drm/vmwgfx: Make the object handles idr-generated
authorThomas Hellstrom <thellstrom@vmware.com>
Wed, 26 Sep 2018 13:50:13 +0000 (15:50 +0200)
committerThomas Hellstrom <thellstrom@vmware.com>
Thu, 27 Sep 2018 14:14:18 +0000 (16:14 +0200)
Instead of generating user-space object handles based on a, possibly
processed, hash of the kernel address of the object, use idr to generate
and lookup those handles. This might improve somewhat on security since
we loose all connections to the object's kernel address. Also idr is
designed to do just this.

As a todo-item, since user-space handles are now generated in sequence,
we can probably use a much simpler hash function to hash them.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Deepak Rawat <drawat@vmware.com>
drivers/gpu/drm/vmwgfx/ttm_lock.c
drivers/gpu/drm/vmwgfx/ttm_object.c
drivers/gpu/drm/vmwgfx/ttm_object.h
drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
drivers/gpu/drm/vmwgfx/vmwgfx_context.c
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
drivers/gpu/drm/vmwgfx/vmwgfx_resource_priv.h
drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
drivers/gpu/drm/vmwgfx/vmwgfx_simple_resource.c
drivers/gpu/drm/vmwgfx/vmwgfx_surface.c

index 0d59f5a19e17c0f82d0be5ffa30eca8faea929ae..16b2083cb9d42e2e1f21e4a0c6bc87020a86dfb3 100644 (file)
@@ -261,7 +261,7 @@ int ttm_vt_lock(struct ttm_lock *lock,
 int ttm_vt_unlock(struct ttm_lock *lock)
 {
        return ttm_ref_object_base_unref(lock->vt_holder,
-                                        lock->base.hash.key, TTM_REF_USAGE);
+                                        lock->base.handle, TTM_REF_USAGE);
 }
 
 void ttm_suspend_unlock(struct ttm_lock *lock)
index 190e2591e1f540c9f6b97a9c84f8a012745f4b18..c5df2d7828e6c30346267a9a91a93a8a2ce0a81a 100644 (file)
@@ -94,6 +94,7 @@ struct ttm_object_device {
        struct dma_buf_ops ops;
        void (*dmabuf_release)(struct dma_buf *dma_buf);
        size_t dma_buf_size;
+       struct idr idr;
 };
 
 /**
@@ -171,14 +172,15 @@ int ttm_base_object_init(struct ttm_object_file *tfile,
        base->ref_obj_release = ref_obj_release;
        base->object_type = object_type;
        kref_init(&base->refcount);
+       idr_preload(GFP_KERNEL);
        spin_lock(&tdev->object_lock);
-       ret = drm_ht_just_insert_please_rcu(&tdev->object_hash,
-                                           &base->hash,
-                                           (unsigned long)base, 31, 0, 0);
+       ret = idr_alloc(&tdev->idr, base, 0, 0, GFP_NOWAIT);
        spin_unlock(&tdev->object_lock);
-       if (unlikely(ret != 0))
-               goto out_err0;
+       idr_preload_end();
+       if (ret < 0)
+               return ret;
 
+       base->handle = ret;
        ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL, false);
        if (unlikely(ret != 0))
                goto out_err1;
@@ -188,9 +190,8 @@ int ttm_base_object_init(struct ttm_object_file *tfile,
        return 0;
 out_err1:
        spin_lock(&tdev->object_lock);
-       (void)drm_ht_remove_item_rcu(&tdev->object_hash, &base->hash);
+       idr_remove(&tdev->idr, base->handle);
        spin_unlock(&tdev->object_lock);
-out_err0:
        return ret;
 }
 
@@ -201,7 +202,7 @@ static void ttm_release_base(struct kref *kref)
        struct ttm_object_device *tdev = base->tfile->tdev;
 
        spin_lock(&tdev->object_lock);
-       (void)drm_ht_remove_item_rcu(&tdev->object_hash, &base->hash);
+       idr_remove(&tdev->idr, base->handle);
        spin_unlock(&tdev->object_lock);
 
