OSDN Git Service

Merge tag 'vmwgfx-next-4.19-2' of git://people.freedesktop.org/~thomash/linux into...
authorDave Airlie <airlied@redhat.com>
Tue, 10 Jul 2018 01:05:46 +0000 (11:05 +1000)
committerDave Airlie <airlied@redhat.com>
Tue, 10 Jul 2018 01:10:34 +0000 (11:10 +1000)
A series of cleanups / reorganizations and modesetting changes that
mostly target atomic state validation.

[airlied: conflicts with SPDX stuff in amdgpu tree]
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1a88485e-e509-b00e-8485-19194f074115@vmware.com
21 files changed:
1  2 
drivers/gpu/drm/vmwgfx/device_include/svga_reg.h
drivers/gpu/drm/vmwgfx/vmwgfx_bo.c
drivers/gpu/drm/vmwgfx/vmwgfx_context.c
drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c
drivers/gpu/drm/vmwgfx/vmwgfx_ioctl.c
drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
drivers/gpu/drm/vmwgfx/vmwgfx_kms.h
drivers/gpu/drm/vmwgfx/vmwgfx_ldu.c
drivers/gpu/drm/vmwgfx/vmwgfx_mob.c
drivers/gpu/drm/vmwgfx/vmwgfx_msg.c
drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
drivers/gpu/drm/vmwgfx/vmwgfx_scrn.c
drivers/gpu/drm/vmwgfx/vmwgfx_shader.c
drivers/gpu/drm/vmwgfx/vmwgfx_stdu.c
drivers/gpu/drm/vmwgfx/vmwgfx_surface.c
drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c

index 0000000,87204ff..2dda033
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,1122 +1,1123 @@@
++// SPDX-License-Identifier: GPL-2.0 OR MIT
+ /**************************************************************************
+  *
+  * Copyright © 2011-2018 VMware, Inc., Palo Alto, CA., USA
+  * All Rights Reserved.
+  *
+  * Permission is hereby granted, free of charge, to any person obtaining a
+  * copy of this software and associated documentation files (the
+  * "Software"), to deal in the Software without restriction, including
+  * without limitation the rights to use, copy, modify, merge, publish,
+  * distribute, sub license, and/or sell copies of the Software, and to
+  * permit persons to whom the Software is furnished to do so, subject to
+  * the following conditions:
+  *
+  * The above copyright notice and this permission notice (including the
+  * next paragraph) shall be included in all copies or substantial portions
+  * of the Software.
+  *
+  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+  * USE OR OTHER DEALINGS IN THE SOFTWARE.
+  *
+  **************************************************************************/
+ #include <drm/ttm/ttm_placement.h>
+ #include <drm/drmP.h>
+ #include "vmwgfx_drv.h"
+ #include "drm/ttm/ttm_object.h"
+ /**
+  * struct vmw_user_buffer_object - User-space-visible buffer object
+  *
+  * @prime: The prime object providing user visibility.
+  * @vbo: The struct vmw_buffer_object
+  */
+ struct vmw_user_buffer_object {
+       struct ttm_prime_object prime;
+       struct vmw_buffer_object vbo;
+ };
+ /**
+  * vmw_buffer_object - Convert a struct ttm_buffer_object to a struct
+  * vmw_buffer_object.
+  *
+  * @bo: Pointer to the TTM buffer object.
+  * Return: Pointer to the struct vmw_buffer_object embedding the
+  * TTM buffer object.
+  */
+ static struct vmw_buffer_object *
+ vmw_buffer_object(struct ttm_buffer_object *bo)
+ {
+       return container_of(bo, struct vmw_buffer_object, base);
+ }
+ /**
+  * vmw_user_buffer_object - Convert a struct ttm_buffer_object to a struct
+  * vmw_user_buffer_object.
+  *
+  * @bo: Pointer to the TTM buffer object.
+  * Return: Pointer to the struct vmw_buffer_object embedding the TTM buffer
+  * object.
+  */
+ static struct vmw_user_buffer_object *
+ vmw_user_buffer_object(struct ttm_buffer_object *bo)
+ {
+       struct vmw_buffer_object *vmw_bo = vmw_buffer_object(bo);
+       return container_of(vmw_bo, struct vmw_user_buffer_object, vbo);
+ }
+ /**
+  * vmw_bo_pin_in_placement - Validate a buffer to placement.
