diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 4076 |
1 files changed, 221 insertions, 3855 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index ad01c92aaf74..190ad54fb072 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -26,7 +26,6 @@ */ #include <drm/drm_vma_manager.h> -#include <drm/drm_pci.h> #include <drm/i915_drm.h> #include <linux/dma-fence-array.h> #include <linux/kthread.h> @@ -39,32 +38,27 @@ #include <linux/dma-buf.h> #include <linux/mman.h> +#include "display/intel_display.h" +#include "display/intel_frontbuffer.h" + +#include "gem/i915_gem_clflush.h" +#include "gem/i915_gem_context.h" +#include "gem/i915_gem_ioctls.h" +#include "gem/i915_gem_pm.h" +#include "gem/i915_gemfs.h" +#include "gt/intel_engine_pm.h" +#include "gt/intel_gt_pm.h" +#include "gt/intel_mocs.h" +#include "gt/intel_reset.h" +#include "gt/intel_workarounds.h" + #include "i915_drv.h" -#include "i915_gem_clflush.h" -#include "i915_gemfs.h" -#include "i915_globals.h" -#include "i915_reset.h" +#include "i915_scatterlist.h" #include "i915_trace.h" #include "i915_vgpu.h" #include "intel_drv.h" -#include "intel_frontbuffer.h" -#include "intel_mocs.h" #include "intel_pm.h" -#include "intel_workarounds.h" - -static void i915_gem_flush_free_objects(struct drm_i915_private *i915); - -static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj) -{ - if (obj->cache_dirty) - return false; - - if (!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE)) - return true; - - return obj->pin_global; /* currently in use by HW, keep flushed */ -} static int insert_mappable_node(struct i915_ggtt *ggtt, @@ -83,124 +77,6 @@ remove_mappable_node(struct drm_mm_node *node) drm_mm_remove_node(node); } -/* some bookkeeping */ -static void i915_gem_info_add_obj(struct drm_i915_private *dev_priv, - u64 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, - u64 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_park(struct drm_i915_private *i915) -{ - intel_wakeref_t wakeref; - - GEM_TRACE("\n"); - - lockdep_assert_held(&i915->drm.struct_mutex); - GEM_BUG_ON(i915->gt.active_requests); - GEM_BUG_ON(!list_empty(&i915->gt.active_rings)); - - if (!i915->gt.awake) - return; - - /* - * Be paranoid and flush a concurrent interrupt to make sure - * we don't reactivate any irq tasklets after parking. - * - * FIXME: Note that even though we have waited for execlists to be idle, - * there may still be an in-flight interrupt even though the CSB - * is now empty. synchronize_irq() makes sure that a residual interrupt - * is completed before we continue, but it doesn't prevent the HW from - * raising a spurious interrupt later. To complete the shield we should - * coordinate disabling the CS irq with flushing the interrupts. - */ - synchronize_irq(i915->drm.irq); - - intel_engines_park(i915); - i915_timelines_park(i915); - - i915_pmu_gt_parked(i915); - i915_vma_parked(i915); - - wakeref = fetch_and_zero(&i915->gt.awake); - GEM_BUG_ON(!wakeref); - - if (INTEL_GEN(i915) >= 6) - gen6_rps_idle(i915); - - intel_display_power_put(i915, POWER_DOMAIN_GT_IRQ, wakeref); - - i915_globals_park(); -} - -void i915_gem_park(struct drm_i915_private *i915) -{ - GEM_TRACE("\n"); - - lockdep_assert_held(&i915->drm.struct_mutex); - GEM_BUG_ON(i915->gt.active_requests); - - if (!i915->gt.awake) - return; - - /* Defer the actual call to __i915_gem_park() to prevent ping-pongs */ - mod_delayed_work(i915->wq, &i915->gt.idle_work, msecs_to_jiffies(100)); -} - -void i915_gem_unpark(struct drm_i915_private *i915) -{ - GEM_TRACE("\n"); - - lockdep_assert_held(&i915->drm.struct_mutex); - GEM_BUG_ON(!i915->gt.active_requests); - assert_rpm_wakelock_held(i915); - - if (i915->gt.awake) - return; - - /* - * It seems that the DMC likes to transition between the DC states a lot - * when there are no connected displays (no active power domains) during - * command submission. - * - * This activity has negative impact on the performance of the chip with - * huge latencies observed in the interrupt handler and elsewhere. - * - * Work around it by grabbing a GT IRQ power domain whilst there is any - * GT activity, preventing any DC state transitions. - */ - i915->gt.awake = intel_display_power_get(i915, POWER_DOMAIN_GT_IRQ); - GEM_BUG_ON(!i915->gt.awake); - - i915_globals_unpark(); - - intel_enable_gt_powersave(i915); - i915_update_gfx_val(i915); - if (INTEL_GEN(i915) >= 6) - gen6_rps_busy(i915); - i915_pmu_gt_unparked(i915); - - intel_engines_unpark(i915); - - i915_queue_hangcheck(i915); - - queue_delayed_work(i915->wq, - &i915->gt.retire_work, - round_jiffies_up_relative(HZ)); -} - int i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, struct drm_file *file) @@ -225,178 +101,14 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, return 0; } -static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj) -{ - struct address_space *mapping = obj->base.filp->f_mapping; - drm_dma_handle_t *phys; - struct sg_table *st; - struct scatterlist *sg; - char *vaddr; - int i; - int err; - - if (WARN_ON(i915_gem_object_needs_bit17_swizzle(obj))) - return -EINVAL; - - /* Always aligning to the object size, allows a single allocation - * to handle all possible callers, and given typical object sizes, - * the alignment of the buddy allocation will naturally match. - */ - phys = drm_pci_alloc(obj->base.dev, - roundup_pow_of_two(obj->base.size), - roundup_pow_of_two(obj->base.size)); - if (!phys) - return -ENOMEM; - - vaddr = phys->vaddr; - 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)) { - err = PTR_ERR(page); - goto err_phys; - } - - src = kmap_atomic(page); - memcpy(vaddr, src, PAGE_SIZE); - drm_clflush_virt_range(vaddr, PAGE_SIZE); - kunmap_atomic(src); - - put_page(page); - vaddr += PAGE_SIZE; - } - - i915_gem_chipset_flush(to_i915(obj->base.dev)); - - st = kmalloc(sizeof(*st), GFP_KERNEL); - if (!st) { - err = -ENOMEM; - goto err_phys; - } - - if (sg_alloc_table(st, 1, GFP_KERNEL)) { - kfree(st); - err = -ENOMEM; - goto err_phys; - } - - sg = st->sgl; - sg->offset = 0; - sg->length = obj->base.size; - - sg_dma_address(sg) = phys->busaddr; - sg_dma_len(sg) = obj->base.size; - - obj->phys_handle = phys; - - __i915_gem_object_set_pages(obj, st, sg->length); - - return 0; - -err_phys: - drm_pci_free(obj->base.dev, phys); - - return err; -} - -static void __start_cpu_write(struct drm_i915_gem_object *obj) -{ - obj->read_domains = I915_GEM_DOMAIN_CPU; - obj->write_domain = I915_GEM_DOMAIN_CPU; - if (cpu_write_needs_clflush(obj)) - obj->cache_dirty = true; -} - -void -__i915_gem_object_release_shmem(struct drm_i915_gem_object *obj, - struct sg_table *pages, - bool needs_clflush) -{ - GEM_BUG_ON(obj->mm.madv == __I915_MADV_PURGED); - - if (obj->mm.madv == I915_MADV_DONTNEED) - obj->mm.dirty = false; - - if (needs_clflush && - (obj->read_domains & I915_GEM_DOMAIN_CPU) == 0 && - !(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_READ)) - drm_clflush_sg(pages); - - __start_cpu_write(obj); -} - -static void -i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj, - struct sg_table *pages) -{ - __i915_gem_object_release_shmem(obj, pages, false); - - if (obj->mm.dirty) { - struct address_space *mapping = obj->base.filp->f_mapping; - char *vaddr = obj->phys_handle->vaddr; - int i; - - for (i = 0; i < obj->base.size / PAGE_SIZE; i++) { - struct page *page; - char *dst; - - page = shmem_read_mapping_page(mapping, i); - if (IS_ERR(page)) - continue; - - dst = kmap_atomic(page); - drm_clflush_virt_range(vaddr, PAGE_SIZE); - memcpy(dst, vaddr, PAGE_SIZE); - kunmap_atomic(dst); - - set_page_dirty(page); - if (obj->mm.madv == I915_MADV_WILLNEED) - mark_page_accessed(page); - put_page(page); - vaddr += PAGE_SIZE; - } - obj->mm.dirty = false; - } - - sg_free_table(pages); - kfree(pages); - - drm_pci_free(obj->base.dev, obj->phys_handle); -} - -static void -i915_gem_object_release_phys(struct drm_i915_gem_object *obj) -{ - i915_gem_object_unpin_pages(obj); -} - -static const struct drm_i915_gem_object_ops i915_gem_phys_ops = { - .get_pages = i915_gem_object_get_pages_phys, - .put_pages = i915_gem_object_put_pages_phys, - .release = i915_gem_object_release_phys, -}; - -static const struct drm_i915_gem_object_ops i915_gem_object_ops; - int i915_gem_object_unbind(struct drm_i915_gem_object *obj) { struct i915_vma *vma; LIST_HEAD(still_in_list); - int ret; + int ret = 0; lockdep_assert_held(&obj->base.dev->struct_mutex); - /* Closed vma are removed from the obj->vma_list - but they may - * still have an active binding on the object. To remove those we - * must wait for all rendering to complete to the object (as unbinding - * must anyway), and retire the requests. - */ - ret = i915_gem_object_set_to_cpu_domain(obj, false); - if (ret) - return ret; - spin_lock(&obj->vma.lock); while (!ret && (vma = list_first_entry_or_null(&obj->vma.list, struct i915_vma, @@ -414,190 +126,6 @@ int i915_gem_object_unbind(struct drm_i915_gem_object *obj) return ret; } -static long -i915_gem_object_wait_fence(struct dma_fence *fence, - unsigned int flags, - long timeout) -{ - struct i915_request *rq; - - BUILD_BUG_ON(I915_WAIT_INTERRUPTIBLE != 0x1); - - if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags)) - return timeout; - - if (!dma_fence_is_i915(fence)) - return dma_fence_wait_timeout(fence, - flags & I915_WAIT_INTERRUPTIBLE, - timeout); - - rq = to_request(fence); - if (i915_request_completed(rq)) - goto out; - - timeout = i915_request_wait(rq, flags, timeout); - -out: - if (flags & I915_WAIT_LOCKED && i915_request_completed(rq)) - i915_request_retire_upto(rq); - - return timeout; -} - -static long -i915_gem_object_wait_reservation(struct reservation_object *resv, - unsigned int flags, - long timeout) -{ - unsigned int seq = __read_seqcount_begin(&resv->seq); - struct dma_fence *excl; - bool prune_fences = false; - - if (flags & I915_WAIT_ALL) { - struct dma_fence **shared; - unsigned int count, i; - int ret; - - ret = reservation_object_get_fences_rcu(resv, - &excl, &count, &shared); - if (ret) - return ret; - - for (i = 0; i < count; i++) { - timeout = i915_gem_object_wait_fence(shared[i], - flags, timeout); - if (timeout < 0) - break; - - dma_fence_put(shared[i]); - } - - for (; i < count; i++) - dma_fence_put(shared[i]); - kfree(shared); - - /* - * If both shared fences and an exclusive fence exist, - * then by construction the shared fences must be later - * than the exclusive fence. If we successfully wait for - * all the shared fences, we know that the exclusive fence - * must all be signaled. If all the shared fences are - * signaled, we can prune the array and recover the - * floating references on the fences/requests. - */ - prune_fences = count && timeout >= 0; - } else { - excl = reservation_object_get_excl_rcu(resv); - } - - if (excl && timeout >= 0) - timeout = i915_gem_object_wait_fence(excl, flags, timeout); - - dma_fence_put(excl); - - /* - * Opportunistically prune the fences iff we know they have *all* been - * signaled and that the reservation object has not been changed (i.e. - * no new fences have been added). - */ - if (prune_fences && !__read_seqcount_retry(&resv->seq, seq)) { - if (reservation_object_trylock(resv)) { - if (!__read_seqcount_retry(&resv->seq, seq)) - reservation_object_add_excl_fence(resv, NULL); - reservation_object_unlock(resv); - } - } - - return timeout; -} - -static void __fence_set_priority(struct dma_fence *fence, - const struct i915_sched_attr *attr) -{ - struct i915_request *rq; - struct intel_engine_cs *engine; - - if (dma_fence_is_signaled(fence) || !dma_fence_is_i915(fence)) - return; - - rq = to_request(fence); - engine = rq->engine; - - local_bh_disable(); - rcu_read_lock(); /* RCU serialisation for set-wedged protection */ - if (engine->schedule) - engine->schedule(rq, attr); - rcu_read_unlock(); - local_bh_enable(); /* kick the tasklets if queues were reprioritised */ -} - -static void fence_set_priority(struct dma_fence *fence, - const struct i915_sched_attr *attr) -{ - /* Recurse once into a fence-array */ - if (dma_fence_is_array(fence)) { - struct dma_fence_array *array = to_dma_fence_array(fence); - int i; - - for (i = 0; i < array->num_fences; i++) - __fence_set_priority(array->fences[i], attr); - } else { - __fence_set_priority(fence, attr); - } -} - -int -i915_gem_object_wait_priority(struct drm_i915_gem_object *obj, - unsigned int flags, - const struct i915_sched_attr *attr) -{ - struct dma_fence *excl; - - if (flags & I915_WAIT_ALL) { - struct dma_fence **shared; - unsigned int count, i; - int ret; - - ret = reservation_object_get_fences_rcu(obj->resv, - &excl, &count, &shared); - if (ret) - return ret; - - for (i = 0; i < count; i++) { - fence_set_priority(shared[i], attr); - dma_fence_put(shared[i]); - } - - kfree(shared); - } else { - excl = reservation_object_get_excl_rcu(obj->resv); - } - - if (excl) { - fence_set_priority(excl, attr); - dma_fence_put(excl); - } - return 0; -} - -/** - * Waits for rendering to the object to be completed - * @obj: i915 gem object - * @flags: how to wait (under a lock, for all rendering or just for writes etc) - * @timeout: how long to wait - */ -int -i915_gem_object_wait(struct drm_i915_gem_object *obj, - unsigned int flags, - long timeout) -{ - might_sleep(); - GEM_BUG_ON(timeout < 0); - - timeout = i915_gem_object_wait_reservation(obj->resv, flags, timeout); - return timeout < 0 ? timeout : 0; -} - static int i915_gem_phys_pwrite(struct drm_i915_gem_object *obj, struct drm_i915_gem_pwrite *args, @@ -636,7 +164,7 @@ i915_gem_create(struct drm_file *file, return -EINVAL; /* Allocate the new object */ - obj = i915_gem_object_create(dev_priv, size); + obj = i915_gem_object_create_shmem(dev_priv, size); if (IS_ERR(obj)) return PTR_ERR(obj); @@ -656,19 +184,36 @@ i915_gem_dumb_create(struct drm_file *file, struct drm_device *dev, struct drm_mode_create_dumb *args) { + int cpp = DIV_ROUND_UP(args->bpp, 8); + u32 format; + + switch (cpp) { + case 1: + format = DRM_FORMAT_C8; + break; + case 2: + format = DRM_FORMAT_RGB565; + break; + case 4: + format = DRM_FORMAT_XRGB8888; + break; + default: + return -EINVAL; + } + /* have to work out size/pitch and return them */ - args->pitch = ALIGN(args->width * DIV_ROUND_UP(args->bpp, 8), 64); + args->pitch = ALIGN(args->width * cpp, 64); + + /* align stride to page size so that we can remap */ + if (args->pitch > intel_plane_fb_max_stride(to_i915(dev), format, + DRM_FORMAT_MOD_LINEAR)) + args->pitch = ALIGN(args->pitch, 4096); + args->size = args->pitch * args->height; return i915_gem_create(file, to_i915(dev), &args->size, &args->handle); } -static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj) -{ - return !