        /*
@@ -248,19 +249,13 @@ struct ttm_base_object *ttm_base_object_lookup(struct ttm_object_file *tfile,
 struct ttm_base_object *
 ttm_base_object_lookup_for_ref(struct ttm_object_device *tdev, uint32_t key)
 {
-       struct ttm_base_object *base = NULL;
-       struct drm_hash_item *hash;
-       struct drm_open_hash *ht = &tdev->object_hash;
-       int ret;
+       struct ttm_base_object *base;
 
        rcu_read_lock();
-       ret = drm_ht_find_item_rcu(ht, key, &hash);
+       base = idr_find(&tdev->idr, key);
 
-       if (likely(ret == 0)) {
-               base = drm_hash_entry(hash, struct ttm_base_object, hash);
-               if (!kref_get_unless_zero(&base->refcount))
-                       base = NULL;
-       }
+       if (base && !kref_get_unless_zero(&base->refcount))
+               base = NULL;
        rcu_read_unlock();
 
        return base;
@@ -284,7 +279,7 @@ bool ttm_ref_object_exists(struct ttm_object_file *tfile,
        struct ttm_ref_object *ref;
 
        rcu_read_lock();
-       if (unlikely(drm_ht_find_item_rcu(ht, base->hash.key, &hash) != 0))
+       if (unlikely(drm_ht_find_item_rcu(ht, base->handle, &hash) != 0))
                goto out_false;
 
        /*
@@ -334,7 +329,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
 
        while (ret == -EINVAL) {
                rcu_read_lock();
-               ret = drm_ht_find_item_rcu(ht, base->hash.key, &hash);
+               ret = drm_ht_find_item_rcu(ht, base->handle, &hash);
 
                if (ret == 0) {
                        ref = drm_hash_entry(hash, struct ttm_ref_object, hash);
@@ -358,7 +353,7 @@ int ttm_ref_object_add(struct ttm_object_file *tfile,
                        return -ENOMEM;
                }
 
-               ref->hash.key = base->hash.key;
+               ref->hash.key = base->handle;
                ref->obj = base;
                ref->tfile = tfile;
                ref->ref_type = ref_type;
@@ -510,6 +505,7 @@ ttm_object_device_init(struct ttm_mem_global *mem_glob,
        if (ret != 0)
                goto out_no_object_hash;
 
+       idr_init(&tdev->idr);
        tdev->ops = *ops;
        tdev->dmabuf_release = tdev->ops.release;
        tdev->ops.release = ttm_prime_dmabuf_release;
@@ -528,6 +524,8 @@ void ttm_object_device_release(struct ttm_object_device **p_tdev)
 
        *p_tdev = NULL;
 
+       WARN_ON_ONCE(!idr_is_empty(&tdev->idr));
+       idr_destroy(&tdev->idr);
        drm_ht_remove(&tdev->object_hash);
 
        kfree(tdev);
@@ -630,7 +628,7 @@ int ttm_prime_fd_to_handle(struct ttm_object_file *tfile,
 
        prime = (struct ttm_prime_object *) dma_buf->priv;
        base = &prime->base;
-       *handle = base->hash.key;
+       *handle = base->handle;
        ret = ttm_ref_object_add(tfile, base, TTM_REF_USAGE, NULL, false);
 
        dma_buf_put(dma_buf);
index 1c1b9cc118f89e395af818a1f32d05c484629bda..7aa213f5d67777a27eb8618638b2ddb0b92e5404 100644 (file)
@@ -124,14 +124,14 @@ struct ttm_object_device;
 
 struct ttm_base_object {
        struct rcu_head rhead;
-       struct drm_hash_item hash;
-       enum ttm_object_type object_type;
-       bool shareable;
        struct ttm_object_file *tfile;
        struct kref refcount;
        void (*refcount_release) (struct ttm_base_object **base);
        void (*ref_obj_release) (struct ttm_base_object *base,
                                 enum ttm_ref_type ref_type);
+       u32 handle;
+       enum ttm_object_type object_type;
+       u32 shareable;
 };
 
