drm/i915: Handle pagefaults in execbuffer user relocations

Currently if we hit a pagefault when applying a user relocation for the
execbuffer, we bail and return EFAULT to the application. Instead, we
need to unwind, drop the dev->struct_mutex, copy all the relocation
entries to a vmalloc array (to avoid any potential circular deadlocks
when resolving the pagefault), retake the mutex and then apply the
relocations.  Afterwards, we need to again drop the lock and copy the
vmalloc array back to userspace.

v2: Incorporate feedback from Daniel Vetter.

Reported-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

1 files changed, 293 insertions, 168 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index bc4164590054..92b097dbe4ff 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3254,192 +3254,230 @@ i915_gem_object_set_cpu_read_domain_range(struct drm_gem_object *obj,
         return 0;
 }
 
-/**
- * Pin an object to the GTT and evaluate the relocations landing in it.
- */
 static int
-i915_gem_execbuffer_relocate(struct drm_i915_gem_object *obj,
-                             struct drm_file *file_priv,
-                             struct drm_i915_gem_exec_object2 *entry)
+i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
+                                   struct drm_file *file_priv,
+                                   struct drm_i915_gem_exec_object2 *entry,
+                                   struct drm_i915_gem_relocation_entry *reloc)
 {
         struct drm_device *dev = obj->base.dev;
-        drm_i915_private_t *dev_priv = dev->dev_private;
-        struct drm_i915_gem_relocation_entry __user *user_relocs;
-        struct drm_gem_object *target_obj = NULL;
-        uint32_t target_handle = 0;
-        int i, ret = 0;
+        struct drm_gem_object *target_obj;
+        uint32_t target_offset;
+        int ret = -EINVAL;
 
-        user_relocs = (void __user *)(uintptr_t)entry->relocs_ptr;
-        for (i = 0; i < entry->relocation_count; i++) {
-                struct drm_i915_gem_relocation_entry reloc;
-                uint32_t target_offset;
+        target_obj = drm_gem_object_lookup(dev, file_priv,
+                                           reloc->target_handle);
+        if (target_obj == NULL)
+                return -ENOENT;
 
-                if (__copy_from_user_inatomic(&reloc,
-                                              user_relocs+i,
-                                              sizeof(reloc))) {
-                        ret = -EFAULT;
-                        break;
-                }
+        target_offset = to_intel_bo(target_obj)->gtt_offset;
 
-                if (reloc.target_handle != target_handle) {
-                        drm_gem_object_unreference(target_obj);
+#if WATCH_RELOC
+        DRM_INFO("%s: obj %p offset %08x target %d "
+                 "read %08x write %08x gtt %08x "
+                 "presumed %08x delta %08x\n",
+                 __func__,
+                 obj,
+                 (int) reloc->offset,
+                 (int) reloc->target_handle,
+                 (int) reloc->read_domains,
+                 (int) reloc->write_domain,
+                 (int) target_offset,
+                 (int) reloc->presumed_offset,
+                 reloc->delta);
+#endif
 
-                        target_obj = drm_gem_object_lookup(dev, file_priv,
-                                                           reloc.target_handle);
-                        if (target_obj == NULL) {
-                                ret = -ENOENT;
-                                break;
-                        }
+        /* The target buffer should have appeared before us in the
+         * exec_object list, so it should have a GTT space bound by now.
+         */
+        if (target_offset == 0) {
+                DRM_ERROR("No GTT space found for object %d\n",
+                          reloc->target_handle);
+                goto err;
+        }
 