(obj->cache_level == I915_CACHE_NONE || - obj->cache_level == I915_CACHE_WT); -} - /** * Creates a new mm object and returns a handle to it. * @dev: drm device pointer @@ -688,13 +233,6 @@ i915_gem_create_ioctl(struct drm_device *dev, void *data, &args->size, &args->handle); } -static inline enum fb_op_origin -fb_write_origin(struct drm_i915_gem_object *obj, unsigned int domain) -{ - return (domain == I915_GEM_DOMAIN_GTT ? - obj->frontbuffer_ggtt_origin : ORIGIN_CPU); -} - void i915_gem_flush_ggtt_writes(struct drm_i915_private *dev_priv) { intel_wakeref_t wakeref; @@ -725,171 +263,14 @@ void i915_gem_flush_ggtt_writes(struct drm_i915_private *dev_priv) i915_gem_chipset_flush(dev_priv); - with_intel_runtime_pm(dev_priv, wakeref) { - spin_lock_irq(&dev_priv->uncore.lock); + with_intel_runtime_pm(&dev_priv->runtime_pm, wakeref) { + struct intel_uncore *uncore = &dev_priv->uncore; - POSTING_READ_FW(RING_HEAD(RENDER_RING_BASE)); - - spin_unlock_irq(&dev_priv->uncore.lock); - } -} - -static void -flush_write_domain(struct drm_i915_gem_object *obj, unsigned int flush_domains) -{ - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); - struct i915_vma *vma; - - if (!(obj->write_domain & flush_domains)) - return; - - switch (obj->write_domain) { - case I915_GEM_DOMAIN_GTT: - i915_gem_flush_ggtt_writes(dev_priv); - - intel_fb_obj_flush(obj, - fb_write_origin(obj, I915_GEM_DOMAIN_GTT)); - - for_each_ggtt_vma(vma, obj) { - if (vma->iomap) - continue; - - i915_vma_unset_ggtt_write(vma); - } - break; - - case I915_GEM_DOMAIN_WC: - wmb(); - break; - - case I915_GEM_DOMAIN_CPU: - i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC); - break; - - case I915_GEM_DOMAIN_RENDER: - if (gpu_write_needs_clflush(obj)) - obj->cache_dirty = true; - break; - } - - obj->write_domain = 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, - unsigned int *needs_clflush) -{ - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - *needs_clflush = 0; - if (!i915_gem_object_has_struct_page(obj)) - return -ENODEV; - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED, - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - ret = i915_gem_object_pin_pages(obj); - if (ret) - return ret; - - if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_READ || - !static_cpu_has(X86_FEATURE_CLFLUSH)) { - ret = i915_gem_object_set_to_cpu_domain(obj, false); - if (ret) - goto err_unpin; - else - goto out; - } - - flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); - - /* If we're not in the cpu read domain, set ourself into the gtt - * read domain and manually flush cachelines (if required). This - * optimizes for the case when the gpu will dirty the data - * anyway again before the next pread happens. - */ - if (!obj->cache_dirty && - !(obj->read_domains & I915_GEM_DOMAIN_CPU)) - *needs_clflush = CLFLUSH_BEFORE; - -out: - /* return with the pages pinned */ - return 0; - -err_unpin: - i915_gem_object_unpin_pages(obj); - return ret; -} - -int i915_gem_obj_prepare_shmem_write(struct drm_i915_gem_object *obj, - unsigned int *needs_clflush) -{ - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - *needs_clflush = 0; - if (!i915_gem_object_has_struct_page(obj)) - return -ENODEV; - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED | - I915_WAIT_ALL, - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - ret = i915_gem_object_pin_pages(obj); - if (ret) - return ret; - - if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE || - !static_cpu_has(X86_FEATURE_CLFLUSH)) { - ret = i915_gem_object_set_to_cpu_domain(obj, true); - if (ret) - goto err_unpin; - else - goto out; + spin_lock_irq(&uncore->lock); + intel_uncore_posting_read_fw(uncore, + RING_HEAD(RENDER_RING_BASE)); + spin_unlock_irq(&uncore->lock); } - - flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); - - /* If we're not in the cpu write domain, set ourself into the - * gtt write domain and manually flush cachelines (as required). - * This optimizes for the case when the gpu will use the data - * right away and we therefore have to clflush anyway. - */ - if (!obj->cache_dirty) { - *needs_clflush |= CLFLUSH_AFTER; - - /* - * Same trick applies to invalidate partially written - * cachelines read before writing. - */ - if (!(obj->read_domains & I915_GEM_DOMAIN_CPU)) - *needs_clflush |= CLFLUSH_BEFORE; - } - -out: - intel_fb_obj_invalidate(obj, ORIGIN_CPU); - obj->mm.dirty = true; - /* return with the pages pinned */ - return 0; - -err_unpin: - i915_gem_object_unpin_pages(obj); - return ret; } static int @@ -915,20 +296,21 @@ static int i915_gem_shmem_pread(struct drm_i915_gem_object *obj, struct drm_i915_gem_pread *args) { - char __user *user_data; - u64 remain; unsigned int needs_clflush; unsigned int idx, offset; + struct dma_fence *fence; + char __user *user_data; + u64 remain; int ret; - ret = mutex_lock_interruptible(&obj->base.dev->struct_mutex); + ret = i915_gem_object_prepare_read(obj, &needs_clflush); if (ret) return ret; - ret = i915_gem_obj_prepare_shmem_read(obj, &needs_clflush); - mutex_unlock(&obj->base.dev->struct_mutex); - if (ret) - return ret; + fence = i915_gem_object_lock_fence(obj); + i915_gem_object_finish_access(obj); + if (!fence) + return -ENOMEM; remain = args->size; user_data = u64_to_user_ptr(args->data_ptr); @@ -947,7 +329,7 @@ i915_gem_shmem_pread(struct drm_i915_gem_object *obj, offset = 0; } - i915_gem_obj_finish_shmem_access(obj); + i915_gem_object_unlock_fence(obj, fence); return ret; } @@ -983,8 +365,9 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, struct i915_ggtt *ggtt = &i915->ggtt; intel_wakeref_t wakeref; struct drm_mm_node node; - struct i915_vma *vma; + struct dma_fence *fence; void __user *user_data; + struct i915_vma *vma; u64 remain, offset; int ret; @@ -992,7 +375,7 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, if (ret) return ret; - wakeref = intel_runtime_pm_get(i915); + wakeref = intel_runtime_pm_get(&i915->runtime_pm); vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE | PIN_NONFAULT | @@ -1013,11 +396,24 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, GEM_BUG_ON(!node.allocated); } - ret = i915_gem_object_set_to_gtt_domain(obj, false); + mutex_unlock(&i915->drm.struct_mutex); + + ret = i915_gem_object_lock_interruptible(obj); if (ret) goto out_unpin; - mutex_unlock(&i915->drm.struct_mutex); + ret = i915_gem_object_set_to_gtt_domain(obj, false); + if (ret) { + i915_gem_object_unlock(obj); + goto out_unpin; + } + + fence = i915_gem_object_lock_fence(obj); + i915_gem_object_unlock(obj); + if (!fence) { + ret = -ENOMEM; + goto out_unpin; + } user_data = u64_to_user_ptr(args->data_ptr); remain = args->size; @@ -1055,8 +451,9 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj, offset += page_length; } - mutex_lock(&i915->drm.struct_mutex); + i915_gem_object_unlock_fence(obj, fence); out_unpin: + mutex_lock(&i915->drm.struct_mutex); if (node.allocated) { wmb(); ggtt->vm.clear_range(&ggtt->vm, node.start, node.size); @@ -1065,7 +462,7 @@ out_unpin: i915_vma_unpin(vma); } out_unlock: - intel_runtime_pm_put(i915, wakeref); + intel_runtime_pm_put(&i915->runtime_pm, wakeref); mutex_unlock(&i915->drm.struct_mutex); return ret; @@ -1165,8 +562,10 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj, { struct drm_i915_private *i915 = to_i915(obj->base.dev); struct i915_ggtt *ggtt = &i915->ggtt; + struct intel_runtime_pm *rpm = &i915->runtime_pm; intel_wakeref_t wakeref; struct drm_mm_node node; + struct dma_fence *fence; struct i915_vma *vma; u64 remain, offset; void __user *user_data; @@ -1184,14 +583,14 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj, * This easily dwarfs any performance advantage from * using the cache bypass of indirect GGTT access. */ - wakeref = intel_runtime_pm_get_if_in_use(i915); + wakeref = intel_runtime_pm_get_if_in_use(rpm); if (!wakeref) { ret = -EFAULT; goto out_unlock; } } else { /* No backing pages, no fallback, we must force GGTT access */ - wakeref = intel_runtime_pm_get(i915); + wakeref = intel_runtime_pm_get(rpm); } vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, @@ -1214,11 +613,24 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj, GEM_BUG_ON(!node.allocated); } - ret = i915_gem_object_set_to_gtt_domain(obj, true); + mutex_unlock(&i915->drm.struct_mutex); + + ret = i915_gem_object_lock_interruptible(obj); if (ret) goto out_unpin; - mutex_unlock(&i915->drm.struct_mutex); + ret = i915_gem_object_set_to_gtt_domain(obj, true); + if (ret) { + i915_gem_object_unlock(obj); + goto out_unpin; + } + + fence = i915_gem_object_lock_fence(obj); + i915_gem_object_unlock(obj); + if (!fence) { + ret = -ENOMEM; + goto out_unpin; + } intel_fb_obj_invalidate(obj, ORIGIN_CPU); @@ -1263,8 +675,9 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj, } intel_fb_obj_flush(obj, ORIGIN_CPU); - mutex_lock(&i915->drm.struct_mutex); + i915_gem_object_unlock_fence(obj, fence); out_unpin: + mutex_lock(&i915->drm.struct_mutex); if (node.allocated) { wmb(); ggtt->vm.clear_range(&ggtt->vm, node.start, node.size); @@ -1273,7 +686,7 @@ out_unpin: i915_vma_unpin(vma); } out_rpm: - intel_runtime_pm_put(i915, wakeref); + intel_runtime_pm_put(rpm, wakeref); out_unlock: mutex_unlock(&i915->drm.struct_mutex); return ret; @@ -1310,22 +723,22 @@ static int i915_gem_shmem_pwrite(struct drm_i915_gem_object *obj, const struct drm_i915_gem_pwrite *args) { - struct drm_i915_private *i915 = to_i915(obj->base.dev); - void __user *user_data; - u64 remain; unsigned int partial_cacheline_write; unsigned int needs_clflush; unsigned int offset, idx; + struct dma_fence *fence; + void __user *user_data; + u64 remain; int ret; - ret = mutex_lock_interruptible(&i915->drm.struct_mutex); + ret = i915_gem_object_prepare_write(obj, &needs_clflush); if (ret) return ret; - ret = i915_gem_obj_prepare_shmem_write(obj, &needs_clflush); - mutex_unlock(&i915->drm.struct_mutex); - if (ret) - return ret; + fence = i915_gem_object_lock_fence(obj); + i915_gem_object_finish_access(obj); + if (!fence) + return -ENOMEM; /* If we don't overwrite a cacheline completely we need to be * careful to have up-to-date data by first clflushing. Don't @@ -1354,7 +767,8 @@ i915_gem_shmem_pwrite(struct drm_i915_gem_object *obj, } intel_fb_obj_flush(obj, ORIGIN_CPU); - i915_gem_obj_finish_shmem_access(obj); + i915_gem_object_unlock_fence(obj, fence); + return ret; } @@ -1443,143 +857,6 @@ err: return ret; } -static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) -{ - struct drm_i915_private *i915 = to_i915(obj->base.dev); - struct list_head *list; - struct i915_vma *vma; - - GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj)); - - mutex_lock(&i915->ggtt.vm.mutex); - for_each_ggtt_vma(vma, obj) { - if (!drm_mm_node_allocated(&vma->node)) - continue; - - list_move_tail(&vma->vm_link, &vma->vm->bound_list); - } - mutex_unlock(&i915->ggtt.vm.mutex); - - spin_lock(&i915->mm.obj_lock); - list = obj->bind_count ? &i915->mm.bound_list : &i915->mm.unbound_list; - list_move_tail(&obj->mm.link, list); - spin_unlock(&i915->mm.obj_lock); -} - -/** - * Called when user space prepares to use an object with the CPU, either - * through the mmap ioctl's mapping or a GTT mapping. - * @dev: drm device - * @data: ioctl data blob - * @file: drm file - */ -int -i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_gem_set_domain *args = data; - struct drm_i915_gem_object *obj; - u32 read_domains = args->read_domains; - u32 write_domain = args->write_domain; - int err; - - /* Only handle setting domains to types used by the CPU. */ - if ((write_domain | read_domains) & I915_GEM_GPU_DOMAINS) - return -EINVAL; - - /* - * Having something in the write domain implies it's in the read - * domain, and only that read domain. Enforce that in the request. - */ - if (write_domain && read_domains != write_domain) - return -EINVAL; - - if (!read_domains) - return 0; - - obj = i915_gem_object_lookup(file, args->handle); - if (!obj) - return -ENOENT; - - /* - * Already in the desired write domain? Nothing for us to do! - * - * We apply a little bit of cunning here to catch a broader set of - * no-ops. If obj->write_domain is set, we must be in the same - * obj->read_domains, and only that domain. Therefore, if that - * obj->write_domain matches the request read_domains, we are - * already in the same read/write domain and can skip the operation, - * without having to further check the requested write_domain. - */ - if (READ_ONCE(obj->write_domain) == read_domains) { - err = 0; - goto out; - } - - /* - * Try to flush the object off the GPU without holding the lock. - * We will repeat the flush holding the lock in the normal manner - * to catch cases where we are gazumped. - */ - err = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_PRIORITY | - (write_domain ? I915_WAIT_ALL : 0), - MAX_SCHEDULE_TIMEOUT); - if (err) - goto out; - - /* - * Proxy objects do not control access to the backing storage, ergo - * they cannot be used as a means to manipulate the cache domain - * tracking for that backing storage. The proxy object is always - * considered to be outside of any cache domain. - */ - if (i915_gem_object_is_proxy(obj)) { - err = -ENXIO; - goto out; - } - - /* - * Flush and acquire obj->pages so that we are coherent through - * direct access in memory with previous cached writes through - * shmemfs and that our cache domain tracking remains valid. - * For example, if the obj->filp was moved to swap without us - * being notified and releasing the pages, we would mistakenly - * continue to assume that the obj remained out of the CPU cached - * domain. - */ - err = i915_gem_object_pin_pages(obj); - if (err) - goto out; - - err = i915_mutex_lock_interruptible(dev); - if (err) - goto out_unpin; - - if (read_domains & I915_GEM_DOMAIN_WC) - err = i915_gem_object_set_to_wc_domain(obj, write_domain); - else if (read_domains & I915_GEM_DOMAIN_GTT) - err = i915_gem_object_set_to_gtt_domain(obj, write_domain); - else - err = i915_gem_object_set_to_cpu_domain(obj, write_domain); - - /* And bump the LRU for this access */ - i915_gem_object_bump_inactive_ggtt(obj); - - mutex_unlock(&dev->struct_mutex); - - if (write_domain != 0) - intel_fb_obj_invalidate(obj, - fb_write_origin(obj, write_domain)); - -out_unpin: - i915_gem_object_unpin_pages(obj); -out: - i915_gem_object_put(obj); - return err; -} - /** * Called when user space has done writes to this buffer * @dev: drm device @@ -1609,421 +886,7 @@ i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, return 0; } -static inline bool -__vma_matches(struct vm_area_struct *vma, struct file *filp, - unsigned long addr, unsigned long size) -{ - if (vma->vm_file != filp) - return false; - - return vma->vm_start == addr && - (vma->vm_end - vma->vm_start) == PAGE_ALIGN(size); -} - -/** - * i915_gem_mmap_ioctl - Maps the contents of an object, returning the address - * it is mapped to. - * @dev: drm device - * @data: ioctl data blob - * @file: drm file - * - * While the mapping holds a reference on the contents of the object, it doesn't - * imply a ref on the object itself. - * - * IMPORTANT: - * - * DRM driver writers who look a this function as an example for how to do GEM - * mmap support, please don't implement mmap support like here. The modern way - * to implement DRM mmap support is with an mmap offset ioctl (like - * i915_gem_mmap_gtt) and then using the mmap syscall on the DRM fd directly. - * That way debug tooling like valgrind will understand what's going on, hiding - * the mmap call in a driver private ioctl will break that. The i915 driver only - * does cpu mmaps