 
@@ -350,4 +350,11 @@ extern int ttm_prime_handle_to_fd(struct ttm_object_file *tfile,
 
 #define ttm_prime_object_kfree(__obj, __prime)         \
        kfree_rcu(__obj, __prime.base.rhead)
+
+/*
+ * Extra memory required by the base object's idr storage, which is allocated
+ * separately from the base object itself. We estimate an on-average 128 bytes
+ * per idr.
+ */
+#define TTM_OBJ_EXTRA_SIZE 128
 #endif
index 36e84acee3ea0e6ee015c5245771de7aa375ccde..3df4e5266cac6b68f6a6ed53af5febd456b5f52a 100644 (file)
@@ -441,7 +441,8 @@ static size_t vmw_bo_acc_size(struct vmw_private *dev_priv, size_t size,
                struct_size = backend_size +
                        ttm_round_pot(sizeof(struct vmw_buffer_object));
                user_struct_size = backend_size +
-                       ttm_round_pot(sizeof(struct vmw_user_buffer_object));
+                 ttm_round_pot(sizeof(struct vmw_user_buffer_object)) +
+                                     TTM_OBJ_EXTRA_SIZE;
        }
 
        if (dev_priv->map_mode == vmw_dma_alloc_coherent)
@@ -631,7 +632,7 @@ int vmw_user_bo_alloc(struct vmw_private *dev_priv,
                *p_base = &user_bo->prime.base;
                kref_get(&(*p_base)->refcount);
        }
-       *handle = user_bo->prime.base.hash.key;
+       *handle = user_bo->prime.base.handle;
 
 out_no_base_object:
        return ret;
@@ -940,7 +941,7 @@ int vmw_user_bo_reference(struct ttm_object_file *tfile,
 
        user_bo = container_of(vbo, struct vmw_user_buffer_object, vbo);
 
-       *handle = user_bo->prime.base.hash.key;
+       *handle = user_bo->prime.base.handle;
        return ttm_ref_object_add(tfile, &user_bo->prime.base,
                                  TTM_REF_USAGE, NULL, false);
 }
index 4d502567d24c8b6c97e63703ac14f315d79085da..24d7c81081aec9448657b525473be380028ad55e 100644 (file)
@@ -755,14 +755,10 @@ static int vmw_context_define(struct drm_device *dev, void *data,
                return -EINVAL;
        }
 
-       /*
-        * Approximate idr memory usage with 128 bytes. It will be limited
-        * by maximum number_of contexts anyway.
-        */
-
        if (unlikely(vmw_user_context_size == 0))
-               vmw_user_context_size = ttm_round_pot(sizeof(*ctx)) + 128 +
-                 ((dev_priv->has_mob) ? vmw_cmdbuf_res_man_size() : 0);
+               vmw_user_context_size = ttm_round_pot(sizeof(*ctx)) +
+                 ((dev_priv->has_mob) ? vmw_cmdbuf_res_man_size() : 0) +
+                 + VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
 
        ret = ttm_read_lock(&dev_priv->reservation_sem, true);
        if (unlikely(ret != 0))
@@ -807,7 +803,7 @@ static int vmw_context_define(struct drm_device *dev, void *data,
                goto out_err;
        }
 