-                        target_handle = reloc.target_handle;
-                }
-                target_offset = to_intel_bo(target_obj)->gtt_offset;
+        /* Validate that the target is in a valid r/w GPU domain */
+        if (reloc->write_domain & (reloc->write_domain - 1)) {
+                DRM_ERROR("reloc with multiple write domains: "
+                          "obj %p target %d offset %d "
+                          "read %08x write %08x",
+                          obj, reloc->target_handle,
+                          (int) reloc->offset,
+                          reloc->read_domains,
+                          reloc->write_domain);
+                goto err;
+        }
+        if (reloc->write_domain & I915_GEM_DOMAIN_CPU ||
+            reloc->read_domains & I915_GEM_DOMAIN_CPU) {
+                DRM_ERROR("reloc with read/write CPU domains: "
+                          "obj %p target %d offset %d "
+                          "read %08x write %08x",
+                          obj, reloc->target_handle,
+                          (int) reloc->offset,
+                          reloc->read_domains,
+                          reloc->write_domain);
+                goto err;
+        }
+        if (reloc->write_domain && target_obj->pending_write_domain &&
+            reloc->write_domain != target_obj->pending_write_domain) {
+                DRM_ERROR("Write domain conflict: "
+                          "obj %p target %d offset %d "
+                          "new %08x old %08x\n",
+                          obj, reloc->target_handle,
+                          (int) reloc->offset,
+                          reloc->write_domain,
+                          target_obj->pending_write_domain);
+                goto err;
+        }
 
-#if WATCH_RELOC
-                DRM_INFO("%s: obj %p offset %08x target %d "
-                         "read %08x write %08x gtt %08x "
-                         "presumed %08x delta %08x\n",
-                         __func__,
-                         obj,
-                         (int) reloc.offset,
-                         (int) reloc.target_handle,
-                         (int) reloc.read_domains,
-                         (int) reloc.write_domain,
-                         (int) target_offset,
-                         (int) reloc.presumed_offset,
-                         reloc.delta);
-#endif
+        target_obj->pending_read_domains |= reloc->read_domains;
+        target_obj->pending_write_domain |= reloc->write_domain;
 
-                /* The target buffer should have appeared before us in the
-                 * exec_object list, so it should have a GTT space bound by now.
-                 */
-                if (target_offset == 0) {
-                        DRM_ERROR("No GTT space found for object %d\n",
-                                  reloc.target_handle);
-                        ret = -EINVAL;
-                        break;
-                }
+        /* If the relocation already has the right value in it, no
+         * more work needs to be done.
+         */
+        if (target_offset == reloc->presumed_offset)
+                goto out;
 
-                /* Validate that the target is in a valid r/w GPU domain */
-                if (reloc.write_domain & (reloc.write_domain - 1)) {
-                        DRM_ERROR("reloc with multiple write domains: "
-                                  "obj %p target %d offset %d "
-                                  "read %08x write %08x",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.offset,
-                                  reloc.read_domains,
-                                  reloc.write_domain);
-                        ret = -EINVAL;
-                        break;
-                }
-                if (reloc.write_domain & I915_GEM_DOMAIN_CPU ||
-                    reloc.read_domains & I915_GEM_DOMAIN_CPU) {
-                        DRM_ERROR("reloc with read/write CPU domains: "
-                                  "obj %p target %d offset %d "
-                                  "read %08x write %08x",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.offset,
-                                  reloc.read_domains,
-                                  reloc.write_domain);
-                        ret = -EINVAL;
-                        break;
-                }
-                if (reloc.write_domain && target_obj->pending_write_domain &&
-                    reloc.write_domain != target_obj->pending_write_domain) {
-                        DRM_ERROR("Write domain conflict: "
-                                  "obj %p target %d offset %d "
-                                  "new %08x old %08x\n",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.offset,
-                                  reloc.write_domain,
-                                  target_obj->pending_write_domain);
-                        ret = -EINVAL;
-                        break;
-                }
+        /* Check that the relocation address is valid... */
+        if (reloc->offset > obj->base.size - 4) {
+                DRM_ERROR("Relocation beyond object bounds: "
+                          "obj %p target %d offset %d size %d.\n",
+                          obj, reloc->target_handle,
+                          (int) reloc->offset,
+                          (int) obj->base.size);
+                goto err;
+        }
+        if (reloc->offset & 3) {
+                DRM_ERROR("Relocation not 4-byte aligned: "
+                          "obj %p target %d offset %d.\n",
+                          obj, reloc->target_handle,
+                          (int) reloc->offset);
+                goto err;
+        }
 
