diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_gem.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 3437 |
1 files changed, 1312 insertions, 2125 deletions
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index a77ce9983f69..947e82c2b175 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -29,10 +29,13 @@ #include <drm/drm_vma_manager.h> #include <drm/i915_drm.h> #include "i915_drv.h" +#include "i915_gem_dmabuf.h" #include "i915_vgpu.h" #include "i915_trace.h" #include "intel_drv.h" +#include "intel_frontbuffer.h" #include "intel_mocs.h" +#include <linux/reservation.h> #include <linux/shmem_fs.h> #include <linux/slab.h> #include <linux/swap.h> @@ -41,10 +44,6 @@ static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj); static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj); -static void -i915_gem_object_retire__write(struct drm_i915_gem_object *obj); -static void -i915_gem_object_retire__read(struct drm_i915_gem_object *obj, int ring); static bool cpu_cache_is_coherent(struct drm_device *dev, enum i915_cache_level level) @@ -139,7 +138,6 @@ int i915_mutex_lock_interruptible(struct drm_device *dev) if (ret) return ret; - WARN_ON(i915_verify_lists(dev)); return 0; } @@ -156,10 +154,10 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data, pinned = 0; mutex_lock(&dev->struct_mutex); list_for_each_entry(vma, &ggtt->base.active_list, vm_link) - if (vma->pin_count) + if (i915_vma_is_pinned(vma)) pinned += vma->node.size; list_for_each_entry(vma, &ggtt->base.inactive_list, vm_link) - if (vma->pin_count) + if (i915_vma_is_pinned(vma)) pinned += vma->node.size; mutex_unlock(&dev->struct_mutex); @@ -281,23 +279,129 @@ static const struct drm_i915_gem_object_ops i915_gem_phys_ops = { .release = i915_gem_object_release_phys, }; -static int -drop_pages(struct drm_i915_gem_object *obj) +int i915_gem_object_unbind(struct drm_i915_gem_object *obj) { - struct i915_vma *vma, *next; + struct i915_vma *vma; + LIST_HEAD(still_in_list); int ret; - drm_gem_object_reference(&obj->base); - list_for_each_entry_safe(vma, next, &obj->vma_list, obj_link) - if (i915_vma_unbind(vma)) - break; + lockdep_assert_held(&obj->base.dev->struct_mutex); - ret = i915_gem_object_put_pages(obj); - drm_gem_object_unreference(&obj->base); + /* 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_wait_rendering(obj, false); + if (ret) + return ret; + + i915_gem_retire_requests(to_i915(obj->base.dev)); + + while ((vma = list_first_entry_or_null(&obj->vma_list, + struct i915_vma, + obj_link))) { + list_move_tail(&vma->obj_link, &still_in_list); + ret = i915_vma_unbind(vma); + if (ret) + break; + } + list_splice(&still_in_list, &obj->vma_list); return ret; } +/** + * Ensures that all rendering to the object has completed and the object is + * safe to unbind from the GTT or access from the CPU. + * @obj: i915 gem object + * @readonly: waiting for just read access or read-write access + */ +int +i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, + bool readonly) +{ + struct reservation_object *resv; + struct i915_gem_active *active; + unsigned long active_mask; + int idx; + + lockdep_assert_held(&obj->base.dev->struct_mutex); + + if (!readonly) { + active = obj->last_read; + active_mask = i915_gem_object_get_active(obj); + } else { + active_mask = 1; + active = &obj->last_write; + } + + for_each_active(active_mask, idx) { + int ret; + + ret = i915_gem_active_wait(&active[idx], + &obj->base.dev->struct_mutex); + if (ret) + return ret; + } + + resv = i915_gem_object_get_dmabuf_resv(obj); + if (resv) { + long err; + + err = reservation_object_wait_timeout_rcu(resv, !readonly, true, + MAX_SCHEDULE_TIMEOUT); + if (err < 0) + return err; + } + + return 0; +} + +/* A nonblocking variant of the above wait. Must be called prior to + * acquiring the mutex for the object, as the object state may change + * during this call. A reference must be held by the caller for the object. + */ +static __must_check int +__unsafe_wait_rendering(struct drm_i915_gem_object *obj, + struct intel_rps_client *rps, + bool readonly) +{ + struct i915_gem_active *active; + unsigned long active_mask; + int idx; + + active_mask = __I915_BO_ACTIVE(obj); + if (!active_mask) + return 0; + + if (!readonly) { + active = obj->last_read; + } else { + active_mask = 1; + active = &obj->last_write; + } + + for_each_active(active_mask, idx) { + int ret; + + ret = i915_gem_active_wait_unlocked(&active[idx], + I915_WAIT_INTERRUPTIBLE, + NULL, rps); + if (ret) + return ret; + } + + return 0; +} + +static struct intel_rps_client *to_rps_client(struct drm_file *file) +{ + struct drm_i915_file_private *fpriv = file->driver_priv; + + return &fpriv->rps; +} + int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align) @@ -318,7 +422,11 @@ i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, if (obj->base.filp == NULL) return -EINVAL; - ret = drop_pages(obj); + ret = i915_gem_object_unbind(obj); + if (ret) + return ret; + + ret = i915_gem_object_put_pages(obj); if (ret) return ret; @@ -408,7 +516,7 @@ i915_gem_create(struct drm_file *file, ret = drm_gem_handle_create(file, &obj->base, &handle); /* drop reference from allocate - handle holds it now */ - drm_gem_object_unreference_unlocked(&obj->base); + i915_gem_object_put_unlocked(obj); if (ret) return ret; @@ -502,33 +610,106 @@ __copy_from_user_swizzled(char *gpu_vaddr, int gpu_offset, * flush the object from the CPU cache. */ int i915_gem_obj_prepare_shmem_read(struct drm_i915_gem_object *obj, - int *needs_clflush) + unsigned int *needs_clflush) { int ret; *needs_clflush = 0; - if (WARN_ON(!i915_gem_object_has_struct_page(obj))) - return -EINVAL; + if (!i915_gem_object_has_struct_page(obj)) + return -ENODEV; + + ret = i915_gem_object_wait_rendering(obj, true); + if (ret) + return ret; + + ret = i915_gem_object_get_pages(obj); + if (ret) + return ret; + + i915_gem_object_pin_pages(obj); + + i915_gem_object_flush_gtt_write_domain(obj); - if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) { - /* If we're not in the cpu read domain, set ourself into the gtt - * read domain and manually flush cachelines (if required). This - * optimizes for the case when the gpu will dirty the data - * anyway again before the next pread happens. */ + /* If we're not in the cpu read domain, set ourself into the gtt + * read domain and manually flush cachelines (if required). This + * optimizes for the case when the gpu will dirty the data + * anyway again before the next pread happens. + */ + if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) *needs_clflush = !cpu_cache_is_coherent(obj->base.dev, obj->cache_level); - ret = i915_gem_object_wait_rendering(obj, true); + + if (*needs_clflush && !static_cpu_has(X86_FEATURE_CLFLUSH)) { + ret = i915_gem_object_set_to_cpu_domain(obj, false); if (ret) - return ret; + goto err_unpin; + + *needs_clflush = 0; } + /* 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; + + *needs_clflush = 0; + if (!i915_gem_object_has_struct_page(obj)) + return -ENODEV; + + ret = i915_gem_object_wait_rendering(obj, false); + if (ret) + return ret; + ret = i915_gem_object_get_pages(obj); if (ret) return ret; i915_gem_object_pin_pages(obj); + i915_gem_object_flush_gtt_write_domain(obj); + + /* If we're not in the cpu write domain, set ourself into the + * gtt write domain and manually flush cachelines (as required). + * This optimizes for the case when the gpu will use the data + * right away and we therefore have to clflush anyway. + */ + if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) + *needs_clflush |= cpu_write_needs_clflush(obj) << 1; + + /* Same trick applies to invalidate partially written cachelines read + * before writing. + */ + if (!(obj->base.read_domains & I915_GEM_DOMAIN_CPU)) + *needs_clflush |= !cpu_cache_is_coherent(obj->base.dev, + obj->cache_level); + + if (*needs_clflush && !static_cpu_has(X86_FEATURE_CLFLUSH)) { + ret = i915_gem_object_set_to_cpu_domain(obj, true); + if (ret) + goto err_unpin; + + *needs_clflush = 0; + } + + if ((*needs_clflush & CLFLUSH_AFTER) == 0) + obj->cache_dirty = true; + + intel_fb_obj_invalidate(obj, ORIGIN_CPU); + obj->dirty = 1; + /* return with the pages pinned */ + return 0; + +err_unpin: + i915_gem_object_unpin_pages(obj); return ret; } @@ -638,14 +819,24 @@ i915_gem_gtt_pread(struct drm_device *dev, { struct drm_i915_private *dev_priv = to_i915(dev); struct i915_ggtt *ggtt = &dev_priv->ggtt; + struct i915_vma *vma; struct drm_mm_node node; char __user *user_data; uint64_t remain; uint64_t offset; int ret; - ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE); - if (ret) { + vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE); + if (!IS_ERR(vma)) { + node.start = i915_ggtt_offset(vma); + node.allocated = false; + ret = i915_vma_put_fence(vma); + if (ret) { + i915_vma_unpin(vma); + vma = ERR_PTR(ret); + } + } + if (IS_ERR(vma)) { ret = insert_mappable_node(dev_priv, &node, PAGE_SIZE); if (ret) goto out; @@ -657,12 +848,6 @@ i915_gem_gtt_pread(struct drm_device *dev, } i915_gem_object_pin_pages(obj); - } else { - node.start = i915_gem_obj_ggtt_offset(obj); - node.allocated = false; - ret = i915_gem_object_put_fence(obj); - if (ret) - goto out_unpin; } ret = i915_gem_object_set_to_gtt_domain(obj, false); @@ -675,7 +860,7 @@ i915_gem_gtt_pread(struct drm_device *dev, mutex_unlock(&dev->struct_mutex); if (likely(!i915.prefault_disable)) { - ret = fault_in_multipages_writeable(user_data, remain); + ret = fault_in_pages_writeable(user_data, remain); if (ret) { mutex_lock(&dev->struct_mutex); goto out_unpin; @@ -707,7 +892,7 @@ i915_gem_gtt_pread(struct drm_device *dev, * and write to user memory which may result into page * faults, and so we cannot perform this under struct_mutex. */ - if (slow_user_access(ggtt->mappable, page_base, + if (slow_user_access(&ggtt->mappable, page_base, page_offset, user_data, page_length, false)) { ret = -EFAULT; @@ -739,7 +924,7 @@ out_unpin: i915_gem_object_unpin_pages(obj); remove_mappable_node(&node); } else { - i915_gem_object_ggtt_unpin(obj); + i915_vma_unpin(vma); } out: return ret; @@ -760,19 +945,14 @@ i915_gem_shmem_pread(struct drm_device *dev, int needs_clflush = 0; struct sg_page_iter sg_iter; - if (!i915_gem_object_has_struct_page(obj)) - return -ENODEV; - - user_data = u64_to_user_ptr(args->data_ptr); - remain = args->size; - - obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); - ret = i915_gem_obj_prepare_shmem_read(obj, &needs_clflush); if (ret) return ret; + obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); + user_data = u64_to_user_ptr(args->data_ptr); offset = args->offset; + remain = args->size; for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, offset >> PAGE_SHIFT) { @@ -803,7 +983,7 @@ i915_gem_shmem_pread(struct drm_device *dev, mutex_unlock(&dev->struct_mutex); if (likely(!i915.prefault_disable) && !prefaulted) { - ret = fault_in_multipages_writeable(user_data, remain); + ret = fault_in_pages_writeable(user_data, remain); /* Userspace is tricking us, but we've already clobbered * its pages with the prefault and promised to write the * data up to the first fault. Hence ignore any errors @@ -828,7 +1008,7 @@ next_page: } out: - i915_gem_object_unpin_pages(obj); + i915_gem_obj_finish_shmem_access(obj); return ret; } @@ -857,25 +1037,27 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data, args->size)) return -EFAULT; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; - - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; /* Bounds check source. */ if (args->offset > obj->base.size || args->size > obj->base.size - args->offset) { ret = -EINVAL; - goto out; + goto err; } trace_i915_gem_object_pread(obj, args->offset, args->size); + ret = __unsafe_wait_rendering(obj, to_rps_client(file), true); + if (ret) + goto err; + + ret = i915_mutex_lock_interruptible(dev); + if (ret) + goto err; + ret = i915_gem_shmem_pread(dev, obj, args, file); /* pread for non shmem backed objects */ @@ -886,10 +1068,13 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data, intel_runtime_pm_put(to_i915(dev)); } -out: - drm_gem_object_unreference(&obj->base); -unlock: + i915_gem_object_put(obj); mutex_unlock(&dev->struct_mutex); + + return ret; + +err: + i915_gem_object_put_unlocked(obj); return ret; } @@ -919,7 +1104,7 @@ fast_user_write(struct io_mapping *mapping, /** * This is the fast pwrite path, where we copy the data directly from the * user into the GTT, uncached. - * @dev: drm device pointer + * @i915: i915 device private data * @obj: i915 gem object * @args: pwrite arguments structure * @file: drm file pointer @@ -932,17 +1117,28 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915, { struct i915_ggtt *ggtt = &i915->ggtt; struct drm_device *dev = obj->base.dev; + struct i915_vma *vma; struct drm_mm_node node; uint64_t remain, offset; char __user *user_data; int ret; bool hit_slow_path = false; - if (obj->tiling_mode != I915_TILING_NONE) + if (i915_gem_object_is_tiled(obj)) return -EFAULT; - ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE | PIN_NONBLOCK); - if (ret) { + vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, + PIN_MAPPABLE | PIN_NONBLOCK); + if (!IS_ERR(vma)) { + node.start = i915_ggtt_offset(vma); + node.allocated = false; + ret = i915_vma_put_fence(vma); + if (ret) { + i915_vma_unpin(vma); + vma = ERR_PTR(ret); + } + } + if (IS_ERR(vma)) { ret = insert_mappable_node(i915, &node, PAGE_SIZE); if (ret) goto out; @@ -954,19 +1150,13 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915, } i915_gem_object_pin_pages(obj); - } else { - node.start = i915_gem_obj_ggtt_offset(obj); - node.allocated = false; - ret = i915_gem_object_put_fence(obj); - if (ret) - goto out_unpin; } ret = i915_gem_object_set_to_gtt_domain(obj, true); if (ret) goto out_unpin; - intel_fb_obj_invalidate(obj, ORIGIN_GTT); + intel_fb_obj_invalidate(obj, ORIGIN_CPU); obj->dirty = true; user_data = u64_to_user_ptr(args->data_ptr); @@ -998,11 +1188,11 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915, * If the object is non-shmem backed, we retry again with the * path that handles page fault. */ - if (fast_user_write(ggtt->mappable, page_base, + if (fast_user_write(&ggtt->mappable, page_base, page_offset, user_data, page_length)) { hit_slow_path = true; mutex_unlock(&dev->struct_mutex); - if (slow_user_access(ggtt->mappable, + if (slow_user_access(&ggtt->mappable, page_base, page_offset, user_data, page_length, true)) { @@ -1033,7 +1223,7 @@ out_flush: } } - intel_fb_obj_flush(obj, false, ORIGIN_GTT); + intel_fb_obj_flush(obj, false, ORIGIN_CPU); out_unpin: if (node.allocated) { wmb(); @@ -1043,7 +1233,7 @@ out_unpin: i915_gem_object_unpin_pages(obj); remove_mappable_node(&node); } else { - i915_gem_object_ggtt_unpin(obj); + i915_vma_unpin(vma); } out: return ret; @@ -1126,41 +1316,17 @@ i915_gem_shmem_pwrite(struct drm_device *dev, int shmem_page_offset, page_length, ret = 0; int obj_do_bit17_swizzling, page_do_bit17_swizzling; int hit_slowpath = 0; - int needs_clflush_after = 0; - int needs_clflush_before = 0; + unsigned int needs_clflush; struct sg_page_iter sg_iter; - user_data = u64_to_user_ptr(args->data_ptr); - remain = args->size; - - obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); - - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) { - /* If we're not in the cpu write domain, set ourself into the gtt - * write domain and manually flush cachelines (if required). This - * optimizes for the case when the gpu will use the data - * right away and we therefore have to clflush anyway. */ - needs_clflush_after = cpu_write_needs_clflush(obj); - ret = i915_gem_object_wait_rendering(obj, false); - if (ret) - return ret; - } - /* Same trick applies to invalidate partially written cachelines read - * before writing. */ - if ((obj->base.read_domains & I915_GEM_DOMAIN_CPU) == 0) - needs_clflush_before = - !cpu_cache_is_coherent(dev, obj->cache_level); - - ret = i915_gem_object_get_pages(obj); + ret = i915_gem_obj_prepare_shmem_write(obj, &needs_clflush); if (ret) return ret; - intel_fb_obj_invalidate(obj, ORIGIN_CPU); - - i915_gem_object_pin_pages(obj); - + obj_do_bit17_swizzling = i915_gem_object_needs_bit17_swizzle(obj); + user_data = u64_to_user_ptr(args->data_ptr); offset = args->offset; - obj->dirty = 1; + remain = args->size; for_each_sg_page(obj->pages->sgl, &sg_iter, obj->pages->nents, offset >> PAGE_SHIFT) { @@ -1184,7 +1350,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev, /* If we don't overwrite a cacheline completely we need to be * careful to have up-to-date data by first clflushing. Don't * overcomplicate things and flush the entire patch. */ - partial_cacheline_write = needs_clflush_before && + partial_cacheline_write = needs_clflush & CLFLUSH_BEFORE && ((shmem_page_offset | page_length) & (boot_cpu_data.x86_clflush_size - 1)); @@ -1194,7 +1360,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev, ret = shmem_pwrite_fast(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, partial_cacheline_write, - needs_clflush_after); + needs_clflush & CLFLUSH_AFTER); if (ret == 0) goto next_page; @@ -1203,7 +1369,7 @@ i915_gem_shmem_pwrite(struct drm_device *dev, ret = shmem_pwrite_slow(page, shmem_page_offset, page_length, user_data, page_do_bit17_swizzling, partial_cacheline_write, - needs_clflush_after); + needs_clflush & CLFLUSH_AFTER); mutex_lock(&dev->struct_mutex); @@ -1217,7 +1383,7 @@ next_page: } out: - i915_gem_object_unpin_pages(obj); + i915_gem_obj_finish_shmem_access(obj); if (hit_slowpath) { /* @@ -1225,17 +1391,15 @@ out: * cachelines in-line while writing and the object moved * out of the cpu write domain while we've dropped the lock. */ - if (!needs_clflush_after && + if (!