-       arg->cid = ctx->base.hash.key;
+       arg->cid = ctx->base.handle;
 out_err:
        vmw_resource_unreference(&res);
 out_unlock:
index 3d546d4093341fd6ba329c06830f173666e2f707..f87261545f2c1e8123846af64d406f32efcb4883 100644 (file)
@@ -306,7 +306,8 @@ struct vmw_fence_manager *vmw_fence_manager_init(struct vmw_private *dev_priv)
        INIT_LIST_HEAD(&fman->cleanup_list);
        INIT_WORK(&fman->work, &vmw_fence_work_func);
        fman->fifo_down = true;
-       fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence));
+       fman->user_fence_size = ttm_round_pot(sizeof(struct vmw_user_fence)) +
+               TTM_OBJ_EXTRA_SIZE;
        fman->fence_size = ttm_round_pot(sizeof(struct vmw_fence_obj));
        fman->event_fence_action_size =
                ttm_round_pot(sizeof(struct vmw_event_fence_action));
@@ -650,7 +651,7 @@ int vmw_user_fence_create(struct drm_file *file_priv,
        }
 
        *p_fence = &ufence->fence;
-       *p_handle = ufence->base.hash.key;
+       *p_handle = ufence->base.handle;
 
        return 0;
 out_err:
@@ -1137,7 +1138,7 @@ int vmw_fence_event_ioctl(struct drm_device *dev, void *data,
                                          "object.\n");
                                goto out_no_ref_obj;
                        }
-                       handle = base->hash.key;
+                       handle = base->handle;
                }
                ttm_base_object_unref(&base);
        }
index 6453708682963c54807fb1130d49901ed90497b4..7e19eba0b0b8cd930be900aa0a4354dc5ac6c714 100644 (file)
 
 #include "vmwgfx_drv.h"
 
+/*
+ * Extra memory required by the resource id's ida storage, which is allocated
+ * separately from the base object itself. We estimate an on-average 128 bytes
+ * per ida.
+ */
 #define VMW_IDA_ACC_SIZE 128
 
 enum vmw_cmdbuf_res_state {
index c72b4351176aa625e4269a60724450b6889e7c5c..6915c8258e6be8def003bd74b1dba505dc50ed63 100644 (file)
@@ -740,13 +740,10 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
        };
        int ret;
 
-       /*
-        * Approximate idr memory usage with 128 bytes. It will be limited
-        * by maximum number_of shaders anyway.
-        */
        if (unlikely(vmw_user_shader_size == 0))
                vmw_user_shader_size =
-                       ttm_round_pot(sizeof(struct vmw_user_shader)) + 128;
+                       ttm_round_pot(sizeof(struct vmw_user_shader)) +
+                       VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
 
        ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
                                   vmw_user_shader_size,
@@ -792,7 +789,7 @@ static int vmw_user_shader_alloc(struct vmw_private *dev_priv,
        }
 
        if (handle)
-               *handle = ushader->base.hash.key;
+               *handle = ushader->base.handle;
 out_err:
        vmw_resource_unreference(&res);
 out:
@@ -814,13 +811,10 @@ static struct vmw_resource *vmw_shader_alloc(struct vmw_private *dev_priv,
        };
        int ret;
 
-       /*
-        * Approximate idr memory usage with 128 bytes. It will be limited
-        * by maximum number_of shaders anyway.
-        */
        if (unlikely(vmw_shader_size == 0))
                vmw_shader_size =
-                       ttm_round_pot(sizeof(struct vmw_shader)) + 128;
+                       ttm_round_pot(sizeof(struct vmw_shader)) +
+                       VMW_IDA_ACC_SIZE;
 
        ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv),
                                   vmw_shader_size,
index 3bd60f7a9d6d72fc9f72c6c483acf7136908eb32..6a6865384e91f21fba1890f69b3889771e683bea 100644 (file)
@@ -159,7 +159,8 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
 
        alloc_size = offsetof(struct vmw_user_simple_resource, simple) +
          func->size;
-       account_size = ttm_round_pot(alloc_size) + VMW_IDA_ACC_SIZE;
+       account_size = ttm_round_pot(alloc_size) + VMW_IDA_ACC_SIZE +
+               TTM_OBJ_EXTRA_SIZE;
 
        ret = ttm_read_lock(&dev_priv->reservation_sem, true);
        if (ret)
@@ -208,7 +209,7 @@ vmw_simple_resource_create_ioctl(struct drm_device *dev, void *data,
                goto out_err;
        }
 