+  *
+  * @dev_priv:  Driver private.
+  * @buf:  DMA buffer to move.
+  * @placement:  The placement to pin it.
+  * @interruptible:  Use interruptible wait.
+  * Return: Zero on success, Negative error code on failure. In particular
+  * -ERESTARTSYS if interrupted by a signal
+  */
+ int vmw_bo_pin_in_placement(struct vmw_private *dev_priv,
+                           struct vmw_buffer_object *buf,
+                           struct ttm_placement *placement,
+                           bool interruptible)
+ {
+       struct ttm_operation_ctx ctx = {interruptible, false };
+       struct ttm_buffer_object *bo = &buf->base;
+       int ret;
+       uint32_t new_flags;
+       ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible);
+       if (unlikely(ret != 0))
+               return ret;
+       vmw_execbuf_release_pinned_bo(dev_priv);
+       ret = ttm_bo_reserve(bo, interruptible, false, NULL);
+       if (unlikely(ret != 0))
+               goto err;
+       if (buf->pin_count > 0)
+               ret = ttm_bo_mem_compat(placement, &bo->mem,
+                                       &new_flags) == true ? 0 : -EINVAL;
+       else
+               ret = ttm_bo_validate(bo, placement, &ctx);
+       if (!ret)
+               vmw_bo_pin_reserved(buf, true);
+       ttm_bo_unreserve(bo);
+ err:
+       ttm_write_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_bo_pin_in_vram_or_gmr - Move a buffer to vram or gmr.
+  *
+  * This function takes the reservation_sem in write mode.
+  * Flushes and unpins the query bo to avoid failures.
+  *
+  * @dev_priv:  Driver private.
+  * @buf:  DMA buffer to move.
+  * @pin:  Pin buffer if true.
+  * @interruptible:  Use interruptible wait.
+  * Return: Zero on success, Negative error code on failure. In particular
+  * -ERESTARTSYS if interrupted by a signal
+  */
+ int vmw_bo_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
+                             struct vmw_buffer_object *buf,
+                             bool interruptible)
+ {
+       struct ttm_operation_ctx ctx = {interruptible, false };
+       struct ttm_buffer_object *bo = &buf->base;
+       int ret;
+       uint32_t new_flags;
+       ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible);
+       if (unlikely(ret != 0))
+               return ret;
+       vmw_execbuf_release_pinned_bo(dev_priv);
+       ret = ttm_bo_reserve(bo, interruptible, false, NULL);
+       if (unlikely(ret != 0))
+               goto err;
+       if (buf->pin_count > 0) {
+               ret = ttm_bo_mem_compat(&vmw_vram_gmr_placement, &bo->mem,
+                                       &new_flags) == true ? 0 : -EINVAL;
+               goto out_unreserve;
+       }
+       ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, &ctx);
+       if (likely(ret == 0) || ret == -ERESTARTSYS)
+               goto out_unreserve;
+       ret = ttm_bo_validate(bo, &vmw_vram_placement, &ctx);
+ out_unreserve:
+       if (!ret)
+               vmw_bo_pin_reserved(buf, true);
+       ttm_bo_unreserve(bo);
+ err:
+       ttm_write_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_bo_pin_in_vram - Move a buffer to vram.
+  *
+  * This function takes the reservation_sem in write mode.
+  * Flushes and unpins the query bo to avoid failures.
+  *
+  * @dev_priv:  Driver private.
+  * @buf:  DMA buffer to move.
+  * @interruptible:  Use interruptible wait.
+  * Return: Zero on success, Negative error code on failure. In particular
+  * -ERESTARTSYS if interrupted by a signal
+  */
+ int vmw_bo_pin_in_vram(struct vmw_private *dev_priv,
+                      struct vmw_buffer_object *buf,
+                      bool interruptible)
+ {
+       return vmw_bo_pin_in_placement(dev_priv, buf, &vmw_vram_placement,
+                                      interruptible);
+ }
+ /**
+  * vmw_bo_pin_in_start_of_vram - Move a buffer to start of vram.
+  *
+  * This function takes the reservation_sem in write mode.
+  * Flushes and unpins the query bo to avoid failures.
+  *
+  * @dev_priv:  Driver private.
+  * @buf:  DMA buffer to pin.
+  * @interruptible:  Use interruptible wait.