this way because we didn't know better. - */ -int -i915_gem_mmap_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_gem_mmap *args = data; - struct drm_i915_gem_object *obj; - unsigned long addr; - - if (args->flags & ~(I915_MMAP_WC)) - return -EINVAL; - - if (args->flags & I915_MMAP_WC && !boot_cpu_has(X86_FEATURE_PAT)) - return -ENODEV; - - obj = i915_gem_object_lookup(file, args->handle); - if (!obj) - return -ENOENT; - - /* prime objects have no backing filp to GEM mmap - * pages from. - */ - if (!obj->base.filp) { - addr = -ENXIO; - goto err; - } - - if (range_overflows(args->offset, args->size, (u64)obj->base.size)) { - addr = -EINVAL; - goto err; - } - - addr = vm_mmap(obj->base.filp, 0, args->size, - PROT_READ | PROT_WRITE, MAP_SHARED, - args->offset); - if (IS_ERR_VALUE(addr)) - goto err; - - if (args->flags & I915_MMAP_WC) { - struct mm_struct *mm = current->mm; - struct vm_area_struct *vma; - - if (down_write_killable(&mm->mmap_sem)) { - addr = -EINTR; - goto err; - } - vma = find_vma(mm, addr); - if (vma && __vma_matches(vma, obj->base.filp, addr, args->size)) - vma->vm_page_prot = - pgprot_writecombine(vm_get_page_prot(vma->vm_flags)); - else - addr = -ENOMEM; - up_write(&mm->mmap_sem); - if (IS_ERR_VALUE(addr)) - goto err; - - /* This may race, but that's ok, it only gets set */ - WRITE_ONCE(obj->frontbuffer_ggtt_origin, ORIGIN_CPU); - } - i915_gem_object_put(obj); - - args->addr_ptr = (u64)addr; - return 0; - -err: - i915_gem_object_put(obj); - return addr; -} - -static unsigned int tile_row_pages(const struct drm_i915_gem_object *obj) -{ - return i915_gem_object_get_tile_row_size(obj) >> PAGE_SHIFT; -} - -/** - * i915_gem_mmap_gtt_version - report the current feature set for GTT mmaps - * - * A history of the GTT mmap interface: - * - * 0 - Everything had to fit into the GTT. Both parties of a memcpy had to - * aligned and suitable for fencing, and still fit into the available - * mappable space left by the pinned display objects. A classic problem - * we called the page-fault-of-doom where we would ping-pong between - * two objects that could not fit inside the GTT and so the memcpy - * would page one object in at the expense of the other between every - * single byte. - * - * 1 - Objects can be any size, and have any compatible fencing (X Y, or none - * as set via i915_gem_set_tiling() [DRM_I915_GEM_SET_TILING]). If the - * object is too large for the available space (or simply too large - * for the mappable aperture!), a view is created instead and faulted - * into userspace. (This view is aligned and sized appropriately for - * fenced access.) - * - * 2 - Recognise WC as a separate cache domain so that we can flush the - * delayed writes via GTT before performing direct access via WC. - * - * 3 - Remove implicit set-domain(GTT) and synchronisation on initial - * pagefault; swapin remains transparent. - * - * Restrictions: - * - * * snoopable objects cannot be accessed via the GTT. It can cause machine - * hangs on some architectures, corruption on others. An attempt to service - * a GTT page fault from a snoopable object will generate a SIGBUS. - * - * * the object must be able to fit into RAM (physical memory, though no - * limited to the mappable aperture). - * - * - * Caveats: - * - * * a new GTT page fault will synchronize rendering from the GPU and flush - * all data to system memory. Subsequent access will not be synchronized. - * - * * all mappings are revoked on runtime device suspend. - * - * * there are only 8, 16 or 32 fence registers to share between all users - * (older machines require fence register for display and blitter access - * as well). Contention of the fence registers will cause the previous users - * to be unmapped and any new access will generate new page faults. - * - * * running out of memory while servicing a fault may generate a SIGBUS, - * rather than the expected SIGSEGV. - */ -int i915_gem_mmap_gtt_version(void) -{ - return 3; -} - -static inline struct i915_ggtt_view -compute_partial_view(const struct drm_i915_gem_object *obj, - pgoff_t page_offset, - unsigned int chunk) -{ - struct i915_ggtt_view view; - - if (i915_gem_object_is_tiled(obj)) - chunk = roundup(chunk, tile_row_pages(obj)); - - view.type = I915_GGTT_VIEW_PARTIAL; - view.partial.offset = rounddown(page_offset, chunk); - view.partial.size = - min_t(unsigned int, chunk, - (obj->base.size >> PAGE_SHIFT) - view.partial.offset); - - /* If the partial covers the entire object, just create a normal VMA. */ - if (chunk >= obj->base.size >> PAGE_SHIFT) - view.type = I915_GGTT_VIEW_NORMAL; - - return view; -} - -/** - * i915_gem_fault - fault a page into the GTT - * @vmf: fault info - * - * The fault handler is set up by drm_gem_mmap() when a object is GTT mapped - * from userspace. The fault handler takes care of binding the object to - * the GTT (if needed), allocating and programming a fence register (again, - * only if needed based on whether the old reg is still valid or the object - * is tiled) and inserting a new PTE into the faulting process. - * - * Note that the faulting process may involve evicting existing objects - * from the GTT and/or fence registers to make room. So performance may - * suffer if the GTT working set is large or there are few fence registers - * left. - * - * The current feature set supported by i915_gem_fault() and thus GTT mmaps - * is exposed via I915_PARAM_MMAP_GTT_VERSION (see i915_gem_mmap_gtt_version). - */ -vm_fault_t i915_gem_fault(struct vm_fault *vmf) -{ -#define MIN_CHUNK_PAGES (SZ_1M >> PAGE_SHIFT) - struct vm_area_struct *area = vmf->vma; - struct drm_i915_gem_object *obj = to_intel_bo(area->vm_private_data); - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - struct i915_ggtt *ggtt = &dev_priv->ggtt; - bool write = area->vm_flags & VM_WRITE; - intel_wakeref_t wakeref; - struct i915_vma *vma; - pgoff_t page_offset; - int srcu; - int ret; - - /* Sanity check that we allow writing into this object */ - if (i915_gem_object_is_readonly(obj) && write) - return VM_FAULT_SIGBUS; - - /* We don't use vmf->pgoff since that has the fake offset */ - page_offset = (vmf->address - area->vm_start) >> PAGE_SHIFT; - - trace_i915_gem_object_fault(obj, page_offset, true, write); - - ret = i915_gem_object_pin_pages(obj); - if (ret) - goto err; - - wakeref = intel_runtime_pm_get(dev_priv); - - srcu = i915_reset_trylock(dev_priv); - if (srcu < 0) { - ret = srcu; - goto err_rpm; - } - - ret = i915_mutex_lock_interruptible(dev); - if (ret) - goto err_reset; - - /* Access to snoopable pages through the GTT is incoherent. */ - if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(dev_priv)) { - ret = -EFAULT; - goto err_unlock; - } - - /* Now pin it into the GTT as needed */ - vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, - PIN_MAPPABLE | - PIN_NONBLOCK | - PIN_NONFAULT); - if (IS_ERR(vma)) { - /* Use a partial view if it is bigger than available space */ - struct i915_ggtt_view view = - compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES); - unsigned int flags; - - flags = PIN_MAPPABLE; - if (view.type == I915_GGTT_VIEW_NORMAL) - flags |= PIN_NONBLOCK; /* avoid warnings for pinned */ - - /* - * Userspace is now writing through an untracked VMA, abandon - * all hope that the hardware is able to track future writes. - */ - obj->frontbuffer_ggtt_origin = ORIGIN_CPU; - - vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags); - if (IS_ERR(vma) && !view.type) { - flags = PIN_MAPPABLE; - view.type = I915_GGTT_VIEW_PARTIAL; - vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags); - } - } - if (IS_ERR(vma)) { - ret = PTR_ERR(vma); - goto err_unlock; - } - - ret = i915_vma_pin_fence(vma); - if (ret) - goto err_unpin; - - /* Finally, remap it using the new GTT offset */ - ret = remap_io_mapping(area, - area->vm_start + (vma->ggtt_view.partial.offset << PAGE_SHIFT), - (ggtt->gmadr.start + vma->node.start) >> PAGE_SHIFT, - min_t(u64, vma->size, area->vm_end - area->vm_start), - &ggtt->iomap); - if (ret) - goto err_fence; - - /* Mark as being mmapped into userspace for later revocation */ - assert_rpm_wakelock_held(dev_priv); - if (!i915_vma_set_userfault(vma) && !obj->userfault_count++) - list_add(&obj->userfault_link, &dev_priv->mm.userfault_list); - GEM_BUG_ON(!obj->userfault_count); - - i915_vma_set_ggtt_write(vma); - -err_fence: - i915_vma_unpin_fence(vma); -err_unpin: - __i915_vma_unpin(vma); -err_unlock: - mutex_unlock(&dev->struct_mutex); -err_reset: - i915_reset_unlock(dev_priv, srcu); -err_rpm: - intel_runtime_pm_put(dev_priv, wakeref); - i915_gem_object_unpin_pages(obj); -err: - switch (ret) { - case -EIO: - /* - * We eat errors when the gpu is terminally wedged to avoid - * userspace unduly crashing (gl has no provisions for mmaps to - * fail). But any other -EIO isn't ours (e.g. swap in failure) - * and so needs to be reported. - */ - if (!i915_terminally_wedged(dev_priv)) - return VM_FAULT_SIGBUS; - /* else: fall through */ - case -EAGAIN: - /* - * 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. - */ - case 0: - case -ERESTARTSYS: - case -EINTR: - case -EBUSY: - /* - * EBUSY is ok: this just means that another thread - * already did the job. - */ - return VM_FAULT_NOPAGE; - case -ENOMEM: - return VM_FAULT_OOM; - case -ENOSPC: - case -EFAULT: - return VM_FAULT_SIGBUS; - default: - WARN_ONCE(ret, "unhandled error in i915_gem_fault: %i\n", ret); - return VM_FAULT_SIGBUS; - } -} - -static void __i915_gem_object_release_mmap(struct drm_i915_gem_object *obj) -{ - struct i915_vma *vma; - - GEM_BUG_ON(!obj->userfault_count); - - obj->userfault_count = 0; - list_del(&obj->userfault_link); - drm_vma_node_unmap(&obj->base.vma_node, - obj->base.dev->anon_inode->i_mapping); - - for_each_ggtt_vma(vma, obj) - i915_vma_unset_userfault(vma); -} - -/** - * i915_gem_release_mmap - remove physical page mappings - * @obj: obj in question - * - * Preserve the reservation of the mmapping with the DRM core code, but - * relinquish ownership of the pages back to the system. - * - * It is vital that we remove the page mapping if we have mapped a tiled - * object through the GTT and then lose the fence register due to - * resource pressure. Similarly if the object has been moved out of the - * aperture, than pages mapped into userspace must be revoked. Removing the - * mapping will then trigger a page fault on the next user access, allowing - * fixup by i915_gem_fault(). - */ -void -i915_gem_release_mmap(struct drm_i915_gem_object *obj) -{ - struct drm_i915_private *i915 = to_i915(obj->base.dev); - intel_wakeref_t wakeref; - - /* Serialisation between user GTT access and our code depends upon - * revoking the CPU's PTE whilst the mutex is held. The next user - * pagefault then has to wait until we release the mutex. - * - * Note that RPM complicates somewhat by adding an additional - * requirement that operations to the GGTT be made holding the RPM - * wakeref. - */ - lockdep_assert_held(&i915->drm.struct_mutex); - wakeref = intel_runtime_pm_get(i915); - - if (!obj->userfault_count) - goto out; - - __i915_gem_object_release_mmap(obj); - - /* Ensure that the CPU's PTE are revoked and there are not outstanding - * memory transactions from userspace before we return. The TLB - * flushing implied above by changing the PTE above *should* be - * sufficient, an extra barrier here just provides us with a bit - * of paranoid documentation about our requirement to serialise - * memory writes before touching registers / GSM. - */ - wmb(); - -out: - intel_runtime_pm_put(i915, wakeref); -} - -void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv) +void i915_gem_runtime_suspend(struct drm_i915_private *i915) { struct drm_i915_gem_object *obj, *on; int i; @@ -2036,17 +899,19 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv) */ list_for_each_entry_safe(obj, on, - &dev_priv->mm.userfault_list, userfault_link) + &i915->ggtt.userfault_list, userfault_link) __i915_gem_object_release_mmap(obj); - /* The fence will be lost when the device powers down. If any were + /* + * The fence will be lost when the device powers down. If any were * in use by hardware (i.e. they are pinned), we should not be powering * down! All other fences will be reacquired by the user upon waking. */ - for (i = 0; i < dev_priv->num_fence_regs; i++) { - struct drm_i915_fence_reg *reg = &dev_priv->fence_regs[i]; + for (i = 0; i < i915->ggtt.num_fences; i++) { + struct i915_fence_reg *reg = &i915->ggtt.fence_regs[i]; - /* Ideally we want to assert that the fence register is not + /* + * Ideally we want to assert that the fence register is not * live at this point (i.e. that no piece of code will be * trying to write through fence + GTT, as that both violates * our tracking of activity and associated locking/barriers, @@ -2065,1056 +930,6 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv) } } -static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj) -{ - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); - int err; - - err = drm_gem_create_mmap_offset(&obj->base); - if (likely(!err)) - return 0; - - /* Attempt to reap some mmap space from dead objects */ - do { - err = i915_gem_wait_for_idle(dev_priv, - I915_WAIT_INTERRUPTIBLE, - MAX_SCHEDULE_TIMEOUT); - if (err) - break; - - i915_gem_drain_freed_objects(dev_priv); - err = drm_gem_create_mmap_offset(&obj->base); - if (!err) - break; - - } while (flush_delayed_work(&dev_priv->gt.retire_work)); - - return err; -} - -static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) -{ - drm_gem_free_mmap_offset(&obj->base); -} - -int -i915_gem_mmap_gtt(struct drm_file *file, - struct drm_device *dev, - u32 handle, - u64 *offset) -{ - struct drm_i915_gem_object *obj; - int ret; - - obj = i915_gem_object_lookup(file, handle); - if (!obj) - return -ENOENT; - - ret = i915_gem_object_create_mmap_offset(obj); - if (ret == 0) - *offset = drm_vma_node_offset_addr(&obj->base.vma_node); - - i915_gem_object_put(obj); - return ret; -} - -/** - * i915_gem_mmap_gtt_ioctl - prepare an object for GTT mmap'ing - * @dev: DRM device - * @data: GTT mapping ioctl data - * @file: GEM object info - * - * Simply returns the fake offset to userspace so it can mmap it. - * The mmap call will end up in drm_gem_mmap(), which will set things - * up so we can get faults in the handler above. - * - * The fault handler will take care of binding the object into the GTT - * (since it may have been evicted to make room for something), allocating - * a fence register, and mapping the appropriate aperture address into - * userspace. - */ -int -i915_gem_mmap_gtt_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_gem_mmap_gtt *args = data; - - return i915_gem_mmap_gtt(file, dev, args->handle, &args->offset); -} - -/* Immediately discard the backing storage */ -static void -i915_gem_object_truncate(struct drm_i915_gem_object *obj) -{ - i915_gem_object_free_mmap_offset(obj); - - if (obj->base.filp == NULL) - return; - - /* Our goal here is to return as much of the memory as - * is possible back to the system as we are called from OOM. - * To do this we must instruct the shmfs to drop