-       func->set_arg_handle(data, usimple->base.hash.key);
+       func->set_arg_handle(data, usimple->base.handle);
 out_err:
        vmw_resource_unreference(&res);
 out_ret:
index bd4cf995089cad39c624f54a1816d12b8a6dd838..a67c8f9af129860029d81277cd4db4b900d93bbe 100644 (file)
@@ -731,7 +731,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
 
        if (unlikely(vmw_user_surface_size == 0))
                vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
-                       128;
+                       VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
 
        num_sizes = 0;
        for (i = 0; i < DRM_VMW_MAX_SURFACE_FACES; ++i) {
@@ -744,7 +744,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
            num_sizes == 0)
                return -EINVAL;
 
-       size = vmw_user_surface_size + 128 +
+       size = vmw_user_surface_size +
                ttm_round_pot(num_sizes * sizeof(struct drm_vmw_size)) +
                ttm_round_pot(num_sizes * sizeof(struct vmw_surface_offset));
 
@@ -886,7 +886,7 @@ int vmw_surface_define_ioctl(struct drm_device *dev, void *data,
                goto out_unlock;
        }
 
-       rep->sid = user_srf->prime.base.hash.key;
+       rep->sid = user_srf->prime.base.handle;
        vmw_resource_unreference(&res);
 
        ttm_read_unlock(&dev_priv->reservation_sem);
@@ -1024,7 +1024,7 @@ int vmw_surface_reference_ioctl(struct drm_device *dev, void *data,
        if (unlikely(ret != 0)) {
                DRM_ERROR("copy_to_user failed %p %u\n",
                          user_sizes, srf->num_sizes);
-               ttm_ref_object_base_unref(tfile, base->hash.key, TTM_REF_USAGE);
+               ttm_ref_object_base_unref(tfile, base->handle, TTM_REF_USAGE);
                ret = -EFAULT;
        }
 
@@ -1609,9 +1609,9 @@ vmw_gb_surface_define_internal(struct drm_device *dev,
 
        if (unlikely(vmw_user_surface_size == 0))
                vmw_user_surface_size = ttm_round_pot(sizeof(*user_srf)) +
-                       128;
+                       VMW_IDA_ACC_SIZE + TTM_OBJ_EXTRA_SIZE;
 
-       size = vmw_user_surface_size + 128;
+       size = vmw_user_surface_size;
 
        /* Define a surface based on the parameters. */
        ret = vmw_surface_gb_priv_define(dev,
@@ -1683,7 +1683,7 @@ vmw_gb_surface_define_internal(struct drm_device *dev,
                goto out_unlock;
        }
 
-       rep->handle      = user_srf->prime.base.hash.key;
+       rep->handle      = user_srf->prime.base.handle;
        rep->backup_size = res->backup_size;
        if (res->backup) {
                rep->buffer_map_handle =
@@ -1745,7 +1745,7 @@ vmw_gb_surface_reference_internal(struct drm_device *dev,
        if (unlikely(ret != 0)) {
                DRM_ERROR("Could not add a reference to a GB surface "
                          "backup buffer.\n");
-               (void) ttm_ref_object_base_unref(tfile, base->hash.key,
+               (void) ttm_ref_object_base_unref(tfile, base->handle,
                                                 TTM_REF_USAGE);
                goto out_bad_resource;
        }
@@ -1759,7 +1759,7 @@ vmw_gb_surface_reference_internal(struct drm_device *dev,
        rep->creq.base.array_size = srf->array_size;
        rep->creq.base.buffer_handle = backup_handle;
        rep->creq.base.base_size = srf->base_size;
-       rep->crep.handle = user_srf->prime.base.hash.key;
+       rep->crep.handle = user_srf->prime.base.handle;
        rep->crep.backup_size = srf->res.backup_size;
        rep->crep.buffer_handle = backup_handle;
        rep->crep.buffer_map_handle =