+  * Return: Zero on success, Negative error code on failure. In particular
+  * -ERESTARTSYS if interrupted by a signal
+  */
+ int vmw_bo_pin_in_start_of_vram(struct vmw_private *dev_priv,
+                               struct vmw_buffer_object *buf,
+                               bool interruptible)
+ {
+       struct ttm_operation_ctx ctx = {interruptible, false };
+       struct ttm_buffer_object *bo = &buf->base;
+       struct ttm_placement placement;
+       struct ttm_place place;
+       int ret = 0;
+       uint32_t new_flags;
+       place = vmw_vram_placement.placement[0];
+       place.lpfn = bo->num_pages;
+       placement.num_placement = 1;
+       placement.placement = &place;
+       placement.num_busy_placement = 1;
+       placement.busy_placement = &place;
+       ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible);
+       if (unlikely(ret != 0))
+               return ret;
+       vmw_execbuf_release_pinned_bo(dev_priv);
+       ret = ttm_bo_reserve(bo, interruptible, false, NULL);
+       if (unlikely(ret != 0))
+               goto err_unlock;
+       /*
+        * Is this buffer already in vram but not at the start of it?
+        * In that case, evict it first because TTM isn't good at handling
+        * that situation.
+        */
+       if (bo->mem.mem_type == TTM_PL_VRAM &&
+           bo->mem.start < bo->num_pages &&
+           bo->mem.start > 0 &&
+           buf->pin_count == 0) {
+               ctx.interruptible = false;
+               (void) ttm_bo_validate(bo, &vmw_sys_placement, &ctx);
+       }
+       if (buf->pin_count > 0)
+               ret = ttm_bo_mem_compat(&placement, &bo->mem,
+                                       &new_flags) == true ? 0 : -EINVAL;
+       else
+               ret = ttm_bo_validate(bo, &placement, &ctx);
+       /* For some reason we didn't end up at the start of vram */
+       WARN_ON(ret == 0 && bo->offset != 0);
+       if (!ret)
+               vmw_bo_pin_reserved(buf, true);
+       ttm_bo_unreserve(bo);
+ err_unlock:
+       ttm_write_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_bo_unpin - Unpin the buffer given buffer, does not move the buffer.
+  *
+  * This function takes the reservation_sem in write mode.
+  *
+  * @dev_priv:  Driver private.
+  * @buf:  DMA buffer to unpin.
+  * @interruptible:  Use interruptible wait.
+  * Return: Zero on success, Negative error code on failure. In particular
+  * -ERESTARTSYS if interrupted by a signal
+  */
+ int vmw_bo_unpin(struct vmw_private *dev_priv,
+                struct vmw_buffer_object *buf,
+                bool interruptible)
+ {
+       struct ttm_buffer_object *bo = &buf->base;
+       int ret;
+       ret = ttm_read_lock(&dev_priv->reservation_sem, interruptible);
+       if (unlikely(ret != 0))
+               return ret;
+       ret = ttm_bo_reserve(bo, interruptible, false, NULL);
+       if (unlikely(ret != 0))
+               goto err;
+       vmw_bo_pin_reserved(buf, false);
+       ttm_bo_unreserve(bo);
+ err:
+       ttm_read_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_bo_get_guest_ptr - Get the guest ptr representing the current placement
+  * of a buffer.
+  *
+  * @bo: Pointer to a struct ttm_buffer_object. Must be pinned or reserved.
+  * @ptr: SVGAGuestPtr returning the result.
+  */
+ void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
+                         SVGAGuestPtr *ptr)
+ {
+       if (bo->mem.mem_type == TTM_PL_VRAM) {
+               ptr->gmrId = SVGA_GMR_FRAMEBUFFER;
+               ptr->offset = bo->offset;
+       } else {
+               ptr->gmrId = bo->mem.start;
+               ptr->offset = 0;
+       }
+ }
+ /**
+  * vmw_bo_pin_reserved - Pin or unpin a buffer object without moving it.
+  *
+  * @vbo: The buffer object. Must be reserved.
+  * @pin: Whether to pin or unpin.