all of its - * backing pages, *now*. - */ - shmem_truncate_range(file_inode(obj->base.filp), 0, (loff_t)-1); - obj->mm.madv = __I915_MADV_PURGED; - obj->mm.pages = ERR_PTR(-EFAULT); -} - -/* Try to discard unwanted pages */ -void __i915_gem_object_invalidate(struct drm_i915_gem_object *obj) -{ - struct address_space *mapping; - - lockdep_assert_held(&obj->mm.lock); - GEM_BUG_ON(i915_gem_object_has_pages(obj)); - - switch (obj->mm.madv) { - case I915_MADV_DONTNEED: - i915_gem_object_truncate(obj); - case __I915_MADV_PURGED: - return; - } - - if (obj->base.filp == NULL) - return; - - mapping = obj->base.filp->f_mapping, - invalidate_mapping_pages(mapping, 0, (loff_t)-1); -} - -/* - * Move pages to appropriate lru and release the pagevec, decrementing the - * ref count of those pages. - */ -static void check_release_pagevec(struct pagevec *pvec) -{ - check_move_unevictable_pages(pvec); - __pagevec_release(pvec); - cond_resched(); -} - -static void -i915_gem_object_put_pages_gtt(struct drm_i915_gem_object *obj, - struct sg_table *pages) -{ - struct sgt_iter sgt_iter; - struct pagevec pvec; - struct page *page; - - __i915_gem_object_release_shmem(obj, pages, true); - i915_gem_gtt_finish_pages(obj, pages); - - if (i915_gem_object_needs_bit17_swizzle(obj)) - i915_gem_object_save_bit_17_swizzle(obj, pages); - - mapping_clear_unevictable(file_inode(obj->base.filp)->i_mapping); - - pagevec_init(&pvec); - for_each_sgt_page(page, sgt_iter, pages) { - if (obj->mm.dirty) - set_page_dirty(page); - - if (obj->mm.madv == I915_MADV_WILLNEED) - mark_page_accessed(page); - - if (!pagevec_add(&pvec, page)) - check_release_pagevec(&pvec); - } - if (pagevec_count(&pvec)) - check_release_pagevec(&pvec); - obj->mm.dirty = false; - - sg_free_table(pages); - kfree(pages); -} - -static void __i915_gem_object_reset_page_iter(struct drm_i915_gem_object *obj) -{ - struct radix_tree_iter iter; - void __rcu **slot; - - rcu_read_lock(); - radix_tree_for_each_slot(slot, &obj->mm.get_page.radix, &iter, 0) - radix_tree_delete(&obj->mm.get_page.radix, iter.index); - rcu_read_unlock(); -} - -static struct sg_table * -__i915_gem_object_unset_pages(struct drm_i915_gem_object *obj) -{ - struct drm_i915_private *i915 = to_i915(obj->base.dev); - struct sg_table *pages; - - pages = fetch_and_zero(&obj->mm.pages); - if (IS_ERR_OR_NULL(pages)) - return pages; - - spin_lock(&i915->mm.obj_lock); - list_del(&obj->mm.link); - spin_unlock(&i915->mm.obj_lock); - - if (obj->mm.mapping) { - void *ptr; - - ptr = page_mask_bits(obj->mm.mapping); - if (is_vmalloc_addr(ptr)) - vunmap(ptr); - else - kunmap(kmap_to_page(ptr)); - - obj->mm.mapping = NULL; - } - - __i915_gem_object_reset_page_iter(obj); - obj->mm.page_sizes.phys = obj->mm.page_sizes.sg = 0; - - return pages; -} - -int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj, - enum i915_mm_subclass subclass) -{ - struct sg_table *pages; - int ret; - - if (i915_gem_object_has_pinned_pages(obj)) - return -EBUSY; - - GEM_BUG_ON(obj->bind_count); - - /* May be called by shrinker from within get_pages() (on another bo) */ - mutex_lock_nested(&obj->mm.lock, subclass); - if (unlikely(atomic_read(&obj->mm.pages_pin_count))) { - ret = -EBUSY; - goto unlock; - } - - /* - * ->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. - */ - pages = __i915_gem_object_unset_pages(obj); - - /* - * XXX Temporary hijinx to avoid updating all backends to handle - * NULL pages. In the future, when we have more asynchronous - * get_pages backends we should be better able to handle the - * cancellation of the async task in a more uniform manner. - */ - if (!pages && !i915_gem_object_needs_async_cancel(obj)) - pages = ERR_PTR(-EINVAL); - - if (!IS_ERR(pages)) - obj->ops->put_pages(obj, pages); - - ret = 0; -unlock: - mutex_unlock(&obj->mm.lock); - - return ret; -} - -bool i915_sg_trim(struct sg_table *orig_st) -{ - struct sg_table new_st; - struct scatterlist *sg, *new_sg; - unsigned int i; - - if (orig_st->nents == orig_st->orig_nents) - return false; - - if (sg_alloc_table(&new_st, orig_st->nents, GFP_KERNEL | __GFP_NOWARN)) - return false; - - new_sg = new_st.sgl; - for_each_sg(orig_st->sgl, sg, orig_st->nents, i) { - sg_set_page(new_sg, sg_page(sg), sg->length, 0); - sg_dma_address(new_sg) = sg_dma_address(sg); - sg_dma_len(new_sg) = sg_dma_len(sg); - - new_sg = sg_next(new_sg); - } - GEM_BUG_ON(new_sg); /* Should walk exactly nents and hit the end */ - - sg_free_table(orig_st); - - *orig_st = new_st; - return true; -} - -static int i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) -{ - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); - const unsigned long page_count = obj->base.size / PAGE_SIZE; - unsigned long i; - struct address_space *mapping; - struct sg_table *st; - struct scatterlist *sg; - struct sgt_iter sgt_iter; - struct page *page; - unsigned long last_pfn = 0; /* suppress gcc warning */ - unsigned int max_segment = i915_sg_segment_size(); - unsigned int sg_page_sizes; - struct pagevec pvec; - gfp_t noreclaim; - int ret; - - /* - * Assert that the object is not currently in any GPU domain. As it - * wasn't in the GTT, there shouldn't be any way it could have been in - * a GPU cache - */ - GEM_BUG_ON(obj->read_domains & I915_GEM_GPU_DOMAINS); - GEM_BUG_ON(obj->write_domain & I915_GEM_GPU_DOMAINS); - - /* - * If there's no chance of allocating enough pages for the whole - * object, bail early. - */ - if (page_count > totalram_pages()) - return -ENOMEM; - - st = kmalloc(sizeof(*st), GFP_KERNEL); - if (st == NULL) - return -ENOMEM; - -rebuild_st: - if (sg_alloc_table(st, page_count, GFP_KERNEL)) { - kfree(st); - return -ENOMEM; - } - - /* - * Get the list of pages out of our struct file. They'll be pinned - * at this point until we release them. - * - * Fail silently without starting the shrinker - */ - mapping = obj->base.filp->f_mapping; - mapping_set_unevictable(mapping); - noreclaim = mapping_gfp_constraint(mapping, ~__GFP_RECLAIM); - noreclaim |= __GFP_NORETRY | __GFP_NOWARN; - - sg = st->sgl; - st->nents = 0; - sg_page_sizes = 0; - for (i = 0; i < page_count; i++) { - const unsigned int shrink[] = { - I915_SHRINK_BOUND | I915_SHRINK_UNBOUND | I915_SHRINK_PURGEABLE, - 0, - }, *s = shrink; - gfp_t gfp = noreclaim; - - do { - cond_resched(); - page = shmem_read_mapping_page_gfp(mapping, i, gfp); - if (!IS_ERR(page)) - break; - - if (!*s) { - ret = PTR_ERR(page); - goto err_sg; - } - - i915_gem_shrink(dev_priv, 2 * page_count, NULL, *s++); - - /* - * We've tried hard to allocate the memory by reaping - * our own buffer, now let the real VM do its job and - * go down in flames if truly OOM. - * - * However, since graphics tend to be disposable, - * defer the oom here by reporting the ENOMEM back - * to userspace. - */ - if (!*s) { - /* reclaim and warn, but no oom */ - gfp = mapping_gfp_mask(mapping); - - /* - * Our bo are always dirty and so we require - * kswapd to reclaim our pages (direct reclaim - * does not effectively begin pageout of our - * buffers on its own). However, direct reclaim - * only waits for kswapd when under allocation - * congestion. So as a result __GFP_RECLAIM is - * unreliable and fails to actually reclaim our - * dirty pages -- unless you try over and over - * again with !__GFP_NORETRY. However, we still - * want to fail this allocation rather than - * trigger the out-of-memory killer and for - * this we want __GFP_RETRY_MAYFAIL. - */ - gfp |= __GFP_RETRY_MAYFAIL; - } - } while (1); - - if (!i || - sg->length >= max_segment || - page_to_pfn(page) != last_pfn + 1) { - if (i) { - sg_page_sizes |= sg->length; - 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); - - /* Check that the i965g/gm workaround works. */ - WARN_ON((gfp & __GFP_DMA32) && (last_pfn >= 0x00100000UL)); - } - if (sg) { /* loop terminated early; short sg table */ - sg_page_sizes |= sg->length; - sg_mark_end(sg); - } - - /* Trim unused sg entries to avoid wasting memory. */ - i915_sg_trim(st); - - ret = i915_gem_gtt_prepare_pages(obj, st); - if (ret) { - /* - * DMA remapping failed? One possible cause is that - * it could not reserve enough large entries, asking - * for PAGE_SIZE chunks instead may be helpful. - */ - if (max_segment > PAGE_SIZE) { - for_each_sgt_page(page, sgt_iter, st) - put_page(page); - sg_free_table(st); - - max_segment = PAGE_SIZE; - goto rebuild_st; - } else { - dev_warn(&dev_priv->drm.pdev->dev, - "Failed to DMA remap %lu pages\n", - page_count); - goto err_pages; - } - } - - if (i915_gem_object_needs_bit17_swizzle(obj)) - i915_gem_object_do_bit_17_swizzle(obj, st); - - __i915_gem_object_set_pages(obj, st, sg_page_sizes); - - return 0; - -err_sg: - sg_mark_end(sg); -err_pages: - mapping_clear_unevictable(mapping); - pagevec_init(&pvec); - for_each_sgt_page(page, sgt_iter, st) { - if (!pagevec_add(&pvec, page)) - check_release_pagevec(&pvec); - } - if (pagevec_count(&pvec)) - check_release_pagevec(&pvec); - sg_free_table(st); - kfree(st); - - /* - * 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 (ret == -ENOSPC) - ret = -ENOMEM; - - return ret; -} - -void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj, - struct sg_table *pages, - unsigned int sg_page_sizes) -{ - struct drm_i915_private *i915 = to_i915(obj->base.dev); - unsigned long supported = INTEL_INFO(i915)->page_sizes; - int i; - - lockdep_assert_held(&obj->mm.lock); - - /* Make the pages coherent with the GPU (flushing any swapin). */ - if (obj->cache_dirty) { - obj->write_domain = 0; - if (i915_gem_object_has_struct_page(obj)) - drm_clflush_sg(pages); - obj->cache_dirty = false; - } - - obj->mm.get_page.sg_pos = pages->sgl; - obj->mm.get_page.sg_idx = 0; - - obj->mm.pages = pages; - - if (i915_gem_object_is_tiled(obj) && - i915->quirks & QUIRK_PIN_SWIZZLED_PAGES) { - GEM_BUG_ON(obj->mm.quirked); - __i915_gem_object_pin_pages(obj); - obj->mm.quirked = true; - } - - GEM_BUG_ON(!sg_page_sizes); - obj->mm.page_sizes.phys = sg_page_sizes; - - /* - * Calculate the supported page-sizes which fit into the given - * sg_page_sizes. This will give us the page-sizes which we may be able - * to use opportunistically when later inserting into the GTT. For - * example if phys=2G, then in theory we should be able to use 1G, 2M, - * 64K or 4K pages, although in practice this will depend on a number of - * other factors. - */ - obj->mm.page_sizes.sg = 0; - for_each_set_bit(i, &supported, ilog2(I915_GTT_MAX_PAGE_SIZE) + 1) { - if (obj->mm.page_sizes.phys & ~0u << i) - obj->mm.page_sizes.sg |= BIT(i); - } - GEM_BUG_ON(!HAS_PAGE_SIZES(i915, obj->mm.page_sizes.sg)); - - spin_lock(&i915->mm.obj_lock); - list_add(&obj->mm.link, &i915->mm.unbound_list); - spin_unlock(&i915->mm.obj_lock); -} - -static int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj) -{ - int err; - - if (unlikely(obj->mm.madv != I915_MADV_WILLNEED)) { - DRM_DEBUG("Attempting to obtain a purgeable object\n"); - return -EFAULT; - } - - err = obj->ops->get_pages(obj); - GEM_BUG_ON(!err && !i915_gem_object_has_pages(obj)); - - return err; -} - -/* Ensure that the associated pages are gathered from the backing storage - * and pinned into our object. i915_gem_object_pin_pages() may be called - * multiple times before they are released by a single call to - * i915_gem_object_unpin_pages() - once the pages are no longer referenced - * either as a result of memory pressure (reaping pages under the shrinker) - * or as the object is itself released. - */ -int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj) -{ - int err; - - err = mutex_lock_interruptible(&obj->mm.lock); - if (err) - return err; - - if (unlikely(!i915_gem_object_has_pages(obj))) { - GEM_BUG_ON(i915_gem_object_has_pinned_pages(obj)); - - err = ____i915_gem_object_get_pages(obj); - if (err) - goto unlock; - - smp_mb__before_atomic(); - } - atomic_inc(&obj->mm.pages_pin_count); - -unlock: - mutex_unlock(&obj->mm.lock); - return err; -} - -/* The 'mapping' part of i915_gem_object_pin_map() below */ -static void *i915_gem_object_map(const struct drm_i915_gem_object *obj, - enum i915_map_type type) -{ - unsigned long n_pages = obj->base.size >> PAGE_SHIFT; - struct sg_table *sgt = obj->mm.pages; - struct sgt_iter sgt_iter; - struct page *page; - struct page *stack_pages[32]; - struct page **pages = stack_pages; - unsigned long i = 0; - pgprot_t pgprot; - void *addr; - - /* A single page can always be kmapped */ - if (n_pages == 1 && type == I915_MAP_WB) - return kmap(sg_page(sgt->sgl)); - - if (n_pages > ARRAY_SIZE(stack_pages)) { - /* Too big for stack -- allocate temporary array instead */ - pages = kvmalloc_array(n_pages, sizeof(*pages), GFP_KERNEL); - if (!pages) - return NULL; - } - - for_each_sgt_page(page, sgt_iter, sgt) - pages[i++] = page; - - /* Check that we have the expected number of pages */ - GEM_BUG_ON(i != n_pages); - - switch (type) { - default: - MISSING_CASE(type); - /* fallthrough to use PAGE_KERNEL anyway */ - case I915_MAP_WB: - pgprot = PAGE_KERNEL; - break; - case I915_MAP_WC: - pgprot = pgprot_writecombine(PAGE_KERNEL_IO); - break; - } - addr = vmap(pages, n_pages, 0, pgprot); - - if (pages != stack_pages) - kvfree(pages); - - return addr; -} - -/* get, pin, and map the pages of the object into kernel space */ -void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj, - enum i915_map_type type) -{ - enum i915_map_type has_type; - bool pinned; - void *ptr; - int ret; - - if (unlikely(!i915_gem_object_has_struct_page(obj))) - return ERR_PTR(-ENXIO); - - ret = mutex_lock_interruptible(&obj->mm.lock); - if (ret) - return ERR_PTR(ret); - - pinned = !