-                target_obj->pending_read_domains |= reloc.read_domains;
-                target_obj->pending_write_domain |= reloc.write_domain;
+        /* and points to somewhere within the target object. */
+        if (reloc->delta >= target_obj->size) {
+                DRM_ERROR("Relocation beyond target object bounds: "
+                          "obj %p target %d delta %d size %d.\n",
+                          obj, reloc->target_handle,
+                          (int) reloc->delta,
+                          (int) target_obj->size);
+                goto err;
+        }
 
-                /* If the relocation already has the right value in it, no
-                 * more work needs to be done.
-                 */
-                if (target_offset == reloc.presumed_offset)
-                        continue;
+        reloc->delta += target_offset;
+        if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) {
+                uint32_t page_offset = reloc->offset & ~PAGE_MASK;
+                char *vaddr;
 
-                /* Check that the relocation address is valid... */
-                if (reloc.offset > obj->base.size - 4) {
-                        DRM_ERROR("Relocation beyond object bounds: "
-                                  "obj %p target %d offset %d size %d.\n",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.offset, (int) obj->base.size);
-                        ret = -EINVAL;
-                        break;
-                }
-                if (reloc.offset & 3) {
-                        DRM_ERROR("Relocation not 4-byte aligned: "
-                                  "obj %p target %d offset %d.\n",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.offset);
-                        ret = -EINVAL;
-                        break;
-                }
+                vaddr = kmap_atomic(obj->pages[reloc->offset >> PAGE_SHIFT]);
+                *(uint32_t *)(vaddr + page_offset) = reloc->delta;
+                kunmap_atomic(vaddr);
+        } else {
+                struct drm_i915_private *dev_priv = dev->dev_private;
+                uint32_t __iomem *reloc_entry;
+                void __iomem *reloc_page;
 
-                /* and points to somewhere within the target object. */
-                if (reloc.delta >= target_obj->size) {
-                        DRM_ERROR("Relocation beyond target object bounds: "
-                                  "obj %p target %d delta %d size %d.\n",
-                                  obj, reloc.target_handle,
-                                  (int) reloc.delta, (int) target_obj->size);
-                        ret = -EINVAL;
-                        break;
-                }
+                ret = i915_gem_object_set_to_gtt_domain(&obj->base, 1);
+                if (ret)
+                        goto err;
 
-                reloc.delta += target_offset;
-                if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) {
-                        uint32_t page_offset = reloc.offset & ~PAGE_MASK;
-                        char *vaddr;
+                /* Map the page containing the relocation we're going to perform.  */
+                reloc->offset += obj->gtt_offset;
+                reloc_page = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
+                                                      reloc->offset & PAGE_MASK);
+                reloc_entry = (uint32_t __iomem *)
+                        (reloc_page + (reloc->offset & ~PAGE_MASK));
+                iowrite32(reloc->delta, reloc_entry);
+                io_mapping_unmap_atomic(reloc_page);
+        }
 
-                        vaddr = kmap_atomic(obj->pages[reloc.offset >> PAGE_SHIFT]);
-                        *(uint32_t *)(vaddr + page_offset) = reloc.delta;
-                        kunmap_atomic(vaddr);
-                } else {
-                        uint32_t __iomem *reloc_entry;
-                        void __iomem *reloc_page;
+        /* and update the user's relocation entry */
+        reloc->presumed_offset = target_offset;
 
-                        ret = i915_gem_object_set_to_gtt_domain(&obj->base, 1);
-                        if (ret)
-                                break;
+out:
+        ret = 0;
+err:
+        drm_gem_object_unreference(target_obj);
+        return ret;
+}
 
