page + * and reports ENOSPC should there be insufficient, along with the usual + * ENOMEM for a genuine allocation failure. + * + * We use ENOSPC in our driver to mean that we have run out of aperture + * space and so want to translate the error from shmemfs back to our + * usual understanding of ENOMEM. + */ + if (PTR_ERR(page) == -ENOSPC) + return -ENOMEM; + else + return PTR_ERR(page); } /* Ensure that the associated pages are gathered from the backing storage @@ -1853,25 +2133,34 @@ i915_gem_object_get_pages(struct drm_i915_gem_object *obj) if (obj->pages) return 0; + if (obj->madv != I915_MADV_WILLNEED) { + DRM_DEBUG("Attempting to obtain a purgeable object\n"); + return -EFAULT; + } + BUG_ON(obj->pages_pin_count); ret = ops->get_pages(obj); if (ret) return ret; - list_add_tail(&obj->gtt_list, &dev_priv->mm.unbound_list); + list_add_tail(&obj->global_list, &dev_priv->mm.unbound_list); return 0; } -void +static void i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, - struct intel_ring_buffer *ring, - u32 seqno) + struct intel_engine_cs *ring) { struct drm_device *dev = obj->base.dev; struct drm_i915_private *dev_priv = dev->dev_private; + u32 seqno = intel_ring_get_seqno(ring); BUG_ON(ring == NULL); + if (obj->ring != ring && obj->last_write_seqno) { + /* Keep the seqno relative to the current ring */ + obj->last_write_seqno = seqno; + } obj->ring = ring; /* Add a reference if we're newly entering the active list. */ @@ -1880,8 +2169,6 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, obj->active = 1; } - /* Move from whatever list we were on to the tail of execution. */ - list_move_tail(&obj->mm_list, &dev_priv->mm.active_list); list_move_tail(&obj->ring_list, &ring->active_list); obj->last_read_seqno = seqno; @@ -1900,19 +2187,28 @@ i915_gem_object_move_to_active(struct drm_i915_gem_object *obj, } } +void i915_vma_move_to_active(struct i915_vma *vma, + struct intel_engine_cs *ring) +{ + list_move_tail(&vma->mm_list, &vma->vm->active_list); + return i915_gem_object_move_to_active(vma->obj, ring); +} + static void i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj) { - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + struct i915_address_space *vm; + struct i915_vma *vma; BUG_ON(obj->base.write_domain & ~I915_GEM_GPU_DOMAINS); BUG_ON(!obj->active); - if (obj->pin_count) /* are we a framebuffer? */ - intel_mark_fb_idle(obj); - - list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); + list_for_each_entry(vm, &dev_priv->vm_list, global_link) { + vma = i915_gem_obj_to_vma(obj, vm); + if (vma && !list_empty(&vma->mm_list)) + list_move_tail(&vma->mm_list, &vm->inactive_list); + } list_del_init(&obj->ring_list); obj->ring = NULL; @@ -1930,40 +2226,100 @@ i915_gem_object_move_to_inactive(struct drm_i915_gem_object *obj) WARN_ON(i915_verify_lists(dev)); } -static u32 -i915_gem_get_seqno(struct drm_device *dev) +static void +i915_gem_object_retire(struct drm_i915_gem_object *obj) { - drm_i915_private_t *dev_priv = dev->dev_private; - u32 seqno = dev_priv->next_seqno; + struct intel_engine_cs *ring = obj->ring; - /* reserve 0 for non-seqno */ - if (++dev_priv->next_seqno == 0) - dev_priv->next_seqno = 1; + if (ring == NULL) + return; - return seqno; + if (i915_seqno_passed(ring->get_seqno(ring, true), + obj->last_read_seqno)) + i915_gem_object_move_to_inactive(obj); } -u32 -i915_gem_next_request_seqno(struct intel_ring_buffer *ring) +static int +i915_gem_init_seqno(struct drm_device *dev, u32 seqno) { - if (ring->outstanding_lazy_request == 0) - ring->outstanding_lazy_request = i915_gem_get_seqno(ring->dev); + struct drm_i915_private *dev_priv = dev->dev_private; + struct intel_engine_cs *ring; + int ret, i, j; + + /* Carefully retire all requests without writing to the rings */ + for_each_ring(ring, dev_priv, i) { + ret = intel_ring_idle(ring); + if (ret) + return ret; + } + i915_gem_retire_requests(dev); + + /* Finally reset hw state */ + for_each_ring(ring, dev_priv, i) { + intel_ring_init_seqno(ring, seqno); + + for (j = 0; j < ARRAY_SIZE(ring->semaphore.sync_seqno); j++) + ring->semaphore.sync_seqno[j] = 0; + } - return ring->outstanding_lazy_request; + return 0; +} + +int i915_gem_set_seqno(struct drm_device *dev, u32 seqno) +{ + struct drm_i915_private *dev_priv = dev->dev_private; + int ret; + + if (seqno == 0) + return -EINVAL; + + /* HWS page needs to be set less than what we + * will inject to ring + */ + ret = i915_gem_init_seqno(dev, seqno - 1); + if (ret) + return ret; + + /* Carefully set the last_seqno value so that wrap + * detection still works + */ + dev_priv->next_seqno = seqno; + dev_priv->last_seqno = seqno - 1; + if (dev_priv->last_seqno == 0) + dev_priv->last_seqno--; + + return 0; } int -i915_add_request(struct intel_ring_buffer *ring, - struct drm_file *file, - u32 *out_seqno) +i915_gem_get_seqno(struct drm_device *dev, u32 *seqno) { - drm_i915_private_t *dev_priv = ring->dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; + + /* reserve 0 for non-seqno */ + if (dev_priv->next_seqno == 0) { + int ret = i915_gem_init_seqno(dev, 0); + if (ret) + return ret; + + dev_priv->next_seqno = 1; + } + + *seqno = dev_priv->last_seqno = dev_priv->next_seqno++; + return 0; +} + +int __i915_add_request(struct intel_engine_cs *ring, + struct drm_file *file, + struct drm_i915_gem_object *obj, + u32 *out_seqno) +{ + struct drm_i915_private *dev_priv = ring->dev->dev_private; struct drm_i915_gem_request *request; - u32 request_ring_position; - u32 seqno; - int was_empty; + u32 request_ring_position, request_start; int ret; + request_start = intel_ring_get_tail(ring); /* * Emit any outstanding flushes - execbuf can fail to emit the flush * after having emitted the batchbuffer command. Hence we need to fix @@ -1975,12 +2331,10 @@ i915_add_request(struct intel_ring_buffer *ring, if (ret) return ret; - request = kmalloc(sizeof(*request), GFP_KERNEL); - if (request == NULL) + request = ring->preallocated_lazy_request; + if (WARN_ON(request == NULL)) return -ENOMEM; - seqno = i915_gem_next_request_seqno(ring); - /* Record the position of the start of the request so that * should we detect the updated seqno part-way through the * GPU processing the request, we never over-estimate the @@ -1988,19 +2342,31 @@ i915_add_request(struct intel_ring_buffer *ring, */ request_ring_position = intel_ring_get_tail(ring); - ret = ring->add_request(ring, &seqno); - if (ret) { - kfree(request); + ret = ring->add_request(ring); + if (ret) return ret; - } - - trace_i915_gem_request_add(ring, seqno); - request->seqno = seqno; + request->seqno = intel_ring_get_seqno(ring); request->ring = ring; + request->head = request_start; request->tail = request_ring_position; + + /* Whilst this request exists, batch_obj will be on the + * active_list, and so will hold the active reference. Only when this + * request is retired will the the batch_obj be moved onto the + * inactive_list and lose its active reference. Hence we do not need + * to explicitly hold another reference here. + */ + request->batch_obj = obj; + + /* Hold a reference to the current context so that we can inspect + * it later in case a hangcheck error event fires. + */ + request->ctx = ring->last_context; + if (request->ctx) + i915_gem_context_reference(request->ctx); + request->emitted_jiffies = jiffies; - was_empty = list_empty(&ring->request_list); list_add_tail(&request->list, &ring->request_list); request->file_priv = NULL; @@ -2014,23 +2380,22 @@ i915_add_request(struct intel_ring_buffer *ring, spin_unlock(&file_priv->mm.lock); } - ring->outstanding_lazy_request = 0; + trace_i915_gem_request_add(ring, request->seqno); + ring->outstanding_lazy_seqno = 0; + ring->preallocated_lazy_request = NULL; - if (!dev_priv->mm.suspended) { - if (i915_enable_hangcheck) { - mod_timer(&dev_priv->hangcheck_timer, - jiffies + - msecs_to_jiffies(DRM_I915_HANGCHECK_PERIOD)); - } - if (was_empty) { - queue_delayed_work(dev_priv->wq, - &dev_priv->mm.retire_work, HZ); - intel_mark_busy(dev_priv->dev); - } + if (!dev_priv->ums.mm_suspended) { + i915_queue_hangcheck(ring->dev); + + cancel_delayed_work_sync(&dev_priv->mm.idle_work); + queue_delayed_work(dev_priv->wq, + &dev_priv->mm.retire_work, + round_jiffies_up_relative(HZ)); + intel_mark_busy(dev_priv->dev); } if (out_seqno) - *out_seqno = seqno; + *out_seqno = request->seqno; return 0; } @@ -2043,28 +2408,106 @@ i915_gem_request_remove_from_client(struct drm_i915_gem_request *request) return; spin_lock(&file_priv->mm.lock); - if (request->file_priv) { - list_del(&request->client_list); - request->file_priv = NULL; - } + list_del(&request->client_list); + request->file_priv = NULL; spin_unlock(&file_priv->mm.lock); } -static void i915_gem_reset_ring_lists(struct drm_i915_private *dev_priv, - struct intel_ring_buffer *ring) +static bool i915_context_is_banned(struct drm_i915_private *dev_priv, + const struct intel_context *ctx) { - while (!list_empty(&ring->request_list)) { - struct drm_i915_gem_request *request; + unsigned long elapsed; - request = list_first_entry(&ring->request_list, - struct drm_i915_gem_request, - list); + elapsed = get_seconds() - ctx->hang_stats.guilty_ts; + + if (ctx->hang_stats.banned) + return true; + + if (elapsed <= DRM_I915_CTX_BAN_PERIOD) { + if (!i915_gem_context_is_default(ctx)) { + DRM_DEBUG("context hanging too fast, banning!\n"); + return true; + } else if (i915_stop_ring_allow_ban(dev_priv)) { + if (i915_stop_ring_allow_warn(dev_priv)) + DRM_ERROR("gpu hanging too fast, banning!\n"); + return true; + } + } + + return false; +} + +static void i915_set_reset_status(struct drm_i915_private *dev_priv, + struct intel_context *ctx, + const bool guilty) +{ + struct i915_ctx_hang_stats *hs; + + if (WARN_ON(!ctx)) + return; + + hs = &ctx->hang_stats; + + if (guilty) { + hs->banned = i915_context_is_banned(dev_priv, ctx); + hs->batch_active++; + hs->guilty_ts = get_seconds(); + } else { + hs->batch_pending++; + } +} + +static void i915_gem_free_request(struct drm_i915_gem_request *request) +{ + list_del(&request->list); + i915_gem_request_remove_from_client(request); + + if (request->ctx) + i915_gem_context_unreference(request->ctx); + + kfree(request); +} + +struct drm_i915_gem_request * +i915_gem_find_active_request(struct intel_engine_cs *ring) +{ + struct drm_i915_gem_request *request; + u32 completed_seqno; - list_del(&request->list); - i915_gem_request_remove_from_client(request); - kfree(request); + completed_seqno = ring->get_seqno(ring, false); + + list_for_each_entry(request, &ring->request_list, list) { + if (i915_seqno_passed(completed_seqno, request->seqno)) + continue; + + return request; } + return NULL; +} + +static void i915_gem_reset_ring_status(struct drm_i915_private *dev_priv, + struct intel_engine_cs *ring) +{ + struct drm_i915_gem_request *request; + bool ring_hung; + + request = i915_gem_find_active_request(ring); + + if (request == NULL) + return; + + ring_hung = ring->hangcheck.score >= HANGCHECK_SCORE_RING_HUNG; + + i915_set_reset_status(dev_priv, request->ctx, ring_hung); + + list_for_each_entry_continue(request, &ring->request_list, list) + i915_set_reset_status(dev_priv, request->ctx, false); +} + +static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv, + struct intel_engine_cs *ring) +{ while (!list_empty(&ring->active_list)) { struct drm_i915_gem_object *obj; @@ -2074,9 +2517,31 @@ static void i915_gem_reset_ring_lists(struct drm_i915_private *dev_priv, i915_gem_object_move_to_inactive(obj); } + + /* + * We must free the requests after all the corresponding objects have + * been moved off active lists. Which is the same order as the normal + * retire_requests function does. This is important if object hold + * implicit references on things like e.g. ppgtt address spaces through + * the request. + */ + while (!list_empty(&ring->request_list)) { + struct drm_i915_gem_request *request; + + request = list_first_entry(&ring->request_list, + struct drm_i915_gem_request, + list); + + i915_gem_free_request(request); + } + + /* These may not have been flush before the reset, do so now */ + kfree(ring->preallocated_lazy_request); + ring->preallocated_lazy_request = NULL; + ring->outstanding_lazy_seqno = 0; } -static void i915_gem_reset_fences(struct drm_device *dev) +void i915_gem_restore_fences(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; int i; @@ -2084,51 +2549,48 @@ static void i915_gem_reset_fences(struct drm_device *dev) for (i = 0; i < dev_priv->num_fence_regs; i++) { struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i]; - i915_gem_write_fence(dev, i, NULL); - - if (reg->obj) - i915_gem_object_fence_lost(reg->obj); - - reg->pin_count = 0; - reg->obj = NULL; - INIT_LIST_HEAD(®->lru_list); + /* + * Commit delayed tiling changes if we have an object still + * attached to the fence, otherwise just clear the fence. + */ + if (reg->obj) { + i915_gem_object_update_fence(reg->obj, reg, + reg->obj->tiling_mode); + } else { + i915_gem_write_fence(dev, i, NULL); + } } - - INIT_LIST_HEAD(&dev_priv->mm.fence_list); } void i915_gem_reset(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; - struct drm_i915_gem_object *obj; - struct intel_ring_buffer *ring; + struct intel_engine_cs *ring; int i; + /* + * Before we free the objects from the requests, we need to inspect + * them for finding the guilty party. As the requests only borrow + * their reference to the objects, the inspection must be done first. + */ for_each_ring(ring, dev_priv, i) - i915_gem_reset_ring_lists(dev_priv, ring); + i915_gem_reset_ring_status(dev_priv, ring); - /* Move everything out of the GPU domains to ensure we do any - * necessary invalidation upon reuse. - */ - list_for_each_entry(obj, - &dev_priv->mm.inactive_list, - mm_list) - { - obj->base.read_domains &= ~I915_GEM_GPU_DOMAINS; - } + for_each_ring(ring, dev_priv, i) + i915_gem_reset_ring_cleanup(dev_priv, ring); + + i915_gem_context_reset(dev); - /* The fence registers are invalidated so clear them out */ - i915_gem_reset_fences(dev); + i915_gem_restore_fences(dev); } /** * This function clears the request list as sequence numbers are passed. */ void -i915_gem_retire_requests_ring(struct intel_ring_buffer *ring) +i915_gem_retire_requests_ring(struct intel_engine_cs *ring) { uint32_t seqno; - int i; if (list_empty(&ring->request_list)) return; @@ -2137,9 +2599,23 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring) seqno = ring->get_seqno(ring, true); - for (i = 0; i < ARRAY_SIZE(ring->sync_seqno); i++) - if (seqno >= ring->sync_seqno[i]) - ring->sync_seqno[i] = 0; + /* Move any buffers on the active list that are no longer referenced + * by the ringbuffer to the flushing/inactive lists as appropriate, + * before we free the context associated with the requests. + */ + while (!list_empty(&ring->active_list)) { + struct drm_i915_gem_object *obj; + + obj = list_first_entry(&ring->active_list, + struct drm_i915_gem_object, + ring_list); + + if (!i915_seqno_passed(seqno, obj->last_read_seqno)) + break; + + i915_gem_object_move_to_inactive(obj); + } + while (!list_empty(&ring->request_list)) { struct drm_i915_gem_request *request; @@ -2157,27 +2633,9 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring) * of tail of the request to update the last known position * of the GPU head. */ - ring->last_retired_head = request->tail; - - list_del(&request->list); - i915_gem_request_remove_from_client(request); - kfree(request); - } - - /* Move any buffers on the active list that are no longer referenced - * by the ringbuffer to the flushing/inactive lists as appropriate. - */ - while (!list_empty(&ring->active_list)) { - struct drm_i915_gem_object *obj; - - obj = list_first_entry(&ring->active_list, - struct drm_i915_gem_object, - ring_list); - - if (!i915_seqno_passed(seqno, obj->last_read_seqno)) - break; + ring->buffer->last_retired_head = request->tail; - i915_gem_object_move_to_inactive(obj); + i915_gem_free_request(request); } if (unlikely(ring->trace_irq_seqno && @@ -2189,55 +2647,53 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring) WARN_ON(i915_verify_lists(ring->dev)); } -void +bool i915_gem_retire_requests(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; - struct intel_ring_buffer *ring; + struct drm_i915_private *dev_priv = dev->dev_private; + struct intel_engine_cs *ring; + bool idle = true; int i; - for_each_ring(ring, dev_priv, i) + for_each_ring(ring, dev_priv, i) { i915_gem_retire_requests_ring(ring); + idle &= list_empty(&ring->request_list); + } + + if (idle) + mod_delayed_work(dev_priv->wq, + &dev_priv->mm.idle_work, + msecs_to_jiffies(100)); + + return idle; } static void i915_gem_retire_work_handler(struct work_struct *work) { - drm_i915_private_t *dev_priv; - struct drm_device *dev; - struct intel_ring_buffer *ring; + struct drm_i915_private *dev_priv = + container_of(work, typeof(*dev_priv), mm.retire_work.work); + struct drm_device *dev = dev_priv->dev; bool idle; - int i; - - dev_priv = container_of(work, drm_i915_private_t, - mm.retire_work.work); - dev = dev_priv->dev; /* Come back later if the device is busy... */ - if (!mutex_trylock(&dev->struct_mutex)) { - queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, HZ); - return; - } - - i915_gem_retire_requests(dev); - - /* Send a periodic flush down the ring so we don't hold onto GEM - * objects indefinitely. - */ - idle = true; - for_each_ring(ring, dev_priv, i) { - if (ring->gpu_caches_dirty) - i915_add_request(ring, NULL, NULL); - - idle &= list_empty(&ring->request_list); + idle = false; + if (mutex_trylock(&dev->struct_mutex)) { + idle = i915_gem_retire_requests(dev); + mutex_unlock(&dev->struct_mutex); } + if (!idle) + queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, + round_jiffies_up_relative(HZ)); +} - if (!dev_priv->mm.suspended && !idle) - queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, HZ); - if (idle) - intel_mark_idle(dev); +static void +i915_gem_idle_work_handler(struct work_struct *work) +{ + struct drm_i915_private *dev_priv = + container_of(work, typeof(*dev_priv), mm.idle_work.work); - mutex_unlock(&dev->struct_mutex); + intel_mark_idle(dev_priv->dev); } /** @@ -2286,10 +2742,12 @@ i915_gem_object_flush_active(struct drm_i915_gem_object *obj) int i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { + struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_gem_wait *args = data; struct drm_i915_gem_object *obj; - struct intel_ring_buffer *ring = NULL; + struct intel_engine_cs *ring = NULL; struct timespec timeout_stack, *timeout = NULL; + unsigned reset_counter; u32 seqno = 0; int ret = 0; @@ -2330,13 +2788,12 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) } drm_gem_object_unreference(&obj->base); + reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter); mutex_unlock(&dev->struct_mutex); - ret = __wait_seqno(ring, seqno, true, timeout); - if (timeout) { - WARN_ON(!timespec_valid(timeout)); + ret = __wait_seqno(ring, seqno, reset_counter, true, timeout, file->driver_priv); + if (timeout) args->timeout_ns = timespec_to_ns(timeout); - } return ret; out: @@ -2359,9 +2816,9 @@ out: */ int i915_gem_object_sync(struct drm_i915_gem_object *obj, - struct intel_ring_buffer *to) + struct intel_engine_cs *to) { - struct intel_ring_buffer *from = obj->ring; + struct intel_engine_cs *from = obj->ring; u32 seqno; int ret, idx; @@ -2374,16 +2831,21 @@ i915_gem_object_sync(struct drm_i915_gem_object *obj, idx = intel_ring_sync_index(from, to); seqno = obj->last_read_seqno; - if (seqno <= from->sync_seqno[idx]) + if (seqno <= from->semaphore.sync_seqno[idx]) return 0; ret = i915_gem_check_olr(obj->ring, seqno); if (ret) return ret; - ret = to->sync_to(to, from, seqno); + trace_i915_gem_ring_sync_to(from, to, seqno); + ret = to->semaphore.sync_to(to, from, seqno); if (!ret) - from->sync_seqno[idx] = seqno; + /* We use last_read_seqno because sync_to() + * might have just caused seqno wrap under + * the radar. + */ + from->semaphore.sync_seqno[idx] = obj->last_read_seqno; return ret; } @@ -2392,15 +2854,15 @@ static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) { u32 old_write_domain, old_read_domains; - /* Act a barrier for all accesses through the GTT */ - mb(); - /* Force a pagefault for domain tracking on next user access */ i915_gem_release_mmap(obj); if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) return; + /* Wait for any direct GTT access to complete */ + mb(); + old_read_domains = obj->base.read_domains; old_write_domain = obj->base.write_domain; @@ -2412,19 +2874,21 @@ static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) old_write_domain); } -/** - * Unbinds an object from the GTT aperture. - */ -int -i915_gem_object_unbind(struct drm_i915_gem_object *obj) +int i915_vma_unbind(struct i915_vma *vma) { - drm_i915_private_t *dev_priv = obj->base.dev->dev_private; - int ret = 0; + struct drm_i915_gem_object *obj = vma->obj; + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + int ret; - if (obj->gtt_space == NULL) + if (list_empty(&vma->vma_link)) return 0; - if (obj->pin_count) + if (!drm_mm_node_allocated(&vma->node)) { + i915_gem_vma_destroy(vma); + return 0; + } + + if (vma->pin_count) return -EBUSY; BUG_ON(obj->pages == NULL); @@ -2437,56 +2901,56 @@ i915_gem_object_unbind(struct drm_i915_gem_object *obj) * cause memory corruption through use-after-free. */ - i915_gem_object_finish_gtt(obj); + if (i915_is_ggtt(vma->vm)) { + i915_gem_object_finish_gtt(obj); - /* release the fence reg _after_ flushing */ - ret = i915_gem_object_put_fence(obj); - if (ret) - return ret; + /* release the fence reg _after_ flushing */ + ret = i915_gem_object_put_fence(obj); + if (ret) + return ret; + } - trace_i915_gem_object_unbind(obj); + trace_i915_vma_unbind(vma); + + vma->unbind_vma(vma); - if (obj->has_global_gtt_mapping) - i915_gem_gtt_unbind_object(obj); - if (obj->has_aliasing_ppgtt_mapping) { - i915_ppgtt_unbind_object(dev_priv->mm.aliasing_ppgtt, obj); - obj->has_aliasing_ppgtt_mapping = 0; - } i915_gem_gtt_finish_object(obj); - list_del(&obj->mm_list); - list_move_tail(&obj->gtt_list, &dev_priv->mm.unbound_list); + list_del_init(&vma->mm_list); /* Avoid an unnecessary call to unbind on rebind. */ - obj->map_and_fenceable = true; + if (i915_is_ggtt(vma->vm)) + obj->map_and_fenceable = true; - drm_mm_put_block(obj->gtt_space); - obj->gtt_space = NULL; - obj->gtt_offset = 0; + drm_mm_remove_node(&vma->node); + i915_gem_vma_destroy(vma); - return 0; -} + /* Since the unbound list is global, only move to that list if + * no more VMAs exist. */ + if (list_empty(&obj->vma_list)) + list_move_tail(&obj->global_list, &dev_priv->mm.unbound_list); -static int i915_ring_idle(struct intel_ring_buffer *ring) -{ - if (list_empty(&ring->active_list)) - return 0; + /* And finally now the object is completely decoupled from this vma, + * we can drop its hold on the backing storage and allow it to be + * reaped by the shrinker. + */ + i915_gem_object_unpin_pages(obj); - return i915_wait_seqno(ring, i915_gem_next_request_seqno(ring)); + return 0; } int i915_gpu_idle(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; - struct intel_ring_buffer *ring; + struct drm_i915_private *dev_priv = dev->dev_private; + struct intel_engine_cs *ring; int ret, i; /* Flush everything onto the inactive list. */ for_each_ring(ring, dev_priv, i) { - ret = i915_switch_context(ring, NULL, DEFAULT_CONTEXT_ID); + ret = i915_switch_context(ring, ring->default_context); if (ret) return ret; - ret = i915_ring_idle(ring); + ret = intel_ring_idle(ring); if (ret) return ret; } @@ -2494,70 +2958,74 @@ int i915_gpu_idle(struct drm_device *dev) return 0; } -static void sandybridge_write_fence_reg(struct drm_device *dev, int reg, - struct drm_i915_gem_object *obj) +static void i965_write_fence_reg(struct drm_device *dev, int reg, + struct drm_i915_gem_object *obj) { - drm_i915_private_t *dev_priv = dev->dev_private; - uint64_t val; - - if (obj) { - u32 size = obj->gtt_space->size; - - val = (uint64_t)((obj->gtt_offset + size - 4096) & - 0xfffff000) << 32; - val |= obj->gtt_offset & 0xfffff000; - val |= (uint64_t)((obj->stride / 128) - 1) << - SANDYBRIDGE_FENCE_PITCH_SHIFT; + struct drm_i915_private *dev_priv = dev->dev_private; + int fence_reg; + int fence_pitch_shift; - if (obj->tiling_mode == I915_TILING_Y) - val |= 1 << I965_FENCE_TILING_Y_SHIFT; - val |= I965_FENCE_REG_VALID; - } else - val = 0; + if (INTEL_INFO(dev)->gen >= 6) { + fence_reg = FENCE_REG_SANDYBRIDGE_0; + fence_pitch_shift = SANDYBRIDGE_FENCE_PITCH_SHIFT; + } else { + fence_reg = FENCE_REG_965_0; + fence_pitch_shift = I965_FENCE_PITCH_SHIFT; + } - I915_WRITE64(FENCE_REG_SANDYBRIDGE_0 + reg * 8, val); - POSTING_READ(FENCE_REG_SANDYBRIDGE_0 + reg * 8); -} + fence_reg += reg * 8; -static void i965_write_fence_reg(struct drm_device *dev, int reg, - struct drm_i915_gem_object *obj) -{ - drm_i915_private_t *dev_priv = dev->dev_private; - uint64_t val; + /* To w/a incoherency with non-atomic 64-bit register updates, + * we split the 64-bit update into two 32-bit writes. In order + * for a partial fence not to be evaluated between writes, we + * precede the update with write to turn off the fence register, + * and only enable the fence as the last step. + * + * For extra levels of paranoia, we make sure each step lands + * before applying the next step. + */ + I915_WRITE(fence_reg, 0); + POSTING_READ(fence_reg); if (obj) { - u32 size = obj->gtt_space->size; + u32 size = i915_gem_obj_ggtt_size(obj); + uint64_t val; - val = (uint64_t)((obj->gtt_offset + size - 4096) & + val = (uint64_t)((i915_gem_obj_ggtt_offset(obj) + size - 4096) & 0xfffff000) << 32; - val |= obj->gtt_offset & 0xfffff000; - val |= ((obj->stride / 128) - 1) << I965_FENCE_PITCH_SHIFT; + val |= i915_gem_obj_ggtt_offset(obj) & 0xfffff000; + val |= (uint64_t)((obj->stride / 128) - 1) << fence_pitch_shift; if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I965_FENCE_TILING_Y_SHIFT; val |= I965_FENCE_REG_VALID; - } else - val = 0; - I915_WRITE64(FENCE_REG_965_0 + reg * 8, val); - POSTING_READ(FENCE_REG_965_0 + reg * 8); + I915_WRITE(fence_reg + 4, val >> 32); + POSTING_READ(fence_reg + 4); + + I915_WRITE(fence_reg + 0, val); + POSTING_READ(fence_reg); + } else { + I915_WRITE(fence_reg + 4, 0); + POSTING_READ(fence_reg + 4); + } } static void i915_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; u32 val; if (obj) { - u32 size = obj->gtt_space->size; + u32 size = i915_gem_obj_ggtt_size(obj); int pitch_val; int tile_width; - WARN((obj->gtt_offset & ~I915_FENCE_START_MASK) || + WARN((i915_gem_obj_ggtt_offset(obj) & ~I915_FENCE_START_MASK) || (size & -size) != size || - (obj->gtt_offset & (size - 1)), - "object 0x%08x [fenceable? %d] not 1M or pot-size (0x%08x) aligned\n", - obj->gtt_offset, obj->map_and_fenceable, size); + (i915_gem_obj_ggtt_offset(obj) & (size - 1)), + "object 0x%08lx [fenceable? %d] not 1M or pot-size (0x%08x) aligned\n", + i915_gem_obj_ggtt_offset(obj), obj->map_and_fenceable, size); if (obj->tiling_mode == I915_TILING_Y && HAS_128_BYTE_Y_TILING(dev)) tile_width = 128; @@ -2568,7 +3036,7 @@ static void i915_write_fence_reg(struct drm_device *dev, int reg, pitch_val = obj->stride / tile_width; pitch_val = ffs(pitch_val) - 1; - val = obj->gtt_offset; + val = i915_gem_obj_ggtt_offset(obj); if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I830_FENCE_TILING_Y_SHIFT; val |= I915_FENCE_SIZE_BITS(size); @@ -2589,23 +3057,23 @@ static void i915_write_fence_reg(struct drm_device *dev, int reg, static void i830_write_fence_reg(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; uint32_t val; if (obj) { - u32 size = obj->gtt_space->size; + u32 size = i915_gem_obj_ggtt_size(obj); uint32_t pitch_val; - WARN((obj->gtt_offset & ~I830_FENCE_START_MASK) || + WARN((i915_gem_obj_ggtt_offset(obj) & ~I830_FENCE_START_MASK) || (size & -size) != size || - (obj->gtt_offset & (size - 1)), - "object 0x%08x not 512K or pot-size 0x%08x aligned\n", - obj->gtt_offset, size); + (i915_gem_obj_ggtt_offset(obj) & (size - 1)), + "object 0x%08lx not 512K or pot-size 0x%08x aligned\n", + i915_gem_obj_ggtt_offset(obj), size); pitch_val = obj->stride / 128; pitch_val = ffs(pitch_val) - 1; - val = obj->gtt_offset; + val = i915_gem_obj_ggtt_offset(obj); if (obj->tiling_mode == I915_TILING_Y) val |= 1 << I830_FENCE_TILING_Y_SHIFT; val |= I830_FENCE_SIZE_BITS(size); @@ -2618,18 +3086,42 @@ static void i830_write_fence_reg(struct drm_device *dev, int reg, POSTING_READ(FENCE_REG_830_0 + reg * 4); } +inline static bool i915_gem_object_needs_mb(struct drm_i915_gem_object *obj) +{ + return obj && obj->base.read_domains & I915_GEM_DOMAIN_GTT; +} + static void i915_gem_write_fence(struct drm_device *dev, int reg, struct drm_i915_gem_object *obj) { + struct drm_i915_private *dev_priv = dev->dev_private; + + /* Ensure that all CPU reads are completed before installing a fence + * and all writes before removing the fence. + */ + if (i915_gem_object_needs_mb(dev_priv->fence_regs[reg].obj)) + mb(); + + WARN(obj && (!obj->stride || !obj->tiling_mode), + "bogus fence setup with stride: 0x%x, tiling mode: %i\n", + obj->stride, obj->tiling_mode); + switch (INTEL_INFO(dev)->gen) { + case 8: case 7: - case 6: sandybridge_write_fence_reg(dev, reg, obj); break; + case 6: case 5: case 4: i965_write_fence_reg(dev, reg, obj); break; case 3: i915_write_fence_reg(dev, reg, obj); break; case 2: i830_write_fence_reg(dev, reg, obj); break; - default: break; + default: BUG(); } + + /* And similarly be paranoid that no direct access to this region + * is reordered to before the fence is installed. + */ + if (i915_gem_object_needs_mb(obj)) + mb(); } static inline int fence_number(struct drm_i915_private *dev_priv, @@ -2656,10 +3148,11 @@ static void i915_gem_object_update_fence(struct drm_i915_gem_object *obj, fence->obj = NULL; list_del_init(&fence->lru_list); } + obj->fence_dirty = false; } static int -i915_gem_object_flush_fence(struct drm_i915_gem_object *obj) +i915_gem_object_wait_fence(struct drm_i915_gem_object *obj) { if (obj->last_fenced_seqno) { int ret = i915_wait_seqno(obj->ring, obj->last_fenced_seqno); @@ -2669,12 +3162,6 @@ i915_gem_object_flush_fence(struct drm_i915_gem_object *obj) obj->last_fenced_seqno = 0; } - /* Ensure that all CPU reads are completed before installing a fence - * and all writes before removing the fence. - */ - if (obj->base.read_domains & I915_GEM_DOMAIN_GTT) - mb(); - obj->fenced_gpu_access = false; return 0; } @@ -2683,19 +3170,23 @@ int i915_gem_object_put_fence(struct drm_i915_gem_object *obj) { struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + struct drm_i915_fence_reg *fence; int ret; - ret = i915_gem_object_flush_fence(obj); + ret = i915_gem_object_wait_fence(obj); if (ret) return ret; if (obj->fence_reg == I915_FENCE_REG_NONE) return 0; - i915_gem_object_update_fence(obj, - &dev_priv->fence_regs[obj->fence_reg], - false); + fence = &dev_priv->fence_regs[obj->fence_reg]; + + if (WARN_ON(fence->pin_count)) + return -EBUSY; + i915_gem_object_fence_lost(obj); + i915_gem_object_update_fence(obj, fence, false); return 0; } @@ -2719,7 +3210,7 @@ i915_find_fence_reg(struct drm_device *dev) } if (avail == NULL) - return NULL; + goto deadlock; /* None available, try to steal one or wait for a user to finish */ list_for_each_entry(reg, &dev_priv->mm.fence_list, lru_list) { @@ -2729,7 +3220,12 @@ i915_find_fence_reg(struct drm_device *dev) return reg; } - return NULL; +deadlock: + /* Wait for completion of pending flips which consume fences */ + if (intel_has_pending_fb_unpin(dev)) + return ERR_PTR(-EAGAIN); + + return ERR_PTR(-EDEADLK); } /** @@ -2759,7 +3255,7 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj) * will need to serialise the write to the associated fence register? */ if (obj->fence_dirty) { - ret = i915_gem_object_flush_fence(obj); + ret = i915_gem_object_wait_fence(obj); if (ret) return ret; } @@ -2774,13 +3270,13 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj) } } else if (enable) { reg = i915_find_fence_reg(dev); - if (reg == NULL) - return -EDEADLK; + if (IS_ERR(reg)) + return PTR_ERR(reg); if (reg->obj) { struct drm_i915_gem_object *old = reg->obj; - ret = i915_gem_object_flush_fence(old); + ret = i915_gem_object_wait_fence(old); if (ret) return ret; @@ -2790,7 +3286,6 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj) return 0; i915_gem_object_update_fence(obj, reg, enable); - obj->fence_dirty = false; return 0; } @@ -2803,12 +3298,12 @@ static bool i915_gem_valid_gtt_space(struct drm_device *dev, /* On non-LLC machines we have to be careful when putting differing * types of snoopable memory together to avoid the prefetcher - * crossing memory domains and dieing. + * crossing memory domains and dying. */ if (HAS_LLC(dev)) return true; - if (gtt_space == NULL) + if (!drm_mm_node_allocated(gtt_space)) return true; if (list_empty(>t_space->node_list)) @@ -2832,7 +3327,7 @@ static void i915_gem_verify_gtt(struct drm_device *dev) struct drm_i915_gem_object *obj; int err = 0; - list_for_each_entry(obj, &dev_priv->mm.gtt_list, gtt_list) { + list_for_each_entry(obj, &dev_priv->mm.gtt_list, global_list) { if (obj->gtt_space == NULL) { printk(KERN_ERR "object found on GTT list with no space reserved\n"); err++; @@ -2841,8 +3336,8 @@ static void i915_gem_verify_gtt(struct drm_device *dev) if (obj->cache_level != obj->gtt_space->color) { printk(KERN_ERR "object reserved space [%08lx, %08lx] with wrong color, cache_level=%x, color=%lx\n", - obj->gtt_space->start, - obj->gtt_space->start + obj->gtt_space->size, + i915_gem_obj_ggtt_offset(obj), + i915_gem_obj_ggtt_offset(obj) + i915_gem_obj_ggtt_size(obj), obj->cache_level, obj->gtt_space->color); err++; @@ -2853,8 +3348,8 @@ static void i915_gem_verify_gtt(struct drm_device *dev) obj->gtt_space, obj->cache_level)) { printk(KERN_ERR "invalid GTT space found at [%08lx, %08lx] - color=%x\n", - obj->gtt_space->start, - obj->gtt_space->start + obj->gtt_space->size, + i915_gem_obj_ggtt_offset(obj), + i915_gem_obj_ggtt_offset(obj) + i915_gem_obj_ggtt_size(obj), obj->cache_level); err++; continue; @@ -2868,140 +3363,142 @@ static void i915_gem_verify_gtt(struct drm_device *dev) /** * Finds free space in the GTT aperture and binds the object there. */ -static int -i915_gem_object_bind_to_gtt(struct drm_i915_gem_object *obj, - unsigned alignment, - bool map_and_fenceable, - bool nonblocking) +static struct i915_vma * +i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj, + struct i915_address_space *vm, + unsigned alignment, + uint64_t flags) { struct drm_device *dev = obj->base.dev; - drm_i915_private_t *dev_priv = dev->dev_private; - struct drm_mm_node *free_space; + struct drm_i915_private *dev_priv = dev->dev_private; u32 size, fence_size, fence_alignment, unfenced_alignment; - bool mappable, fenceable; + unsigned long start = + flags & PIN_OFFSET_BIAS ? flags & PIN_OFFSET_MASK : 0; + unsigned long end = + flags & PIN_MAPPABLE ? dev_priv->gtt.mappable_end : vm->total; + struct i915_vma *vma; int ret; - if (obj->madv != I915_MADV_WILLNEED) { - DRM_ERROR("Attempting to bind a purgeable object\n"); - return -EINVAL; - } - fence_size = i915_gem_get_gtt_size(dev, obj->base.size, obj->tiling_mode); fence_alignment = i915_gem_get_gtt_alignment(dev, obj->base.size, - obj->tiling_mode); + obj->tiling_mode, true); unfenced_alignment = - i915_gem_get_unfenced_gtt_alignment(dev, - obj->base.size, - obj->tiling_mode); + i915_gem_get_gtt_alignment(dev, + obj->base.size, + obj->tiling_mode, false); if (alignment == 0) - alignment = map_and_fenceable ? fence_alignment : + alignment = flags & PIN_MAPPABLE ? fence_alignment : unfenced_alignment; - if (map_and_fenceable && alignment & (fence_alignment - 1)) { - DRM_ERROR("Invalid object alignment requested %u\n", alignment); - return -EINVAL; + if (flags & PIN_MAPPABLE && alignment & (fence_alignment - 1)) { + DRM_DEBUG("Invalid object alignment requested %u\n", alignment); + return ERR_PTR(-EINVAL); } - size = map_and_fenceable ? fence_size : obj->base.size; + size = flags & PIN_MAPPABLE ? fence_size : obj->base.size; /* If the object is bigger than the entire aperture, reject it early * before evicting everything in a vain attempt to find space. */ - if (obj->base.size > - (map_and_fenceable ? dev_priv->mm.gtt_mappable_end : dev_priv->mm.gtt_total)) { - DRM_ERROR("Attempting to bind an object larger than the aperture\n"); - return -E2BIG; + if (obj->base.size > end) { + DRM_DEBUG("Attempting to bind an object larger than the aperture: object=%zd > %s aperture=%lu\n", + obj->base.size, + flags & PIN_MAPPABLE ? "mappable" : "total", + end); + return ERR_PTR(-E2BIG); } ret = i915_gem_object_get_pages(obj); if (ret) - return ret; + return ERR_PTR(ret); - search_free: - if (map_and_fenceable) - free_space = - drm_mm_search_free_in_range_color(&dev_priv->mm.gtt_space, - size, alignment, obj->cache_level, - 0, dev_priv->mm.gtt_mappable_end, - false); - else - free_space = drm_mm_search_free_color(&dev_priv->mm.gtt_space, - size, alignment, obj->cache_level, - false); - - if (free_space != NULL) { - if (map_and_fenceable) - obj->gtt_space = - drm_mm_get_block_range_generic(free_space, - size, alignment, obj->cache_level, - 0, dev_priv->mm.gtt_mappable_end, - false); - else - obj->gtt_space = - drm_mm_get_block_generic(free_space, - size, alignment, obj->cache_level, - false); - } - if (obj->gtt_space == NULL) { - ret = i915_gem_evict_something(dev, size, alignment, + i915_gem_object_pin_pages(obj); + + vma = i915_gem_obj_lookup_or_create_vma(obj, vm); + if (IS_ERR(vma)) + goto err_unpin; + +search_free: + ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node, + size, alignment, + obj->cache_level, + start, end, + DRM_MM_SEARCH_DEFAULT, + DRM_MM_CREATE_DEFAULT); + if (ret) { + ret = i915_gem_evict_something(dev, vm, size, alignment, obj->cache_level, - map_and_fenceable, - nonblocking); - if (ret) - return ret; + start, end, + flags); + if (ret == 0) + goto search_free; - goto search_free; + goto err_free_vma; } - if (WARN_ON(!i915_gem_valid_gtt_space(dev, - obj->gtt_space, + if (WARN_ON(!i915_gem_valid_gtt_space(dev, &vma->node, obj->cache_level))) { - drm_mm_put_block(obj->gtt_space); - obj->gtt_space = NULL; - return -EINVAL; + ret = -EINVAL; + goto err_remove_node; } - ret = i915_gem_gtt_prepare_object(obj); - if (ret) { - drm_mm_put_block(obj->gtt_space); - obj->gtt_space = NULL; - return ret; - } + if (ret) + goto err_remove_node; - if (!dev_priv->mm.aliasing_ppgtt) - i915_gem_gtt_bind_object(obj, obj->cache_level); + list_move_tail(&obj->global_list, &dev_priv->mm.bound_list); + list_add_tail(&vma->mm_list, &vm->inactive_list); - list_move_tail(&obj->gtt_list, &dev_priv->mm.bound_list); - list_add_tail(&obj->mm_list, &dev_priv->mm.inactive_list); + if (i915_is_ggtt(vm)) { + bool mappable, fenceable; - obj->gtt_offset = obj->gtt_space->start; + fenceable = (vma->node.size == fence_size && + (vma->node.start & (fence_alignment - 1)) == 0); - fenceable = - obj->gtt_space->size == fence_size && - (obj->gtt_space->start & (fence_alignment - 1)) == 0; + mappable = (vma->node.start + obj->base.size <= + dev_priv->gtt.mappable_end); + + obj->map_and_fenceable = mappable && fenceable; + } - mappable = - obj->gtt_offset + obj->base.size <= dev_priv->mm.gtt_mappable_end; + WARN_ON(flags & PIN_MAPPABLE && !obj->map_and_fenceable); - obj->map_and_fenceable = mappable && fenceable; + trace_i915_vma_bind(vma, flags); + vma->bind_vma(vma, obj->cache_level, + flags & (PIN_MAPPABLE | PIN_GLOBAL) ? GLOBAL_BIND : 0); - trace_i915_gem_object_bind(obj, map_and_fenceable); i915_gem_verify_gtt(dev); - return 0; + return vma; + +err_remove_node: + drm_mm_remove_node(&vma->node); +err_free_vma: + i915_gem_vma_destroy(vma); + vma = ERR_PTR(ret); +err_unpin: + i915_gem_object_unpin_pages(obj); + return vma; } -void -i915_gem_clflush_object(struct drm_i915_gem_object *obj) +bool +i915_gem_clflush_object(struct drm_i915_gem_object *obj, + bool force) { /* If we don't have a page list set up, then we're not pinned * to GPU, and we can ignore the cache flush because it'll happen * again at bind time. */ if (obj->pages == NULL) - return; + return false; + + /* + * Stolen memory is always coherent with the GPU as it is explicitly + * marked as wc by the system, or the system is cache-coherent. + */ + if (obj->stolen) + return false; /* If the GPU is snooping the contents of the CPU cache, * we do not need to manually clear the CPU cache lines. However, @@ -3011,12 +3508,13 @@ i915_gem_clflush_object(struct drm_i915_gem_object *obj) * snooping behaviour occurs naturally as the result of our domain * tracking. */ - if (obj->cache_level != I915_CACHE_NONE) - return; + if (!force && cpu_cache_is_coherent(obj->base.dev, obj->cache_level)) + return false; trace_i915_gem_object_clflush(obj); - drm_clflush_sg(obj->pages); + + return true; } /** Flushes the GTT write domain for the object if it's dirty. */ @@ -3048,15 +3546,17 @@ i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj) /** 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) +i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj, + bool force) { uint32_t old_write_domain; if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) return; - i915_gem_clflush_object(obj); - intel_gtt_chipset_flush(); + if (i915_gem_clflush_object(obj, force)) + i915_gem_chipset_flush(obj->base.dev); + old_write_domain = obj->base.write_domain; obj->base.write_domain = 0; @@ -3074,12 +3574,12 @@ i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj) int i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) { - drm_i915_private_t *dev_priv = obj->base.dev->dev_private; + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; uint32_t old_write_domain, old_read_domains; int ret; /* Not valid to be called on unbound objects. */ - if (obj->gtt_space == NULL) + if (!i915_gem_obj_bound_any(obj)) return -EINVAL; if (obj->base.write_domain == I915_GEM_DOMAIN_GTT) @@ -3089,7 +3589,15 @@ 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); + i915_gem_object_retire(obj); + i915_gem_object_flush_cpu_write_domain(obj, false); + + /* Serialise direct access to this object with the barriers for + * coherent writes from the GPU, by effectively invalidating the + * GTT domain upon first access. + */ + if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) + mb(); old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains; @@ -3110,8 +3618,13 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain); /* And bump the LRU for this access */ - if (i915_gem_object_is_inactive(obj)) - list_move_tail(&obj->mm_list, &dev_priv->mm.inactive_list); + if (i915_gem_object_is_inactive(obj)) { + struct i915_vma *vma = i915_gem_obj_to_ggtt(obj); + if (vma) + list_move_tail(&vma->mm_list, + &dev_priv->gtt.base.inactive_list); + + } return 0; } @@ -3120,24 +3633,26 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, enum i915_cache_level cache_level) { struct drm_device *dev = obj->base.dev; - drm_i915_private_t *dev_priv = dev->dev_private; + struct i915_vma *vma, *next; int ret; if (obj->cache_level == cache_level) return 0; - if (obj->pin_count) { + if (i915_gem_obj_is_pinned(obj)) { DRM_DEBUG("can not change the cache level of pinned objects\n"); return -EBUSY; } - if (!i915_gem_valid_gtt_space(dev, obj->gtt_space, cache_level)) { - ret = i915_gem_object_unbind(obj); - if (ret) - return ret; + list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link) { + if (!i915_gem_valid_gtt_space(dev, &vma->node, cache_level)) { + ret = i915_vma_unbind(vma); + if (ret) + return ret; + } } - if (obj->gtt_space) { + if (i915_gem_obj_bound_any(obj)) { ret = i915_gem_object_finish_gpu(obj); if (ret) return ret; @@ -3154,16 +3669,17 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, return ret; } - if (obj->has_global_gtt_mapping) - i915_gem_gtt_bind_object(obj, cache_level); - if (obj->has_aliasing_ppgtt_mapping) - i915_ppgtt_bind_object(dev_priv->mm.aliasing_ppgtt, - obj, cache_level); - - obj->gtt_space->color = cache_level; + list_for_each_entry(vma, &obj->vma_list, vma_link) + if (drm_mm_node_allocated(&vma->node)) + vma->bind_vma(vma, cache_level, + obj->has_global_gtt_mapping ? GLOBAL_BIND : 0); } - if (cache_level == I915_CACHE_NONE) { + list_for_each_entry(vma, &obj->vma_list, vma_link) + vma->node.color = cache_level; + obj->cache_level = cache_level; + + if (cpu_write_needs_clflush(obj)) { u32 old_read_domains, old_write_domain; /* If we're coming from LLC cached, then we haven't @@ -3172,8 +3688,8 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, * in obj->write_domain and have been skipping the clflushes. * Just set it to the CPU cache for now. */ + i915_gem_object_retire(obj); WARN_ON(obj->base.write_domain & ~I915_GEM_DOMAIN_CPU); - WARN_ON(obj->base.read_domains & ~I915_GEM_DOMAIN_CPU); old_read_domains = obj->base.read_domains; old_write_domain = obj->base.write_domain; @@ -3186,7 +3702,6 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, old_write_domain); } - obj->cache_level = cache_level; i915_gem_verify_gtt(dev); return 0; } @@ -3208,7 +3723,20 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, goto