(type & I915_MAP_OVERRIDE); - type &= ~I915_MAP_OVERRIDE; - - if (!atomic_inc_not_zero(&obj->mm.pages_pin_count)) { - if (unlikely(!i915_gem_object_has_pages(obj))) { - GEM_BUG_ON(i915_gem_object_has_pinned_pages(obj)); - - ret = ____i915_gem_object_get_pages(obj); - if (ret) - goto err_unlock; - - smp_mb__before_atomic(); - } - atomic_inc(&obj->mm.pages_pin_count); - pinned = false; - } - GEM_BUG_ON(!i915_gem_object_has_pages(obj)); - - ptr = page_unpack_bits(obj->mm.mapping, &has_type); - if (ptr && has_type != type) { - if (pinned) { - ret = -EBUSY; - goto err_unpin; - } - - if (is_vmalloc_addr(ptr)) - vunmap(ptr); - else - kunmap(kmap_to_page(ptr)); - - ptr = obj->mm.mapping = NULL; - } - - if (!ptr) { - ptr = i915_gem_object_map(obj, type); - if (!ptr) { - ret = -ENOMEM; - goto err_unpin; - } - - obj->mm.mapping = page_pack_bits(ptr, type); - } - -out_unlock: - mutex_unlock(&obj->mm.lock); - return ptr; - -err_unpin: - atomic_dec(&obj->mm.pages_pin_count); -err_unlock: - ptr = ERR_PTR(ret); - goto out_unlock; -} - -void __i915_gem_object_flush_map(struct drm_i915_gem_object *obj, - unsigned long offset, - unsigned long size) -{ - enum i915_map_type has_type; - void *ptr; - - GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj)); - GEM_BUG_ON(range_overflows_t(typeof(obj->base.size), - offset, size, obj->base.size)); - - obj->mm.dirty = true; - - if (obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE) - return; - - ptr = page_unpack_bits(obj->mm.mapping, &has_type); - if (has_type == I915_MAP_WC) - return; - - drm_clflush_virt_range(ptr + offset, size); - if (size == obj->base.size) { - obj->write_domain &= ~I915_GEM_DOMAIN_CPU; - obj->cache_dirty = false; - } -} - -static int -i915_gem_object_pwrite_gtt(struct drm_i915_gem_object *obj, - const struct drm_i915_gem_pwrite *arg) -{ - struct address_space *mapping = obj->base.filp->f_mapping; - char __user *user_data = u64_to_user_ptr(arg->data_ptr); - u64 remain, offset; - unsigned int pg; - - /* Caller already validated user args */ - GEM_BUG_ON(!access_ok(user_data, arg->size)); - - /* - * Before we instantiate/pin the backing store for our use, we - * can prepopulate the shmemfs filp efficiently using a write into - * the pagecache. We avoid the penalty of instantiating all the - * pages, important if the user is just writing to a few and never - * uses the object on the GPU, and using a direct write into shmemfs - * allows it to avoid the cost of retrieving a page (either swapin - * or clearing-before-use) before it is overwritten. - */ - if (i915_gem_object_has_pages(obj)) - return -ENODEV; - - if (obj->mm.madv != I915_MADV_WILLNEED) - return -EFAULT; - - /* - * Before the pages are instantiated the object is treated as being - * in the CPU domain. The pages will be clflushed as required before - * use, and we can freely write into the pages directly. If userspace - * races pwrite with any other operation; corruption will ensue - - * that is userspace's prerogative! - */ - - remain = arg->size; - offset = arg->offset; - pg = offset_in_page(offset); - - do { - unsigned int len, unwritten; - struct page *page; - void *data, *vaddr; - int err; - char c; - - len = PAGE_SIZE - pg; - if (len > remain) - len = remain; - - /* Prefault the user page to reduce potential recursion */ - err = __get_user(c, user_data); - if (err) - return err; - - err = __get_user(c, user_data + len - 1); - if (err) - return err; - - err = pagecache_write_begin(obj->base.filp, mapping, - offset, len, 0, - &page, &data); - if (err < 0) - return err; - - vaddr = kmap_atomic(page); - unwritten = __copy_from_user_inatomic(vaddr + pg, - user_data, - len); - kunmap_atomic(vaddr); - - err = pagecache_write_end(obj->base.filp, mapping, - offset, len, len - unwritten, - page, data); - if (err < 0) - return err; - - /* We don't handle -EFAULT, leave it to the caller to check */ - if (unwritten) - return -ENODEV; - - remain -= len; - user_data += len; - offset += len; - pg = 0; - } while (remain); - - return 0; -} - -static void -i915_gem_retire_work_handler(struct work_struct *work) -{ - struct drm_i915_private *dev_priv = - container_of(work, typeof(*dev_priv), gt.retire_work.work); - struct drm_device *dev = &dev_priv->drm; - - /* Come back later if the device is busy... */ - if (mutex_trylock(&dev->struct_mutex)) { - i915_retire_requests(dev_priv); - mutex_unlock(&dev->struct_mutex); - } - - /* - * Keep the retire handler running until we are finally idle. - * We do not need to do this test under locking as in the worst-case - * we queue the retire worker once too often. - */ - if (READ_ONCE(dev_priv->gt.awake)) - queue_delayed_work(dev_priv->wq, - &dev_priv->gt.retire_work, - round_jiffies_up_relative(HZ)); -} - -static bool switch_to_kernel_context_sync(struct drm_i915_private *i915, - unsigned long mask) -{ - bool result = true; - - /* - * Even if we fail to switch, give whatever is running a small chance - * to save itself before we report the failure. Yes, this may be a - * false positive due to e.g. ENOMEM, caveat emptor! - */ - if (i915_gem_switch_to_kernel_context(i915, mask)) - result = false; - - if (i915_gem_wait_for_idle(i915, - I915_WAIT_LOCKED | - I915_WAIT_FOR_IDLE_BOOST, - I915_GEM_IDLE_TIMEOUT)) - result = false; - - if (!result) { - if (i915_modparams.reset) { /* XXX hide warning from gem_eio */ - dev_err(i915->drm.dev, - "Failed to idle engines, declaring wedged!\n"); - GEM_TRACE_DUMP(); - } - - /* Forcibly cancel outstanding work and leave the gpu quiet. */ - i915_gem_set_wedged(i915); - } - - i915_retire_requests(i915); /* ensure we flush after wedging */ - return result; -} - -static bool load_power_context(struct drm_i915_private *i915) -{ - /* Force loading the kernel context on all engines */ - if (!switch_to_kernel_context_sync(i915, ALL_ENGINES)) - return false; - - /* - * Immediately park the GPU so that we enable powersaving and - * treat it as idle. The next time we issue a request, we will - * unpark and start using the engine->pinned_default_state, otherwise - * it is in limbo and an early reset may fail. - */ - __i915_gem_park(i915); - - return true; -} - -static void -i915_gem_idle_work_handler(struct work_struct *work) -{ - struct drm_i915_private *i915 = - container_of(work, typeof(*i915), gt.idle_work.work); - bool rearm_hangcheck; - - if (!READ_ONCE(i915->gt.awake)) - return; - - if (READ_ONCE(i915->gt.active_requests)) - return; - - rearm_hangcheck = - cancel_delayed_work_sync(&i915->gpu_error.hangcheck_work); - - if (!mutex_trylock(&i915->drm.struct_mutex)) { - /* Currently busy, come back later */ - mod_delayed_work(i915->wq, - &i915->gt.idle_work, - msecs_to_jiffies(50)); - goto out_rearm; - } - - /* - * Flush out the last user context, leaving only the pinned - * kernel context resident. Should anything unfortunate happen - * while we are idle (such as the GPU being power cycled), no users - * will be harmed. - */ - if (!work_pending(&i915->gt.idle_work.work) && - !i915->gt.active_requests) { - ++i915->gt.active_requests; /* don't requeue idle */ - - switch_to_kernel_context_sync(i915, i915->gt.active_engines); - - if (!--i915->gt.active_requests) { - __i915_gem_park(i915); - rearm_hangcheck = false; - } - } - - mutex_unlock(&i915->drm.struct_mutex); - -out_rearm: - if (rearm_hangcheck) { - GEM_BUG_ON(!i915->gt.awake); - i915_queue_hangcheck(i915); - } -} - -void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file) -{ - struct drm_i915_private *i915 = to_i915(gem->dev); - struct drm_i915_gem_object *obj = to_intel_bo(gem); - struct drm_i915_file_private *fpriv = file->driver_priv; - struct i915_lut_handle *lut, *ln; - - mutex_lock(&i915->drm.struct_mutex); - - list_for_each_entry_safe(lut, ln, &obj->lut_list, obj_link) { - struct i915_gem_context *ctx = lut->ctx; - struct i915_vma *vma; - - GEM_BUG_ON(ctx->file_priv == ERR_PTR(-EBADF)); - if (ctx->file_priv != fpriv) - continue; - - vma = radix_tree_delete(&ctx->handles_vma, lut->handle); - GEM_BUG_ON(vma->obj != obj); - - /* We allow the process to have multiple handles to the same - * vma, in the same fd namespace, by virtue of flink/open. - */ - GEM_BUG_ON(!vma->open_count); - if (!--vma->open_count && !i915_vma_is_ggtt(vma)) - i915_vma_close(vma); - - list_del(&lut->obj_link); - list_del(&lut->ctx_link); - - i915_lut_handle_free(lut); - __i915_gem_object_release_unless_active(obj); - } - - mutex_unlock(&i915->drm.struct_mutex); -} - -static unsigned long to_wait_timeout(s64 timeout_ns) -{ - if (timeout_ns < 0) - return MAX_SCHEDULE_TIMEOUT; - - if (timeout_ns == 0) - return 0; - - return nsecs_to_jiffies_timeout(timeout_ns); -} - -/** - * i915_gem_wait_ioctl - implements DRM_IOCTL_I915_GEM_WAIT - * @dev: drm device pointer - * @data: ioctl data blob - * @file: drm file pointer - * - * Returns 0 if successful, else an error is returned with the remaining time in - * the timeout parameter. - * -ETIME: object is still busy after timeout - * -ERESTARTSYS: signal interrupted the wait - * -ENONENT: object doesn't exist - * Also possible, but rare: - * -EAGAIN: incomplete, restart syscall - * -ENOMEM: damn - * -ENODEV: Internal IRQ fail - * -E?: The add request failed - * - * The wait ioctl with a timeout of 0 reimplements the busy ioctl. With any - * non-zero timeout parameter the wait ioctl will wait for the given number of - * nanoseconds on an object becoming unbusy. Since the wait itself does so - * without holding struct_mutex the object may become re-busied before this - * function completes. A similar but shorter * race condition exists in the busy - * ioctl - */ -int -i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) -{ - struct drm_i915_gem_wait *args = data; - struct drm_i915_gem_object *obj; - ktime_t start; - long ret; - - if (args->flags != 0) - return -EINVAL; - - obj = i915_gem_object_lookup(file, args->bo_handle); - if (!obj) - return -ENOENT; - - start = ktime_get(); - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_PRIORITY | - I915_WAIT_ALL, - to_wait_timeout(args->timeout_ns)); - - if (args->timeout_ns > 0) { - args->timeout_ns -= ktime_to_ns(ktime_sub(ktime_get(), start)); - if (args->timeout_ns < 0) - args->timeout_ns = 0; - - /* - * Apparently ktime isn't accurate enough and occasionally has a - * bit of mismatch in the jiffies<->nsecs<->ktime loop. So patch - * things up to make the test happy. We allow up to 1 jiffy. - * - * This is a regression from the timespec->ktime conversion. - */ - if (ret == -ETIME && !nsecs_to_jiffies(args->timeout_ns)) - args->timeout_ns = 0; - - /* Asked to wait beyond the jiffie/scheduler precision? */ - if (ret == -ETIME && args->timeout_ns) - ret = -EAGAIN; - } - - i915_gem_object_put(obj); - return ret; -} - static int wait_for_engines(struct drm_i915_private *i915) { if (wait_for(intel_engines_are_idle(i915), I915_IDLE_ENGINES_TIMEOUT)) { @@ -3135,9 +950,6 @@ wait_for_timelines(struct drm_i915_private *i915, struct i915_gt_timelines *gt = &i915->gt.timelines; struct i915_timeline *tl; - if (!READ_ONCE(i915->gt.active_requests)) - return timeout; - mutex_lock(>->mutex); list_for_each_entry(tl, >->active_list, link) { struct i915_request *rq; @@ -3177,9 +989,10 @@ wait_for_timelines(struct drm_i915_private *i915, int i915_gem_wait_for_idle(struct drm_i915_private *i915, unsigned int flags, long timeout) { - GEM_TRACE("flags=%x (%s), timeout=%ld%s\n", + GEM_TRACE("flags=%x (%s), timeout=%ld%s, awake?=%s\n", flags, flags & I915_WAIT_LOCKED ? "locked" : "unlocked", - timeout, timeout == MAX_SCHEDULE_TIMEOUT ? " (forever)" : ""); + timeout, timeout == MAX_SCHEDULE_TIMEOUT ? " (forever)" : "", + yesno(i915->gt.awake)); /* If the device is asleep, we have no requests outstanding */ if (!READ_ONCE(i915->gt.awake)) @@ -3204,565 +1017,6 @@ int i915_gem_wait_for_idle(struct drm_i915_private *i915, return 0; } -static void __i915_gem_object_flush_for_display(struct drm_i915_gem_object *obj) -{ - /* - * We manually flush the CPU domain so that we can override and - * force the flush for the display, and perform it asyncrhonously. - */ - flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); - if (obj->cache_dirty) - i915_gem_clflush_object(obj, I915_CLFLUSH_FORCE); - obj->write_domain = 0; -} - -void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj) -{ - if (!READ_ONCE(obj->pin_global)) - return; - - mutex_lock(&obj->base.dev->struct_mutex); - __i915_gem_object_flush_for_display(obj); - mutex_unlock(&obj->base.dev->struct_mutex); -} - -/** - * Moves a single object to the WC read, and possibly write domain. - * @obj: object to act on - * @write: ask for write access or read only - * - * This function returns when the move is complete, including waiting on - * flushes to occur. - */ -int -i915_gem_object_set_to_wc_domain(struct drm_i915_gem_object *obj, bool write) -{ - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED | - (write ? I915_WAIT_ALL : 0), - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - if (obj->write_domain == I915_GEM_DOMAIN_WC) - return 0; - - /* Flush and acquire obj->pages so that we are coherent through - * direct access in memory with previous cached writes through - * shmemfs and that our cache domain tracking remains valid. - * For example, if the obj->filp was moved to swap without us - * being notified and releasing the pages, we would mistakenly - * continue to assume that the obj remained out of the CPU cached - * domain. - */ - ret = i915_gem_object_pin_pages(obj); - if (ret) - return ret; - - flush_write_domain(obj, ~I915_GEM_DOMAIN_WC); - - /* Serialise direct access to this object with the barriers for - * coherent writes from the GPU, by effectively invalidating the - * WC domain upon first access. - */ - if ((obj->read_domains & I915_GEM_DOMAIN_WC) == 0) - mb(); - - /* It should now be out of any other write domains, and we can update - * the domain values for our changes. - */ - GEM_BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_WC) != 0); - obj->read_domains |= I915_GEM_DOMAIN_WC; - if (write) { - obj->read_domains = I915_GEM_DOMAIN_WC; - obj->write_domain = I915_GEM_DOMAIN_WC; - obj->mm.dirty = true; - } - - i915_gem_object_unpin_pages(obj); - return 0; -} - -/** - * Moves a single object to the GTT read, and possibly write domain. - * @obj: object to act on - * @write: ask for write access or read only - * - * This function returns when the move is complete, including waiting on - * flushes to occur. - */ -int -i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) -{ - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED | - (write ? I915_WAIT_ALL : 0), - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - if (obj->write_domain == I915_GEM_DOMAIN_GTT) - return 0; - - /* Flush and acquire obj->pages so that we are coherent through - * direct access in memory with previous cached writes through - * shmemfs and that our cache domain tracking remains valid. - * For example, if the obj->filp was moved to swap without us - * being notified and releasing the pages, we would mistakenly - * continue to assume that the obj remained out of the CPU cached - * domain. - */ - ret = i915_gem_object_pin_pages(obj); - if (ret) - return ret; - - flush_write_domain(obj, ~I915_GEM_DOMAIN_GTT); - - /* 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->read_domains & I915_GEM_DOMAIN_GTT) == 0) - mb(); - - /* It should now be out of any other write domains, and we can update - * the domain values for our changes. - */ - GEM_BUG_ON((obj->write_domain & ~I915_GEM_DOMAIN_GTT) != 0); - obj->read_domains |= I915_GEM_DOMAIN_GTT; - if (write) { - obj->read_domains = I915_GEM_DOMAIN_GTT; - obj->write_domain = I915_GEM_DOMAIN_GTT; - obj->mm.dirty = true; - } - - i915_gem_object_unpin_pages(obj); - return 0; -} - -/** - * Changes the cache-level of an object across all VMA. - * @obj: object to act on - * @cache_level: new cache level to set for the object - * - * After this function returns, the object will be in the new cache-level - * across all GTT and the contents of the backing storage will be coherent, - * with respect to the new cache-level. In order to keep the backing storage - * coherent for all users, we only allow a single cache level to be set - * globally on the object and prevent it from being changed whilst the - * hardware is reading from the object. That is if the object is currently - * on the scanout it will be set to uncached (or equivalent display - * cache coherency) and all non-MOCS GPU access will also be uncached so - * that all direct access to the scanout remains coherent. - */ -int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, - enum i915_cache_level cache_level) -{ - struct i915_vma *vma; - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - if (obj->cache_level == cache_level) - return 0; - - /* Inspect the list of currently bound VMA and unbind any that would - * be invalid given the new cache-level. This is principally to - * catch the issue of the CS prefetch crossing page boundaries and - * reading an invalid PTE on older architectures. - */ -restart: - list_for_each_entry(vma, &obj->vma.list, obj_link) { - if (!drm_mm_node_allocated(&vma->node)) - continue; - - if (i915_vma_is_pinned(vma)) { - DRM_DEBUG("can not change the cache level of pinned objects\n"); - return -EBUSY; - } - - if (!i915_vma_is_closed(vma) && - i915_gem_valid_gtt_space(vma, cache_level)) - continue; - - ret = i915_vma_unbind(vma); - if (ret) - return ret; - - /* As unbinding may affect other elements in the - * obj->vma_list (due to side-effects from retiring - * an active vma), play safe and restart the iterator. - */ - goto restart; - } - - /* We can reuse the existing drm_mm nodes but need to change the - * cache-level on the PTE. We could simply unbind them all and - * rebind with the correct cache-level on next use. However since - * we already have a valid slot, dma mapping, pages etc, we may as - * rewrite the PTE in the belief that doing so tramples upon less - * state and so involves less work. - */ - if (obj->bind_count) { - /* Before we change the PTE, the GPU must not be accessing it. - * If we wait upon the object, we know that all the bound - * VMA are no longer active. - */ - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED | - I915_WAIT_ALL, - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - if (!HAS_LLC(to_i915(obj->base.dev)) && - cache_level != I915_CACHE_NONE) { - /* Access to snoopable pages through the GTT is - * incoherent and on some machines causes a hard - * lockup. Relinquish the CPU mmaping to force - * userspace to refault in the pages and we can - * then double check if the GTT mapping is still - * valid for that pointer access. - */ - i915_gem_release_mmap(obj); - - /* As we no longer need a fence for GTT access, - * we can relinquish it now (and so prevent having - * to steal a fence from someone else on the next - * fence request). Note GPU activity would have - * dropped the fence as all snoopable access is - * supposed to be linear. - */ - for_each_ggtt_vma(vma, obj) { - ret = i915_vma_put_fence(vma); - if (ret) - return ret; - } - } else { - /* We either have incoherent backing store and - * so no GTT access or the architecture is fully - * coherent. In such cases, existing GTT mmaps - * ignore the cache bit in the PTE and we can - * rewrite it without confusing the GPU or having - * to force userspace to fault back in its mmaps. - */ - } - - list_for_each_entry(vma, &obj->vma.list, obj_link) { - if (!drm_mm_node_allocated(&vma->node)) - continue; - - ret = i915_vma_bind(vma, cache_level, PIN_UPDATE); - if (ret) - return ret; - } - } - - list_for_each_entry(vma, &obj->vma.list, obj_link) - vma->node.color = cache_level; - i915_gem_object_set_cache_coherency(obj, cache_level); - obj->cache_dirty = true; /* Always invalidate stale cachelines */ - - return 0; -} - -int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_gem_caching *args = data; - struct drm_i915_gem_object *obj; - int err = 0; - - rcu_read_lock(); - obj = i915_gem_object_lookup_rcu(file, args->handle); - if (!obj) { - err = -ENOENT; - goto out; - } - - 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; - } -out: - rcu_read_unlock(); - return err; -} - -int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_private *i915 = to_i915(dev); - struct drm_i915_gem_caching *args = data; - struct drm_i915_gem_object *obj; - enum i915_cache_level level; - int ret = 0; - - switch (args->caching) { - case I915_CACHING_NONE: - level = I915_CACHE_NONE; - break; - case I915_CACHING_CACHED: - /* - * Due to a HW issue on BXT A stepping, GPU stores via a - * snooped mapping may leave stale data in a corresponding CPU - * cacheline, whereas normally such cachelines would get - * invalidated. - */ - if (!HAS_LLC(i915) && !HAS_SNOOP(i915)) - return -ENODEV; - - level = I915_CACHE_LLC; - break; - case I915_CACHING_DISPLAY: - level = HAS_WT(i915) ? I915_CACHE_WT : I915_CACHE_NONE; - break; - default: - return -EINVAL; - } - - obj = i915_gem_object_lookup(file, args->handle); - if (!obj) - return -ENOENT; - - /* - * The caching mode of proxy object is handled by its generator, and - * not allowed to be changed by userspace. - */ - if (i915_gem_object_is_proxy(obj)) { - ret = -ENXIO; - goto out; - } - - if (obj->cache_level == level) - goto out; - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE, - MAX_SCHEDULE_TIMEOUT); - if (ret) - goto out; - - ret = i915_mutex_lock_interruptible(dev); - if (ret) - goto out; - - ret = i915_gem_object_set_cache_level(obj, level); - mutex_unlock(&dev->struct_mutex); - -out: - i915_gem_object_put(obj); - return ret; -} - -/* - * Prepare buffer for display plane (scanout, cursors, etc). Can be called from - * an uninterruptible phase (modesetting) and allows any flushes to be pipelined - * (for pageflips). We only flush the caches while preparing the buffer for - * display, the callers are responsible for frontbuffer flush. - */ -struct i915_vma * -i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, - u32 alignment, - const struct i915_ggtt_view *view, - unsigned int flags) -{ - struct i915_vma *vma; - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - /* Mark the global pin early so that we account for the - * display coherency whilst setting up the cache domains. - */ - obj->pin_global++; - - /* 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 - * chipsets. - * - * However for gen6+, we could do better by using the GFDT bit instead - * 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, - HAS_WT(to_i915(obj->base.dev)) ? - I915_CACHE_WT : I915_CACHE_NONE); - if (ret) { - vma = ERR_PTR(ret); - goto err_unpin_global; - } - - /* 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. However, - * it may simply be too big to fit into mappable, in which case - * put it anyway and hope that userspace can cope (but always first - * try to preserve the existing ABI). - */ - vma = ERR_PTR(-ENOSPC); - if ((flags & PIN_MAPPABLE) == 0 && - (!view || view->type == I915_GGTT_VIEW_NORMAL)) - vma = i915_gem_object_ggtt_pin(obj, view, 0, alignment, - flags | - PIN_MAPPABLE | - PIN_NONBLOCK); - if (IS_ERR(vma)) - vma = i915_gem_object_ggtt_pin(obj, view, 0, alignment, flags); - if (IS_ERR(vma)) - goto err_unpin_global; - - vma->display_alignment = max_t(u64, vma->display_alignment, alignment); - - __i915_gem_object_flush_for_display(obj); - - /* It should now be out of any other write domains, and we can update - * the domain values for our changes. - */ - obj->read_domains |= I915_GEM_DOMAIN_GTT; - - return vma; - -err_unpin_global: - obj->pin_global--; - return vma; -} - -void -i915_gem_object_unpin_from_display_plane(struct i915_vma *vma) -{ - lockdep_assert_held(&vma->vm->i915->drm.struct_mutex); - - if (WARN_ON(vma->obj->pin_global == 0)) - return; - - if (--vma->obj->pin_global == 0) - vma->display_alignment = I915_GTT_MIN_ALIGNMENT; - - /* Bump the LRU to try and avoid premature eviction whilst flipping */ - i915_gem_object_bump_inactive_ggtt(vma->obj); - - i915_vma_unpin(vma); -} - -/** - * Moves a single object to the CPU read, and possibly write domain. - * @obj: object to act on - * @write: requesting write or read-only access - * - * This function returns when the move is complete, including waiting on - * flushes to occur. - */ -int -i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) -{ - int ret; - - lockdep_assert_held(&obj->base.dev->struct_mutex); - - ret = i915_gem_object_wait(obj, - I915_WAIT_INTERRUPTIBLE | - I915_WAIT_LOCKED | - (write ? I915_WAIT_ALL : 0), - MAX_SCHEDULE_TIMEOUT); - if (ret) - return ret; - - flush_write_domain(obj, ~I915_GEM_DOMAIN_CPU); - - /* Flush the CPU cache if it's still invalid. */ - if ((obj->read_domains & I915_GEM_DOMAIN_CPU) == 0) { - i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC); - obj->read_domains |= I915_GEM_DOMAIN_CPU; - } - - /* It should now be out of any other write domains, and we can update - * the domain values for our changes. - */ - GEM_BUG_ON(obj->write_domain & ~I915_GEM_DOMAIN_CPU); - - /* If we're writing through the CPU, then the GPU read domains will - * need to be invalidated at next use. - */ - if (write) - __start_cpu_write(obj); - - return 0; -} - -/* Throttle our rendering by waiting until the ring has completed our requests - * emitted over 20 msec ago. - * - * Note that if we were to use the current jiffies each time around the loop, - * we wouldn't escape the function with any frames outstanding if the time to - * render a frame was over 20ms. - * - * This should get us reasonable parallelism between CPU and GPU but also - * relatively low latency when blocking on a particular request to finish. - */ -static int -i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) -{ - struct drm_i915_private *dev_priv = to_i915(dev); - struct drm_i915_file_private *file_priv = file->driver_priv; - unsigned long recent_enough = jiffies - DRM_I915_THROTTLE_JIFFIES; - struct i915_request *request, *target = NULL; - long ret; - - /* ABI: return -EIO if already wedged */ - ret = i915_terminally_wedged(dev_priv); - if (ret) - return ret; - - spin_lock(&file_priv->mm.lock); - list_for_each_entry(request, &file_priv->mm.request_list, client_link) { - if (time_after_eq(request->emitted_jiffies, recent_enough)) - break; - - if (target) { - list_del(&target->client_link); - target->file_priv = NULL; - } - - target = request; - } - if (target) - i915_request_get(target); - spin_unlock(&file_priv->mm.lock); - - if (target == NULL) - return 0; - - ret = i915_request_wait(target, - I915_WAIT_INTERRUPTIBLE, - MAX_SCHEDULE_TIMEOUT); - i915_request_put(target); - - return ret < 0 ? ret : 0; -} - struct i915_vma * i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, const struct i915_ggtt_view *view, @@ -3842,146 +1096,11 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, return vma; } -static __always_inline u32 __busy_read_flag(u8 id) -{ - if (id == (u8)I915_ENGINE_CLASS_INVALID) - return 0xffff0000u; - - GEM_BUG_ON(id >= 16); - return 0x10000u << id; -} - -static __always_inline u32 __busy_write_id(u8 id) -{ - /* - * The uABI guarantees an active writer is also amongst the read - * engines. This would be true if we accessed the activity tracking - * under the lock, but as we perform the lookup of the object and - * its activity locklessly we can not guarantee that the last_write - * being active implies that we have set the same engine flag from - * last_read - hence we always set both read and write busy for - * last_write. - */ - if (id == (u8)I915_ENGINE_CLASS_INVALID) - return 0xffffffffu; - - return (id + 1) | __busy_read_flag(id); -} - -static __always_inline unsigned int -__busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u8 id)) -{ - const struct i915_request *rq; - - /* - * We have to check the current hw status of the fence as the uABI - * guarantees forward progress. We could rely on the idle worker - * to eventually flush us, but to minimise latency just ask the - * hardware. - * - * Note we only report on the status of native fences. - */ - if (!dma_fence_is_i915(fence)) - return 0; - - /* opencode to_request() in order to avoid const warnings */ - rq = container_of(fence, const struct i915_request, fence); - if (i915_request_completed(rq)) - return 0; - - /* Beware type-expansion follies! */ - BUILD_BUG_ON(!typecheck(u8, rq->engine->uabi_class)); - return flag(rq->engine->uabi_class); -} - -static __always_inline unsigned int -busy_check_reader(const struct dma_fence *fence) -{ - return __busy_set_if_active(fence, __busy_read_flag); -} - -static __always_inline unsigned int -busy_check_writer(const struct dma_fence *fence) -{ - if (!fence) - return 0; - - return __busy_set_if_active(fence, __busy_write_id); -} - -int -i915_gem_busy_ioctl(struct drm_device *dev, void *data, - struct drm_file *file) -{ - struct drm_i915_gem_busy *args = data; - struct drm_i915_gem_object *obj; - struct reservation_object_list *list; - unsigned int seq; - int err; - - err = -ENOENT; - rcu_read_lock(); - obj = i915_gem_object_lookup_rcu(file, args->handle); - if (!obj) - goto out; - - /* - * A discrepancy here is that we do not report the status of - * non-i915 fences, i.e. even though we may report the object as idle, - * a call to set-domain may still stall waiting for foreign rendering. - * This also means that wait-ioctl may report an object as busy, - * where busy-ioctl considers it idle. - * - * We trade the ability to warn of foreign fences to report on which - * i915 engines are active for the object. - * - * Alternatively, we can trade that extra information on read/write - * activity with - * args->busy = - * !reservation_object_test_signaled_rcu(obj->resv, true); - * to report the overall busyness. This is what the wait-ioctl does. - * - */ -retry: - seq = raw_read_seqcount(&obj->resv->seq); - - /* Translate the exclusive fence to the READ *and* WRITE engine */ - args->busy = busy_check_writer(rcu_dereference(obj->resv->fence_excl)); - - /* Translate shared fences to READ set of engines */ - list = rcu_dereference(obj->resv->fence); - if (list) { - unsigned int shared_count = list->shared_count, i; - - for (i = 0; i < shared_count; ++i) { - struct dma_fence *fence = - rcu_dereference(list->shared[i]); - - args->busy |= busy_check_reader(fence); - } - } - - if (args->busy && read_seqcount_retry(&obj->resv->seq, seq)) - goto retry; - - err = 0; -out: - rcu_read_unlock(); - return err; -} - -int -i915_gem_throttle_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - return i915_gem_ring_throttle(dev, file_priv); -} - int i915_gem_madvise_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { - struct drm_i915_private *dev_priv = to_i915(dev); + struct drm_i915_private *i915 = to_i915(dev); struct drm_i915_gem_madvise *args = data; struct drm_i915_gem_object *obj; int err; @@ -4004,7 +1123,7 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data, if (i915_gem_object_has_pages(obj) && i915_gem_object_is_tiled(obj) && - dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) { + i915->quirks & QUIRK_PIN_SWIZZLED_PAGES) { if (obj->mm.madv == I915_MADV_WILLNEED) { GEM_BUG_ON(!obj->mm.quirked); __i915_gem_object_unpin_pages(obj); @@ -4020,6 +1139,24 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data, if (obj->mm.madv != __I915_MADV_PURGED) obj->mm.madv = args->madv; + if (i915_gem_object_has_pages(obj)) { + struct list_head *list; + + if (i915_gem_object_is_shrinkable(obj)) { + unsigned long flags; + + spin_lock_irqsave(&i915->mm.obj_lock, flags); + + if (obj->mm.madv != I915_MADV_WILLNEED) + list = &i915->mm.purge_list; + else + list = &i915->mm.shrink_list; + list_move_tail(&obj->mm.link, list); + + spin_unlock_irqrestore(&i915->mm.obj_lock, flags); + } + } + /* if the object is no longer