(needs_clflush & CLFLUSH_AFTER) && obj->base.write_domain != I915_GEM_DOMAIN_CPU) { if (i915_gem_clflush_object(obj, obj->pin_display)) - needs_clflush_after = true; + needs_clflush |= CLFLUSH_AFTER; } } - if (needs_clflush_after) + if (needs_clflush & CLFLUSH_AFTER) i915_gem_chipset_flush(to_i915(dev)); - else - obj->cache_dirty = true; intel_fb_obj_flush(obj, false, ORIGIN_CPU); return ret; @@ -1267,33 +1431,35 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, return -EFAULT; if (likely(!i915.prefault_disable)) { - ret = fault_in_multipages_readable(u64_to_user_ptr(args->data_ptr), + ret = fault_in_pages_readable(u64_to_user_ptr(args->data_ptr), args->size); if (ret) return -EFAULT; } - intel_runtime_pm_get(dev_priv); - - ret = i915_mutex_lock_interruptible(dev); - if (ret) - goto put_rpm; - - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; /* Bounds check destination. */ if (args->offset > obj->base.size || args->size > obj->base.size - args->offset) { ret = -EINVAL; - goto out; + goto err; } trace_i915_gem_object_pwrite(obj, args->offset, args->size); + ret = __unsafe_wait_rendering(obj, to_rps_client(file), false); + if (ret) + goto err; + + intel_runtime_pm_get(dev_priv); + + ret = i915_mutex_lock_interruptible(dev); + if (ret) + goto err_rpm; + ret = -EFAULT; /* We can only do the GTT pwrite on untiled buffers, as otherwise * it would end up going through the fenced access, and we'll get @@ -1312,505 +1478,28 @@ i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, if (ret == -EFAULT || ret == -ENOSPC) { if (obj->phys_handle) ret = i915_gem_phys_pwrite(obj, args, file); - else if (i915_gem_object_has_struct_page(obj)) - ret = i915_gem_shmem_pwrite(dev, obj, args, file); else - ret = -ENODEV; + ret = i915_gem_shmem_pwrite(dev, obj, args, file); } -out: - drm_gem_object_unreference(&obj->base); -unlock: + i915_gem_object_put(obj); mutex_unlock(&dev->struct_mutex); -put_rpm: intel_runtime_pm_put(dev_priv); return ret; -} - -static int -i915_gem_check_wedge(unsigned reset_counter, bool interruptible) -{ - if (__i915_terminally_wedged(reset_counter)) - return -EIO; - - if (__i915_reset_in_progress(reset_counter)) { - /* Non-interruptible callers can't handle -EAGAIN, hence return - * -EIO unconditionally for these. */ - if (!interruptible) - return -EIO; - - return -EAGAIN; - } - - return 0; -} - -static unsigned long local_clock_us(unsigned *cpu) -{ - unsigned long t; - - /* Cheaply and approximately convert from nanoseconds to microseconds. - * The result and subsequent calculations are also defined in the same - * approximate microseconds units. The principal source of timing - * error here is from the simple truncation. - * - * Note that local_clock() is only defined wrt to the current CPU; - * the comparisons are no longer valid if we switch CPUs. Instead of - * blocking preemption for the entire busywait, we can detect the CPU - * switch and use that as indicator of system load and a reason to - * stop busywaiting, see busywait_stop(). - */ - *cpu = get_cpu(); - t = local_clock() >> 10; - put_cpu(); - - return t; -} - -static bool busywait_stop(unsigned long timeout, unsigned cpu) -{ - unsigned this_cpu; - - if (time_after(local_clock_us(&this_cpu), timeout)) - return true; - - return this_cpu != cpu; -} - -bool __i915_spin_request(const struct drm_i915_gem_request *req, - int state, unsigned long timeout_us) -{ - unsigned cpu; - - /* When waiting for high frequency requests, e.g. during synchronous - * rendering split between the CPU and GPU, the finite amount of time - * required to set up the irq and wait upon it limits the response - * rate. By busywaiting on the request completion for a short while we - * can service the high frequency waits as quick as possible. However, - * if it is a slow request, we want to sleep as quickly as possible. - * The tradeoff between waiting and sleeping is roughly the time it - * takes to sleep on a request, on the order of a microsecond. - */ - - timeout_us += local_clock_us(&cpu); - do { - if (i915_gem_request_completed(req)) - return true; - - if (signal_pending_state(state, current)) - break; - - if (busywait_stop(timeout_us, cpu)) - break; - - cpu_relax_lowlatency(); - } while (!need_resched()); - - return false; -} - -/** - * __i915_wait_request - wait until execution of request has finished - * @req: duh! - * @interruptible: do an interruptible wait (normally yes) - * @timeout: in - how long to wait (NULL forever); out - how much time remaining - * @rps: RPS client - * - * Note: It is of utmost importance that the passed in seqno and reset_counter - * values have been read by the caller in an smp safe manner. Where read-side - * locks are involved, it is sufficient to read the reset_counter before - * unlocking the lock that protects the seqno. For lockless tricks, the - * reset_counter _must_ be read before, and an appropriate smp_rmb must be - * inserted. - * - * Returns 0 if the request was found within the alloted time. Else returns the - * errno with remaining time filled in timeout argument. - */ -int __i915_wait_request(struct drm_i915_gem_request *req, - bool interruptible, - s64 *timeout, - struct intel_rps_client *rps) -{ - int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE; - DEFINE_WAIT(reset); - struct intel_wait wait; - unsigned long timeout_remain; - s64 before = 0; /* Only to silence a compiler warning. */ - int ret = 0; - - might_sleep(); - - if (list_empty(&req->list)) - return 0; - - if (i915_gem_request_completed(req)) - return 0; - - timeout_remain = MAX_SCHEDULE_TIMEOUT; - if (timeout) { - if (WARN_ON(*timeout < 0)) - return -EINVAL; - - if (*timeout == 0) - return -ETIME; - - timeout_remain = nsecs_to_jiffies_timeout(*timeout); - - /* - * Record current time in case interrupted by signal, or wedged. - */ - before = ktime_get_raw_ns(); - } - - trace_i915_gem_request_wait_begin(req); - - /* This client is about to stall waiting for the GPU. In many cases - * this is undesirable and limits the throughput of the system, as - * many clients cannot continue processing user input/output whilst - * blocked. RPS autotuning may take tens of milliseconds to respond - * to the GPU load and thus incurs additional latency for the client. - * We can circumvent that by promoting the GPU frequency to maximum - * before we wait. This makes the GPU throttle up much more quickly - * (good for benchmarks and user experience, e.g. window animations), - * but at a cost of spending more power processing the workload - * (bad for battery). Not all clients even want their results - * immediately and for them we should just let the GPU select its own - * frequency to maximise efficiency. To prevent a single client from - * forcing the clocks too high for the whole system, we only allow - * each client to waitboost once in a busy period. - */ - if (INTEL_INFO(req->i915)->gen >= 6) - gen6_rps_boost(req->i915, rps, req->emitted_jiffies); - - /* Optimistic spin for the next ~jiffie before touching IRQs */ - if (i915_spin_request(req, state, 5)) - goto complete; - - set_current_state(state); - add_wait_queue(&req->i915->gpu_error.wait_queue, &reset); - - intel_wait_init(&wait, req->seqno); - if (intel_engine_add_wait(req->engine, &wait)) - /* In order to check that we haven't missed the interrupt - * as we enabled it, we need to kick ourselves to do a - * coherent check on the seqno before we sleep. - */ - goto wakeup; - - for (;;) { - if (signal_pending_state(state, current)) { - ret = -ERESTARTSYS; - break; - } - - timeout_remain = io_schedule_timeout(timeout_remain); - if (timeout_remain == 0) { - ret = -ETIME; - break; - } - - if (intel_wait_complete(&wait)) - break; - - set_current_state(state); - -wakeup: - /* Carefully check if the request is complete, giving time - * for the seqno to be visible following the interrupt. - * We also have to check in case we are kicked by the GPU - * reset in order to drop the struct_mutex. - */ - if (__i915_request_irq_complete(req)) - break; - - /* Only spin if we know the GPU is processing this request */ - if (i915_spin_request(req, state, 2)) - break; - } - remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset); - - intel_engine_remove_wait(req->engine, &wait); - __set_current_state(TASK_RUNNING); -complete: - trace_i915_gem_request_wait_end(req); - - if (timeout) { - s64 tres = *timeout - (ktime_get_raw_ns() - before); - - *timeout = tres < 0 ? 0 : tres; - - /* - * 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 regrssion from the timespec->ktime conversion. - */ - if (ret == -ETIME && *timeout < jiffies_to_usecs(1)*1000) - *timeout = 0; - } - - if (rps && req->seqno == req->engine->last_submitted_seqno) { - /* The GPU is now idle and this client has stalled. - * Since no other client has submitted a request in the - * meantime, assume that this client is the only one - * supplying work to the GPU but is unable to keep that - * work supplied because it is waiting. Since the GPU is - * then never kept fully busy, RPS autoclocking will - * keep the clocks relatively low, causing further delays. - * Compensate by giving the synchronous client credit for - * a waitboost next time. - */ - spin_lock(&req->i915->rps.client_lock); - list_del_init(&rps->link); - spin_unlock(&req->i915->rps.client_lock); - } - - return ret; -} - -int i915_gem_request_add_to_client(struct drm_i915_gem_request *req, - struct drm_file *file) -{ - struct drm_i915_file_private *file_priv; - - WARN_ON(!req || !file || req->file_priv); - - if (!req || !file) - return -EINVAL; - - if (req->file_priv) - return -EINVAL; - - file_priv = file->driver_priv; - - spin_lock(&file_priv->mm.lock); - req->file_priv = file_priv; - list_add_tail(&req->client_list, &file_priv->mm.request_list); - spin_unlock(&file_priv->mm.lock); - - req->pid = get_pid(task_pid(current)); - - return 0; -} - -static inline void -i915_gem_request_remove_from_client(struct drm_i915_gem_request *request) -{ - struct drm_i915_file_private *file_priv = request->file_priv; - - if (!file_priv) - return; - - spin_lock(&file_priv->mm.lock); - list_del(&request->client_list); - request->file_priv = NULL; - spin_unlock(&file_priv->mm.lock); - - put_pid(request->pid); - request->pid = NULL; -} - -static void i915_gem_request_retire(struct drm_i915_gem_request *request) -{ - trace_i915_gem_request_retire(request); - - /* We know the GPU must have read the request to have - * sent us the seqno + interrupt, so use the position - * of tail of the request to update the last known position - * of the GPU head. - * - * Note this requires that we are always called in request - * completion order. - */ - request->ringbuf->last_retired_head = request->postfix; - - list_del_init(&request->list); - i915_gem_request_remove_from_client(request); - - if (request->previous_context) { - if (i915.enable_execlists) - intel_lr_context_unpin(request->previous_context, - request->engine); - } - - i915_gem_context_unreference(request->ctx); - i915_gem_request_unreference(request); -} - -static void -__i915_gem_request_retire__upto(struct drm_i915_gem_request *req) -{ - struct intel_engine_cs *engine = req->engine; - struct drm_i915_gem_request *tmp; - - lockdep_assert_held(&engine->i915->drm.struct_mutex); - - if (list_empty(&req->list)) - return; - - do { - tmp = list_first_entry(&engine->request_list, - typeof(*tmp), list); - - i915_gem_request_retire(tmp); - } while (tmp != req); - - WARN_ON(i915_verify_lists(engine->dev)); -} - -/** - * Waits for a request to be signaled, and cleans up the - * request and object lists appropriately for that event. - * @req: request to wait on - */ -int -i915_wait_request(struct drm_i915_gem_request *req) -{ - struct drm_i915_private *dev_priv = req->i915; - bool interruptible; - int ret; - - interruptible = dev_priv->mm.interruptible; - - BUG_ON(!mutex_is_locked(&dev_priv->drm.struct_mutex)); - - ret = __i915_wait_request(req, interruptible, NULL, NULL); - if (ret) - return ret; - - /* If the GPU hung, we want to keep the requests to find the guilty. */ - if (!i915_reset_in_progress(&dev_priv->gpu_error)) - __i915_gem_request_retire__upto(req); - - return 0; -} - -/** - * Ensures that all rendering to the object has completed and the object is - * safe to unbind from the GTT or access from the CPU. - * @obj: i915 gem object - * @readonly: waiting for read access or write - */ -int -i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj, - bool readonly) -{ - int ret, i; - - if (!obj->active) - return 0; - - if (readonly) { - if (obj->last_write_req != NULL) { - ret = i915_wait_request(obj->last_write_req); - if (ret) - return ret; - - i = obj->last_write_req->engine->id; - if (obj->last_read_req[i] == obj->last_write_req) - i915_gem_object_retire__read(obj, i); - else - i915_gem_object_retire__write(obj); - } - } else { - for (i = 0; i < I915_NUM_ENGINES; i++) { - if (obj->last_read_req[i] == NULL) - continue; - - ret = i915_wait_request(obj->last_read_req[i]); - if (ret) - return ret; - - i915_gem_object_retire__read(obj, i); - } - GEM_BUG_ON(obj->active); - } - - return 0; -} - -static void -i915_gem_object_retire_request(struct drm_i915_gem_object *obj, - struct drm_i915_gem_request *req) -{ - int ring = req->engine->id; - - if (obj->last_read_req[ring] == req) - i915_gem_object_retire__read(obj, ring); - else if (obj->last_write_req == req) - i915_gem_object_retire__write(obj); - - if (!i915_reset_in_progress(&req->i915->gpu_error)) - __i915_gem_request_retire__upto(req); -} - -/* A nonblocking variant of the above wait. This is a highly dangerous routine - * as the object state may change during this call. - */ -static __must_check int -i915_gem_object_wait_rendering__nonblocking(struct drm_i915_gem_object *obj, - struct intel_rps_client *rps, - bool readonly) -{ - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - struct drm_i915_gem_request *requests[I915_NUM_ENGINES]; - int ret, i, n = 0; - - BUG_ON(!mutex_is_locked(&dev->struct_mutex)); - BUG_ON(!dev_priv->mm.interruptible); - - if (!obj->active) - return 0; - - if (readonly) { - struct drm_i915_gem_request *req; - - req = obj->last_write_req; - if (req == NULL) - return 0; - - requests[n++] = i915_gem_request_reference(req); - } else { - for (i = 0; i < I915_NUM_ENGINES; i++) { - struct drm_i915_gem_request *req; - - req = obj->last_read_req[i]; - if (req == NULL) - continue; - - requests[n++] = i915_gem_request_reference(req); - } - } - - mutex_unlock(&dev->struct_mutex); - ret = 0; - for (i = 0; ret == 0 && i < n; i++) - ret = __i915_wait_request(requests[i], true, NULL, rps); - mutex_lock(&dev->struct_mutex); - - for (i = 0; i < n; i++) { - if (ret == 0) - i915_gem_object_retire_request(obj, requests[i]); - i915_gem_request_unreference(requests[i]); - } +err_rpm: + intel_runtime_pm_put(dev_priv); +err: + i915_gem_object_put_unlocked(obj); return ret; } -static struct intel_rps_client *to_rps_client(struct drm_file *file) -{ - struct drm_i915_file_private *fpriv = file->driver_priv; - return &fpriv->rps; -} - -static enum fb_op_origin +static inline enum fb_op_origin write_origin(struct drm_i915_gem_object *obj, unsigned domain) { - return domain == I915_GEM_DOMAIN_GTT && !obj->has_wc_mmap ? - ORIGIN_GTT : ORIGIN_CPU; + return (domain == I915_GEM_DOMAIN_GTT ? + obj->frontbuffer_ggtt_origin : ORIGIN_CPU); } /** @@ -1831,10 +1520,7 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, int ret; /* Only handle setting domains to types used by the CPU. */ - if (write_domain & I915_GEM_GPU_DOMAINS) - return -EINVAL; - - if (read_domains & I915_GEM_GPU_DOMAINS) + if ((write_domain | read_domains) & I915_GEM_GPU_DOMAINS) return -EINVAL; /* Having something in the write domain implies it's in the read @@ -1843,25 +1529,21 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, if (write_domain != 0 && read_domains != write_domain) return -EINVAL; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; - - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; /* 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. */ - ret = i915_gem_object_wait_rendering__nonblocking(obj, - to_rps_client(file), - !write_domain); + ret = __unsafe_wait_rendering(obj, to_rps_client(file), !write_domain); if (ret) - goto unref; + goto err; + + ret = i915_mutex_lock_interruptible(dev); + if (ret) + goto err; if (read_domains & I915_GEM_DOMAIN_GTT) ret = i915_gem_object_set_to_gtt_domain(obj, write_domain != 0); @@ -1871,11 +1553,13 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, if (write_domain != 0) intel_fb_obj_invalidate(obj, write_origin(obj, write_domain)); -unref: - drm_gem_object_unreference(&obj->base); -unlock: + i915_gem_object_put(obj); mutex_unlock(&dev->struct_mutex); return ret; + +err: + i915_gem_object_put_unlocked(obj); + return ret; } /** @@ -1890,26 +1574,23 @@ i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, { struct drm_i915_gem_sw_finish *args = data; struct drm_i915_gem_object *obj; - int ret = 0; + int err = 0; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; - - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; /* Pinned buffers may be scanout, so flush the cache */ - if (obj->pin_display) - i915_gem_object_flush_cpu_write_domain(obj); + if (READ_ONCE(obj->pin_display)) { + err = i915_mutex_lock_interruptible(dev); + if (!err) { + i915_gem_object_flush_cpu_write_domain(obj); + mutex_unlock(&dev->struct_mutex); + } + } - drm_gem_object_unreference(&obj->base); -unlock: - mutex_unlock(&dev->struct_mutex); - return ret; + i915_gem_object_put_unlocked(obj); + return err; } /** @@ -1937,7 +1618,7 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_mmap *args = data; - struct drm_gem_object *obj; + struct drm_i915_gem_object *obj; unsigned long addr; if (args->flags & ~(I915_MMAP_WC)) @@ -1946,19 +1627,19 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, if (args->flags & I915_MMAP_WC && !boot_cpu_has(X86_FEATURE_PAT)) return -ENODEV; - obj = drm_gem_object_lookup(file, args->handle); - if (obj == NULL) + 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->filp) { - drm_gem_object_unreference_unlocked(obj); + if (!obj->base.filp) { + i915_gem_object_put_unlocked(obj); return -EINVAL; } - addr = vm_mmap(obj->filp, 0, args->size, + addr = vm_mmap(obj->base.filp, 0, args->size, PROT_READ | PROT_WRITE, MAP_SHARED, args->offset); if (args->flags & I915_MMAP_WC) { @@ -1966,7 +1647,7 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, struct vm_area_struct *vma; if (down_write_killable(&mm->mmap_sem)) { - drm_gem_object_unreference_unlocked(obj); + i915_gem_object_put_unlocked(obj); return -EINTR; } vma = find_vma(mm, addr); @@ -1978,9 +1659,9 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, up_write(&mm->mmap_sem); /* This may race, but that's ok, it only gets set */ - WRITE_ONCE(to_intel_bo(obj)->has_wc_mmap, true); + WRITE_ONCE(obj->frontbuffer_ggtt_origin, ORIGIN_CPU); } - drm_gem_object_unreference_unlocked(obj); + i915_gem_object_put_unlocked(obj); if (IS_ERR((void *)addr)) return addr; @@ -1989,9 +1670,69 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, return 0; } +static unsigned int tile_row_pages(struct drm_i915_gem_object *obj) +{ + u64 size; + + size = i915_gem_object_get_stride(obj); + size *= i915_gem_object_get_tiling(obj) == I915_TILING_Y ? 32 : 8; + + return size >> 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.) + * + * 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 1; +} + /** * i915_gem_fault - fault a page into the GTT - * @vma: VMA in question + * @area: CPU VMA in question * @vmf: fault info * * The fault handler is set up by drm_gem_mmap() when a object is GTT mapped @@ -2004,122 +1745,120 @@ i915_gem_mmap_ioctl(struct drm_device *dev, void *data, * 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). */ -int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +int i915_gem_fault(struct vm_area_struct *area, struct vm_fault *vmf) { - struct drm_i915_gem_object *obj = to_intel_bo(vma->vm_private_data); +#define MIN_CHUNK_PAGES ((1 << 20) >> PAGE_SHIFT) /* 1 MiB */ + 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; - struct i915_ggtt_view view = i915_ggtt_view_normal; - pgoff_t page_offset; - unsigned long pfn; - int ret = 0; bool write = !!(vmf->flags & FAULT_FLAG_WRITE); - - intel_runtime_pm_get(dev_priv); + struct i915_vma *vma; + pgoff_t page_offset; + unsigned int flags; + int ret; /* We don't use vmf->pgoff since that has the fake offset */ - page_offset = ((unsigned long)vmf->virtual_address - vma->vm_start) >> + page_offset = ((unsigned long)vmf->virtual_address - area->vm_start) >> PAGE_SHIFT; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - goto out; - trace_i915_gem_object_fault(obj, page_offset, true, write); /* Try to flush the object off the GPU first without holding the lock. - * Upon reacquiring the lock, we will perform our sanity checks and then + * Upon acquiring the lock, we will perform our sanity checks and then * repeat the flush holding the lock in the normal manner to catch cases * where we are gazumped. */ - ret = i915_gem_object_wait_rendering__nonblocking(obj, NULL, !write); + ret = __unsafe_wait_rendering(obj, NULL, !write); if (ret) - goto unlock; + goto err; + + intel_runtime_pm_get(dev_priv); + + ret = i915_mutex_lock_interruptible(dev); + if (ret) + goto err_rpm; /* Access to snoopable pages through the GTT is incoherent. */ if (obj->cache_level != I915_CACHE_NONE && !HAS_LLC(dev)) { ret = -EFAULT; - goto unlock; + goto err_unlock; } - /* Use a partial view if the object is bigger than the aperture. */ - if (obj->base.size >= ggtt->mappable_end && - obj->tiling_mode == I915_TILING_NONE) { - static const unsigned int chunk_size = 256; // 1 MiB + /* If the object is smaller than a couple of partial vma, it is + * not worth only creating a single partial vma - we may as well + * clear enough space for the full object. + */ + flags = PIN_MAPPABLE; + if (obj->base.size > 2 * MIN_CHUNK_PAGES << PAGE_SHIFT) + flags |= PIN_NONBLOCK | PIN_NONFAULT; + + /* Now pin it into the GTT as needed */ + vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, flags); + if (IS_ERR(vma)) { + struct i915_ggtt_view view; + unsigned int chunk_size; + + /* Use a partial view if it is bigger than available space */ + chunk_size = MIN_CHUNK_PAGES; + if (i915_gem_object_is_tiled(obj)) + chunk_size = max(chunk_size, tile_row_pages(obj)); memset(&view, 0, sizeof(view)); view.type = I915_GGTT_VIEW_PARTIAL; view.params.partial.offset = rounddown(page_offset, chunk_size); view.params.partial.size = - min_t(unsigned int, - chunk_size, - (vma->vm_end - vma->vm_start)/PAGE_SIZE - + min_t(unsigned int, chunk_size, + (area->vm_end - area->vm_start) / PAGE_SIZE - view.params.partial.offset); - } - /* Now pin it into the GTT if needed */ - ret = i915_gem_object_ggtt_pin(obj, &view, 0, PIN_MAPPABLE); - if (ret) - goto unlock; + /* If the partial covers the entire object, just create a + * normal VMA. + */ + if (chunk_size >= obj->base.size >> PAGE_SHIFT) + view.type = I915_GGTT_VIEW_NORMAL; + + /* 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, PIN_MAPPABLE); + } + if (IS_ERR(vma)) { + ret = PTR_ERR(vma); + goto err_unlock; + } ret = i915_gem_object_set_to_gtt_domain(obj, write); if (ret) - goto unpin; + goto err_unpin; - ret = i915_gem_object_get_fence(obj); + ret = i915_vma_get_fence(vma); if (ret) - goto unpin; + goto err_unpin; /* Finally, remap it using the new GTT offset */ - pfn = ggtt->mappable_base + - i915_gem_obj_ggtt_offset_view(obj, &view); - pfn >>= PAGE_SHIFT; - - if (unlikely(view.type == I915_GGTT_VIEW_PARTIAL)) { - /* Overriding existing pages in partial view does not cause - * us any trouble as TLBs are still valid because the fault - * is due to userspace losing part of the mapping or never - * having accessed it before (at this partials' range). - */ - unsigned long base = vma->vm_start + - (view.params.partial.offset << PAGE_SHIFT); - unsigned int i; - - for (i = 0; i < view.params.partial.size; i++) { - ret = vm_insert_pfn(vma, base + i * PAGE_SIZE, pfn + i); - if (ret) - break; - } - - obj->fault_mappable = true; - } else { - if (!obj->fault_mappable) { - unsigned long size = min_t(unsigned long, - vma->vm_end - vma->vm_start, - obj->base.size); - int i; - - for (i = 0; i < size >> PAGE_SHIFT; i++) { - ret = vm_insert_pfn(vma, - (unsigned long)vma->vm_start + i * PAGE_SIZE, - pfn + i); - if (ret) - break; - } + ret = remap_io_mapping(area, + area->vm_start + (vma->ggtt_view.params.partial.offset << PAGE_SHIFT), + (ggtt->mappable_base + vma->node.start) >> PAGE_SHIFT, + min_t(u64, vma->size, area->vm_end - area->vm_start), + &ggtt->mappable); + if (ret) + goto err_unpin; - obj->fault_mappable = true; - } else - ret = vm_insert_pfn(vma, - (unsigned long)vmf->virtual_address, - pfn + page_offset); - } -unpin: - i915_gem_object_ggtt_unpin_view(obj, &view); -unlock: + obj->fault_mappable = true; +err_unpin: + __i915_vma_unpin(vma); +err_unlock: mutex_unlock(&dev->struct_mutex); -out: +err_rpm: + intel_runtime_pm_put(dev_priv); +err: switch (ret) { case -EIO: /* @@ -2160,8 +1899,6 @@ out: ret = VM_FAULT_SIGBUS; break; } - - intel_runtime_pm_put(dev_priv); return ret; } @@ -2215,46 +1952,58 @@ i915_gem_release_all_mmaps(struct drm_i915_private *dev_priv) i915_gem_release_mmap(obj); } -uint32_t -i915_gem_get_gtt_size(struct drm_device *dev, uint32_t size, int tiling_mode) +/** + * i915_gem_get_ggtt_size - return required global GTT size for an object + * @dev_priv: i915 device + * @size: object size + * @tiling_mode: tiling mode + * + * Return the required global GTT size for an object, taking into account + * potential fence register mapping. + */ +u64 i915_gem_get_ggtt_size(struct drm_i915_private *dev_priv, + u64 size, int tiling_mode) { - uint32_t gtt_size; + u64 ggtt_size; - if (INTEL_INFO(dev)->gen >= 4 || + GEM_BUG_ON(size == 0); + + if (INTEL_GEN(dev_priv) >= 4 || tiling_mode == I915_TILING_NONE) return size; /* Previous chips need a power-of-two fence region when tiling */ - if (IS_GEN3(dev)) - gtt_size = 1024*1024; + if (IS_GEN3(dev_priv)) + ggtt_size = 1024*1024; else - gtt_size = 512*1024; + ggtt_size = 512*1024; - while (gtt_size < size) - gtt_size <<= 1; + while (ggtt_size < size) + ggtt_size <<= 1; - return gtt_size; + return ggtt_size; } /** - * i915_gem_get_gtt_alignment - return required GTT alignment for an object - * @dev: drm device + * i915_gem_get_ggtt_alignment - return required global GTT alignment + * @dev_priv: i915 device * @size: object size * @tiling_mode: tiling mode - * @fenced: is fenced alignemned required or not + * @fenced: is fenced alignment required or not * - * Return the required GTT alignment for an object, taking into account + * Return the required global GTT alignment for an object, taking into account * potential fence register mapping. */ -uint32_t -i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size, - int tiling_mode, bool fenced) +u64 i915_gem_get_ggtt_alignment(struct drm_i915_private *dev_priv, u64 size, + int tiling_mode, bool fenced) { + GEM_BUG_ON(size == 0); + /* * Minimum alignment is 4k (GTT page size), but might be greater * if a fence register is needed for the object. */ - if (INTEL_INFO(dev)->gen >= 4 || (!fenced && IS_G33(dev)) || + if (INTEL_GEN(dev_priv) >= 4 || (!fenced && IS_G33(dev_priv)) || tiling_mode == I915_TILING_NONE) return 4096; @@ -2262,42 +2011,34 @@ i915_gem_get_gtt_alignment(struct drm_device *dev, uint32_t size, * Previous chips need to be aligned to the size of the smallest * fence register that can contain the object. */ - return i915_gem_get_gtt_size(dev, size, tiling_mode); + return i915_gem_get_ggtt_size(dev_priv, size, tiling_mode); } 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 ret; - - dev_priv->mm.shrinker_no_lock_stealing = true; + int err; - ret = drm_gem_create_mmap_offset(&obj->base); - if (ret != -ENOSPC) - goto out; + err = drm_gem_create_mmap_offset(&obj->base); + if (!err) + return 0; - /* Badly fragmented mmap space? The only way we can recover - * space is by destroying unwanted objects. We can't randomly release - * mmap_offsets as userspace expects them to be persistent for the - * lifetime of the objects. The closest we can is to release the - * offsets on purgeable objects by truncating it and marking it purged, - * which prevents userspace from ever using that object again. + /* We can idle the GPU locklessly to flush stale objects, but in order + * to claim that space for ourselves, we need to take the big + * struct_mutex to free the requests+objects and allocate our slot. */ - i915_gem_shrink(dev_priv, - obj->base.size >> PAGE_SHIFT, - I915_SHRINK_BOUND | - I915_SHRINK_UNBOUND | - I915_SHRINK_PURGEABLE); - ret = drm_gem_create_mmap_offset(&obj->base); - if (ret != -ENOSPC) - goto out; + err = i915_gem_wait_for_idle(dev_priv, I915_WAIT_INTERRUPTIBLE); + if (err) + return err; - i915_gem_shrink_all(dev_priv); - ret = drm_gem_create_mmap_offset(&obj->base); -out: - dev_priv->mm.shrinker_no_lock_stealing = false; + err = i915_mutex_lock_interruptible(&dev_priv->drm); + if (!err) { + i915_gem_retire_requests(dev_priv); + err = drm_gem_create_mmap_offset(&obj->base); + mutex_unlock(&dev_priv->drm.struct_mutex); + } - return ret; + return err; } static void i915_gem_object_free_mmap_offset(struct drm_i915_gem_object *obj) @@ -2314,32 +2055,15 @@ i915_gem_mmap_gtt(struct drm_file *file, struct drm_i915_gem_object *obj; int ret; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; - - obj = to_intel_bo(drm_gem_object_lookup(file, handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } - - if (obj->madv != I915_MADV_WILLNEED) { - DRM_DEBUG("Attempting to mmap a purgeable buffer\n"); - ret = -EFAULT; - goto out; - } + obj = i915_gem_object_lookup(file, handle); + if (!obj) + return -ENOENT; ret = i915_gem_object_create_mmap_offset(obj); - if (ret) - goto out; - - *offset = drm_vma_node_offset_addr(&obj->base.vma_node); + if (ret == 0) + *offset = drm_vma_node_offset_addr(&obj->base.vma_node); -out: - drm_gem_object_unreference(&obj->base); -unlock: - mutex_unlock(&dev->struct_mutex); + i915_gem_object_put_unlocked(obj); return ret; } @@ -2457,7 +2181,7 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj) if (obj->pages_pin_count) return -EBUSY; - BUG_ON(i915_gem_obj_bound_any(obj)); + GEM_BUG_ON(obj->bind_count); /* ->put_pages might need to allocate memory for the bit17 swizzle * array, hence protect them from being reaped by removing them from gtt @@ -2465,10 +2189,14 @@ i915_gem_object_put_pages(struct drm_i915_gem_object *obj) list_del(&obj->global_list); if (obj->mapping) { - if (is_vmalloc_addr(obj->mapping)) - vunmap(obj->mapping); + void *ptr; + + ptr = ptr_mask_bits(obj->mapping); + if (is_vmalloc_addr(ptr)) + vunmap(ptr); else - kunmap(kmap_to_page(obj->mapping)); + kunmap(kmap_to_page(ptr)); + obj->mapping = NULL; } @@ -2577,7 +2305,7 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj) if (i915_gem_object_needs_bit17_swizzle(obj)) i915_gem_object_do_bit_17_swizzle(obj); - if (obj->tiling_mode != I915_TILING_NONE && + if (i915_gem_object_is_tiled(obj) && dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) i915_gem_object_pin_pages(obj); @@ -2641,7 +2369,8 @@ i915_gem_object_get_pages(struct drm_i915_gem_object *obj) } /* The 'mapping' part of i915_gem_object_pin_map() below */ -static void *i915_gem_object_map(const struct drm_i915_gem_object *obj) +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->pages; @@ -2650,10 +2379,11 @@ static void *i915_gem_object_map(const struct drm_i915_gem_object *obj) 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) + if (n_pages == 1 && type == I915_MAP_WB) return kmap(sg_page(sgt->sgl)); if (n_pages > ARRAY_SIZE(stack_pages)) { @@ -2669,7 +2399,15 @@ static void *i915_gem_object_map(const struct drm_i915_gem_object *obj) /* Check that we have the expected number of pages */ GEM_BUG_ON(i != n_pages); - addr = vmap(pages, n_pages, 0, PAGE_KERNEL); + switch (type) { + 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) drm_free_large(pages); @@ -2678,276 +2416,89 @@ static void *i915_gem_object_map(const struct drm_i915_gem_object *obj) } /* get, pin, and map the pages of the object into kernel space */ -void *i915_gem_object_pin_map(struct drm_i915_gem_object *obj) +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; lockdep_assert_held(&obj->base.dev->struct_mutex); + GEM_BUG_ON(!i915_gem_object_has_struct_page(obj)); ret = i915_gem_object_get_pages(obj); if (ret) return ERR_PTR(ret); i915_gem_object_pin_pages(obj); + pinned = obj->pages_pin_count > 1; - if (!obj->mapping) { - obj->mapping = i915_gem_object_map(obj); - if (!obj->mapping) { - i915_gem_object_unpin_pages(obj); - return ERR_PTR(-ENOMEM); + ptr = ptr_unpack_bits(obj->mapping, has_type); + if (ptr && has_type != type) { + if (pinned) { + ret = -EBUSY; + goto err; } - } - return obj->mapping; -} + if (is_vmalloc_addr(ptr)) + vunmap(ptr); + else + kunmap(kmap_to_page(ptr)); -void i915_vma_move_to_active(struct i915_vma *vma, - struct drm_i915_gem_request *req) -{ - struct drm_i915_gem_object *obj = vma->obj; - struct intel_engine_cs *engine; + ptr = obj->mapping = NULL; + } - engine = i915_gem_request_get_engine(req); + if (!ptr) { + ptr = i915_gem_object_map(obj, type); + if (!ptr) { + ret = -ENOMEM; + goto err; + } - /* Add a reference if we're newly entering the active list. */ - if (obj->active == 0) - drm_gem_object_reference(&obj->base); - obj->active |= intel_engine_flag(engine); + obj->mapping = ptr_pack_bits(ptr, type); + } - list_move_tail(&obj->engine_list[engine->id], &engine->active_list); - i915_gem_request_assign(&obj->last_read_req[engine->id], req); + return ptr; - list_move_tail(&vma->vm_link, &vma->vm->active_list); +err: + i915_gem_object_unpin_pages(obj); + return ERR_PTR(ret); } static void -i915_gem_object_retire__write(struct drm_i915_gem_object *obj) +i915_gem_object_retire__write(struct i915_gem_active *active, + struct drm_i915_gem_request *request) { - GEM_BUG_ON(obj->last_write_req == NULL); - GEM_BUG_ON(!(obj->active & intel_engine_flag(obj->last_write_req->engine))); + struct drm_i915_gem_object *obj = + container_of(active, struct drm_i915_gem_object, last_write); - i915_gem_request_assign(&obj->last_write_req, NULL); intel_fb_obj_flush(obj, true, ORIGIN_CS); } static void -i915_gem_object_retire__read(struct drm_i915_gem_object *obj, int ring) +i915_gem_object_retire__read(struct i915_gem_active *active, + struct drm_i915_gem_request *request) { - struct i915_vma *vma; - - GEM_BUG_ON(obj->last_read_req[ring] == NULL); - GEM_BUG_ON(!(obj->active & (1 << ring))); - - list_del_init(&obj->engine_list[ring]); - i915_gem_request_assign(&obj->last_read_req[ring], NULL); + int idx = request->engine->id; + struct drm_i915_gem_object *obj = + container_of(active, struct drm_i915_gem_object, last_read[idx]); - if (obj->last_write_req && obj->last_write_req->engine->id == ring) - i915_gem_object_retire__write(obj); + GEM_BUG_ON(!i915_gem_object_has_active_engine(obj, idx)); - obj->active &= ~(1 << ring); - if (obj->active) + i915_gem_object_clear_active(obj, idx); + if (i915_gem_object_is_active(obj)) return; /* Bump our place on the bound list to keep it roughly in LRU order * so that we don't steal from recently used but inactive objects * (unless we are forced to ofc!) */ - list_move_tail(&obj->global_list, - &to_i915(obj->base.dev)->mm.bound_list); + if (obj->bind_count) + list_move_tail(&obj->global_list, + &request->i915->mm.bound_list); - list_for_each_entry(vma, &obj->vma_list, obj_link) { - if (!list_empty(&vma->vm_link)) - list_move_tail(&vma->vm_link, &vma->vm->inactive_list); - } - - i915_gem_request_assign(&obj->last_fenced_req, NULL); - drm_gem_object_unreference(&obj->base); -} - -static int -i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno) -{ - struct intel_engine_cs *engine; - int ret; - - /* Carefully retire all requests without writing to the rings */ - for_each_engine(engine, dev_priv) { - ret = intel_engine_idle(engine); - if (ret) - return ret; - } - i915_gem_retire_requests(dev_priv); - - /* If the seqno wraps around, we need to clear the breadcrumb rbtree */ - if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) { - while (intel_kick_waiters(dev_priv) || - intel_kick_signalers(dev_priv)) - yield(); - } - - /* Finally reset hw state */ - for_each_engine(engine, dev_priv) - intel_ring_init_seqno(engine, seqno); - - return 0; -} - -int i915_gem_set_seqno(struct drm_device *dev, u32 seqno) -{ - struct drm_i915_private *dev_priv = to_i915(dev); - int ret; - - if (seqno == 0) - return -EINVAL; - - /* HWS page needs to be set less than what we - * will inject to ring - */ - ret = i915_gem_init_seqno(dev_priv, seqno - 1); - if (ret) - return ret; - - /* Carefully set the last_seqno value so that wrap - * detection still works - */ - dev_priv->next_seqno = seqno; - dev_priv->last_seqno = seqno - 1; - if (dev_priv->last_seqno == 0) - dev_priv->last_seqno--; - - return 0; -} - -int -i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno) -{ - /* reserve 0 for non-seqno */ - if (dev_priv->next_seqno == 0) { - int ret = i915_gem_init_seqno(dev_priv, 0); - if (ret) - return ret; - - dev_priv->next_seqno = 1; - } - - *seqno = dev_priv->last_seqno = dev_priv->next_seqno++; - return 0; -} - -static void i915_gem_mark_busy(const struct intel_engine_cs *engine) -{ - struct drm_i915_private *dev_priv = engine->i915; - - dev_priv->gt.active_engines |= intel_engine_flag(engine); - if (dev_priv->gt.awake) - return; - - intel_runtime_pm_get_noresume(dev_priv); - dev_priv->gt.awake = true; - - i915_update_gfx_val(dev_priv); - if (INTEL_GEN(dev_priv) >= 6) - gen6_rps_busy(dev_priv); - - queue_delayed_work(dev_priv->wq, - &dev_priv->gt.retire_work, - round_jiffies_up_relative(HZ)); -} - -/* - * NB: This function is not allowed to fail. Doing so would mean the the - * request is not being tracked for completion but the work itself is - * going to happen on the hardware. This would be a Bad Thing(tm). - */ -void __i915_add_request(struct drm_i915_gem_request *request, - struct drm_i915_gem_object *obj, - bool flush_caches) -{ - struct intel_engine_cs *engine; - struct intel_ringbuffer *ringbuf; - u32 request_start; - u32 reserved_tail; - int ret; - - if (WARN_ON(request == NULL)) - return; - - engine = request->engine; - ringbuf = request->ringbuf; - - /* - * To ensure that this call will not fail, space for its emissions - * should already have been reserved in the ring buffer. Let the ring - * know that it is time to use that space up. - */ - request_start = intel_ring_get_tail(ringbuf); - reserved_tail = request->reserved_space; - request->reserved_space = 0; - - /* - * Emit any outstanding flushes - execbuf can fail to emit the flush - * after having emitted the batchbuffer command. Hence we need to fix - * things up similar to emitting the lazy request. The difference here - * is that the flush _must_ happen before the next request, no matter - * what. - */ - if (flush_caches) { - if (i915.enable_execlists) - ret = logical_ring_flush_all_caches(request); - else - ret = intel_ring_flush_all_caches(request); - /* Not allowed to fail! */ - WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret); - } - - trace_i915_gem_request_add(request); - - request->head = request_start; - - /* Whilst this request exists, batch_obj will be on the - * active_list, and so will hold the active reference. Only when this - * request is retired will the the batch_obj be moved onto the - * inactive_list and lose its active reference. Hence we do not need - * to explicitly hold another reference here. - */ - request->batch_obj = obj; - - /* Seal the request and mark it as pending execution. Note that - * we may inspect this state, without holding any locks, during - * hangcheck. Hence we apply the barrier to ensure that we do not - * see a more recent value in the hws than we are tracking. - */ - request->emitted_jiffies = jiffies; - request->previous_seqno = engine->last_submitted_seqno; - smp_store_mb(engine->last_submitted_seqno, request->seqno); - list_add_tail(&request->list, &engine->request_list); - - /* Record the position of the start of the request so that - * should we detect the updated seqno part-way through the - * GPU processing the request, we never over-estimate the - * position of the head. - */ - request->postfix = intel_ring_get_tail(ringbuf); - - if (i915.enable_execlists) - ret = engine->emit_request(request); - else { - ret = engine->add_request(request); - - request->tail = intel_ring_get_tail(ringbuf); - } - /* Not allowed to fail! */ - WARN(ret, "emit|add_request failed: %d!\n", ret); - /* Sanity check that the reserved size was large enough. */ - ret = intel_ring_get_tail(ringbuf) - request_start; - if (ret < 0) - ret += ringbuf->size; - WARN_ONCE(ret > reserved_tail, - "Not enough space reserved (%d bytes) " - "for adding the request (%d bytes)\n", - reserved_tail, ret); - - i915_gem_mark_busy(engine); + i915_gem_object_put(obj); } static bool i915_context_is_banned(const struct i915_gem_context *ctx) @@ -2981,101 +2532,6 @@ static void i915_set_reset_status(struct i915_gem_context *ctx, } } -void i915_gem_request_free(struct kref *req_ref) -{ - struct drm_i915_gem_request *req = container_of(req_ref, - typeof(*req), ref); - kmem_cache_free(req->i915->requests, req); -} - -static inline int -__i915_gem_request_alloc(struct intel_engine_cs *engine, - struct i915_gem_context *ctx, - struct drm_i915_gem_request **req_out) -{ - struct drm_i915_private *dev_priv = engine->i915; - unsigned reset_counter = i915_reset_counter(&dev_priv->gpu_error); - struct drm_i915_gem_request *req; - int ret; - - if (!req_out) - return -EINVAL; - - *req_out = NULL; - - /* ABI: Before userspace accesses the GPU (e.g. execbuffer), report - * EIO if the GPU is already wedged, or EAGAIN to drop the struct_mutex - * and restart. - */ - ret = i915_gem_check_wedge(reset_counter, dev_priv->mm.interruptible); - if (ret) - return ret; - - req = kmem_cache_zalloc(dev_priv->requests, GFP_KERNEL); - if (req == NULL) - return -ENOMEM; - - ret = i915_gem_get_seqno(engine->i915, &req->seqno); - if (ret) - goto err; - - kref_init(&req->ref); - req->i915 = dev_priv; - req->engine = engine; - req->ctx = ctx; - i915_gem_context_reference(req->ctx); - - /* - * Reserve space in the ring buffer for all the commands required to - * eventually emit this request. This is to guarantee that the - * i915_add_request() call can't fail. Note that the reserve may need - * to be redone if the request is not actually submitted straight - * away, e.g. because a GPU scheduler has deferred it. - */ - req->reserved_space = MIN_SPACE_FOR_ADD_REQUEST; - - if (i915.enable_execlists) - ret = intel_logical_ring_alloc_request_extras(req); - else - ret = intel_ring_alloc_request_extras(req); - if (ret) - goto err_ctx; - - *req_out = req; - return 0; - -err_ctx: - i915_gem_context_unreference(ctx); -err: - kmem_cache_free(dev_priv->requests, req); - return ret; -} - -/** - * i915_gem_request_alloc - allocate a request structure - * - * @engine: engine that we wish to issue the request on. - * @ctx: context that the request will be associated with. - * This can be NULL if the request is not directly related to - * any specific user context, in which case this function will - * choose an appropriate context to use. - * - * Returns a pointer to the allocated request if successful, - * or an error code if not. - */ -struct drm_i915_gem_request * -i915_gem_request_alloc(struct intel_engine_cs *engine, - struct i915_gem_context *ctx) -{ - struct drm_i915_gem_request *req; - int err; - - if (ctx == NULL) - ctx = engine->i915->kernel_context; - err = __i915_gem_request_alloc(engine, ctx, &req); - return err ? ERR_PTR(err) : req; -} - struct drm_i915_gem_request * i915_gem_find_active_request(struct intel_engine_cs *engine) { @@ -3089,185 +2545,143 @@ i915_gem_find_active_request(struct intel_engine_cs *engine) * extra delay for a recent interrupt is pointless. Hence, we do * not need an engine->irq_seqno_barrier() before the seqno reads. */ - list_for_each_entry(request, &engine->request_list, list) { + list_for_each_entry(request, &engine->request_list, link) { if (i915_gem_request_completed(request)) continue; + if (!i915_sw_fence_done(&request->submit)) + break; + return request; } return NULL; } -static void i915_gem_reset_engine_status(struct intel_engine_cs *engine) +static void reset_request(struct drm_i915_gem_request *request) +{ + void *vaddr = request->ring->vaddr; + u32 head; + + /* As this request likely depends on state from the lost + * context, clear out all the user operations leaving the + * breadcrumb at the end (so we get the fence notifications). + */ + head = request->head; + if (request->postfix < head) { + memset(vaddr + head, 0, request->ring->size - head); + head = 0; + } + memset(vaddr + head, 0, request->postfix - head); +} + +static void i915_gem_reset_engine(struct intel_engine_cs *engine) { struct drm_i915_gem_request *request; + struct i915_gem_context *incomplete_ctx; bool ring_hung; + /* Ensure irq handler finishes, and not run again. */ + tasklet_kill(&engine->irq_tasklet); + if (engine->irq_seqno_barrier) + engine->irq_seqno_barrier(engine); + request = i915_gem_find_active_request(engine); - if (request == NULL) + if (!request) return; ring_hung = engine->hangcheck.score >= HANGCHECK_SCORE_RING_HUNG; - i915_set_reset_status(request->ctx, ring_hung); - list_for_each_entry_continue(request, &engine->request_list, list) - i915_set_reset_status(request->ctx, false); -} - -static void i915_gem_reset_engine_cleanup(struct intel_engine_cs *engine) -{ - struct intel_ringbuffer *buffer; - - while (!list_empty(&engine->active_list)) { - struct drm_i915_gem_object *obj; - - obj = list_first_entry(&engine->active_list, - struct drm_i915_gem_object, - engine_list[engine->id]); - - i915_gem_object_retire__read(obj, engine->id); - } - - /* - * Clear the execlists queue up before freeing the requests, as those - * are the ones that keep the context and ringbuffer backing objects - * pinned in place. - */ - - if (i915.enable_execlists) { - /* Ensure irq handler finishes or is cancelled. */ - tasklet_kill(&engine->irq_tasklet); - - intel_execlists_cancel_requests(engine); - } - - /* - * We must free the requests after all the corresponding objects have - * been moved off active lists. Which is the same order as the normal - * retire_requests function does. This is important if object hold - * implicit references on things like e.g. ppgtt address spaces through - * the request. - */ - while (!list_empty(&engine->request_list)) { - struct drm_i915_gem_request *request; + if (!ring_hung) + return; - request = list_first_entry(&engine->request_list, - struct drm_i915_gem_request, - list); + DRM_DEBUG_DRIVER("resetting %s to restart from tail of request 0x%x\n", + engine->name, request->fence.seqno); - i915_gem_request_retire(request); - } + /* Setup