unlock; } - args->caching = obj->cache_level != I915_CACHE_NONE; + switch (obj->cache_level) { + case I915_CACHE_LLC: + case I915_CACHE_L3_LLC: + args->caching = I915_CACHING_CACHED; + break; + + case I915_CACHE_WT: + args->caching = I915_CACHING_DISPLAY; + break; + + default: + args->caching = I915_CACHING_NONE; + break; + } drm_gem_object_unreference(&obj->base); unlock: @@ -3231,6 +3759,9 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, case I915_CACHING_CACHED: level = I915_CACHE_LLC; break; + case I915_CACHING_DISPLAY: + level = HAS_WT(dev) ? I915_CACHE_WT : I915_CACHE_NONE; + break; default: return -EINVAL; } @@ -3253,6 +3784,31 @@ unlock: return ret; } +static bool is_pin_display(struct drm_i915_gem_object *obj) +{ + struct i915_vma *vma; + + if (list_empty(&obj->vma_list)) + return false; + + vma = i915_gem_obj_to_ggtt(obj); + if (!vma) + return false; + + /* There are 3 sources that pin objects: + * 1. The display engine (scanouts, sprites, cursors); + * 2. Reservations for execbuffer; + * 3. The user. + * + * We can ignore reservations as we hold the struct_mutex and + * are only called outside of the reservation path. The user + * can only increment pin_count once, and so if after + * subtracting the potential reference by the user, any pin_count + * remains, it must be due to another use by the display engine. + */ + return vma->pin_count - !!obj->user_pin_count; +} + /* * Prepare buffer for display plane (scanout, cursors, etc). * Can be called from an uninterruptible phase (modesetting) and allows @@ -3261,9 +3817,10 @@ unlock: int i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, u32 alignment, - struct intel_ring_buffer *pipelined) + struct intel_engine_cs *pipelined) { u32 old_read_domains, old_write_domain; + bool was_pin_display; int ret; if (pipelined != obj->ring) { @@ -3272,6 +3829,12 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, return ret; } + /* Mark the pin_display early so that we account for the + * display coherency whilst setting up the cache domains. + */ + was_pin_display = obj->pin_display; + obj->pin_display = true; + /* The display engine is not coherent with the LLC cache on gen6. As * a result, we make sure that the pinning that is about to occur is * done with uncached PTEs. This is lowest common denominator for all @@ -3281,19 +3844,20 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, * of uncaching, which would allow us to flush all the LLC-cached data * with that bit in the PTE to main memory with just one PIPE_CONTROL. */ - ret = i915_gem_object_set_cache_level(obj, I915_CACHE_NONE); + ret = i915_gem_object_set_cache_level(obj, + HAS_WT(obj->base.dev) ? I915_CACHE_WT : I915_CACHE_NONE); if (ret) - return ret; + goto err_unpin_display; /* As the user may map the buffer once pinned in the display plane * (e.g. libkms for the bootup splash), we have to ensure that we * always use map_and_fenceable for all scanout buffers. */ - ret = i915_gem_object_pin(obj, alignment, true, false); + ret = i915_gem_obj_ggtt_pin(obj, alignment, PIN_MAPPABLE); if (ret) - return ret; + goto err_unpin_display; - i915_gem_object_flush_cpu_write_domain(obj); + i915_gem_object_flush_cpu_write_domain(obj, true); old_write_domain = obj->base.write_domain; old_read_domains = obj->base.read_domains; @@ -3309,6 +3873,18 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, old_write_domain); return 0; + +err_unpin_display: + WARN_ON(was_pin_display != is_pin_display(obj)); + obj->pin_display = was_pin_display; + return ret; +} + +void +i915_gem_object_unpin_from_display_plane(struct drm_i915_gem_object *obj) +{ + i915_gem_object_ggtt_unpin(obj); + obj->pin_display = is_pin_display(obj); } int @@ -3347,6 +3923,7 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) if (ret) return ret; + i915_gem_object_retire(obj); i915_gem_object_flush_gtt_write_domain(obj); old_write_domain = obj->base.write_domain; @@ -3354,7 +3931,7 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) /* Flush the CPU cache if it's still invalid. */ if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) { - i915_gem_clflush_object(obj); + i915_gem_clflush_object(obj, false); obj->base.read_domains |= I915_GEM_DOMAIN_CPU; } @@ -3396,12 +3973,18 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) struct drm_i915_file_private *file_priv = file->driver_priv; unsigned long recent_enough = jiffies - msecs_to_jiffies(20); struct drm_i915_gem_request *request; - struct intel_ring_buffer *ring = NULL; + struct intel_engine_cs *ring = NULL; + unsigned reset_counter; u32 seqno = 0; int ret; - if (atomic_read(&dev_priv->mm.wedged)) - return -EIO; + ret = i915_gem_wait_for_error(&dev_priv->gpu_error); + if (ret) + return ret; + + ret = i915_gem_check_wedge(&dev_priv->gpu_error, false); + if (ret) + return ret; spin_lock(&file_priv->mm.lock); list_for_each_entry(request, &file_priv->mm.request_list, client_list) { @@ -3411,72 +3994,130 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) ring = request->ring; seqno = request->seqno; } + reset_counter = atomic_read(&dev_priv->gpu_error.reset_counter); spin_unlock(&file_priv->mm.lock); if (seqno == 0) return 0; - ret = __wait_seqno(ring, seqno, true, NULL); + ret = __wait_seqno(ring, seqno, reset_counter, true, NULL, NULL); if (ret == 0) queue_delayed_work(dev_priv->wq, &dev_priv->mm.retire_work, 0); return ret; } +static bool +i915_vma_misplaced(struct i915_vma *vma, uint32_t alignment, uint64_t flags) +{ + struct drm_i915_gem_object *obj = vma->obj; + + if (alignment && + vma->node.start & (alignment - 1)) + return true; + + if (flags & PIN_MAPPABLE && !obj->map_and_fenceable) + return true; + + if (flags & PIN_OFFSET_BIAS && + vma->node.start < (flags & PIN_OFFSET_MASK)) + return true; + + return false; +} + int i915_gem_object_pin(struct drm_i915_gem_object *obj, + struct i915_address_space *vm, uint32_t alignment, - bool map_and_fenceable, - bool nonblocking) + uint64_t flags) { + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + struct i915_vma *vma; int ret; - if (WARN_ON(obj->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT)) - return -EBUSY; + if (WARN_ON(vm == &dev_priv->mm.aliasing_ppgtt->base)) + return -ENODEV; - if (obj->gtt_space != NULL) { - if ((alignment && obj->gtt_offset & (alignment - 1)) || - (map_and_fenceable && !obj->map_and_fenceable)) { - WARN(obj->pin_count, + if (WARN_ON(flags & (PIN_GLOBAL | PIN_MAPPABLE) && !i915_is_ggtt(vm))) + return -EINVAL; + + vma = i915_gem_obj_to_vma(obj, vm); + if (vma) { + if (WARN_ON(vma->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT)) + return -EBUSY; + + if (i915_vma_misplaced(vma, alignment, flags)) { + WARN(vma->pin_count, "bo is already pinned with incorrect alignment:" - " offset=%x, req.alignment=%x, req.map_and_fenceable=%d," + " offset=%lx, req.alignment=%x, req.map_and_fenceable=%d," " obj->map_and_fenceable=%d\n", - obj->gtt_offset, alignment, - map_and_fenceable, + i915_gem_obj_offset(obj, vm), alignment, + !!(flags & PIN_MAPPABLE), obj->map_and_fenceable); - ret = i915_gem_object_unbind(obj); + ret = i915_vma_unbind(vma); if (ret) return ret; + + vma = NULL; } } - if (obj->gtt_space == NULL) { - ret = i915_gem_object_bind_to_gtt(obj, alignment, - map_and_fenceable, - nonblocking); - if (ret) - return ret; + if (vma == NULL || !drm_mm_node_allocated(&vma->node)) { + vma = i915_gem_object_bind_to_vm(obj, vm, alignment, flags); + if (IS_ERR(vma)) + return PTR_ERR(vma); } - if (!obj->has_global_gtt_mapping && map_and_fenceable) - i915_gem_gtt_bind_object(obj, obj->cache_level); + if (flags & PIN_GLOBAL && !obj->has_global_gtt_mapping) + vma->bind_vma(vma, obj->cache_level, GLOBAL_BIND); - obj->pin_count++; - obj->pin_mappable |= map_and_fenceable; + vma->pin_count++; + if (flags & PIN_MAPPABLE) + obj->pin_mappable |= true; return 0; } void -i915_gem_object_unpin(struct drm_i915_gem_object *obj) +i915_gem_object_ggtt_unpin(struct drm_i915_gem_object *obj) { - BUG_ON(obj->pin_count == 0); - BUG_ON(obj->gtt_space == NULL); + struct i915_vma *vma = i915_gem_obj_to_ggtt(obj); + + BUG_ON(!vma); + BUG_ON(vma->pin_count == 0); + BUG_ON(!i915_gem_obj_ggtt_bound(obj)); - if (--obj->pin_count == 0) + if (--vma->pin_count == 0) obj->pin_mappable = false; } +bool +i915_gem_object_pin_fence(struct drm_i915_gem_object *obj) +{ + if (obj->fence_reg != I915_FENCE_REG_NONE) { + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + struct i915_vma *ggtt_vma = i915_gem_obj_to_ggtt(obj); + + WARN_ON(!ggtt_vma || + dev_priv->fence_regs[obj->fence_reg].pin_count > + ggtt_vma->pin_count); + dev_priv->fence_regs[obj->fence_reg].pin_count++; + return true; + } else + return false; +} + +void +i915_gem_object_unpin_fence(struct drm_i915_gem_object *obj) +{ + if (obj->fence_reg != I915_FENCE_REG_NONE) { + struct drm_i915_private *dev_priv = obj->base.dev->dev_private; + WARN_ON(dev_priv->fence_regs[obj->fence_reg].pin_count <= 0); + dev_priv->fence_regs[obj->fence_reg].pin_count--; + } +} + int i915_gem_pin_ioctl(struct drm_device *dev, void *data, struct drm_file *file) @@ -3485,6 +4126,9 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data, struct drm_i915_gem_object *obj; int ret; + if (INTEL_INFO(dev)->gen >= 6) + return -ENODEV; + ret = i915_mutex_lock_interruptible(dev); if (ret) return ret; @@ -3496,31 +4140,33 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data, } if (obj->madv != I915_MADV_WILLNEED) { - DRM_ERROR("Attempting to pin a purgeable buffer\n"); - ret = -EINVAL; + DRM_DEBUG("Attempting to pin a purgeable buffer\n"); + ret = -EFAULT; goto out; } if (obj->pin_filp != NULL && obj->pin_filp != file) { - DRM_ERROR("Already pinned in i915_gem_pin_ioctl(): %d\n", + DRM_DEBUG("Already pinned in i915_gem_pin_ioctl(): %d\n", args->handle); ret = -EINVAL; goto out; } - obj->user_pin_count++; - obj->pin_filp = file; - if (obj->user_pin_count == 1) { - ret = i915_gem_object_pin(obj, args->alignment, true, false); + if (obj->user_pin_count == ULONG_MAX) { + ret = -EBUSY; + goto out; + } + + if (obj->user_pin_count == 0) { + ret = i915_gem_obj_ggtt_pin(obj, args->alignment, PIN_MAPPABLE); if (ret) goto out; } - /* XXX - flush the CPU caches for pinned objects - * as the X server doesn't manage domains yet - */ - i915_gem_object_flush_cpu_write_domain(obj); - args->offset = obj->gtt_offset; + obj->user_pin_count++; + obj->pin_filp = file; + + args->offset = i915_gem_obj_ggtt_offset(obj); out: drm_gem_object_unreference(&obj->base); unlock: @@ -3547,7 +4193,7 @@ i915_gem_unpin_ioctl(struct drm_device *dev, void *data, } if (obj->pin_filp != file) { - DRM_ERROR("Not pinned by caller in i915_gem_pin_ioctl(): %d\n", + DRM_DEBUG("Not pinned by caller in i915_gem_pin_ioctl(): %d\n", args->handle); ret = -EINVAL; goto out; @@ -3555,7 +4201,7 @@ i915_gem_unpin_ioctl(struct drm_device *dev, void *data, obj->user_pin_count--; if (obj->user_pin_count == 0) { obj->pin_filp = NULL; - i915_gem_object_unpin(obj); + i915_gem_object_ggtt_unpin(obj); } out: @@ -3635,7 +4281,7 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data, goto unlock; } - if (obj->pin_count) { + if (i915_gem_obj_is_pinned(obj)) { ret = -EINVAL; goto out; } @@ -3659,10 +4305,10 @@ unlock: void i915_gem_object_init(struct drm_i915_gem_object *obj, const