-                        /* Map the page containing the relocation we're going to perform.  */
-                        reloc.offset += obj->gtt_offset;
-                        reloc_page = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
-                                                              reloc.offset & PAGE_MASK);
-                        reloc_entry = (uint32_t __iomem *)
-                                (reloc_page + (reloc.offset & ~PAGE_MASK));
-                        iowrite32(reloc.delta, reloc_entry);
-                        io_mapping_unmap_atomic(reloc_page);
-                }
+static int
+i915_gem_execbuffer_relocate_object(struct drm_i915_gem_object *obj,
+                                    struct drm_file *file_priv,
+                                    struct drm_i915_gem_exec_object2 *entry)
+{
+        struct drm_i915_gem_relocation_entry __user *user_relocs;
+        int i, ret;
+
+        user_relocs = (void __user *)(uintptr_t)entry->relocs_ptr;
+        for (i = 0; i < entry->relocation_count; i++) {
+                struct drm_i915_gem_relocation_entry reloc;
+
+                if (__copy_from_user_inatomic(&reloc,
+                                              user_relocs+i,
+                                              sizeof(reloc)))
+                        return -EFAULT;
+
+                ret = i915_gem_execbuffer_relocate_entry(obj, file_priv, entry, &reloc);
+                if (ret)
+                        return ret;
 
-                /* and update the user's relocation entry */
-                reloc.presumed_offset = target_offset;
                 if (__copy_to_user_inatomic(&user_relocs[i].presumed_offset,
-                                              &reloc.presumed_offset,
-                                              sizeof(reloc.presumed_offset))) {
-                    ret = -EFAULT;
-                    break;
-                }
+                                            &reloc.presumed_offset,
+                                            sizeof(reloc.presumed_offset)))
+                        return -EFAULT;
         }
 
-        drm_gem_object_unreference(target_obj);
-        return ret;
+        return 0;
+}
+
+static int
+i915_gem_execbuffer_relocate_object_slow(struct drm_i915_gem_object *obj,
+                                         struct drm_file *file_priv,
+                                         struct drm_i915_gem_exec_object2 *entry,
+                                         struct drm_i915_gem_relocation_entry *relocs)
+{
+        int i, ret;
+
+        for (i = 0; i < entry->relocation_count; i++) {
+                ret = i915_gem_execbuffer_relocate_entry(obj, file_priv, entry, &relocs[i]);
+                if (ret)
+                        return ret;
+        }
+
+        return 0;
+}
+
+static int
+i915_gem_execbuffer_relocate(struct drm_device *dev,
+                             struct drm_file *file,
+                             struct drm_gem_object **object_list,
+                             struct drm_i915_gem_exec_object2 *exec_list,
+                             int count)
+{
+        int i, ret;
+
+        for (i = 0; i < count; i++) {
+                struct drm_i915_gem_object *obj = to_intel_bo(object_list[i]);
+                obj->base.pending_read_domains = 0;
+                obj->base.pending_write_domain = 0;
+                ret = i915_gem_execbuffer_relocate_object(obj, file,
+                                                          &exec_list[i]);
+                if (ret)
+                        return ret;
+        }
+
+        return 0;
 }
 
 static int
-i915_gem_execbuffer_pin(struct drm_device *dev,
-                        struct drm_file *file,
-                        struct drm_gem_object **object_list,
-                        struct drm_i915_gem_exec_object2 *exec_list,
-                        int count)
+i915_gem_execbuffer_reserve(struct drm_device *dev,
+                            struct drm_file *file,
+                            struct drm_gem_object **object_list,
+                            struct drm_i915_gem_exec_object2 *exec_list,
+                            int count)
 {
         struct drm_i915_private *dev_priv = dev->dev_private;
         int ret, i, retry;
@@ -3502,6 +3540,87 @@ i915_gem_execbuffer_pin(struct drm_device *dev,
 }
 