the CS to resume from the breadcrumb of the hung request */ + engine->reset_hw(engine, request); - /* Having flushed all requests from all queues, we know that all - * ringbuffers must now be empty. However, since we do not reclaim - * all space when retiring the request (to prevent HEADs colliding - * with rapid ringbuffer wraparound) the amount of available space - * upon reset is less than when we start. Do one more pass over - * all the ringbuffers to reset last_retired_head. + /* Users of the default context do not rely on logical state + * preserved between batches. They have to emit full state on + * every batch and so it is safe to execute queued requests following + * the hang. + * + * Other contexts preserve state, now corrupt. We want to skip all + * queued requests that reference the corrupt context. */ - list_for_each_entry(buffer, &engine->buffers, link) { - buffer->last_retired_head = buffer->tail; - intel_ring_update_space(buffer); - } - - intel_ring_init_seqno(engine, engine->last_submitted_seqno); + incomplete_ctx = request->ctx; + if (i915_gem_context_is_default(incomplete_ctx)) + return; - engine->i915->gt.active_engines &= ~intel_engine_flag(engine); + list_for_each_entry_continue(request, &engine->request_list, link) + if (request->ctx == incomplete_ctx) + reset_request(request); } -void i915_gem_reset(struct drm_device *dev) +void i915_gem_reset(struct drm_i915_private *dev_priv) { - struct drm_i915_private *dev_priv = to_i915(dev); struct intel_engine_cs *engine; - /* - * Before we free the objects from the requests, we need to inspect - * them for finding the guilty party. As the requests only borrow - * their reference to the objects, the inspection must be done first. - */ - for_each_engine(engine, dev_priv) - i915_gem_reset_engine_status(engine); + i915_gem_retire_requests(dev_priv); for_each_engine(engine, dev_priv) - i915_gem_reset_engine_cleanup(engine); - mod_delayed_work(dev_priv->wq, &dev_priv->gt.idle_work, 0); + i915_gem_reset_engine(engine); - i915_gem_context_reset(dev); + i915_gem_restore_fences(&dev_priv->drm); - i915_gem_restore_fences(dev); + if (dev_priv->gt.awake) { + intel_sanitize_gt_powersave(dev_priv); + intel_enable_gt_powersave(dev_priv); + if (INTEL_GEN(dev_priv) >= 6) + gen6_rps_busy(dev_priv); + } +} - WARN_ON(i915_verify_lists(dev)); +static void nop_submit_request(struct drm_i915_gem_request *request) +{ } -/** - * This function clears the request list as sequence numbers are passed. - * @engine: engine to retire requests on - */ -void -i915_gem_retire_requests_ring(struct intel_engine_cs *engine) +static void i915_gem_cleanup_engine(struct intel_engine_cs *engine) { - WARN_ON(i915_verify_lists(engine->dev)); + engine->submit_request = nop_submit_request; - /* Retire requests first as we use it above for the early return. - * If we retire requests last, we may use a later seqno and so clear - * the requests lists without clearing the active list, leading to - * confusion. + /* Mark all pending requests as complete so that any concurrent + * (lockless) lookup doesn't try and wait upon the request as we + * reset it. */ - while (!list_empty(&engine->request_list)) { - struct drm_i915_gem_request *request; - - request = list_first_entry(&engine->request_list, - struct drm_i915_gem_request, - list); + intel_engine_init_seqno(engine, engine->last_submitted_seqno); - if (!i915_gem_request_completed(request)) - break; - - i915_gem_request_retire(request); - } - - /* Move any buffers on the active list that are no longer referenced - * by the ringbuffer to the flushing/inactive lists as appropriate, - * before we free the context associated with the requests. + /* + * Clear the execlists queue up before freeing the requests, as those + * are the ones that keep the context and ringbuffer backing objects + * pinned in place. */ - while (!list_empty(&engine->active_list)) { - struct drm_i915_gem_object *obj; - obj = list_first_entry(&engine->active_list, - struct drm_i915_gem_object, - engine_list[engine->id]); - - if (!list_empty(&obj->last_read_req[engine->id]->list)) - break; - - i915_gem_object_retire__read(obj, engine->id); + if (i915.enable_execlists) { + spin_lock(&engine->execlist_lock); + INIT_LIST_HEAD(&engine->execlist_queue); + i915_gem_request_put(engine->execlist_port[0].request); + i915_gem_request_put(engine->execlist_port[1].request); + memset(engine->execlist_port, 0, sizeof(engine->execlist_port)); + spin_unlock(&engine->execlist_lock); } - WARN_ON(i915_verify_lists(engine->dev)); + engine->i915->gt.active_engines &= ~intel_engine_flag(engine); } -void i915_gem_retire_requests(struct drm_i915_private *dev_priv) +void i915_gem_set_wedged(struct drm_i915_private *dev_priv) { struct intel_engine_cs *engine; lockdep_assert_held(&dev_priv->drm.struct_mutex); + set_bit(I915_WEDGED, &dev_priv->gpu_error.flags); - if (dev_priv->gt.active_engines == 0) - return; - - GEM_BUG_ON(!dev_priv->gt.awake); - - for_each_engine(engine, dev_priv) { - i915_gem_retire_requests_ring(engine); - if (list_empty(&engine->request_list)) - dev_priv->gt.active_engines &= ~intel_engine_flag(engine); - } + i915_gem_context_lost(dev_priv); + for_each_engine(engine, dev_priv) + i915_gem_cleanup_engine(engine); + mod_delayed_work(dev_priv->wq, &dev_priv->gt.idle_work, 0); - if (dev_priv->gt.active_engines == 0) - queue_delayed_work(dev_priv->wq, - &dev_priv->gt.idle_work, - msecs_to_jiffies(100)); + i915_gem_retire_requests(dev_priv); } static void @@ -3287,10 +2701,12 @@ i915_gem_retire_work_handler(struct work_struct *work) * 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)) + if (READ_ONCE(dev_priv->gt.awake)) { + i915_queue_hangcheck(dev_priv); queue_delayed_work(dev_priv->wq, &dev_priv->gt.retire_work, round_jiffies_up_relative(HZ)); + } } static void @@ -3300,7 +2716,6 @@ i915_gem_idle_work_handler(struct work_struct *work) container_of(work, typeof(*dev_priv), gt.idle_work.work); struct drm_device *dev = &dev_priv->drm; struct intel_engine_cs *engine; - unsigned int stuck_engines; bool rearm_hangcheck; if (!READ_ONCE(dev_priv->gt.awake)) @@ -3330,12 +2745,6 @@ i915_gem_idle_work_handler(struct work_struct *work) dev_priv->gt.awake = false; rearm_hangcheck = false; - stuck_engines = intel_kick_waiters(dev_priv); - if (unlikely(stuck_engines)) { - DRM_DEBUG_DRIVER("kicked stuck waiters...missed irq\n"); - dev_priv->gpu_error.missed_irq_rings |= stuck_engines; - } - if (INTEL_GEN(dev_priv) >= 6) gen6_rps_idle(dev_priv); intel_runtime_pm_put(dev_priv); @@ -3349,32 +2758,17 @@ out_rearm: } } -/** - * Ensures that an object will eventually get non-busy by flushing any required - * write domains, emitting any outstanding lazy request and retiring and - * completed requests. - * @obj: object to flush - */ -static int -i915_gem_object_flush_active(struct drm_i915_gem_object *obj) +void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file) { - int i; - - if (!obj->active) - return 0; - - for (i = 0; i < I915_NUM_ENGINES; i++) { - struct drm_i915_gem_request *req; - - req = obj->last_read_req[i]; - if (req == NULL) - continue; - - if (i915_gem_request_completed(req)) - i915_gem_object_retire__read(obj, i); - } + struct drm_i915_gem_object *obj = to_intel_bo(gem); + struct drm_i915_file_private *fpriv = file->driver_priv; + struct i915_vma *vma, *vn; - return 0; + mutex_lock(&obj->base.dev->struct_mutex); + list_for_each_entry_safe(vma, vn, &obj->vma_list, obj_link) + if (vma->vm->file == fpriv) + i915_vma_close(vma); + mutex_unlock(&obj->base.dev->struct_mutex); } /** @@ -3405,219 +2799,35 @@ int i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) { struct drm_i915_gem_wait *args = data; + struct intel_rps_client *rps = to_rps_client(file); struct drm_i915_gem_object *obj; - struct drm_i915_gem_request *req[I915_NUM_ENGINES]; - int i, n = 0; - int ret; + unsigned long active; + int idx, ret = 0; if (args->flags != 0) return -EINVAL; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; - - obj = to_intel_bo(drm_gem_object_lookup(file, args->bo_handle)); - if (&obj->base == NULL) { - mutex_unlock(&dev->struct_mutex); + obj = i915_gem_object_lookup(file, args->bo_handle); + if (!obj) return -ENOENT; - } - - /* Need to make sure the object gets inactive eventually. */ - ret = i915_gem_object_flush_active(obj); - if (ret) - goto out; - if (!obj->active) - goto out; - - /* Do this after OLR check to make sure we make forward progress polling - * on this IOCTL with a timeout == 0 (like busy ioctl) - */ - if (args->timeout_ns == 0) { - ret = -ETIME; - goto out; - } - - drm_gem_object_unreference(&obj->base); - - for (i = 0; i < I915_NUM_ENGINES; i++) { - if (obj->last_read_req[i] == NULL) - continue; - - req[n++] = i915_gem_request_reference(obj->last_read_req[i]); - } - - mutex_unlock(&dev->struct_mutex); - - for (i = 0; i < n; i++) { - if (ret == 0) - ret = __i915_wait_request(req[i], true, - args->timeout_ns > 0 ? &args->timeout_ns : NULL, - to_rps_client(file)); - i915_gem_request_unreference(req[i]); - } - return ret; - -out: - drm_gem_object_unreference(&obj->base); - mutex_unlock(&dev->struct_mutex); - return ret; -} - -static int -__i915_gem_object_sync(struct drm_i915_gem_object *obj, - struct intel_engine_cs *to, - struct drm_i915_gem_request *from_req, - struct drm_i915_gem_request **to_req) -{ - struct intel_engine_cs *from; - int ret; - - from = i915_gem_request_get_engine(from_req); - if (to == from) - return 0; - - if (i915_gem_request_completed(from_req)) - return 0; - - if (!i915_semaphore_is_enabled(to_i915(obj->base.dev))) { - struct drm_i915_private *i915 = to_i915(obj->base.dev); - ret = __i915_wait_request(from_req, - i915->mm.interruptible, - NULL, - &i915->rps.semaphores); - if (ret) - return ret; - - i915_gem_object_retire_request(obj, from_req); - } else { - int idx = intel_ring_sync_index(from, to); - u32 seqno = i915_gem_request_get_seqno(from_req); - - WARN_ON(!to_req); - - if (seqno <= from->semaphore.sync_seqno[idx]) - return 0; - - if (*to_req == NULL) { - struct drm_i915_gem_request *req; - - req = i915_gem_request_alloc(to, NULL); - if (IS_ERR(req)) - return PTR_ERR(req); - - *to_req = req; - } - - trace_i915_gem_ring_sync_to(*to_req, from, from_req); - ret = to->semaphore.sync_to(*to_req, from, seqno); - if (ret) - return ret; - - /* We use last_read_req because sync_to() - * might have just caused seqno wrap under - * the radar. - */ - from->semaphore.sync_seqno[idx] = - i915_gem_request_get_seqno(obj->last_read_req[from->id]); - } - - return 0; -} - -/** - * i915_gem_object_sync - sync an object to a ring. - * - * @obj: object which may be in use on another ring. - * @to: ring we wish to use the object on. May be NULL. - * @to_req: request we wish to use the object for. See below. - * This will be allocated and returned if a request is - * required but not passed in. - * - * This code is meant to abstract object synchronization with the GPU. - * Calling with NULL implies synchronizing the object with the CPU - * rather than a particular GPU ring. Conceptually we serialise writes - * between engines inside the GPU. We only allow one engine to write - * into a buffer at any time, but multiple readers. To ensure each has - * a coherent view of memory, we must: - * - * - If there is an outstanding write request to the object, the new - * request must wait for it to complete (either CPU or in hw, requests - * on the same ring will be naturally ordered). - * - * - If we are a write request (pending_write_domain is set), the new - * request must wait for outstanding read requests to complete. - * - * For CPU synchronisation (NULL to) no request is required. For syncing with - * rings to_req must be non-NULL. However, a request does not have to be - * pre-allocated. If *to_req is NULL and sync commands will be emitted then a - * request will be allocated automatically and returned through *to_req. Note - * that it is not guaranteed that commands will be emitted (because the system - * might already be idle). Hence there is no need to create a request that - * might never have any work submitted. Note further that if a request is - * returned in *to_req, it is the responsibility of the caller to submit - * that request (after potentially adding more work to it). - * - * Returns 0 if successful, else propagates up the lower layer error. - */ -int -i915_gem_object_sync(struct drm_i915_gem_object *obj, - struct intel_engine_cs *to, - struct drm_i915_gem_request **to_req) -{ - const bool readonly = obj->base.pending_write_domain == 0; - struct drm_i915_gem_request *req[I915_NUM_ENGINES]; - int ret, i, n; - - if (!obj->active) - return 0; - - if (to == NULL) - return i915_gem_object_wait_rendering(obj, readonly); - - n = 0; - if (readonly) { - if (obj->last_write_req) - req[n++] = obj->last_write_req; - } else { - for (i = 0; i < I915_NUM_ENGINES; i++) - if (obj->last_read_req[i]) - req[n++] = obj->last_read_req[i]; - } - for (i = 0; i < n; i++) { - ret = __i915_gem_object_sync(obj, to, req[i], to_req); + active = __I915_BO_ACTIVE(obj); + for_each_active(active, idx) { + s64 *timeout = args->timeout_ns >= 0 ? &args->timeout_ns : NULL; + ret = i915_gem_active_wait_unlocked(&obj->last_read[idx], + I915_WAIT_INTERRUPTIBLE, + timeout, rps); if (ret) - return ret; + break; } - return 0; -} - -static void i915_gem_object_finish_gtt(struct drm_i915_gem_object *obj) -{ - u32 old_write_domain, old_read_domains; - - /* Force a pagefault for domain tracking on next user access */ - i915_gem_release_mmap(obj); - - if ((obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) - return; - - old_read_domains = obj->base.read_domains; - old_write_domain = obj->base.write_domain; - - obj->base.read_domains &= ~I915_GEM_DOMAIN_GTT; - obj->base.write_domain &= ~I915_GEM_DOMAIN_GTT; - - trace_i915_gem_object_change_domain(obj, - old_read_domains, - old_write_domain); + i915_gem_object_put_unlocked(obj); + return ret; } static void __i915_vma_iounmap(struct i915_vma *vma) { - GEM_BUG_ON(vma->pin_count); + GEM_BUG_ON(i915_vma_is_pinned(vma)); if (vma->iomap == NULL) return; @@ -3626,65 +2836,83 @@ static void __i915_vma_iounmap(struct i915_vma *vma) vma->iomap = NULL; } -static int __i915_vma_unbind(struct i915_vma *vma, bool wait) +int i915_vma_unbind(struct i915_vma *vma) { struct drm_i915_gem_object *obj = vma->obj; - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); + unsigned long active; int ret; - if (list_empty(&vma->obj_link)) - return 0; - - if (!drm_mm_node_allocated(&vma->node)) { - i915_gem_vma_destroy(vma); - return 0; - } - - if (vma->pin_count) - return -EBUSY; + /* First wait upon any activity as retiring the request may + * have side-effects such as unpinning or even unbinding this vma. + */ + active = i915_vma_get_active(vma); + if (active) { + int idx; + + /* When a closed VMA is retired, it is unbound - eek. + * In order to prevent it from being recursively closed, + * take a pin on the vma so that the second unbind is + * aborted. + */ + __i915_vma_pin(vma); - BUG_ON(obj->pages == NULL); + for_each_active(active, idx) { + ret = i915_gem_active_retire(&vma->last_read[idx], + &vma->vm->dev->struct_mutex); + if (ret) + break; + } - if (wait) { - ret = i915_gem_object_wait_rendering(obj, false); + __i915_vma_unpin(vma); if (ret) return ret; + + GEM_BUG_ON(i915_vma_is_active(vma)); } - if (vma->is_ggtt && vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) { - i915_gem_object_finish_gtt(obj); + if (i915_vma_is_pinned(vma)) + return -EBUSY; + + if (!drm_mm_node_allocated(&vma->node)) + goto destroy; + GEM_BUG_ON(obj->bind_count == 0); + GEM_BUG_ON(!obj->pages); + + if (i915_vma_is_map_and_fenceable(vma)) { /* release the fence reg _after_ flushing */ - ret = i915_gem_object_put_fence(obj); + ret = i915_vma_put_fence(vma); if (ret) return ret; + /* Force a pagefault for domain tracking on next user access */ + i915_gem_release_mmap(obj); + __i915_vma_iounmap(vma); + vma->flags &= ~I915_VMA_CAN_FENCE; } - trace_i915_vma_unbind(vma); - - vma->vm->unbind_vma(vma); - vma->bound = 0; - - list_del_init(&vma->vm_link); - if (vma->is_ggtt) { - if (vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) { - obj->map_and_fenceable = false; - } else if (vma->ggtt_view.pages) { - sg_free_table(vma->ggtt_view.pages); - kfree(vma->ggtt_view.pages); - } - vma->ggtt_view.pages = NULL; + if (likely(!vma->vm->closed)) { + trace_i915_vma_unbind(vma); + vma->vm->unbind_vma(vma); } + vma->flags &= ~(I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND); drm_mm_remove_node(&vma->node); - i915_gem_vma_destroy(vma); + list_move_tail(&vma->vm_link, &vma->vm->unbound_list); + + if (vma->pages != obj->pages) { + GEM_BUG_ON(!vma->pages); + sg_free_table(vma->pages); + kfree(vma->pages); + } + vma->pages = NULL; /* Since the unbound list is global, only move to that list if * no more VMAs exist. */ - if (list_empty(&obj->vma_list)) - list_move_tail(&obj->global_list, &dev_priv->mm.unbound_list); + if (--obj->bind_count == 0) + list_move_tail(&obj->global_list, + &to_i915(obj->base.dev)->mm.unbound_list); /* And finally now the object is completely decoupled from this vma, * we can drop its hold on the backing storage and allow it to be @@ -3692,36 +2920,28 @@ static int __i915_vma_unbind(struct i915_vma *vma, bool wait) */ i915_gem_object_unpin_pages(obj); - return 0; -} +destroy: + if (unlikely(i915_vma_is_closed(vma))) + i915_vma_destroy(vma); -int i915_vma_unbind(struct i915_vma *vma) -{ - return __i915_vma_unbind(vma, true); -} - -int __i915_vma_unbind_no_wait(struct i915_vma *vma) -{ - return __i915_vma_unbind(vma, false); + return 0; } -int i915_gem_wait_for_idle(struct drm_i915_private *dev_priv) +int i915_gem_wait_for_idle(struct drm_i915_private *dev_priv, + unsigned int flags) { struct intel_engine_cs *engine; int ret; - lockdep_assert_held(&dev_priv->drm.struct_mutex); - for_each_engine(engine, dev_priv) { if (engine->last_context == NULL) continue; - ret = intel_engine_idle(engine); + ret = intel_engine_idle(engine, flags); if (ret) return ret; } - WARN_ON(i915_verify_lists(dev)); return 0; } @@ -3759,128 +2979,87 @@ static bool i915_gem_valid_gtt_space(struct i915_vma *vma, } /** - * Finds free space in the GTT aperture and binds the object or a view of it - * there. - * @obj: object to bind - * @vm: address space to bind into - * @ggtt_view: global gtt view if applicable - * @alignment: requested alignment + * i915_vma_insert - finds a slot for the vma in its address space + * @vma: the vma + * @size: requested size in bytes (can be larger than the VMA) + * @alignment: required alignment * @flags: mask of PIN_* flags to use + * + * First we try to allocate some free space that meets the requirements for + * the VMA. Failiing that, if the flags permit, it will evict an old VMA, + * preferrably the oldest idle entry to make room for the new VMA. + * + * Returns: + * 0 on success, negative error code otherwise. */ -static struct i915_vma * -i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj, - struct i915_address_space *vm, - const struct i915_ggtt_view *ggtt_view, - unsigned alignment, - uint64_t flags) +static int +i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) { - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - struct i915_ggtt *ggtt = &dev_priv->ggtt; - u32 fence_alignment, unfenced_alignment; - u32 search_flag, alloc_flag; + struct drm_i915_private *dev_priv = to_i915(vma->vm->dev); + struct drm_i915_gem_object *obj = vma->obj; u64 start, end; - u64 size, fence_size; - struct i915_vma *vma; int ret; - if (i915_is_ggtt(vm)) { - u32 view_size; - - if (WARN_ON(!ggtt_view)) - return ERR_PTR(-EINVAL); - - view_size = i915_ggtt_view_size(obj, ggtt_view); - - fence_size = i915_gem_get_gtt_size(dev, - view_size, - obj->tiling_mode); - fence_alignment = i915_gem_get_gtt_alignment(dev, - view_size, - obj->tiling_mode, - true); - unfenced_alignment = i915_gem_get_gtt_alignment(dev, - view_size, - obj->tiling_mode, - false); - size = flags & PIN_MAPPABLE ? fence_size : view_size; - } else { - fence_size = i915_gem_get_gtt_size(dev, - obj->base.size, - obj->tiling_mode); - fence_alignment = i915_gem_get_gtt_alignment(dev, - obj->base.size, - obj->tiling_mode, - true); - unfenced_alignment = - i915_gem_get_gtt_alignment(dev, - obj->base.size, - obj->tiling_mode, - false); - size = flags & PIN_MAPPABLE ? fence_size : obj->base.size; - } + GEM_BUG_ON(vma->flags & (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND)); + GEM_BUG_ON(drm_mm_node_allocated(&vma->node)); + + size = max(size, vma->size); + if (flags & PIN_MAPPABLE) + size = i915_gem_get_ggtt_size(dev_priv, size, + i915_gem_object_get_tiling(obj)); + + alignment = max(max(alignment, vma->display_alignment), + i915_gem_get_ggtt_alignment(dev_priv, size, + i915_gem_object_get_tiling(obj), + flags & PIN_MAPPABLE)); start = flags & PIN_OFFSET_BIAS ? flags & PIN_OFFSET_MASK : 0; - end = vm->total; + + end = vma->vm->total; if (flags & PIN_MAPPABLE) - end = min_t(u64, end, ggtt->mappable_end); + end = min_t(u64, end, dev_priv->ggtt.mappable_end); if (flags & PIN_ZONE_4G) end = min_t(u64, end, (1ULL << 32) - PAGE_SIZE); - if (alignment == 0) - alignment = flags & PIN_MAPPABLE ? fence_alignment : - unfenced_alignment; - if (flags & PIN_MAPPABLE && alignment & (fence_alignment - 1)) { - DRM_DEBUG("Invalid object (view type=%u) alignment requested %u\n", - ggtt_view ? ggtt_view->type : 0, - alignment); - return ERR_PTR(-EINVAL); - } - /* If binding the object/GGTT view requires more space than the entire * aperture has, reject it early before evicting everything in a vain * attempt to find space. */ if (size > end) { - DRM_DEBUG("Attempting to bind an object (view type=%u) larger than the aperture: size=%llu > %s aperture=%llu\n", - ggtt_view ? ggtt_view->type : 0, - size, + DRM_DEBUG("Attempting to bind an object larger than the aperture: request=%llu [object=%zd] > %s aperture=%llu\n", + size, obj->base.size, flags & PIN_MAPPABLE ? "mappable" : "total", end); - return ERR_PTR(-E2BIG); + return -E2BIG; } ret = i915_gem_object_get_pages(obj); if (ret) - return ERR_PTR(ret); + return ret; i915_gem_object_pin_pages(obj); - vma = ggtt_view ? i915_gem_obj_lookup_or_create_ggtt_vma(obj, ggtt_view) : - i915_gem_obj_lookup_or_create_vma(obj, vm); - - if (IS_ERR(vma)) - goto err_unpin; - if (flags & PIN_OFFSET_FIXED) { - uint64_t offset = flags & PIN_OFFSET_MASK; - - if (offset & (alignment - 1) || offset + size > end) { + u64 offset = flags & PIN_OFFSET_MASK; + if (offset & (alignment - 1) || offset > end - size) { ret = -EINVAL; - goto err_free_vma; + goto err_unpin; } + vma->node.start = offset; vma->node.size = size; vma->node.color = obj->cache_level; - ret = drm_mm_reserve_node(&vm->mm, &vma->node); + ret = drm_mm_reserve_node(&vma->vm->mm, &vma->node); if (ret) { ret = i915_gem_evict_for_vma(vma); if (ret == 0) - ret = drm_mm_reserve_node(&vm->mm, &vma->node); + ret = drm_mm_reserve_node(&vma->vm->mm, &vma->node); + if (ret) + goto err_unpin; } - if (ret) - goto err_free_vma; } else { + u32 search_flag, alloc_flag; + if (flags & PIN_HIGH) { search_flag = DRM_MM_SEARCH_BELOW; alloc_flag = DRM_MM_CREATE_TOP; @@ -3889,47 +3068,45 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj, alloc_flag = DRM_MM_CREATE_DEFAULT; } + /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks, + * so we know that we always have a minimum alignment of 4096. + * The drm_mm range manager is optimised to return results + * with zero alignment, so where possible use the optimal + * path. + */ + if (alignment <= 4096) + alignment = 0; + search_free: - ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node, + ret = drm_mm_insert_node_in_range_generic(&vma->vm->mm, + &vma->node, size, alignment, obj->cache_level, start, end, search_flag, alloc_flag); if (ret) { - ret = i915_gem_evict_something(dev, vm, size, alignment, + ret = i915_gem_evict_something(vma->vm, size, alignment, obj->cache_level, start, end, flags); if (ret == 0) goto search_free; - goto err_free_vma; + goto err_unpin; } } - if (WARN_ON(!i915_gem_valid_gtt_space(vma, obj->cache_level))) { - ret = -EINVAL; - goto err_remove_node; - } - - trace_i915_vma_bind(vma, flags); - ret = i915_vma_bind(vma, obj->cache_level, flags); - if (ret) - goto err_remove_node; + GEM_BUG_ON(!i915_gem_valid_gtt_space(vma, obj->cache_level)); list_move_tail(&obj->global_list, &dev_priv->mm.bound_list); - list_add_tail(&vma->vm_link, &vm->inactive_list); + list_move_tail(&vma->vm_link, &vma->vm->inactive_list); + obj->bind_count++; - return vma; + return 0; -err_remove_node: - drm_mm_remove_node(&vma->node); -err_free_vma: - i915_gem_vma_destroy(vma); - vma = ERR_PTR(ret); err_unpin: i915_gem_object_unpin_pages(obj); - return vma; + return ret; } bool @@ -3974,51 +3151,72 @@ i915_gem_clflush_object(struct drm_i915_gem_object *obj, static void i915_gem_object_flush_gtt_write_domain(struct drm_i915_gem_object *obj) { - uint32_t old_write_domain; + struct drm_i915_private *dev_priv = to_i915(obj->base.dev); if (obj->base.write_domain != I915_GEM_DOMAIN_GTT) return; /* No actual flushing is required for the GTT write domain. Writes - * to it immediately go to main memory as far as we know, so there's + * to it "immediately" go to main memory as far as we know, so there's * no chipset flush. It also doesn't land in render cache. * * However, we do have to enforce the order so that all writes through * the GTT land before any writes to the device, such as updates to * the GATT itself. + * + * We also have to wait a bit for the writes to land from the GTT. + * An uncached read (i.e. mmio) seems to be ideal for the round-trip + * timing. This issue has only been observed when switching quickly + * between GTT writes and CPU reads from inside the kernel on recent hw, + * and it appears to only affect discrete GTT blocks (i.e. on LLC + * system agents we cannot reproduce this behaviour). */ wmb(); + if (INTEL_GEN(dev_priv) >= 6 && !HAS_LLC(dev_priv)) + POSTING_READ(RING_ACTHD(dev_priv->engine[RCS].mmio_base)); - old_write_domain = obj->base.write_domain; - obj->base.write_domain = 0; - - intel_fb_obj_flush(obj, false, ORIGIN_GTT); + intel_fb_obj_flush(obj, false, write_origin(obj, I915_GEM_DOMAIN_GTT)); + obj->base.write_domain = 0; trace_i915_gem_object_change_domain(obj, obj->base.read_domains, - old_write_domain); + I915_GEM_DOMAIN_GTT); } /** Flushes the CPU write domain for the object if it's dirty. */ static void i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj) { - uint32_t old_write_domain; - if (obj->base.write_domain != I915_GEM_DOMAIN_CPU) return; if (i915_gem_clflush_object(obj, obj->pin_display)) i915_gem_chipset_flush(to_i915(obj->base.dev)); - old_write_domain = obj->base.write_domain; - obj->base.write_domain = 0; - intel_fb_obj_flush(obj, false, ORIGIN_CPU); + obj->base.write_domain = 0; trace_i915_gem_object_change_domain(obj, obj->base.read_domains, - old_write_domain); + I915_GEM_DOMAIN_CPU); +} + +static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj) +{ + struct i915_vma *vma; + + list_for_each_entry(vma, &obj->vma_list, obj_link) { + if (!i915_vma_is_ggtt(vma)) + continue; + + if (i915_vma_is_active(vma)) + continue; + + if (!drm_mm_node_allocated(&vma->node)) + continue; + + list_move_tail(&vma->vm_link, &vma->vm->inactive_list); + } } /** @@ -4032,20 +3230,16 @@ i915_gem_object_flush_cpu_write_domain(struct drm_i915_gem_object *obj) int i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) { - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - struct i915_ggtt *ggtt = &dev_priv->ggtt; uint32_t old_write_domain, old_read_domains; - struct i915_vma *vma; int ret; - if (obj->base.write_domain == I915_GEM_DOMAIN_GTT) - return 0; - ret = i915_gem_object_wait_rendering(obj, !write); if (ret) return ret; + if (obj->base.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. @@ -4086,10 +3280,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) old_write_domain); /* And bump the LRU for this access */ - vma = i915_gem_obj_to_ggtt(obj); - if (vma && drm_mm_node_allocated(&vma->node) && !obj->active) - list_move_tail(&vma->vm_link, - &ggtt->base.inactive_list); + i915_gem_object_bump_inactive_ggtt(obj); return 0; } @@ -4112,9 +3303,7 @@ i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj, bool write) int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, enum i915_cache_level cache_level) { - struct drm_device *dev = obj->base.dev; - struct i915_vma *vma, *next; - bool bound = false; + struct i915_vma *vma; int ret = 0; if (obj->cache_level == cache_level) @@ -4125,21 +3314,28 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, * catch the issue of the CS prefetch crossing page boundaries and * reading an invalid PTE on older architectures. */ - list_for_each_entry_safe(vma, next, &obj->vma_list, obj_link) { +restart: + list_for_each_entry(vma, &obj->vma_list, obj_link) { if (!drm_mm_node_allocated(&vma->node)) continue; - if (vma->pin_count) { + if (i915_vma_is_pinned(vma)) { DRM_DEBUG("can not change the cache level of pinned objects\n"); return -EBUSY; } - if (!i915_gem_valid_gtt_space(vma, cache_level)) { - ret = i915_vma_unbind(vma); - if (ret) - return ret; - } else - bound = true; + if (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 @@ -4149,7 +3345,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, * rewrite the PTE in the belief that doing so tramples upon less * state and so involves less work. */ - if (bound) { + 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. @@ -4158,7 +3354,7 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, if (ret) return ret; - if (!HAS_LLC(dev) && cache_level != I915_CACHE_NONE) { + if (!HAS_LLC(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 @@ -4175,9 +3371,11 @@ int i915_gem_object_set_cache_level(struct drm_i915_gem_object *obj, * dropped the fence as all snoopable access is * supposed to be linear. */ - ret = i915_gem_object_put_fence(obj); - if (ret) - return ret; + list_for_each_entry(vma, &obj->vma_list, obj_link) { + 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 @@ -4221,8 +3419,8 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, struct drm_i915_gem_caching *args = data; struct drm_i915_gem_object *obj; - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) return -ENOENT; switch (obj->cache_level) { @@ -4240,7 +3438,7 @@ int i915_gem_get_caching_ioctl(struct drm_device *dev, void *data, break; } - drm_gem_object_unreference_unlocked(&obj->base); + i915_gem_object_put_unlocked(obj); return 0; } @@ -4282,15 +3480,15 @@ int i915_gem_set_caching_ioctl(struct drm_device *dev, void *data, if (ret) goto rpm_put; - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) { ret = -ENOENT; goto unlock; } ret = i915_gem_object_set_cache_level(obj, level); - drm_gem_object_unreference(&obj->base); + i915_gem_object_put(obj); unlock: mutex_unlock(&dev->struct_mutex); rpm_put: @@ -4304,11 +3502,12 @@ rpm_put: * Can be called from an uninterruptible phase (modesetting) and allows * any flushes to be pipelined (for pageflips). */ -int +struct i915_vma * i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, u32 alignment, const struct i915_ggtt_view *view) { + struct i915_vma *vma; u32 old_read_domains, old_write_domain; int ret; @@ -4328,19 +3527,31 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, */ ret = i915_gem_object_set_cache_level(obj, HAS_WT(obj->base.dev) ? I915_CACHE_WT : I915_CACHE_NONE); - if (ret) + if (ret) { + vma = ERR_PTR(ret); goto err_unpin_display; + } /* As the user may map the buffer once pinned in the display plane * (e.g. libkms for the bootup splash), we have to ensure that we - * always use map_and_fenceable for all scanout buffers. + * 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). */ - ret = i915_gem_object_ggtt_pin(obj, view, alignment, - view->type == I915_GGTT_VIEW_NORMAL ? - PIN_MAPPABLE : 0); - if (ret) + vma = ERR_PTR(-ENOSPC); + if (view->type == I915_GGTT_VIEW_NORMAL) + vma = i915_gem_object_ggtt_pin(obj, view, 0, alignment, + PIN_MAPPABLE | PIN_NONBLOCK); + if (IS_ERR(vma)) + vma = i915_gem_object_ggtt_pin(obj, view, 0, alignment, 0); + if (IS_ERR(vma)) goto err_unpin_display; + vma->display_alignment = max_t(u64, vma->display_alignment, alignment); + + WARN_ON(obj->pin_display > i915_vma_pin_count(vma)); + i915_gem_object_flush_cpu_write_domain(obj); old_write_domain = obj->base.write_domain; @@ -4356,23 +3567,28 @@ i915_gem_object_pin_to_display_plane(struct drm_i915_gem_object *obj, old_read_domains, old_write_domain); - return 0; + return vma; err_unpin_display: obj->pin_display--; - return ret; + return vma; } void -i915_gem_object_unpin_from_display_plane(struct drm_i915_gem_object *obj, - const struct i915_ggtt_view *view) +i915_gem_object_unpin_from_display_plane(struct i915_vma *vma) { - if (WARN_ON(obj->pin_display == 0)) + if (WARN_ON(vma->obj->pin_display == 0)) return; - i915_gem_object_ggtt_unpin_view(obj, view); + if (--vma->obj->pin_display == 0) + vma->display_alignment = 0; - obj->pin_display--; + /* Bump the LRU to try and avoid premature eviction whilst flipping */ + if (!i915_vma_is_active(vma)) + list_move_tail(&vma->vm_link, &vma->vm->inactive_list); + + i915_vma_unpin(vma); + WARN_ON(vma->obj->pin_display > i915_vma_pin_count(vma)); } /** @@ -4389,13 +3605,13 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_object *obj, bool write) uint32_t old_write_domain, old_read_domains; int ret; - if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) - return 0; - ret = i915_gem_object_wait_rendering(obj, !write); if (ret) return ret; + if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) + return 0; + i915_gem_object_flush_gtt_write_domain(obj); old_write_domain = obj->base.write_domain; @@ -4470,28 +3686,31 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file) target = request; } if (target) - i915_gem_request_reference(target); + i915_gem_request_get(target); spin_unlock(&file_priv->mm.lock); if (target == NULL) return 0; - ret = __i915_wait_request(target, true, NULL, NULL); - i915_gem_request_unreference(target); + ret = i915_wait_request(target, I915_WAIT_INTERRUPTIBLE, NULL, NULL); + i915_gem_request_put(target); return ret; } static bool -i915_vma_misplaced(struct i915_vma *vma, uint32_t alignment, uint64_t flags) +i915_vma_misplaced(struct i915_vma *vma, u64 size, u64 alignment, u64 flags) { - struct drm_i915_gem_object *obj = vma->obj; + if (!drm_mm_node_allocated(&vma->node)) + return false; + + if (vma->node.size < size) + return true; - if (alignment && - vma->node.start & (alignment - 1)) + if (alignment && vma->node.start & (alignment - 1)) return true; - if (flags & PIN_MAPPABLE && !obj->map_and_fenceable) + if (flags & PIN_MAPPABLE && !i915_vma_is_map_and_fenceable(vma)) return true; if (flags & PIN_OFFSET_BIAS && @@ -4508,135 +3727,208 @@ i915_vma_misplaced(struct i915_vma *vma, uint32_t alignment, uint64_t flags) void __i915_vma_set_map_and_fenceable(struct i915_vma *vma) { struct drm_i915_gem_object *obj = vma->obj; + struct drm_i915_private *dev_priv = to_i915(obj->base.dev); bool mappable, fenceable; u32 fence_size, fence_alignment; - fence_size = i915_gem_get_gtt_size(obj->base.dev, - obj->base.size, - obj->tiling_mode); - fence_alignment = i915_gem_get_gtt_alignment(obj->base.dev, - obj->base.size, - obj->tiling_mode, - true); + fence_size = i915_gem_get_ggtt_size(dev_priv, + vma->size, + i915_gem_object_get_tiling(obj)); + fence_alignment = i915_gem_get_ggtt_alignment(dev_priv, + vma->size, + i915_gem_object_get_tiling(obj), + true); fenceable = (vma->node.size == fence_size && (vma->node.start & (fence_alignment - 1)) == 0); mappable = (vma->node.start + fence_size <= - to_i915(obj->base.dev)->ggtt.mappable_end); + dev_priv->ggtt.mappable_end); - obj->map_and_fenceable = mappable && fenceable; + if (mappable && fenceable) + vma->flags |= I915_VMA_CAN_FENCE; + else + vma->flags &= ~I915_VMA_CAN_FENCE; } -static int -i915_gem_object_do_pin(struct drm_i915_gem_object *obj, - struct i915_address_space *vm, - const struct i915_ggtt_view *ggtt_view, - uint32_t alignment, - uint64_t flags) +int __i915_vma_do_pin(struct i915_vma *vma, + u64 size, u64 alignment, u64 flags) { - struct drm_i915_private *dev_priv = to_i915(obj->base.dev); - struct i915_vma *vma; - unsigned bound; + unsigned int bound = vma->flags; int ret; - if (WARN_ON(vm == &dev_priv->mm.aliasing_ppgtt->base)) - return -ENODEV; + GEM_BUG_ON((flags & (PIN_GLOBAL | PIN_USER)) == 0); + GEM_BUG_ON((flags & PIN_GLOBAL) && !i915_vma_is_ggtt(vma)); - if (WARN_ON(flags & (PIN_GLOBAL | PIN_MAPPABLE) && !i915_is_ggtt(vm))) - return -EINVAL; + if (WARN_ON(bound & I915_VMA_PIN_OVERFLOW)) { + ret = -EBUSY; + goto err; + } - if (WARN_ON((flags & (PIN_MAPPABLE | PIN_GLOBAL)) == PIN_MAPPABLE)) - return -EINVAL; + if ((bound & I915_VMA_BIND_MASK) == 0) { + ret = i915_vma_insert(vma, size, alignment, flags); + if (ret) + goto err; + } - if (WARN_ON(i915_is_ggtt(vm) != !!ggtt_view)) - return -EINVAL; + ret = i915_vma_bind(vma, vma->obj->cache_level, flags); + if (ret) + goto err; - vma = ggtt_view ? i915_gem_obj_to_ggtt_view(obj, ggtt_view) : - i915_gem_obj_to_vma(obj, vm); + if ((bound ^ vma->flags) & I915_VMA_GLOBAL_BIND) + __i915_vma_set_map_and_fenceable(vma); - if (vma) { - if (WARN_ON(vma->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT)) - return -EBUSY; + GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags)); + return 0; - if (i915_vma_misplaced(vma, alignment, flags)) { - WARN(vma->pin_count, - "bo is already pinned in %s with incorrect alignment:" - " offset=%08x %08x, req.alignment=%x, req.map_and_fenceable=%d," - " obj->map_and_fenceable=%d\n", - ggtt_view ? "ggtt" : "ppgtt", - upper_32_bits(vma->node.start), - lower_32_bits(vma->node.start), - alignment, - !!(flags & PIN_MAPPABLE), - obj->map_and_fenceable); - ret = i915_vma_unbind(vma); - if (ret) - return ret; +err: + __i915_vma_unpin(vma); + return ret; +} - vma = NULL; - } - } +struct i915_vma * +i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, + const struct i915_ggtt_view *view, + u64 size, + u64 alignment, + u64 flags) +{ + struct i915_address_space *vm = &to_i915(obj->base.dev)->ggtt.base; + struct i915_vma *vma; + int ret; - bound = vma ? vma->bound : 0; - if (vma == NULL || !drm_mm_node_allocated(&vma->node)) { - vma = i915_gem_object_bind_to_vm(obj, vm, ggtt_view, alignment, - flags); - if (IS_ERR(vma)) - return PTR_ERR(vma); - } else { - ret = i915_vma_bind(vma, obj->cache_level, flags); + vma = i915_gem_obj_lookup_or_create_vma(obj, vm, view); + if (IS_ERR(vma)) + return vma; + + if (i915_vma_misplaced(vma, size, alignment, flags)) { + if (flags & PIN_NONBLOCK && + (i915_vma_is_pinned(vma) || i915_vma_is_active(vma))) + return ERR_PTR(-ENOSPC); + + WARN(i915_vma_is_pinned(vma), + "bo is already pinned in ggtt with incorrect alignment:" + " offset=%08x, req.alignment=%llx," + " req.map_and_fenceable=%d, vma->map_and_fenceable=%d\n", + i915_ggtt_offset(vma), alignment, + !!(flags & PIN_MAPPABLE), + i915_vma_is_map_and_fenceable(vma)); + ret = i915_vma_unbind(vma); if (ret) - return ret; + return ERR_PTR(ret); } - if (ggtt_view && ggtt_view->type == I915_GGTT_VIEW_NORMAL && - (bound ^ vma->bound) & GLOBAL_BIND) { - __i915_vma_set_map_and_fenceable(vma); - WARN_ON(flags & PIN_MAPPABLE && !obj->map_and_fenceable); - } + ret = i915_vma_pin(vma, size, alignment, flags | PIN_GLOBAL); + if (ret) + return ERR_PTR(ret); - vma->pin_count++; - return 0; + return vma; } -int -i915_gem_object_pin(struct drm_i915_gem_object *obj, - struct i915_address_space *vm, - uint32_t alignment, - uint64_t flags) +static __always_inline unsigned int __busy_read_flag(unsigned int id) { - return i915_gem_object_do_pin(obj, vm, - i915_is_ggtt(vm) ? &i915_ggtt_view_normal : NULL, - alignment, flags); + /* Note that we could alias engines in the execbuf API, but + * that would be very unwise as it prevents userspace from + * fine control over engine selection. Ahem. + * + * This should be something like EXEC_MAX_ENGINE instead of + * I915_NUM_ENGINES. + */ + BUILD_BUG_ON(I915_NUM_ENGINES > 16); + return 0x10000 << id; } -int -i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj, - const struct i915_ggtt_view *view, - uint32_t alignment, - uint64_t flags) +static __always_inline unsigned int __busy_write_id(unsigned int id) { - struct drm_device *dev = obj->base.dev; - struct drm_i915_private *dev_priv = to_i915(dev); - struct i915_ggtt *ggtt = &dev_priv->ggtt; + /* 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. + */ + return id | __busy_read_flag(id); +} - BUG_ON(!view); +static __always_inline unsigned int +__busy_set_if_active(const struct i915_gem_active *active, + unsigned int (*flag)(unsigned int id)) +{ + struct drm_i915_gem_request *request; - return i915_gem_object_do_pin(obj, &ggtt->base, view, - alignment, flags | PIN_GLOBAL); + request = rcu_dereference(active->request); + if (!request || i915_gem_request_completed(request)) + return 0; + + /* This is racy. See __i915_gem_active_get_rcu() for an in detail + * discussion of how to handle the race correctly, but for reporting + * the busy state we err on the side of potentially reporting the + * wrong engine as being busy (but we guarantee that the result + * is at least self-consistent). + * + * As we use SLAB_DESTROY_BY_RCU, the request may be reallocated + * whilst we are inspecting it, even under the RCU read lock as we are. + * This means that there is a small window for the engine and/or the + * seqno to have been overwritten. The seqno will always be in the + * future compared to the intended, and so we know that if that + * seqno is idle (on whatever engine) our request is idle and the + * return 0 above is correct. + * + * The issue is that if the engine is switched, it is just as likely + * to report that it is busy (but since the switch happened, we know + * the request should be idle). So there is a small chance that a busy + * result is actually the wrong engine. + * + * So why don't we care? + * + * For starters, the busy ioctl is a heuristic that is by definition + * racy. Even with perfect serialisation in the driver, the hardware + * state is constantly advancing - the state we report to the user + * is stale. + * + * The critical information for the busy-ioctl is whether the object + * is idle as userspace relies on that to detect whether its next + * access will stall, or if it has missed submitting commands to + * the hardware allowing the GPU to stall. We never generate a + * false-positive for idleness, thus busy-ioctl is reliable at the + * most fundamental level, and we maintain the guarantee that a + * busy object left to itself will eventually become idle (and stay + * idle!). + * + * We allow ourselves the leeway of potentially misreporting the busy + * state because that is an optimisation heuristic that is constantly + * in flux. Being quickly able to detect the busy/idle state is much + * more important than accurate logging of exactly which engines were + * busy. + * + * For accuracy in reporting the engine, we could use + * + * result = 0; + * request = __i915_gem_active_get_rcu(active); + * if (request) { + * if (!i915_gem_request_completed(request)) + * result = flag(request->engine->exec_id); + * i915_gem_request_put(request); + * } + * + * but that still remains susceptible to both hardware and userspace + * races. So we accept making the result of that race slightly worse, + * given the rarity of the race and its low impact on the result. + */ + return flag(READ_ONCE(request->engine->exec_id)); } -void -i915_gem_object_ggtt_unpin_view(struct drm_i915_gem_object *obj, - const struct i915_ggtt_view *view) +static __always_inline unsigned int +busy_check_reader(const struct i915_gem_active *active) { - struct i915_vma *vma = i915_gem_obj_to_ggtt_view(obj, view); - - WARN_ON(vma->pin_count == 0); - WARN_ON(!i915_gem_obj_ggtt_bound_view(obj, view)); + return __busy_set_if_active(active, __busy_read_flag); +} - --vma->pin_count; +static __always_inline unsigned int +busy_check_writer(const struct i915_gem_active *active) +{ + return __busy_set_if_active(active, __busy_write_id); } int @@ -4645,47 +3937,64 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, { struct drm_i915_gem_busy *args = data; struct drm_i915_gem_object *obj; - int ret; + unsigned long active; - ret = i915_mutex_lock_interruptible(dev); - if (ret) - return ret; + obj = i915_gem_object_lookup(file, args->handle); + if (!obj) + return -ENOENT; - obj = to_intel_bo(drm_gem_object_lookup(file, args->handle)); - if (&obj->base == NULL) { - ret = -ENOENT; - goto unlock; - } + args->busy = 0; + active = __I915_BO_ACTIVE(obj); + if (active) { + int idx; - /* Count all active objects as busy, even if they are currently not used - * by the gpu. Users of this interface expect objects to eventually - * become non-busy without any further actions, therefore emit any - * necessary flushes here. - */ - ret = i915_gem_object_flush_active(obj); - if (ret) - goto unref; + /* Yes, the lookups are intentionally racy. + * + * First, we cannot simply rely on __I915_BO_ACTIVE. We have + * to regard the value as stale and as our ABI guarantees + * forward progress, we confirm the status of each active + * request with the hardware. + * + * Even though we guard the pointer lookup by RCU, that only + * guarantees that the pointer and its contents remain + * dereferencable and does *not* mean that the request we + * have is the same as the one being tracked by the object. + * + * Consider that we lookup the request just as it is being + * retired and freed. We take a local copy of the pointer, + * but before we add its engine into the busy set, the other + * thread reallocates it and assigns it to a task on another + * engine with a fresh and incomplete seqno. Guarding against + * that requires careful serialisation and reference counting, + * i.e. using __i915_gem_active_get_request_rcu(). We don't, + * instead we expect that if the result is busy, which engines + * are busy is not completely reliable - we only guarantee + * that the object was busy. + */ + rcu_read_lock(); - args->busy = 0; - if (obj->active) { - int i; + for_each_active(active, idx) + args->busy |= busy_check_reader(&obj->last_read[idx]); - for (i = 0; i < I915_NUM_ENGINES; i++) { - struct drm_i915_gem_request *req; + /* For ABI sanity, we only care that the write engine is in + * the set of read engines. This should be ensured by the + * ordering of setting last_read/last_write in + * i915_vma_move_to_active(), and then in reverse in retire. + * However, for good measure, we always report the last_write + * request as a busy read as well as being a busy write. + * + * We don't care that the set of active read/write engines + * may change during construction of the result, as it is + * equally liable to change before userspace can inspect + * the result. + */ + args->busy |= busy_check_writer(&obj->last_write); - req = obj->last_read_req[i]; - if (req) - args->busy |= 1 << (16 + req->engine->exec_id); - } - if (obj->last_write_req) - args->busy |= obj->last_write_req->engine->exec_id; + rcu_read_unlock(); } -unref: - drm_gem_object_unreference(&obj->base); -unlock: - mutex_unlock(&dev->struct_mutex); - return ret; + i915_gem_object_put_unlocked(obj); + return 0; } int @@ -4716,19 +4025,14 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data, if (ret) return ret; - obj = to_intel_bo(drm_gem_object_lookup(file_priv, args->handle)); - if (&obj->base == NULL) { + obj = i915_gem_object_lookup(file_priv, args->handle); + if (!obj) { ret = -ENOENT; goto unlock; } - if (i915_gem_obj_is_pinned(obj)) { - ret = -EINVAL; - goto out; - } - if (obj->pages && - obj->tiling_mode != I915_TILING_NONE && + i915_gem_object_is_tiled(obj) && dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES) { if (obj->madv == I915_MADV_WILLNEED) i915_gem_object_unpin_pages(obj); @@ -4745,8 +4049,7 @@ i915_gem_madvise_ioctl(struct drm_device *dev, void *data, args->retained = obj->madv != __I915_MADV_PURGED; -out: - drm_gem_object_unreference(&obj->base); + i915_gem_object_put(obj); unlock: mutex_unlock(&dev->struct_mutex); return ret; @@ -4759,14 +4062,17 @@ void i915_gem_object_init(struct drm_i915_gem_object *obj, INIT_LIST_HEAD(&obj->global_list); for (i = 0; i < I915_NUM_ENGINES; i++) - INIT_LIST_HEAD(&obj->engine_list[i]); + init_request_active(&obj->last_read[i], + i915_gem_object_retire__read); + init_request_active(&obj->last_write, + i915_gem_object_retire__write); INIT_LIST_HEAD(&obj->obj_exec_link); INIT_LIST_HEAD(&obj->vma_list); INIT_LIST_HEAD(&obj->batch_pool_link); obj->ops = ops; - obj->fence_reg = I915_FENCE_REG_NONE; + obj->frontbuffer_ggtt_origin = ORIGIN_GTT; obj->madv = I915_MADV_WILLNEED; i915_gem_info_add_obj(to_i915(obj->base.dev), obj->base.size); @@ -4871,33 +4177,31 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj) trace_i915_gem_object_destroy(obj); + /* 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. + */ list_for_each_entry_safe(vma, next, &obj->vma_list, obj_link) { - int ret; - - vma->pin_count = 0; - ret = i915_vma_unbind(vma); - if (WARN_ON(ret == -ERESTARTSYS)) { - bool was_interruptible; - - was_interruptible = dev_priv->mm.interruptible; - dev_priv->mm.interruptible = false; - - WARN_ON(i915_vma_unbind(vma)); - - dev_priv->mm.interruptible = was_interruptible; - } + GEM_BUG_ON(!i915_vma_is_ggtt(vma)); + GEM_BUG_ON(i915_vma_is_active(vma)); + vma->flags &= ~I915_VMA_PIN_MASK; + i915_vma_close(vma); } + GEM_BUG_ON(obj->bind_count); /* Stolen objects don't hold a ref, but do hold pin count. Fix that up * before progressing. */ if (obj->stolen) i915_gem_object_unpin_pages(obj); - WARN_ON(obj->frontbuffer_bits); + WARN_ON(atomic_read(&obj->frontbuffer_bits)); if (obj->pages && obj->madv == I915_MADV_WILLNEED && dev_priv->quirks & QUIRK_PIN_SWIZZLED_PAGES && - obj->tiling_mode != I915_TILING_NONE) + i915_gem_object_is_tiled(obj)) i915_gem_object_unpin_pages(obj); if (WARN_ON(obj->pages_pin_count)) @@ -4905,7 +4209,6 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj) if (discard_backing_storage(obj)) obj->madv = I915_MADV_DONTNEED; i915_gem_object_put_pages(obj); - i915_gem_object_free_mmap_offset(obj); BUG_ON(obj->pages); @@ -4924,71 +4227,35 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj) intel_runtime_pm_put(dev_priv); } -struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj, - struct i915_address_space *vm) -{ - struct i915_vma *vma; - list_for_each_entry(vma, &obj->vma_list, obj_link) { - if (vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL && - vma->vm == vm) - return vma; - } - return NULL; -} - -struct i915_vma *i915_gem_obj_to_ggtt_view(struct drm_i915_gem_object *obj, - const struct i915_ggtt_view *view) -{ - struct i915_vma *vma; - - GEM_BUG_ON(!view); - - list_for_each_entry(vma, &obj->vma_list, obj_link) - if (vma->is_ggtt && i915_ggtt_view_equal(&vma->ggtt_view, view)) - return vma; - return NULL; -} - -void i915_gem_vma_destroy(struct i915_vma *vma) -{ - WARN_ON(vma->node.allocated); - - /* Keep the vma as a placeholder in the execbuffer reservation lists */ - if (!list_empty(&vma->exec_list)) - return; - - if (!vma->is_ggtt) - i915_ppgtt_put(i915_vm_to_ppgtt(vma->vm)); - - list_del(&vma->obj_link); - - kmem_cache_free(to_i915(vma->obj->base.dev)->vmas, vma); -} - -static void -i915_gem_stop_engines(struct drm_device *dev) +int i915_gem_suspend(struct drm_device *dev) { struct drm_i915_private *dev_priv = to_i915(dev); - struct intel_engine_cs *engine; + int ret; - for_each_engine(engine, dev_priv) - dev_priv->gt.stop_engine(engine); -} - -int -i915_gem_suspend(struct drm_device *dev) -{ - struct drm_i915_private *dev_priv = to_i915(dev); - int ret = 0; + intel_suspend_gt_powersave(dev_priv); mutex_lock(&dev->struct_mutex); - ret = i915_gem_wait_for_idle(dev_priv); + + /* 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 dev_priv->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. + */ + ret = i915_gem_switch_to_kernel_context(dev_priv); + if (ret) + goto err; + + ret = i915_gem_wait_for_idle(dev_priv, + I915_WAIT_INTERRUPTIBLE | + I915_WAIT_LOCKED); if (ret) goto err; i915_gem_retire_requests(dev_priv); - i915_gem_stop_engines(dev); i915_gem_context_lost(dev_priv); mutex_unlock(&dev->struct_mutex); @@ -5008,6 +4275,22 @@ err: return ret; } +void i915_gem_resume(struct drm_device *dev) +{ + struct drm_i915_private *dev_priv = to_i915(dev); + + mutex_lock(&dev->struct_mutex); + i915_gem_restore_gtt_mappings(dev); + + /* 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. + */ + dev_priv->gt.resume(dev_priv); + + mutex_unlock(&dev->struct_mutex); +} + void i915_gem_init_swizzling(struct drm_device *dev) { struct drm_i915_private *dev_priv = to_i915(dev); @@ -5060,53 +4343,6 @@ static void init_unused_rings(struct drm_device *dev) } } -int i915_gem_init_engines(struct drm_device *dev) -{ - struct drm_i915_private *dev_priv = to_i915(dev); - int ret; - - ret = intel_init_render_ring_buffer(dev); - if (ret) - return ret; - - if (HAS_BSD(dev)) { - ret = intel_init_bsd_ring_buffer(dev); - if (ret) - goto cleanup_render_ring; - } - - if (HAS_BLT(dev)) { - ret = intel_init_blt_ring_buffer(dev); - if (ret) - goto cleanup_bsd_ring; - } - - if (HAS_VEBOX(dev)) { - ret = intel_init_vebox_ring_buffer(dev); - if (ret) - goto cleanup_blt_ring; - } - - if (HAS_BSD2(dev)) { - ret = intel_init_bsd2_ring_buffer(dev); - if (ret) - goto cleanup_vebox_ring; - } - - return 0; - -cleanup_vebox_ring: - intel_cleanup_engine(&dev_priv->engine[VECS]); -cleanup_blt_ring: - intel_cleanup_engine(&dev_priv->engine[BCS]); -cleanup_bsd_ring: - intel_cleanup_engine(&dev_priv->engine[VCS]); -cleanup_render_ring: - intel_cleanup_engine(&dev_priv->engine[RCS]); - - return ret; -} - int i915_gem_init_hw(struct drm_device *dev) { @@ -5173,6 +4409,27 @@ out: return ret; } +bool intel_sanitize_semaphores(struct drm_i915_private *dev_priv, int value) +{ + if (INTEL_INFO(dev_priv)->gen < 6) + return false; + + /* TODO: make semaphores and Execlists play nicely together */ + if (i915.enable_execlists) + return false; + + if (value >= 0) + return value; + +#ifdef CONFIG_INTEL_IOMMU + /* Enable semaphores on SNB when IO remapping is off */ + if (INTEL_INFO(dev_priv)->gen == 6 && intel_iommu_gfx_mapped) + return false; +#endif + + return true; +} + int i915_gem_init(struct drm_device *dev) { struct drm_i915_private *dev_priv = to_i915(dev); @@ -5181,15 +4438,11 @@ int i915_gem_init(struct drm_device *dev) mutex_lock(&dev->struct_mutex); if (!i915.enable_execlists) { - dev_priv->gt.execbuf_submit = i915_gem_ringbuffer_submission; - dev_priv->gt.init_engines = i915_gem_init_engines; - dev_priv->gt.cleanup_engine = intel_cleanup_engine; - dev_priv->gt.stop_engine = intel_stop_engine; + dev_priv->gt.resume = intel_legacy_submission_resume; + dev_priv->gt.cleanup_engine = intel_engine_cleanup; } else { - dev_priv->gt.execbuf_submit = intel_execlists_submission; - dev_priv->gt.init_engines = intel_logical_rings_init; + dev_priv->gt.resume = intel_lr_context_resume; dev_priv->gt.cleanup_engine = intel_logical_ring_cleanup; - dev_priv->gt.stop_engine = intel_logical_ring_stop; } /* This is just a security blanket to placate dragons. @@ -5201,24 +4454,27 @@ int i915_gem_init(struct drm_device *dev) intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); i915_gem_init_userptr(dev_priv); - i915_gem_init_ggtt(dev); + + ret = i915_gem_init_ggtt(dev_priv); + if (ret) + goto out_unlock; ret = i915_gem_context_init(dev); if (ret) goto out_unlock; - ret = dev_priv->gt.init_engines(dev); + ret = intel_engines_init(dev); if (ret) goto out_unlock; ret = i915_gem_init_hw(dev); if (ret == -EIO) { - /* Allow ring initialisation to fail by marking the GPU as + /* Allow engine initialisation to fail by marking the GPU as * wedged. But we only want to do this where the GPU is angry, * for all other failure, such as an allocation failure, bail. */ DRM_ERROR("Failed to initialize GPU, declaring it wedged\n"); - atomic_or(I915_WEDGED, &dev_priv->gpu_error.reset_counter); + i915_gem_set_wedged(dev_priv); ret = 0; } @@ -5242,7 +4498,6 @@ i915_gem_cleanup_engines(struct drm_device *dev) static void init_engine_lists(struct intel_engine_cs *engine) { - INIT_LIST_HEAD(&engine->active_list); INIT_LIST_HEAD(&engine->request_list); } @@ -5250,6 +4505,7 @@ void i915_gem_load_init_fences(struct drm_i915_private *dev_priv) { struct drm_device *dev = &dev_priv->drm; + int i; if (INTEL_INFO(dev_priv)->gen >= 7 && !IS_VALLEYVIEW(dev_priv) && !IS_CHERRYVIEW(dev_priv)) @@ -5265,6 +4521,13 @@ i915_gem_load_init_fences(struct drm_i915_private *dev_priv) 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); i915_gem_detect_bit_6_swizzle(dev); @@ -5289,18 +4552,17 @@ i915_gem_load_init(struct drm_device *dev) dev_priv->requests = kmem_cache_create("i915_gem_request", sizeof(struct drm_i915_gem_request), 0, - SLAB_HWCACHE_ALIGN, + SLAB_HWCACHE_ALIGN | + SLAB_RECLAIM_ACCOUNT | + SLAB_DESTROY_BY_RCU, NULL); - INIT_LIST_HEAD(&dev_priv->vm_list); INIT_LIST_HEAD(&dev_priv->context_list); INIT_LIST_HEAD(&dev_priv->mm.unbound_list); INIT_LIST_HEAD(&dev_priv->mm.bound_list); INIT_LIST_HEAD(&dev_priv->mm.fence_list); for (i = 0; i < I915_NUM_ENGINES; i++) init_engine_lists(&dev_priv->engine[i]); - for (i = 0; i < I915_MAX_NUM_FENCES; i++) - INIT_LIST_HEAD(&dev_priv->fence_regs[i].lru_list); INIT_DELAYED_WORK(&dev_priv->gt.retire_work, i915_gem_retire_work_handler); INIT_DELAYED_WORK(&dev_priv->gt.idle_work, @@ -5310,13 +4572,13 @@ i915_gem_load_init(struct drm_device *dev) dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL; - INIT_LIST_HEAD(&dev_priv->mm.fence_list); - init_waitqueue_head(&dev_priv->pending_flip_queue); dev_priv->mm.interruptible = true; - mutex_init(&dev_priv->fb_tracking.lock); + atomic_set(&dev_priv->mm.bsd_engine_dispatch_index, 0); + + spin_lock_init(&dev_priv->fb_tracking.lock); } void i915_gem_load_cleanup(struct drm_device *dev) @@ -5326,11 +4588,32 @@ void i915_gem_load_cleanup(struct drm_device *dev) kmem_cache_destroy(dev_priv->requests); kmem_cache_destroy(dev_priv->vmas); kmem_cache_destroy(dev_priv->objects); + + /* And ensure that our DESTROY_BY_RCU slabs are truly destroyed */ + rcu_barrier(); +} + +int i915_gem_freeze(struct drm_i915_private *dev_priv) +{ + intel_runtime_pm_get(dev_priv); + + mutex_lock(&dev_priv->drm.struct_mutex); + i915_gem_shrink_all(dev_priv); + mutex_unlock(&dev_priv->drm.struct_mutex); + + intel_runtime_pm_put(dev_priv); + + return 0; } int i915_gem_freeze_late(struct drm_i915_private *dev_priv) { struct drm_i915_gem_object *obj; + struct list_head *phases[] = { + &dev_priv->mm.unbound_list, + &dev_priv->mm.bound_list, + NULL + }, **p; /* Called just before we write the hibernation image. * @@ -5341,17 +4624,21 @@ int i915_gem_freeze_late(struct drm_i915_private *dev_priv) * * To make sure the hibernation image contains the latest state, * we update that state just before writing out the image. + * + * To try and reduce the hibernation image, we manually shrink + * the objects as well. */ - list_for_each_entry(obj, &dev_priv->mm.unbound_list, global_list) { - obj->base.read_domains = I915_GEM_DOMAIN_CPU; - obj->base.write_domain = I915_GEM_DOMAIN_CPU; - } + mutex_lock(&dev_priv->drm.struct_mutex); + i915_gem_shrink(dev_priv, -1UL, I915_SHRINK_UNBOUND); - list_for_each_entry(obj, &dev_priv->mm.bound_list, global_list) { - obj->base.read_domains = I915_GEM_DOMAIN_CPU; - obj->base.write_domain = I915_GEM_DOMAIN_CPU; + for (p = phases; *p; p++) { + list_for_each_entry(obj, *p, global_list) { + obj->base.read_domains = I915_GEM_DOMAIN_CPU; + obj->base.write_domain = I915_GEM_DOMAIN_CPU; + } } + mutex_unlock(&dev_priv->drm.struct_mutex); return 0; } @@ -5359,21 +4646,15 @@ int i915_gem_freeze_late(struct drm_i915_private *dev_priv) void i915_gem_release(struct drm_device *dev, struct drm_file *file) { struct drm_i915_file_private *file_priv = file->driver_priv; + struct drm_i915_gem_request *request; /* Clean up our request list when the client is going away, so that * later retire_requests won't dereference our soon-to-be-gone * file_priv. */ spin_lock(&file_priv->mm.lock); - while (!list_empty(&file_priv->mm.request_list)) { - struct drm_i915_gem_request *request; - - request = list_first_entry(&file_priv->mm.request_list, - struct drm_i915_gem_request, - client_list); - list_del(&request->client_list); + list_for_each_entry(request, &file_priv->mm.request_list, client_list) request->file_priv = NULL; - } spin_unlock(&file_priv->mm.lock); if (!list_empty(&file_priv->rps.link)) { @@ -5402,7 +4683,7 @@ int i915_gem_open(struct drm_device *dev, struct drm_file *file) spin_lock_init(&file_priv->mm.lock); INIT_LIST_HEAD(&file_priv->mm.request_list); - file_priv->bsd_ring = -1; + file_priv->bsd_engine = -1; ret = i915_gem_context_open(dev, file); if (ret) @@ -5424,120 +4705,26 @@ void i915_gem_track_fb(struct drm_i915_gem_object *old, struct drm_i915_gem_object *new, unsigned frontbuffer_bits) { + /* Control of individual bits within the mask are guarded by + * the owning plane->mutex, i.e. we can never see concurrent + * manipulation of individual bits. But since the bitfield as a whole + * is updated using RMW, we need to use atomics in order to update + * the bits. + */ + BUILD_BUG_ON(INTEL_FRONTBUFFER_BITS_PER_PIPE * I915_MAX_PIPES > + sizeof(atomic_t) * BITS_PER_BYTE); + if (old) { - WARN_ON(!mutex_is_locked(&old->base.dev->struct_mutex)); - WARN_ON(!(old->frontbuffer_bits & frontbuffer_bits)); - old->frontbuffer_bits &= ~frontbuffer_bits; + WARN_ON(!(atomic_read(&old->frontbuffer_bits) & frontbuffer_bits)); + atomic_andnot(frontbuffer_bits, &old->frontbuffer_bits); } if (new) { - WARN_ON(!mutex_is_locked(&new->base.dev->struct_mutex)); - WARN_ON(new->frontbuffer_bits & frontbuffer_bits); - new->frontbuffer_bits |= frontbuffer_bits; + WARN_ON(atomic_read(&new->frontbuffer_bits) & frontbuffer_bits); + atomic_or(frontbuffer_bits, &new->frontbuffer_bits); } } -/* All the new VM stuff */ -u64 i915_gem_obj_offset(struct drm_i915_gem_object *o, - struct i915_address_space *vm) -{ - struct drm_i915_private *dev_priv = to_i915(o->base.dev); - struct i915_vma *vma; - - WARN_ON(vm == &dev_priv->mm.aliasing_ppgtt->base); - - list_for_each_entry(vma, &o->vma_list, obj_link) { - if (vma->is_ggtt && - vma->ggtt_view.type != I915_GGTT_VIEW_NORMAL) - continue; - if (vma->vm == vm) - return vma->node.start; - } - - WARN(1, "%s vma for this object not found.\n", - i915_is_ggtt(vm) ? "global" : "ppgtt"); - return -1; -} - -u64 i915_gem_obj_ggtt_offset_view(struct drm_i915_gem_object *o, - const struct i915_ggtt_view *view) -{ - struct i915_vma *vma; - - list_for_each_entry(vma, &o->vma_list, obj_link) - if (vma->is_ggtt && i915_ggtt_view_equal(&vma->ggtt_view, view)) - return vma->node.start; - - WARN(1, "global vma for this object not found. (view=%u)\n", view->type); - return -1; -} - -bool i915_gem_obj_bound(struct drm_i915_gem_object *o, - struct i915_address_space *vm) -{ - struct i915_vma *vma; - - list_for_each_entry(vma, &o->vma_list, obj_link) { - if (vma->is_ggtt && - vma->ggtt_view.type != I915_GGTT_VIEW_NORMAL) - continue; - if (vma->vm == vm && drm_mm_node_allocated(&vma->node)) - return true; - } - - return false; -} - -bool i915_gem_obj_ggtt_bound_view(struct drm_i915_gem_object *o, - const struct i915_ggtt_view *view) -{ - struct i915_vma *vma; - - list_for_each_entry(vma, &o->vma_list, obj_link) - if (vma->is_ggtt && - i915_ggtt_view_equal(&vma->ggtt_view, view) && - drm_mm_node_allocated(&vma->node)) - return true; - - return false; -} - -bool i915_gem_obj_bound_any(struct drm_i915_gem_object *o) -{ - struct i915_vma *vma; - - list_for_each_entry(vma, &o->vma_list, obj_link) - if (drm_mm_node_allocated(&vma->node)) - return true; - - return false; -} - -unsigned long i915_gem_obj_ggtt_size(struct drm_i915_gem_object *o) -{ - struct i915_vma *vma; - - GEM_BUG_ON(list_empty(&o->vma_list)); - - list_for_each_entry(vma, &o->vma_list, obj_link) { - if (vma->is_ggtt && - vma->ggtt_view.type == I915_GGTT_VIEW_NORMAL) - return vma->node.size; - } - - return 0; -} - -bool i915_gem_obj_is_pinned(struct drm_i915_gem_object *obj) -{ - struct i915_vma *vma; - list_for_each_entry(vma, &obj->vma_list, obj_link) - if (vma->pin_count > 0) - return true; - - return false; -} - /* 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, int n) @@ -5590,6 +4777,6 @@ i915_gem_object_create_from_data(struct drm_device *dev, return obj; fail: - drm_gem_object_unreference(&obj->base); + i915_gem_object_put(obj); return ERR_PTR(ret); } |