attached, discard its backing storage */ if (obj->mm.madv == I915_MADV_DONTNEED && !i915_gem_object_has_pages(obj)) @@ -4033,355 +1170,13 @@ out: return err; } -static void -frontbuffer_retire(struct i915_active_request *active, - struct i915_request *request) -{ - struct drm_i915_gem_object *obj = - container_of(active, typeof(*obj), frontbuffer_write); - - intel_fb_obj_flush(obj, ORIGIN_CS); -} - -void i915_gem_object_init(struct drm_i915_gem_object *obj, - const struct drm_i915_gem_object_ops *ops) -{ - mutex_init(&obj->mm.lock); - - spin_lock_init(&obj->vma.lock); - INIT_LIST_HEAD(&obj->vma.list); - - INIT_LIST_HEAD(&obj->lut_list); - INIT_LIST_HEAD(&obj->batch_pool_link); - - init_rcu_head(&obj->rcu); - - obj->ops = ops; - - reservation_object_init(&obj->__builtin_resv); - obj->resv = &obj->__builtin_resv; - - obj->frontbuffer_ggtt_origin = ORIGIN_GTT; - i915_active_request_init(&obj->frontbuffer_write, - NULL, frontbuffer_retire); - - obj->mm.madv = I915_MADV_WILLNEED; - INIT_RADIX_TREE(&obj->mm.get_page.radix, GFP_KERNEL | __GFP_NOWARN); - mutex_init(&obj->mm.get_page.lock); - - i915_gem_info_add_obj(to_i915(obj->base.dev), obj->base.size); -} - -static const struct drm_i915_gem_object_ops i915_gem_object_ops = { - .flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE | - I915_GEM_OBJECT_IS_SHRINKABLE, - - .get_pages = i915_gem_object_get_pages_gtt, - .put_pages = i915_gem_object_put_pages_gtt, - - .pwrite = i915_gem_object_pwrite_gtt, -}; - -static int i915_gem_object_create_shmem(struct drm_device *dev, - struct drm_gem_object *obj, - size_t size) -{ - struct drm_i915_private *i915 = to_i915(dev); - unsigned long flags = VM_NORESERVE; - struct file *filp; - - drm_gem_private_object_init(dev, obj, size); - - if (i915->mm.gemfs) - filp = shmem_file_setup_with_mnt(i915->mm.gemfs, "i915", size, - flags); - else - filp = shmem_file_setup("i915", size, flags); - - if (IS_ERR(filp)) - return PTR_ERR(filp); - - obj->filp = filp; - - return 0; -} - -struct drm_i915_gem_object * -i915_gem_object_create(struct drm_i915_private *dev_priv, u64 size) -{ - struct drm_i915_gem_object *obj; - struct address_space *mapping; - unsigned int cache_level; - gfp_t mask; - int ret; - - /* There is a prevalence of the assumption that we fit the object's - * page count inside a 32bit _signed_ variable. Let's document this and - * catch if we ever need to fix it. In the meantime, if you do spot - * such a local variable, please consider fixing! - */ - if (size >> PAGE_SHIFT > INT_MAX) - return ERR_PTR(-E2BIG); - - if (overflows_type(size, obj->base.size)) - return ERR_PTR(-E2BIG); - - obj = i915_gem_object_alloc(); - if (obj == NULL) - return ERR_PTR(-ENOMEM); - - ret = i915_gem_object_create_shmem(&dev_priv->drm, &obj->base, size); - if (ret) - goto fail; - - mask = GFP_HIGHUSER | __GFP_RECLAIMABLE; - if (IS_I965GM(dev_priv) || IS_I965G(dev_priv)) { - /* 965gm cannot relocate objects above 4GiB. */ - mask &= ~__GFP_HIGHMEM; - mask |= __GFP_DMA32; - } - - mapping = obj->base.filp->f_mapping; - mapping_set_gfp_mask(mapping, mask); - GEM_BUG_ON(!(mapping_gfp_mask(mapping) & __GFP_RECLAIM)); - - i915_gem_object_init(obj, &i915_gem_object_ops); - - obj->write_domain = I915_GEM_DOMAIN_CPU; - obj->read_domains = I915_GEM_DOMAIN_CPU; - - if (HAS_LLC(dev_priv)) - /* On some devices, we can have the GPU use the LLC (the CPU - * cache) for about a 10% performance improvement - * compared to uncached. Graphics requests other than - * display scanout are coherent with the CPU in - * accessing this cache. This means in this mode we - * don't need to clflush on the CPU side, and on the - * GPU side we only need to flush internal caches to - * get data visible to the CPU. - * - * However, we maintain the display planes as UC, and so - * need to rebind when first used as such. - */ - cache_level = I915_CACHE_LLC; - else - cache_level = I915_CACHE_NONE; - - i915_gem_object_set_cache_coherency(obj, cache_level); - - trace_i915_gem_object_create(obj); - - return obj; - -fail: - i915_gem_object_free(obj); - return ERR_PTR(ret); -} - -static bool discard_backing_storage(struct drm_i915_gem_object *obj) -{ - /* 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. - */ - - if (obj->mm.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 file_count(obj->base.filp) == 1; -} - -static void __i915_gem_free_objects(struct drm_i915_private *i915, - struct llist_node *freed) -{ - struct drm_i915_gem_object *obj, *on; - intel_wakeref_t wakeref; - - wakeref = intel_runtime_pm_get(i915); - llist_for_each_entry_safe(obj, on, freed, freed) { - struct i915_vma *vma, *vn; - - trace_i915_gem_object_destroy(obj); - - mutex_lock(&i915->drm.struct_mutex); - - GEM_BUG_ON(i915_gem_object_is_active(obj)); - list_for_each_entry_safe(vma, vn, &obj->vma.list, obj_link) { - GEM_BUG_ON(i915_vma_is_active(vma)); - vma->flags &= ~I915_VMA_PIN_MASK; - i915_vma_destroy(vma); - } - GEM_BUG_ON(!list_empty(&obj->vma.list)); - GEM_BUG_ON(!RB_EMPTY_ROOT(&obj->vma.tree)); - - /* This serializes freeing with the shrinker. Since the free - * is delayed, first by RCU then by the workqueue, we want the - * shrinker to be able to free pages of unreferenced objects, - * or else we may oom whilst there are plenty of deferred - * freed objects. - */ - if (i915_gem_object_has_pages(obj)) { - spin_lock(&i915->mm.obj_lock); - list_del_init(&obj->mm.link); - spin_unlock(&i915->mm.obj_lock); - } - - mutex_unlock(&i915->drm.struct_mutex); - - GEM_BUG_ON(obj->bind_count); - GEM_BUG_ON(obj->userfault_count); - GEM_BUG_ON(atomic_read(&obj->frontbuffer_bits)); - GEM_BUG_ON(!list_empty(&obj->lut_list)); - - if (obj->ops->release) - obj->ops->release(obj); - - if (WARN_ON(i915_gem_object_has_pinned_pages(obj))) - atomic_set(&obj->mm.pages_pin_count, 0); - __i915_gem_object_put_pages(obj, I915_MM_NORMAL); - GEM_BUG_ON(i915_gem_object_has_pages(obj)); - - if (obj->base.import_attach) - drm_prime_gem_destroy(&obj->base, NULL); - - reservation_object_fini(&obj->__builtin_resv); - drm_gem_object_release(&obj->base); - i915_gem_info_remove_obj(i915, obj->base.size); - - bitmap_free(obj->bit_17); - i915_gem_object_free(obj); - - GEM_BUG_ON(!atomic_read(&i915->mm.free_count)); - atomic_dec(&i915->mm.free_count); - - if (on) - cond_resched(); - } - intel_runtime_pm_put(i915, wakeref); -} - -static void i915_gem_flush_free_objects(struct drm_i915_private *i915) -{ - struct llist_node *freed; - - /* Free the oldest, most stale object to keep the free_list short */ - freed = NULL; - if (!llist_empty(&i915->mm.free_list)) { /* quick test for hotpath */ - /* Only one consumer of llist_del_first() allowed */ - spin_lock(&i915->mm.free_lock); - freed = llist_del_first(&i915->mm.free_list); - spin_unlock(&i915->mm.free_lock); - } - if (unlikely(freed)) { - freed->next = NULL; - __i915_gem_free_objects(i915, freed); - } -} - -static void __i915_gem_free_work(struct work_struct *work) -{ - struct drm_i915_private *i915 = - container_of(work, struct drm_i915_private, mm.free_work); - struct llist_node *freed; - - /* - * All file-owned VMA should have been released by this point through - * i915_gem_close_object(), or earlier by i915_gem_context_close(). - * However, the object may also be bound into the global GTT (e.g. - * older GPUs without per-process support, or for direct access through - * the GTT either for the user or for scanout). Those VMA still need to - * unbound now. - */ - - spin_lock(&i915->mm.free_lock); - while ((freed = llist_del_all(&i915->mm.free_list))) { - spin_unlock(&i915->mm.free_lock); - - __i915_gem_free_objects(i915, freed); - if (need_resched()) - return; - - spin_lock(&i915->mm.free_lock); - } - spin_unlock(&i915->mm.free_lock); -} - -static void __i915_gem_free_object_rcu(struct rcu_head *head) -{ - struct drm_i915_gem_object *obj = - container_of(head, typeof(*obj), rcu); - struct drm_i915_private *i915 = to_i915(obj->base.dev); - - /* - * We reuse obj->rcu for the freed list, so we had better not treat - * it like a rcu_head from this point forwards. And we expect all - * objects to be freed via this path. - */ - destroy_rcu_head(&obj->rcu); - - /* - * Since we require blocking on struct_mutex to unbind the freed - * object from the GPU before releasing resources back to the - * system, we can not do that directly from the RCU callback (which may - * be a softirq context), but must instead then defer that work onto a - * kthread. We use the RCU callback rather than move the freed object - * directly onto the work queue so that we can mix between using the - * worker and performing frees directly from subsequent allocations for - * crude but effective memory throttling. - */ - if (llist_add(&obj->freed, &i915->mm.free_list)) - queue_work(i915->wq, &i915->mm.free_work); -} - -void i915_gem_free_object(struct drm_gem_object *gem_obj) -{ - struct drm_i915_gem_object *obj = to_intel_bo(gem_obj); - - if (obj->mm.quirked) - __i915_gem_object_unpin_pages(obj); - - if (discard_backing_storage(obj)) - obj->mm.madv = I915_MADV_DONTNEED; - - /* - * Before we free the object, make sure any pure RCU-only - * read-side critical sections are complete, e.g. - * i915_gem_busy_ioctl(). For the corresponding synchronized - * lookup see i915_gem_object_lookup_rcu(). - */ - atomic_inc(&to_i915(obj->base.dev)->mm.free_count); - call_rcu(&obj->rcu, __i915_gem_free_object_rcu); -} - -void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj) -{ - lockdep_assert_held(&obj->base.dev->struct_mutex); - - if (!i915_gem_object_has_active_reference(obj) && - i915_gem_object_is_active(obj)) - i915_gem_object_set_active_reference(obj); - else - i915_gem_object_put(obj); -} - void i915_gem_sanitize(struct drm_i915_private *i915) { intel_wakeref_t wakeref; GEM_TRACE("\n"); - wakeref = intel_runtime_pm_get(i915); + wakeref = intel_runtime_pm_get(&i915->runtime_pm); intel_uncore_forcewake_get(&i915->uncore, FORCEWAKE_ALL); /* @@ -4401,141 +1196,10 @@ void i915_gem_sanitize(struct drm_i915_private *i915) * it may impact the display and we are uncertain about the stability * of the reset, so this could be applied to even earlier gen. */ - intel_engines_sanitize(i915, false); - - intel_uncore_forcewake_put(&i915->uncore, FORCEWAKE_ALL); - intel_runtime_pm_put(i915, wakeref); - - mutex_lock(&i915->drm.struct_mutex); - i915_gem_contexts_lost(i915); - mutex_unlock(&i915->drm.struct_mutex); -} - -void i915_gem_suspend(struct drm_i915_private *i915) -{ - intel_wakeref_t wakeref; - - GEM_TRACE("\n"); - - wakeref = intel_runtime_pm_get(i915); - - flush_workqueue(i915->wq); - - mutex_lock(&i915->drm.struct_mutex); - - /* - * We have to flush all the executing contexts to main memory so - * that they can saved in the hibernation image. To ensure the last - * context image is coherent, we have to switch away from it. That - * leaves the i915->kernel_context still active when - * we actually suspend, and its image in memory may not match the GPU - * state. Fortunately, the kernel_context is disposable and we do - * not rely on its state. - */ - switch_to_kernel_context_sync(i915, i915->gt.active_engines); - - mutex_unlock(&i915->drm.struct_mutex); - i915_reset_flush(i915); - - drain_delayed_work(&i915->gt.retire_work); - - /* - * As the idle_work is rearming if it detects a race, play safe and - * repeat the flush until it is definitely idle. - */ - drain_delayed_work(&i915->gt.idle_work); - - /* - * Assert that we successfully flushed all the work and - * reset the GPU back to its idle, low power state. - */ - GEM_BUG_ON(i915->gt.awake); - - intel_uc_suspend(i915); - - intel_runtime_pm_put(i915, wakeref); -} - -void i915_gem_suspend_late(struct drm_i915_private *i915) -{ - struct drm_i915_gem_object *obj; - struct list_head *phases[] = { - &i915->mm.unbound_list, - &i915->mm.bound_list, - NULL - }, **phase; - - /* - * Neither the BIOS, ourselves or any other kernel - * expects the system to be in execlists mode on startup, - * so we need to reset the GPU back to legacy mode. And the only - * known way to disable logical contexts is through a GPU reset. - * - * So in order to leave the system in a known default configuration, - * always reset the GPU upon unload and suspend. Afterwards we then - * clean up the GEM state tracking, flushing off the requests and - * leaving the system in a known idle state. - * - * Note that is of the upmost importance that the GPU is idle and - * all stray writes are flushed *before* we dismantle the backing - * storage for the pinned objects. - * - * However, since we are uncertain that resetting the GPU on older - * machines is a good idea, we don't - just in case it leaves the - * machine in an unusable condition. - */ - - mutex_lock(&i915->drm.struct_mutex); - for (phase = phases; *phase; phase++) { - list_for_each_entry(obj, *phase, mm.link) - WARN_ON(i915_gem_object_set_to_gtt_domain(obj, false)); - } - mutex_unlock(&i915->drm.struct_mutex); + intel_gt_sanitize(i915, false); - intel_uc_sanitize(i915); - i915_gem_sanitize(i915); -} - -void i915_gem_resume(struct drm_i915_private *i915) -{ - GEM_TRACE("\n"); - - WARN_ON(i915->gt.awake); - - mutex_lock(&i915->drm.struct_mutex); - intel_uncore_forcewake_get(&i915->uncore, FORCEWAKE_ALL); - - i915_gem_restore_gtt_mappings(i915); - i915_gem_restore_fences(i915); - - /* - * As we didn't flush the kernel context before suspend, we cannot - * guarantee that the context image is complete. So let's just reset - * it and start again. - */ - intel_gt_resume(i915); - - if (i915_gem_init_hw(i915)) - goto err_wedged; - - intel_uc_resume(i915); - - /* Always reload a context for powersaving. */ - if (!load_power_context(i915)) - goto err_wedged; - -out_unlock: intel_uncore_forcewake_put(&i915->uncore, FORCEWAKE_ALL); - mutex_unlock(&i915->drm.struct_mutex); - return; - -err_wedged: - if (!i915_reset_failed(i915)) { - dev_err(i915->drm.dev, - "Failed to re-initialize GPU, declaring it wedged!