struct drm_i915_gem_object_ops *ops) { - INIT_LIST_HEAD(&obj->mm_list); - INIT_LIST_HEAD(&obj->gtt_list); + INIT_LIST_HEAD(&obj->global_list); INIT_LIST_HEAD(&obj->ring_list); - INIT_LIST_HEAD(&obj->exec_list); + INIT_LIST_HEAD(&obj->obj_exec_link); + INIT_LIST_HEAD(&obj->vma_list); obj->ops = ops; @@ -3684,14 +4330,14 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, { struct drm_i915_gem_object *obj; struct address_space *mapping; - u32 mask; + gfp_t mask; - obj = kzalloc(sizeof(*obj), GFP_KERNEL); + obj = i915_gem_object_alloc(dev); if (obj == NULL) return NULL; if (drm_gem_object_init(dev, &obj->base, size) != 0) { - kfree(obj); + i915_gem_object_free(obj); return NULL; } @@ -3702,7 +4348,7 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, mask |= __GFP_DMA32; } - mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping; + mapping = file_inode(obj->base.filp)->i_mapping; mapping_set_gfp_mask(mapping, mask); i915_gem_object_init(obj, &i915_gem_object_ops); @@ -3727,134 +4373,206 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, } else obj->cache_level = I915_CACHE_NONE; + trace_i915_gem_object_create(obj); + return obj; } -int i915_gem_init_object(struct drm_gem_object *obj) +static bool discard_backing_storage(struct drm_i915_gem_object *obj) { - BUG(); + /* If we are the last user of the backing storage (be it shmemfs + * pages or stolen etc), we know that the pages are going to be + * immediately released. In this case, we can then skip copying + * back the contents from the GPU. + */ - return 0; + if (obj->madv != I915_MADV_WILLNEED) + return false; + + if (obj->base.filp == NULL) + return true; + + /* At first glance, this looks racy, but then again so would be + * userspace racing mmap against close. However, the first external + * reference to the filp can only be obtained through the + * i915_gem_mmap_ioctl() which safeguards us against the user + * acquiring such a reference whilst we are in the middle of + * freeing the object. + */ + return atomic_long_read(&obj->base.filp->f_count) == 1; } void i915_gem_free_object(struct drm_gem_object *gem_obj) { struct drm_i915_gem_object *obj = to_intel_bo(gem_obj); struct drm_device *dev = obj->base.dev; - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; + struct i915_vma *vma, *next; + + intel_runtime_pm_get(dev_priv); trace_i915_gem_object_destroy(obj); - if (obj->phys_obj) - i915_gem_detach_phys_object(dev, obj); + list_for_each_entry_safe(vma, next, &obj->vma_list, vma_link) { + int ret; - obj->pin_count = 0; - if (WARN_ON(i915_gem_object_unbind(obj) == -ERESTARTSYS)) { - bool was_interruptible; + vma->pin_count = 0; + ret = i915_vma_unbind(vma); + if (WARN_ON(ret == -ERESTARTSYS)) { + bool was_interruptible; - was_interruptible = dev_priv->mm.interruptible; - dev_priv->mm.interruptible = false; + was_interruptible = dev_priv->mm.interruptible; + dev_priv->mm.interruptible = false; - WARN_ON(i915_gem_object_unbind(obj)); + WARN_ON(i915_vma_unbind(vma)); - dev_priv->mm.interruptible = was_interruptible; + dev_priv->mm.interruptible = was_interruptible; + } } - obj->pages_pin_count = 0; + i915_gem_object_detach_phys(obj); + + /* Stolen objects don't hold a ref, but do hold pin count. Fix that up + * before progressing. */ + if (obj->stolen) + i915_gem_object_unpin_pages(obj); + + if (WARN_ON(obj->pages_pin_count)) + obj->pages_pin_count = 0; + if (discard_backing_storage(obj)) + obj->madv = I915_MADV_DONTNEED; i915_gem_object_put_pages(obj); i915_gem_object_free_mmap_offset(obj); + i915_gem_object_release_stolen(obj); BUG_ON(obj->pages); if (obj->base.import_attach) drm_prime_gem_destroy(&obj->base, NULL); + if (obj->ops->release) + obj->ops->release(obj); + drm_gem_object_release(&obj->base); i915_gem_info_remove_obj(dev_priv, obj->base.size); kfree(obj->bit_17); - kfree(obj); + i915_gem_object_free(obj); + + intel_runtime_pm_put(dev_priv); +} + +struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj, + struct i915_address_space *vm) +{ + struct i915_vma *vma; + list_for_each_entry(vma, &obj->vma_list, vma_link) + if (vma->vm == vm) + return vma; + + return NULL; +} + +void i915_gem_vma_destroy(struct i915_vma *vma) +{ + WARN_ON(vma->node.allocated); + + /* Keep the vma as a placeholder in the execbuffer reservation lists */ + if (!list_empty(&vma->exec_list)) + return; + + list_del(&vma->vma_link); + + kfree(vma); +} + +static void +i915_gem_stop_ringbuffers(struct drm_device *dev) +{ + struct drm_i915_private *dev_priv = dev->dev_private; + struct intel_engine_cs *ring; + int i; + + for_each_ring(ring, dev_priv, i) + intel_stop_ring_buffer(ring); } int -i915_gem_idle(struct drm_device *dev) +i915_gem_suspend(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; - int ret; + struct drm_i915_private *dev_priv = dev->dev_private; + int ret = 0; mutex_lock(&dev->struct_mutex); - - if (dev_priv->mm.suspended) { - mutex_unlock(&dev->struct_mutex); - return 0; - } + if (dev_priv->ums.mm_suspended) + goto err; ret = i915_gpu_idle(dev); - if (ret) { - mutex_unlock(&dev->struct_mutex); - return ret; - } + if (ret) + goto err; + i915_gem_retire_requests(dev); /* Under UMS, be paranoid and evict. */ if (!drm_core_check_feature(dev, DRIVER_MODESET)) i915_gem_evict_everything(dev); - i915_gem_reset_fences(dev); + i915_kernel_lost_context(dev); + i915_gem_stop_ringbuffers(dev); /* Hack! Don't let anybody do execbuf while we don't control the chip. * We need to replace this with a semaphore, or something. - * And not confound mm.suspended! + * And not confound ums.mm_suspended! */ - dev_priv->mm.suspended = 1; - del_timer_sync(&dev_priv->hangcheck_timer); - - i915_kernel_lost_context(dev); - i915_gem_cleanup_ringbuffer(dev); - + dev_priv->ums.mm_suspended = !drm_core_check_feature(dev, + DRIVER_MODESET); mutex_unlock(&dev->struct_mutex); - /* Cancel the retire work handler, which should be idle now. */ + del_timer_sync(&dev_priv->gpu_error.hangcheck_timer); cancel_delayed_work_sync(&dev_priv->mm.retire_work); + cancel_delayed_work_sync(&dev_priv->mm.idle_work); return 0; + +err: + mutex_unlock(&dev->struct_mutex); + return ret; } -void i915_gem_l3_remap(struct drm_device *dev) +int i915_gem_l3_remap(struct intel_engine_cs *ring, int slice) { - drm_i915_private_t *dev_priv = dev->dev_private; - u32 misccpctl; - int i; - - if (!IS_IVYBRIDGE(dev)) - return; + struct drm_device *dev = ring->dev; + struct drm_i915_private *dev_priv = dev->dev_private; + u32 reg_base = GEN7_L3LOG_BASE + (slice * 0x200); + u32 *remap_info = dev_priv->l3_parity.remap_info[slice]; + int i, ret; - if (!dev_priv->mm.l3_remap_info) - return; + if (!HAS_L3_DPF(dev) || !remap_info) + return 0; - misccpctl = I915_READ(GEN7_MISCCPCTL); - I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE); - POSTING_READ(GEN7_MISCCPCTL); + ret = intel_ring_begin(ring, GEN7_L3LOG_SIZE / 4 * 3); + if (ret) + return ret; + /* + * Note: We do not worry about the concurrent register cacheline hang + * here because no other code should access these registers other than + * at initialization time. + */ for (i = 0; i < GEN7_L3LOG_SIZE; i += 4) { - u32 remap = I915_READ(GEN7_L3LOG_BASE + i); - if (remap && remap != dev_priv->mm.l3_remap_info[i/4]) - DRM_DEBUG("0x%x was already programmed to %x\n", - GEN7_L3LOG_BASE + i, remap); - if (remap && !dev_priv->mm.l3_remap_info[i/4]) - DRM_DEBUG_DRIVER("Clearing remapped register\n"); - I915_WRITE(GEN7_L3LOG_BASE + i, dev_priv->mm.l3_remap_info[i/4]); + intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); + intel_ring_emit(ring, reg_base + i); + intel_ring_emit(ring, remap_info[i/4]); } - /* Make sure all the writes land before disabling dop clock gating */ - POSTING_READ(GEN7_L3LOG_BASE); + intel_ring_advance(ring); - I915_WRITE(GEN7_MISCCPCTL, misccpctl); + return ret; } void i915_gem_init_swizzling(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; if (INTEL_INFO(dev)->gen < 5 || dev_priv->mm.bit_6_swizzle_x == I915_BIT_6_SWIZZLE_NONE) @@ -3869,70 +4587,12 @@ void i915_gem_init_swizzling(struct drm_device *dev) I915_WRITE(TILECTL, I915_READ(TILECTL) | TILECTL_SWZCTL); if (IS_GEN6(dev)) I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_SNB)); - else + else if (IS_GEN7(dev)) I915_WRITE(ARB_MODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_IVB)); -} - -void i915_gem_init_ppgtt(struct drm_device *dev) -{ - drm_i915_private_t *dev_priv = dev->dev_private; - uint32_t pd_offset; - struct intel_ring_buffer *ring; - struct i915_hw_ppgtt *ppgtt = dev_priv->mm.aliasing_ppgtt; - uint32_t __iomem *pd_addr; - uint32_t pd_entry; - int i; - - if (!dev_priv->mm.aliasing_ppgtt) - return; - - - pd_addr = dev_priv->mm.gtt->gtt + ppgtt->pd_offset/sizeof(uint32_t); - for (i = 0; i < ppgtt->num_pd_entries; i++) { - dma_addr_t pt_addr; - - if (dev_priv->mm.gtt->needs_dmar) - pt_addr = ppgtt->pt_dma_addr[i]; - else - pt_addr = page_to_phys(ppgtt->pt_pages[i]); - - pd_entry = GEN6_PDE_ADDR_ENCODE(pt_addr); - pd_entry |= GEN6_PDE_VALID; - - writel(pd_entry, pd_addr + i); - } - readl(pd_addr); - - pd_offset = ppgtt->pd_offset; - pd_offset /= 64; /* in cachelines, */ - pd_offset <<= 16; - - if (INTEL_INFO(dev)->gen == 6) { - uint32_t ecochk, gab_ctl, ecobits; - - ecobits = I915_READ(GAC_ECO_BITS); - I915_WRITE(GAC_ECO_BITS, ecobits | ECOBITS_PPGTT_CACHE64B); - - gab_ctl = I915_READ(GAB_CTL); - I915_WRITE(GAB_CTL, gab_ctl | GAB_CTL_CONT_AFTER_PAGEFAULT); - - ecochk = I915_READ(GAM_ECOCHK); - I915_WRITE(GAM_ECOCHK, ecochk | ECOCHK_SNB_BIT | - ECOCHK_PPGTT_CACHE64B); - I915_WRITE(GFX_MODE, _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); - } else if (INTEL_INFO(dev)->gen >= 7) { - I915_WRITE(GAM_ECOCHK, ECOCHK_PPGTT_CACHE64B); - /* GFX_MODE is per-ring on gen7+ */ - } - - for_each_ring(ring, dev_priv, i) { - if (INTEL_INFO(dev)->gen >= 7) - I915_WRITE(RING_MODE_GEN7(ring), - _MASKED_BIT_ENABLE(GFX_PPGTT_ENABLE)); - - I915_WRITE(RING_PP_DIR_DCLV(ring), PP_DIR_DCLV_2G); - I915_WRITE(RING_PP_DIR_BASE(ring), pd_offset); - } + else if (IS_GEN8(dev)) + I915_WRITE(GAMTARBMODE, _MASKED_BIT_ENABLE(ARB_MODE_SWIZZLE_BDW)); + else + BUG(); } static bool @@ -3951,22 +4611,11 @@ intel_enable_blt(struct drm_device *dev) return true; } -int -i915_gem_init_hw(struct drm_device *dev) +static int i915_gem_init_rings(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; int ret; - if (!intel_enable_gtt()) - return -EIO; - - if (IS_HASWELL(dev) && (I915_READ(0x120010) == 1)) - I915_WRITE(0x9008, I915_READ(0x9008) | 0xf0000); - - i915_gem_l3_remap(dev); - - i915_gem_init_swizzling(dev); - ret = intel_init_render_ring_buffer(dev); if (ret) return ret; @@ -3983,94 +4632,138 @@ i915_gem_init_hw(struct drm_device *dev) goto cleanup_bsd_ring; } - dev_priv->next_seqno = 1; + if (HAS_VEBOX(dev)) { + ret = intel_init_vebox_ring_buffer(dev); + if (ret) + goto cleanup_blt_ring; + } - /* - * XXX: There was some w/a described somewhere suggesting loading - * contexts before PPGTT. - */ - i915_gem_context_init(dev); - i915_gem_init_ppgtt(dev); + if (HAS_BSD2(dev)) { + ret = intel_init_bsd2_ring_buffer(dev); + if (ret) + goto cleanup_vebox_ring; + } + + ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000)); + if (ret) + goto cleanup_bsd2_ring; return 0; +cleanup_bsd2_ring: + intel_cleanup_ring_buffer(&dev_priv->ring[VCS2]); +cleanup_vebox_ring: + intel_cleanup_ring_buffer(&dev_priv->ring[VECS]); +cleanup_blt_ring: + intel_cleanup_ring_buffer(&dev_priv->ring[BCS]); cleanup_bsd_ring: intel_cleanup_ring_buffer(&dev_priv->ring[VCS]); cleanup_render_ring: intel_cleanup_ring_buffer(&dev_priv->ring[RCS]); + return ret; } -static bool -intel_enable_ppgtt(struct drm_device *dev) +int +i915_gem_init_hw(struct drm_device *dev) { - if (i915_enable_ppgtt >= 0) - return i915_enable_ppgtt; + struct drm_i915_private *dev_priv = dev->dev_private; + int ret, i; -#ifdef CONFIG_INTEL_IOMMU - /* Disable ppgtt on SNB if VT-d is on. */ - if (INTEL_INFO(dev)->gen == 6 && intel_iommu_gfx_mapped) - return false; -#endif + if (INTEL_INFO(dev)->gen < 6 && !intel_enable_gtt()) + return -EIO; - return true; + if (dev_priv->ellc_size) + I915_WRITE(HSW_IDICR, I915_READ(HSW_IDICR) | IDIHASHMSK(0xf)); + + if (IS_HASWELL(dev)) + I915_WRITE(MI_PREDICATE_RESULT_2, IS_HSW_GT3(dev) ? + LOWER_SLICE_ENABLED : LOWER_SLICE_DISABLED); + + if (HAS_PCH_NOP(dev)) { + if (IS_IVYBRIDGE(dev)) { + u32 temp = I915_READ(GEN7_MSG_CTL); + temp &= ~(WAIT_FOR_PCH_FLR_ACK | WAIT_FOR_PCH_RESET_ACK); + I915_WRITE(GEN7_MSG_CTL, temp); + } else if (INTEL_INFO(dev)->gen >= 7) { + u32 temp = I915_READ(HSW_NDE_RSTWRN_OPT); + temp &= ~RESET_PCH_HANDSHAKE_ENABLE; + I915_WRITE(HSW_NDE_RSTWRN_OPT, temp); + } + } + + i915_gem_init_swizzling(dev); + + ret = i915_gem_init_rings(dev); + if (ret) + return ret; + + for (i = 0; i < NUM_L3_SLICES(dev); i++) + i915_gem_l3_remap(&dev_priv->ring[RCS], i); + + /* + * XXX: Contexts should only be initialized once. Doing a switch to the + * default context switch however is something we'd like to do after + * reset or thaw (the latter may not actually be necessary for HW, but + * goes with our code better). Context switching requires rings (for + * the do_switch), but before enabling PPGTT. So don't move this. + */ + ret = i915_gem_context_enable(dev_priv); + if (ret && ret != -EIO) { + DRM_ERROR("Context enable failed %d\n", ret); + i915_gem_cleanup_ringbuffer(dev); + } + + return ret; } int i915_gem_init(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; - unsigned long gtt_size, mappable_size; int ret; - gtt_size = dev_priv->mm.gtt->gtt_total_entries << PAGE_SHIFT; - mappable_size = dev_priv->mm.gtt->gtt_mappable_entries << PAGE_SHIFT; - mutex_lock(&dev->struct_mutex); - if (intel_enable_ppgtt(dev) && HAS_ALIASING_PPGTT(dev)) { - /* PPGTT pdes are stolen from global gtt ptes, so shrink the - * aperture accordingly when using aliasing ppgtt. */ - gtt_size -= I915_PPGTT_PD_ENTRIES*PAGE_SIZE; - i915_gem_init_global_gtt(dev, 0, mappable_size, gtt_size); - - ret = i915_gem_init_aliasing_ppgtt(dev); - if (ret) { - mutex_unlock(&dev->struct_mutex); - return ret; - } - } else { - /* Let GEM Manage all of the aperture. - * - * However, leave one page at the end still bound to the scratch - * page. There are a number of places where the hardware - * apparently prefetches past the end of the object, and we've - * seen multiple hangs with the GPU head pointer stuck in a - * batchbuffer bound at the last page of the aperture. One page - * should be enough to keep any prefetching inside of the - * aperture. - */ - i915_gem_init_global_gtt(dev, 0, mappable_size, - gtt_size); + if (IS_VALLEYVIEW(dev)) { + /* VLVA0 (potential hack), BIOS isn't actually waking us */ + I915_WRITE(VLV_GTLC_WAKE_CTRL, VLV_GTLC_ALLOWWAKEREQ); + if (wait_for((I915_READ(VLV_GTLC_PW_STATUS) & + VLV_GTLC_ALLOWWAKEACK), 10)) + DRM_DEBUG_DRIVER("allow wake ack timed out\n"); } - ret = i915_gem_init_hw(dev); - mutex_unlock(&dev->struct_mutex); + i915_gem_init_userptr(dev); + i915_gem_init_global_gtt(dev); + + ret = i915_gem_context_init(dev); if (ret) { - i915_gem_cleanup_aliasing_ppgtt(dev); + mutex_unlock(&dev->struct_mutex); return ret; } + ret = i915_gem_init_hw(dev); + if (ret == -EIO) { + /* Allow ring initialisation to fail by marking the GPU as + * wedged. But we only want to do this where the GPU is angry, + * for all other failure, such as an allocation failure, bail. + */ + DRM_ERROR("Failed to initialize GPU, declaring it wedged\n"); + atomic_set_mask(I915_WEDGED, &dev_priv->gpu_error.reset_counter); + ret = 0; + } + mutex_unlock(&dev->struct_mutex); + /* Allow hardware batchbuffers unless told otherwise, but not for KMS. */ if (!drm_core_check_feature(dev, DRIVER_MODESET)) dev_priv->dri1.allow_batchbuffer = 1; - return 0; + return ret; } void i915_gem_cleanup_ringbuffer(struct drm_device *dev) { - drm_i915_private_t *dev_priv = dev->dev_private; - struct intel_ring_buffer *ring; + struct drm_i915_private *dev_priv = dev->dev_private; + struct intel_engine_cs *ring; int i; for_each_ring(ring, dev_priv, i) @@ -4081,19 +4774,19 @@ int i915_gem_entervt_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { - drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_private *dev_priv = dev->dev_private; int ret; if (drm_core_check_feature(dev, DRIVER_MODESET)) return 0; - if (atomic_read(&dev_priv->mm.wedged)) { + if (i915_reset_in_progress(&dev_priv->gpu_error)) { DRM_ERROR("Reenabling wedged hardware, good luck\n"); - atomic_set(&dev_priv->mm.wedged, 0); + atomic_set(&dev_priv->gpu_error.reset_counter, 0); } mutex_lock(&dev->struct_mutex); - dev_priv->mm.suspended = 0; + dev_priv->ums.mm_suspended = 0; ret = i915_gem_init_hw(dev); if (ret != 0) { @@ -4101,19 +4794,18 @@ i915_gem_entervt_ioctl(struct drm_device *dev, void *data, return ret; } - BUG_ON(!list_empty(&dev_priv->mm.active_list)); - mutex_unlock(&dev->struct_mutex); + BUG_ON(!list_empty(&dev_priv->gtt.base.active_list)); - ret = drm_irq_install(dev); + ret = drm_irq_install(dev, dev->pdev->irq); if (ret) goto cleanup_ringbuffer; + mutex_unlock(&dev->struct_mutex); return 0; cleanup_ringbuffer: - mutex_lock(&dev->struct_mutex); i915_gem_cleanup_ringbuffer(dev); - dev_priv->mm.suspended = 1; + dev_priv->ums.mm_suspended = 1; mutex_unlock(&dev->struct_mutex); return ret; @@ -4126,8 +4818,11 @@ i915_gem_leavevt_ioctl(struct drm_device *dev, void *data, if (drm_core_check_feature(dev, DRIVER_MODESET)) return 0; + mutex_lock(&dev->struct_mutex); drm_irq_uninstall(dev); - return i915_gem_idle(dev); + mutex_unlock(&dev->struct_mutex); + + return i915_gem_suspend(dev); } void @@ -4138,26 +4833,46 @@ i915_gem_lastclose(struct drm_device *dev) if (drm_core_check_feature(dev, DRIVER_MODESET)) return; - ret = i915_gem_idle(dev); + ret = i915_gem_suspend(dev); if (ret) DRM_ERROR("failed to idle hardware: %d\n", ret); } static void -init_ring_lists(struct intel_ring_buffer *ring) +init_ring_lists(struct intel_engine_cs *ring) { INIT_LIST_HEAD(&ring->active_list); INIT_LIST_HEAD(&ring->request_list); } +void i915_init_vm(struct drm_i915_private *dev_priv, + struct i915_address_space *vm) +{ + if (!i915_is_ggtt(vm)) + drm_mm_init(&vm->mm, vm->start, vm->total); + vm->dev = dev_priv->dev; + INIT_LIST_HEAD(&vm->active_list); + INIT_LIST_HEAD(&vm->inactive_list); + INIT_LIST_HEAD(&vm->global_link); + list_add_tail(&vm->global_link, &dev_priv->vm_list); +} + void i915_gem_load(struct drm_device *dev) { + struct drm_i915_private *dev_priv = dev->dev_private; int i; - drm_i915_private_t *dev_priv = dev->dev_private; - INIT_LIST_HEAD(&dev_priv->mm.active_list); - INIT_LIST_HEAD(&dev_priv->mm.inactive_list); + dev_priv->slab = + kmem_cache_create("i915_gem_object", + sizeof(struct drm_i915_gem_object), 0, + SLAB_HWCACHE_ALIGN, + NULL); + + INIT_LIST_HEAD(&dev_priv->vm_list); + i915_init_vm(dev_priv, &dev_priv->gtt.base); + + INIT_LIST_HEAD(&dev_priv->context_list); INIT_LIST_HEAD(&dev_priv->mm.unbound_list); INIT_LIST_HEAD(&dev_priv->mm.bound_list); INIT_LIST_HEAD(&dev_priv->mm.fence_list); @@ -4167,10 +4882,12 @@ i915_gem_load(struct drm_device *dev) INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list); INIT_DELAYED_WORK(&dev_priv->mm.retire_work, i915_gem_retire_work_handler); - init_completion(&dev_priv->error_completion); + INIT_DELAYED_WORK(&dev_priv->mm.idle_work, + i915_gem_idle_work_handler); + init_waitqueue_head(&dev_priv->gpu_error.reset_queue); /* On GEN3 we really need to make sure the ARB C3 LP bit is set */ - if (IS_GEN3(dev)) { + if (!drm_core_check_feature(dev, DRIVER_MODESET) && IS_GEN3(dev)) { I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE)); } @@ -4181,258 +4898,323 @@ i915_gem_load(struct drm_device *dev) if (!drm_core_check_feature(dev, DRIVER_MODESET)) dev_priv->fence_reg_start = 3; - if (INTEL_INFO(dev)->gen >= 4 || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev)) + if (INTEL_INFO(dev)->gen >= 7 && !IS_VALLEYVIEW(dev)) + dev_priv->num_fence_regs = 32; + else if (INTEL_INFO(dev)->gen >= 4 || IS_I945G(dev) || IS_I945GM(dev) || IS_G33(dev)) dev_priv->num_fence_regs = 16; else dev_priv->num_fence_regs = 8; /* Initialize fence registers to zero */ - i915_gem_reset_fences(dev); + INIT_LIST_HEAD(&dev_priv->mm.fence_list); + i915_gem_restore_fences(dev); i915_gem_detect_bit_6_swizzle(dev); init_waitqueue_head(&dev_priv->pending_flip_queue); dev_priv->mm.interruptible = true; - dev_priv->mm.inactive_shrinker.shrink = i915_gem_inactive_shrink; - dev_priv->mm.inactive_shrinker.seeks = DEFAULT_SEEKS; - register_shrinker(&dev_priv->mm.inactive_shrinker); + dev_priv->mm.shrinker.scan_objects = i915_gem_shrinker_scan; + dev_priv->mm.shrinker.count_objects = i915_gem_shrinker_count; + dev_priv->mm.shrinker.seeks = DEFAULT_SEEKS; + register_shrinker(&dev_priv->mm.shrinker); + + dev_priv->mm.oom_notifier.notifier_call = i915_gem_shrinker_oom; + register_oom_notifier(&dev_priv->mm.oom_notifier); } -/* - * Create a physically contiguous memory object for this object - * e.g. for cursor + overlay regs - */ -static int i915_gem_init_phys_object(struct drm_device *dev, - int id, int size, int align) +void i915_gem_release(struct drm_device *dev, struct drm_file *file) { - drm_i915_private_t *dev_priv = dev->dev_private; - struct drm_i915_gem_phys_object *phys_obj; + struct drm_i915_file_private *file_priv = file->driver_priv; + + cancel_delayed_work_sync(&file_priv->mm.idle_work); + + /* Clean up our request list when the client is going away, so that + * later retire_requests won't dereference our soon-to-be-gone + * file_priv. + */ + spin_lock(&file_priv->mm.lock); + while (!list_empty(&file_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + + request = list_first_entry(&file_priv->mm.request_list, + struct drm_i915_gem_request, + client_list); + list_del(&request->client_list); + request->file_priv = NULL; + } + spin_unlock(&file_priv->mm.lock); +} + +static void +i915_gem_file_idle_work_handler(struct work_struct *work) +{ + struct drm_i915_file_private *file_priv = + container_of(work, typeof(*file_priv), mm.idle_work.work); + + atomic_set(&file_priv->rps_wait_boost, false); +} + +int i915_gem_open(struct drm_device *dev, struct drm_file *file) +{ + struct drm_i915_file_private *file_priv; int ret; - if (dev_priv->mm.phys_objs[id - 1] || !size) - return 0; + DRM_DEBUG_DRIVER("\n"); - phys_obj = kzalloc(sizeof(struct drm_i915_gem_phys_object), GFP_KERNEL); - if (!phys_obj) + file_priv = kzalloc(sizeof(*file_priv), GFP_KERNEL); + if (!file_priv) return -ENOMEM; - phys_obj->id = id; + file->driver_priv = file_priv; + file_priv->dev_priv = dev->dev_private; + file_priv->file = file; - phys_obj->handle = drm_pci_alloc(dev, size, align); - if (!phys_obj->handle) { - ret = -ENOMEM; - goto kfree_obj; - } -#ifdef CONFIG_X86 - set_memory_wc((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE); -#endif + spin_lock_init(&file_priv->mm.lock); + INIT_LIST_HEAD(&file_priv->mm.request_list); + INIT_DELAYED_WORK(&file_priv->mm.idle_work, + i915_gem_file_idle_work_handler); - dev_priv->mm.phys_objs[id - 1] = phys_obj; + ret = i915_gem_context_open(dev, file); + if (ret) + kfree(file_priv); - return 0; -kfree_obj: - kfree(phys_obj); return ret; } -static void i915_gem_free_phys_object(struct drm_device *dev, int id) +static bool mutex_is_locked_by(struct mutex *mutex, struct task_struct *task) { - drm_i915_private_t *dev_priv = dev->dev_private; - struct drm_i915_gem_phys_object *phys_obj; - - if (!dev_priv->mm.phys_objs[id - 1]) - return; - - phys_obj = dev_priv->mm.phys_objs[id - 1]; - if (phys_obj->cur_obj) { - i915_gem_detach_phys_object(dev, phys_obj->cur_obj); - } + if (!mutex_is_locked(mutex)) + return false; -#ifdef CONFIG_X86 - set_memory_wb((unsigned long)phys_obj->handle->vaddr, phys_obj->handle->size / PAGE_SIZE); +#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_MUTEXES) + return mutex->owner == task; +#else + /* Since UP may be pre-empted, we cannot assume that we own the lock */ + return false; #endif - drm_pci_free(dev, phys_obj->handle); - kfree(phys_obj); - dev_priv->mm.phys_objs[id - 1] = NULL; } -void i915_gem_free_all_phys_object(struct drm_device *dev) +static bool i915_gem_shrinker_lock(struct drm_device *dev, bool *unlock) { - int i; + if (!mutex_trylock(&dev->struct_mutex)) { + if (!mutex_is_locked_by(&dev->struct_mutex, current)) + return false; - for (i = I915_GEM_PHYS_CURSOR_0; i <= I915_MAX_PHYS_OBJECT; i++) - i915_gem_free_phys_object(dev, i); + if (to_i915(dev)->mm.shrinker_no_lock_stealing) + return false; + + *unlock = false; + } else + *unlock = true; + + return true; } -void i915_gem_detach_phys_object(struct drm_device *dev, - struct drm_i915_gem_object *obj) +static int num_vma_bound(struct drm_i915_gem_object *obj) { - struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping; - char *vaddr; - int i; - int page_count; + struct i915_vma *vma; + int count = 0; - if (!obj->phys_obj) - return; - vaddr = obj->phys_obj->handle->vaddr; + list_for_each_entry(vma, &obj->vma_list, vma_link) + if (drm_mm_node_allocated(&vma->node)) + count++; - page_count = obj->base.size / PAGE_SIZE; - for (i = 0; i < page_count; i++) { - struct page *page = shmem_read_mapping_page(mapping, i); - if (!IS_ERR(page)) { - char *dst = kmap_atomic(page); - memcpy(dst, vaddr + i*PAGE_SIZE, PAGE_SIZE); - kunmap_atomic(dst); + return count; +} - drm_clflush_pages(&page, 1); +static unsigned long +i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc) +{ + struct drm_i915_private *dev_priv = + container_of(shrinker, struct drm_i915_private, mm.shrinker); + struct drm_device *dev = dev_priv->dev; + struct drm_i915_gem_object *obj; + unsigned long count; + bool unlock; - set_page_dirty(page); - mark_page_accessed(page); - page_cache_release(page); - } + if (!i915_gem_shrinker_lock(dev, &unlock)) + return 0; + + count = 0; + list_for_each_entry(obj, &dev_priv->mm.unbound_list, global_list) + if (obj->pages_pin_count == 0) + count += obj->base.size >> PAGE_SHIFT; + + list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) { + if (!i915_gem_obj_is_pinned(obj) && + obj->pages_pin_count == num_vma_bound(obj)) + count += obj->base.size >> PAGE_SHIFT; } - intel_gtt_chipset_flush(); - obj->phys_obj->cur_obj = NULL; - obj->phys_obj = NULL; + if (unlock) + mutex_unlock(&dev->struct_mutex); + + return count; } -int -i915_gem_attach_phys_object(struct drm_device *dev, - struct drm_i915_gem_object *obj, - int id, - int align) +/* All the new VM stuff */ +unsigned long i915_gem_obj_offset(struct drm_i915_gem_object *o, + struct i915_address_space *vm) { - struct address_space *mapping = obj->base.filp->f_path.dentry->d_inode->i_mapping; - drm_i915_private_t *dev_priv = dev->dev_private; - int ret = 0; - int page_count; - int i; + struct drm_i915_private *dev_priv = o->base.dev->dev_private; + struct i915_vma *vma; - if (id > I915_MAX_PHYS_OBJECT) - return -EINVAL; + if (!dev_priv->mm.aliasing_ppgtt || + vm == &dev_priv->mm.aliasing_ppgtt->base) + vm = &dev_priv->gtt.base; - if (obj->phys_obj) { - if (obj->phys_obj->id == id) - return 0; - i915_gem_detach_phys_object(dev, obj); - } + BUG_ON(list_empty(&o->vma_list)); + list_for_each_entry(vma, &o->vma_list, vma_link) { + if (vma->vm == vm) + return vma->node.start; - /* create a new object */ - if (!dev_priv->mm.phys_objs[id - 1]) { - ret = i915_gem_init_phys_object(dev, id, - obj->base.size, align); - if (ret) { - DRM_ERROR("failed to init phys object %d size: %zu\n", - id, obj->base.size); - return ret; - } } + return -1; +} - /* bind to the object */ - obj->phys_obj = dev_priv->mm.phys_objs[id - 1]; - obj->phys_obj->cur_obj = obj; - - page_count = obj->base.size / PAGE_SIZE; +bool i915_gem_obj_bound(struct drm_i915_gem_object *o, + struct i915_address_space *vm) +{ + struct i915_vma *vma; - for (i = 0; i < page_count; i++) { - struct page *page; - char *dst, *src; + list_for_each_entry(vma, &o->vma_list, vma_link) + if (vma->vm == vm && drm_mm_node_allocated(&vma->node)) + return true; - page = shmem_read_mapping_page(mapping, i); - if (IS_ERR(page)) - return PTR_ERR(page); + return false; +} - src = kmap_atomic(page); - dst = obj->phys_obj->handle->vaddr + (i * PAGE_SIZE); - memcpy(dst, src, PAGE_SIZE); - kunmap_atomic(src); +bool i915_gem_obj_bound_any(struct drm_i915_gem_object *o) +{ + struct i915_vma *vma; - mark_page_accessed(page); - page_cache_release(page); - } + list_for_each_entry(vma, &o->vma_list, vma_link) + if (drm_mm_node_allocated(&vma->node)) + return true; - return 0; + return false; } -static int -i915_gem_phys_pwrite(struct drm_device *dev, - struct drm_i915_gem_object *obj, - struct drm_i915_gem_pwrite *args, - struct drm_file *file_priv) +unsigned long i915_gem_obj_size(struct drm_i915_gem_object *o, + struct i915_address_space *vm) { - void *vaddr = obj->phys_obj->handle->vaddr + args->offset; - char __user *user_data = (char __user *) (uintptr_t) args->data_ptr; + struct drm_i915_private *dev_priv = o->base.dev->dev_private; + struct i915_vma *vma; - if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) { - unsigned long unwritten; + if (!dev_priv->mm.aliasing_ppgtt || + vm == &dev_priv->mm.aliasing_ppgtt->base) + vm = &dev_priv->gtt.base; - /* The physical object once assigned is fixed for the lifetime - * of the obj, so we can safely drop the lock and continue - * to access vaddr. - */ - mutex_unlock(&dev->struct_mutex); - unwritten = copy_from_user(vaddr, user_data, args->size); - mutex_lock(&dev->struct_mutex); - if (unwritten) - return -EFAULT; - } + BUG_ON(list_empty(&o->vma_list)); + + list_for_each_entry(vma, &o->vma_list, vma_link) + if (vma->vm == vm) + return vma->node.size; - intel_gtt_chipset_flush(); return 0; } -void i915_gem_release(struct drm_device *dev, struct drm_file *file) +static unsigned long +i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) { - struct drm_i915_file_private *file_priv = file->driver_priv; - - /* Clean up our request list when the client is going away, so that - * later retire_requests won't dereference our soon-to-be-gone - * file_priv. - */ - spin_lock(&file_priv->mm.lock); - while (!list_empty(&file_priv->mm.request_list)) { - struct drm_i915_gem_request *request; + struct drm_i915_private *dev_priv = + container_of(shrinker, struct drm_i915_private, mm.shrinker); + struct drm_device *dev = dev_priv->dev; + unsigned long freed; + bool unlock; + + if (!i915_gem_shrinker_lock(dev, &unlock)) + return SHRINK_STOP; + + freed = i915_gem_purge(dev_priv, sc->nr_to_scan); + if (freed < sc->nr_to_scan) + freed += __i915_gem_shrink(dev_priv, + sc->nr_to_scan - freed, + false); + if (unlock) + mutex_unlock(&dev->struct_mutex); - request = list_first_entry(&file_priv->mm.request_list, - struct drm_i915_gem_request, - client_list); - list_del(&request->client_list); - request->file_priv = NULL; - } - spin_unlock(&file_priv->mm.lock); + return freed; } static int -i915_gem_inactive_shrink(struct shrinker *shrinker, struct shrink_control *sc) +i915_gem_shrinker_oom(struct notifier_block *nb, unsigned long event, void *ptr) { struct drm_i915_private *dev_priv = - container_of(shrinker, - struct drm_i915_private, - mm.inactive_shrinker); + container_of(nb, struct drm_i915_private, mm.oom_notifier); struct drm_device *dev = dev_priv->dev; struct drm_i915_gem_object *obj; - int nr_to_scan = sc->nr_to_scan; - int cnt; + unsigned long timeout = msecs_to_jiffies(5000) + 1; + unsigned long pinned, bound, unbound, freed; + bool was_interruptible; + bool unlock; - if (!mutex_trylock(&dev->struct_mutex)) - return 0; + while (!i915_gem_shrinker_lock(dev, &unlock) && --timeout) + schedule_timeout_killable(1); + if (timeout == 0) { + pr_err("Unable to purge GPU memory due lock contention.\n"); + return NOTIFY_DONE; + } - if (nr_to_scan) { - nr_to_scan -= i915_gem_purge(dev_priv, nr_to_scan); - if (nr_to_scan > 0) - i915_gem_shrink_all(dev_priv); + was_interruptible = dev_priv->mm.interruptible; + dev_priv->mm.interruptible = false; + + freed = i915_gem_shrink_all(dev_priv); + + dev_priv->mm.interruptible = was_interruptible; + + /* Because we may be allocating inside our own driver, we cannot + * assert that there are no objects with pinned pages that are not + * being pointed to by hardware. + */ + unbound = bound = pinned = 0; + list_for_each_entry(obj, &dev_priv->mm.unbound_list, global_list) { + if (!obj->base.filp) /* not backed by a freeable object */ + continue; + + if (obj->pages_pin_count) + pinned += obj->base.size; + else + unbound += obj->base.size; } + list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) { + if (!obj->base.filp) + continue; - cnt = 0; - list_for_each_entry(obj, &dev_priv->mm.unbound_list, gtt_list) - if (obj->pages_pin_count == 0) - cnt += obj->base.size >> PAGE_SHIFT; - list_for_each_entry(obj, &dev_priv->mm.bound_list, gtt_list) - if (obj->pin_count == 0 && obj->pages_pin_count == 0) - cnt += obj->base.size >> PAGE_SHIFT; + if (obj->pages_pin_count) + pinned += obj->base.size; + else + bound += obj->base.size; + } - mutex_unlock(&dev->struct_mutex); - return cnt; + if (unlock) + mutex_unlock(&dev->struct_mutex); + + pr_info("Purging GPU memory, %lu bytes freed, %lu bytes still pinned.\n", + freed, pinned); + if (unbound || bound) + pr_err("%lu and %lu bytes still available in the " + "bound and unbound GPU page lists.\n", + bound, unbound); + + *(unsigned long *)ptr += freed; + return NOTIFY_DONE; +} + +struct i915_vma *i915_gem_obj_to_ggtt(struct drm_i915_gem_object *obj) +{ + struct i915_vma *vma; + + /* This WARN has probably outlived its usefulness (callers already + * WARN if they don't find the GGTT vma they expect). When removing, + * remember to remove the pre-check in is_pin_display() as well */ + if (WARN_ON(list_empty(&obj->vma_list))) + return NULL; + + vma = list_first_entry(&obj->vma_list, typeof(*vma), vma_link); + if (vma->vm != obj_to_ggtt(obj)) + return NULL; + + return vma; } |