 static int
+i915_gem_execbuffer_relocate_slow(struct drm_device *dev,
+                                  struct drm_file *file,
+                                  struct drm_gem_object **object_list,
+                                  struct drm_i915_gem_exec_object2 *exec_list,
+                                  int count)
+{
+        struct drm_i915_gem_relocation_entry *reloc;
+        int i, total, ret;
+
+        for (i = 0; i < count; i++) {
+                struct drm_i915_gem_object *obj = to_intel_bo(object_list[i]);
+                obj->in_execbuffer = false;
+        }
+
+        mutex_unlock(&dev->struct_mutex);
+
+        total = 0;
+        for (i = 0; i < count; i++)
+                total += exec_list[i].relocation_count;
+
+        reloc = drm_malloc_ab(total, sizeof(*reloc));
+        if (reloc == NULL) {
+                mutex_lock(&dev->struct_mutex);
+                return -ENOMEM;
+        }
+
+        total = 0;
+        for (i = 0; i < count; i++) {
+                struct drm_i915_gem_relocation_entry __user *user_relocs;
+
+                user_relocs = (void __user *)(uintptr_t)exec_list[i].relocs_ptr;
+
+                if (copy_from_user(reloc+total, user_relocs,
+                                   exec_list[i].relocation_count *
+                                   sizeof(*reloc))) {
+                        ret = -EFAULT;
+                        mutex_lock(&dev->struct_mutex);
+                        goto err;
+                }
+
+                total += exec_list[i].relocation_count;
+        }
+
+        ret = i915_mutex_lock_interruptible(dev);
+        if (ret) {
+                mutex_lock(&dev->struct_mutex);
+                goto err;
+        }
+
+        ret = i915_gem_execbuffer_reserve(dev, file,
+                                          object_list, exec_list,
+                                          count);
+        if (ret)
+                goto err;
+
+        total = 0;
+        for (i = 0; i < count; i++) {
+                struct drm_i915_gem_object *obj = to_intel_bo(object_list[i]);
+                obj->base.pending_read_domains = 0;
+                obj->base.pending_write_domain = 0;
+                ret = i915_gem_execbuffer_relocate_object_slow(obj, file,
+                                                               &exec_list[i],
+                                                               reloc + total);
+                if (ret)
+                        goto err;
+
+                total += exec_list[i].relocation_count;
+        }
+
+        /* Leave the user relocations as are, this is the painfully slow path,
+         * and we want to avoid the complication of dropping the lock whilst
+         * having buffers reserved in the aperture and so causing spurious
+         * ENOSPC for random operations.
+         */
+
+err:
+        drm_free_large(reloc);
+        return ret;
+}
+
+static int
 i915_gem_execbuffer_move_to_gpu(struct drm_device *dev,
                                 struct drm_file *file,
                                 struct intel_ring_buffer *ring,
@@ -3781,18 +3900,24 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
         }
 
         /* Move the objects en-masse into the GTT, evicting if necessary. */
-        ret = i915_gem_execbuffer_pin(dev, file,
-                                      object_list, exec_list,
-                                      args->buffer_count);
+        ret = i915_gem_execbuffer_reserve(dev, file,
+                                          object_list, exec_list,
+                                          args->buffer_count);
         if (ret)
                 goto err;
 
         /* The objects are in their final locations, apply the relocations. */
-        for (i = 0; i < args->buffer_count; i++) {
-                struct drm_i915_gem_object *obj = to_intel_bo(object_list[i]);
-                obj->base.pending_read_domains = 0;
-                obj->base.pending_write_domain = 0;
-                ret = i915_gem_execbuffer_relocate(obj, file, &exec_list[i]);
+        ret = i915_gem_execbuffer_relocate(dev, file,
+                                           object_list, exec_list,
+                                           args->buffer_count);
+        if (ret) {
+                if (ret == -EFAULT) {
+                        ret = i915_gem_execbuffer_relocate_slow(dev, file,
+                                                                object_list,
+                                                                exec_list,
+                                                                args->buffer_count);
+                        BUG_ON(!mutex_is_locked(&dev->struct_mutex));
+                }
                 if (ret)
                         goto err;
         }