\n"); - i915_gem_set_wedged(i915); - } - goto out_unlock; + intel_runtime_pm_put(&i915->runtime_pm, wakeref); } void i915_gem_init_swizzling(struct drm_i915_private *dev_priv) @@ -4586,27 +1250,6 @@ static void init_unused_rings(struct drm_i915_private *dev_priv) } } -static int __i915_gem_restart_engines(void *data) -{ - struct drm_i915_private *i915 = data; - struct intel_engine_cs *engine; - enum intel_engine_id id; - int err; - - for_each_engine(engine, i915, id) { - err = engine->init_hw(engine); - if (err) { - DRM_ERROR("Failed to restart %s (%d)\n", - engine->name, err); - return err; - } - } - - intel_engines_set_scheduler_caps(i915); - - return 0; -} - int i915_gem_init_hw(struct drm_i915_private *dev_priv) { int ret; @@ -4665,12 +1308,13 @@ int i915_gem_init_hw(struct drm_i915_private *dev_priv) intel_mocs_init_l3cc_table(dev_priv); /* Only when the HW is re-initialised, can we replay the requests */ - ret = __i915_gem_restart_engines(dev_priv); + ret = intel_engines_resume(dev_priv); if (ret) goto cleanup_uc; intel_uncore_forcewake_put(&dev_priv->uncore, FORCEWAKE_ALL); + intel_engines_set_scheduler_caps(dev_priv); return 0; cleanup_uc: @@ -4683,8 +1327,9 @@ out: static int __intel_engines_record_defaults(struct drm_i915_private *i915) { - struct i915_gem_context *ctx; struct intel_engine_cs *engine; + struct i915_gem_context *ctx; + struct i915_gem_engines *e; enum intel_engine_id id; int err = 0; @@ -4701,18 +1346,21 @@ static int __intel_engines_record_defaults(struct drm_i915_private *i915) if (IS_ERR(ctx)) return PTR_ERR(ctx); + e = i915_gem_context_lock_engines(ctx); + for_each_engine(engine, i915, id) { + struct intel_context *ce = e->engines[id]; struct i915_request *rq; - rq = i915_request_alloc(engine, ctx); + rq = intel_context_create_request(ce); if (IS_ERR(rq)) { err = PTR_ERR(rq); - goto out_ctx; + goto err_active; } err = 0; - if (engine->init_context) - err = engine->init_context(rq); + if (rq->engine->init_context) + err = rq->engine->init_context(rq); i915_request_add(rq); if (err) @@ -4720,21 +1368,16 @@ static int __intel_engines_record_defaults(struct drm_i915_private *i915) } /* Flush the default context image to memory, and enable powersaving. */ - if (!load_power_context(i915)) { + if (!i915_gem_load_power_context(i915)) { err = -EIO; goto err_active; } for_each_engine(engine, i915, id) { - struct intel_context *ce; - struct i915_vma *state; + struct intel_context *ce = e->engines[id]; + struct i915_vma *state = ce->state; void *vaddr; - ce = intel_context_lookup(ctx, engine); - if (!ce) - continue; - - state = ce->state; if (!state) continue; @@ -4752,7 +1395,9 @@ static int __intel_engines_record_defaults(struct drm_i915_private *i915) if (err) goto err_active; + i915_gem_object_lock(state->obj); err = i915_gem_object_set_to_cpu_domain(state->obj, false); + i915_gem_object_unlock(state->obj); if (err) goto err_active; @@ -4790,6 +1435,7 @@ static int __intel_engines_record_defaults(struct drm_i915_private *i915) } out_ctx: + i915_gem_context_unlock_engines(ctx); i915_gem_context_set_closed(ctx); i915_gem_context_put(ctx); return err; @@ -4842,6 +1488,23 @@ static void i915_gem_fini_scratch(struct drm_i915_private *i915) i915_vma_unpin_and_release(&i915->gt.scratch, 0); } +static int intel_engines_verify_workarounds(struct drm_i915_private *i915) +{ + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err = 0; + + if (!IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) + return 0; + + for_each_engine(engine, i915, id) { + if (intel_engine_verify_workarounds(engine, "load")) + err = -EIO; + } + + return err; +} + int i915_gem_init(struct drm_i915_private *dev_priv) { int ret; @@ -4853,11 +1516,6 @@ int i915_gem_init(struct drm_i915_private *dev_priv) dev_priv->mm.unordered_timeline = dma_fence_context_alloc(1); - if (HAS_LOGICAL_RING_CONTEXTS(dev_priv)) - dev_priv->gt.cleanup_engine = intel_logical_ring_cleanup; - else - dev_priv->gt.cleanup_engine = intel_engine_cleanup; - i915_timelines_init(dev_priv); ret = i915_gem_init_userptr(dev_priv); @@ -4894,6 +1552,12 @@ int i915_gem_init(struct drm_i915_private *dev_priv) goto err_ggtt; } + ret = intel_engines_setup(dev_priv); + if (ret) { + GEM_BUG_ON(ret == -EIO); + goto err_unlock; + } + ret = i915_gem_contexts_init(dev_priv); if (ret) { GEM_BUG_ON(ret == -EIO); @@ -4927,6 +1591,10 @@ int i915_gem_init(struct drm_i915_private *dev_priv) */ intel_init_clock_gating(dev_priv); + ret = intel_engines_verify_workarounds(dev_priv); + if (ret) + goto err_init_hw; + ret = __intel_engines_record_defaults(dev_priv); if (ret) goto err_init_hw; @@ -4955,6 +1623,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv) err_init_hw: mutex_unlock(&dev_priv->drm.struct_mutex); + i915_gem_set_wedged(dev_priv); i915_gem_suspend(dev_priv); i915_gem_suspend_late(dev_priv); @@ -4967,7 +1636,7 @@ err_uc_init: err_pm: if (ret != -EIO) { intel_cleanup_gt_powersave(dev_priv); - i915_gem_cleanup_engines(dev_priv); + intel_engines_cleanup(dev_priv); } err_context: if (ret != -EIO) @@ -5014,8 +1683,12 @@ err_uc_misc: return ret; } -void i915_gem_fini(struct drm_i915_private *dev_priv) +void i915_gem_fini_hw(struct drm_i915_private *dev_priv) { + GEM_BUG_ON(dev_priv->gt.awake); + + intel_wakeref_auto_fini(&dev_priv->ggtt.userfault_wakeref); + i915_gem_suspend_late(dev_priv); intel_disable_gt_powersave(dev_priv); @@ -5025,7 +1698,15 @@ void i915_gem_fini(struct drm_i915_private *dev_priv) mutex_lock(&dev_priv->drm.struct_mutex); intel_uc_fini_hw(dev_priv); intel_uc_fini(dev_priv); - i915_gem_cleanup_engines(dev_priv); + mutex_unlock(&dev_priv->drm.struct_mutex); + + i915_gem_drain_freed_objects(dev_priv); +} + +void i915_gem_fini(struct drm_i915_private *dev_priv) +{ + mutex_lock(&dev_priv->drm.struct_mutex); + intel_engines_cleanup(dev_priv); i915_gem_contexts_fini(dev_priv); i915_gem_fini_scratch(dev_priv); mutex_unlock(&dev_priv->drm.struct_mutex); @@ -5048,77 +1729,32 @@ void i915_gem_init_mmio(struct drm_i915_private *i915) i915_gem_sanitize(i915); } -void -i915_gem_cleanup_engines(struct drm_i915_private *dev_priv) -{ - struct intel_engine_cs *engine; - enum intel_engine_id id; - - for_each_engine(engine, dev_priv, id) - dev_priv->gt.cleanup_engine(engine); -} - -void -i915_gem_load_init_fences(struct drm_i915_private *dev_priv) -{ - int i; - - if (INTEL_GEN(dev_priv) >= 7 && !IS_VALLEYVIEW(dev_priv) && - !IS_CHERRYVIEW(dev_priv)) - dev_priv->num_fence_regs = 32; - else if (INTEL_GEN(dev_priv) >= 4 || - IS_I945G(dev_priv) || IS_I945GM(dev_priv) || - IS_G33(dev_priv) || IS_PINEVIEW(dev_priv)) - dev_priv->num_fence_regs = 16; - else - dev_priv->num_fence_regs = 8; - - if (intel_vgpu_active(dev_priv)) - dev_priv->num_fence_regs = - I915_READ(vgtif_reg(avail_rs.fence_num)); - - /* Initialize fence registers to zero */ - for (i = 0; i < dev_priv->num_fence_regs; i++) { - struct drm_i915_fence_reg *fence = &dev_priv->fence_regs[i]; - - fence->i915 = dev_priv; - fence->id = i; - list_add_tail(&fence->link, &dev_priv->mm.fence_list); - } - i915_gem_restore_fences(dev_priv); - - i915_gem_detect_bit_6_swizzle(dev_priv); -} - static void i915_gem_init__mm(struct drm_i915_private *i915) { - spin_lock_init(&i915->mm.object_stat_lock); spin_lock_init(&i915->mm.obj_lock); spin_lock_init(&i915->mm.free_lock); init_llist_head(&i915->mm.free_list); - INIT_LIST_HEAD(&i915->mm.unbound_list); - INIT_LIST_HEAD(&i915->mm.bound_list); - INIT_LIST_HEAD(&i915->mm.fence_list); - INIT_LIST_HEAD(&i915->mm.userfault_list); + INIT_LIST_HEAD(&i915->mm.purge_list); + INIT_LIST_HEAD(&i915->mm.shrink_list); - INIT_WORK(&i915->mm.free_work, __i915_gem_free_work); + i915_gem_init__objects(i915); } int i915_gem_init_early(struct drm_i915_private *dev_priv) { int err; + intel_gt_pm_init(dev_priv); + INIT_LIST_HEAD(&dev_priv->gt.active_rings); INIT_LIST_HEAD(&dev_priv->gt.closed_vma); + spin_lock_init(&dev_priv->gt.closed_lock); i915_gem_init__mm(dev_priv); + i915_gem_init__pm(dev_priv); - INIT_DELAYED_WORK(&dev_priv->gt.retire_work, - i915_gem_retire_work_handler); - INIT_DELAYED_WORK(&dev_priv->gt.idle_work, - i915_gem_idle_work_handler); init_waitqueue_head(&dev_priv->gpu_error.wait_queue); init_waitqueue_head(&dev_priv->gpu_error.reset_queue); mutex_init(&dev_priv->gpu_error.wedge_mutex); @@ -5140,7 +1776,7 @@ void i915_gem_cleanup_early(struct drm_i915_private *dev_priv) i915_gem_drain_freed_objects(dev_priv); GEM_BUG_ON(!llist_empty(&dev_priv->mm.free_list)); GEM_BUG_ON(atomic_read(&dev_priv->mm.free_count)); - WARN_ON(dev_priv->mm.object_count); + WARN_ON(dev_priv->mm.shrink_count); cleanup_srcu_struct(&dev_priv->gpu_error.reset_backoff_srcu); @@ -5160,11 +1796,7 @@ int i915_gem_freeze(struct drm_i915_private *dev_priv) int i915_gem_freeze_late(struct drm_i915_private *i915) { struct drm_i915_gem_object *obj; - struct list_head *phases[] = { - &i915->mm.unbound_list, - &i915->mm.bound_list, - NULL - }, **phase; + intel_wakeref_t wakeref; /* * Called just before we write the hibernation image. @@ -5181,15 +1813,18 @@ int i915_gem_freeze_late(struct drm_i915_private *i915) * the objects as well, see i915_gem_freeze() */ - i915_gem_shrink(i915, -1UL, NULL, I915_SHRINK_UNBOUND); + wakeref = intel_runtime_pm_get(&i915->runtime_pm); + + i915_gem_shrink(i915, -1UL, NULL, ~0); i915_gem_drain_freed_objects(i915); - mutex_lock(&i915->drm.struct_mutex); - for (phase = phases; *phase; phase++) { - list_for_each_entry(obj, *phase, mm.link) - WARN_ON(i915_gem_object_set_to_cpu_domain(obj, true)); + list_for_each_entry(obj, &i915->mm.shrink_list, mm.link) { + i915_gem_object_lock(obj); + WARN_ON(i915_gem_object_set_to_cpu_domain(obj, true)); + i915_gem_object_unlock(obj); } - mutex_unlock(&i915->drm.struct_mutex); + + intel_runtime_pm_put(&i915->runtime_pm, wakeref); return 0; } @@ -5270,276 +1905,7 @@ void i915_gem_track_fb(struct drm_i915_gem_object *old, } } -/* Allocate a new GEM object and fill it with the supplied data */ -struct drm_i915_gem_object * -i915_gem_object_create_from_data(struct drm_i915_private *dev_priv, - const void *data, size_t size) -{ - struct drm_i915_gem_object *obj; - struct file *file; - size_t offset; - int err; - - obj = i915_gem_object_create(dev_priv, round_up(size, PAGE_SIZE)); - if (IS_ERR(obj)) - return obj; - - GEM_BUG_ON(obj->write_domain != I915_GEM_DOMAIN_CPU); - - file = obj->base.filp; - offset = 0; - do { - unsigned int len = min_t(typeof(size), size, PAGE_SIZE); - struct page *page; - void *pgdata, *vaddr; - - err = pagecache_write_begin(file, file->f_mapping, - offset, len, 0, - &page, &pgdata); - if (err < 0) - goto fail; - - vaddr = kmap(page); - memcpy(vaddr, data, len); - kunmap(page); - - err = pagecache_write_end(file, file->f_mapping, - offset, len, len, - page, pgdata); - if (err < 0) - goto fail; - - size -= len; - data += len; - offset += len; - } while (size); - - return obj; - -fail: - i915_gem_object_put(obj); - return ERR_PTR(err); -} - -struct scatterlist * -i915_gem_object_get_sg(struct drm_i915_gem_object *obj, - unsigned int n, - unsigned int *offset) -{ - struct i915_gem_object_page_iter *iter = &obj->mm.get_page; - struct scatterlist *sg; - unsigned int idx, count; - - might_sleep(); - GEM_BUG_ON(n >= obj->base.size >> PAGE_SHIFT); - GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj)); - - /* As we iterate forward through the sg, we record each entry in a - * radixtree for quick repeated (backwards) lookups. If we have seen - * this index previously, we will have an entry for it. - * - * Initial lookup is O(N), but this is amortized to O(1) for - * sequential page access (where each new request is consecutive - * to the previous one). Repeated lookups are O(lg(obj->base.size)), - * i.e. O(1) with a large constant! - */ - if (n < READ_ONCE(iter->sg_idx)) - goto lookup; - - mutex_lock(&iter->lock); - - /* We prefer to reuse the last sg so that repeated lookup of this - * (or the subsequent) sg are fast - comparing against the last - * sg is faster than going through the radixtree. - */ - - sg = iter->sg_pos; - idx = iter->sg_idx; - count = __sg_page_count(sg); - - while (idx + count <= n) { - void *entry; - unsigned long i; - int ret; - - /* If we cannot allocate and insert this entry, or the - * individual pages from this range, cancel updating the - * sg_idx so that on this lookup we are forced to linearly - * scan onwards, but on future lookups we will try the - * insertion again (in which case we need to be careful of - * the error return reporting that we have already inserted - * this index). - */ - ret = radix_tree_insert(&iter->radix, idx, sg); - if (ret && ret != -EEXIST) - goto scan; - - entry = xa_mk_value(idx); - for (i = 1; i < count; i++) { - ret = radix_tree_insert(&iter->radix, idx + i, entry); - if (ret && ret != -EEXIST) - goto scan; - } - - idx += count; - sg = ____sg_next(sg); - count = __sg_page_count(sg); - } - -scan: - iter->sg_pos = sg; - iter->sg_idx = idx; - - mutex_unlock(&iter->lock); - - if (unlikely(n < idx)) /* insertion completed by another thread */ - goto lookup; - - /* In case we failed to insert the entry into the radixtree, we need - * to look beyond the current sg. - */ - while (idx + count <= n) { - idx += count; - sg = ____sg_next(sg); - count = __sg_page_count(sg); - } - - *offset = n - idx; - return sg; - -lookup: - rcu_read_lock(); - - sg = radix_tree_lookup(&iter->radix, n); - GEM_BUG_ON(!sg); - - /* If this index is in the middle of multi-page sg entry, - * the radix tree will contain a value entry that points - * to the start of that range. We will return the pointer to - * the base page and the offset of this page within the - * sg entry's range. - */ - *offset = 0; - if (unlikely(xa_is_value(sg))) { - unsigned long base = xa_to_value(sg); - - sg = radix_tree_lookup(&iter->radix, base); - GEM_BUG_ON(!sg); - - *offset = n - base; - } - - rcu_read_unlock(); - - return sg; -} - -struct page * -i915_gem_object_get_page(struct drm_i915_gem_object *obj, unsigned int n) -{ - struct scatterlist *sg; - unsigned int offset; - - GEM_BUG_ON(!i915_gem_object_has_struct_page(obj)); - - sg = i915_gem_object_get_sg(obj, n, &offset); - return nth_page(sg_page(sg), offset); -} - -/* Like i915_gem_object_get_page(), but mark the returned page dirty */ -struct page * -i915_gem_object_get_dirty_page(struct drm_i915_gem_object *obj, - unsigned int n) -{ - struct page *page; - - page = i915_gem_object_get_page(obj, n); - if (!obj->mm.dirty) - set_page_dirty(page); - - return page; -} - -dma_addr_t -i915_gem_object_get_dma_address(struct drm_i915_gem_object *obj, - unsigned long n) -{ - struct scatterlist *sg; - unsigned int offset; - - sg = i915_gem_object_get_sg(obj, n, &offset); - return sg_dma_address(sg) + (offset << PAGE_SHIFT); -} - -int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align) -{ - struct sg_table *pages; - int err; - - if (align > obj->base.size) - return -EINVAL; - - if (obj->ops == &i915_gem_phys_ops) - return 0; - - if (obj->ops != &i915_gem_object_ops) - return -EINVAL; - - err = i915_gem_object_unbind(obj); - if (err) - return err; - - mutex_lock(&obj->mm.lock); - - if (obj->mm.madv != I915_MADV_WILLNEED) { - err = -EFAULT; - goto err_unlock; - } - - if (obj->mm.quirked) { - err = -EFAULT; - goto err_unlock; - } - - if (obj->mm.mapping) { - err = -EBUSY; - goto err_unlock; - } - - pages = __i915_gem_object_unset_pages(obj); - - obj->ops = &i915_gem_phys_ops; - - err = ____i915_gem_object_get_pages(obj); - if (err) - goto err_xfer; - - /* Perma-pin (until release) the physical set of pages */ - __i915_gem_object_pin_pages(obj); - - if (!IS_ERR_OR_NULL(pages)) - i915_gem_object_ops.put_pages(obj, pages); - mutex_unlock(&obj->mm.lock); - return 0; - -err_xfer: - obj->ops = &i915_gem_object_ops; - if (!IS_ERR_OR_NULL(pages)) { - unsigned int sg_page_sizes = i915_sg_page_sizes(pages->sgl); - - __i915_gem_object_set_pages(obj, pages, sg_page_sizes); - } -err_unlock: - mutex_unlock(&obj->mm.lock); - return err; -} - #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) -#include "selftests/scatterlist.c" #include "selftests/mock_gem_device.c" -#include "selftests/huge_gem_object.c" -#include "selftests/huge_pages.c" -#include "selftests/i915_gem_object.c" -#include "selftests/i915_gem_coherency.c" #include "selftests/i915_gem.c" #endif |