+  *
+  */
+ void vmw_bo_pin_reserved(struct vmw_buffer_object *vbo, bool pin)
+ {
+       struct ttm_operation_ctx ctx = { false, true };
+       struct ttm_place pl;
+       struct ttm_placement placement;
+       struct ttm_buffer_object *bo = &vbo->base;
+       uint32_t old_mem_type = bo->mem.mem_type;
+       int ret;
+       lockdep_assert_held(&bo->resv->lock.base);
+       if (pin) {
+               if (vbo->pin_count++ > 0)
+                       return;
+       } else {
+               WARN_ON(vbo->pin_count <= 0);
+               if (--vbo->pin_count > 0)
+                       return;
+       }
+       pl.fpfn = 0;
+       pl.lpfn = 0;
+       pl.flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | VMW_PL_FLAG_MOB
+               | TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED;
+       if (pin)
+               pl.flags |= TTM_PL_FLAG_NO_EVICT;
+       memset(&placement, 0, sizeof(placement));
+       placement.num_placement = 1;
+       placement.placement = &pl;
+       ret = ttm_bo_validate(bo, &placement, &ctx);
+       BUG_ON(ret != 0 || bo->mem.mem_type != old_mem_type);
+ }
+ /**
+  * vmw_bo_map_and_cache - Map a buffer object and cache the map
+  *
+  * @vbo: The buffer object to map
+  * Return: A kernel virtual address or NULL if mapping failed.
+  *
+  * This function maps a buffer object into the kernel address space, or
+  * returns the virtual kernel address of an already existing map. The virtual
+  * address remains valid as long as the buffer object is pinned or reserved.
+  * The cached map is torn down on either
+  * 1) Buffer object move
+  * 2) Buffer object swapout
+  * 3) Buffer object destruction
+  *
+  */
+ void *vmw_bo_map_and_cache(struct vmw_buffer_object *vbo)
+ {
+       struct ttm_buffer_object *bo = &vbo->base;
+       bool not_used;
+       void *virtual;
+       int ret;
+       virtual = ttm_kmap_obj_virtual(&vbo->map, &not_used);
+       if (virtual)
+               return virtual;
+       ret = ttm_bo_kmap(bo, 0, bo->num_pages, &vbo->map);
+       if (ret)
+               DRM_ERROR("Buffer object map failed: %d.\n", ret);
+       return ttm_kmap_obj_virtual(&vbo->map, &not_used);
+ }
+ /**
+  * vmw_bo_unmap - Tear down a cached buffer object map.
+  *
+  * @vbo: The buffer object whose map we are tearing down.
+  *
+  * This function tears down a cached map set up using
+  * vmw_buffer_object_map_and_cache().
+  */
+ void vmw_bo_unmap(struct vmw_buffer_object *vbo)
+ {
+       if (vbo->map.bo == NULL)
+               return;
+       ttm_bo_kunmap(&vbo->map);
+ }
+ /**
+  * vmw_bo_acc_size - Calculate the pinned memory usage of buffers
+  *
+  * @dev_priv: Pointer to a struct vmw_private identifying the device.
+  * @size: The requested buffer size.
+  * @user: Whether this is an ordinary dma buffer or a user dma buffer.
+  */
+ static size_t vmw_bo_acc_size(struct vmw_private *dev_priv, size_t size,
+                             bool user)
+ {
+       static size_t struct_size, user_struct_size;
+       size_t num_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
+       size_t page_array_size = ttm_round_pot(num_pages * sizeof(void *));
+       if (unlikely(struct_size == 0)) {
+               size_t backend_size = ttm_round_pot(vmw_tt_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));
+       }
+       if (dev_priv->map_mode == vmw_dma_alloc_coherent)
+               page_array_size +=
+                       ttm_round_pot(num_pages * sizeof(dma_addr_t));
+       return ((user) ? user_struct_size : struct_size) +
+               page_array_size;
+ }
+ /**
+  * vmw_bo_bo_free - vmw buffer object destructor
+  *
+  * @bo: Pointer to the embedded struct ttm_buffer_object
+  */
+ void vmw_bo_bo_free(struct ttm_buffer_object *bo)
+ {
+       struct vmw_buffer_object *vmw_bo = vmw_buffer_object(bo);
+       vmw_bo_unmap(vmw_bo);
+       kfree(vmw_bo);
+ }
+ /**
+  * vmw_user_bo_destroy - vmw buffer object destructor
+  *
+  * @bo: Pointer to the embedded struct ttm_buffer_object
+  */
+ static void vmw_user_bo_destroy(struct ttm_buffer_object *bo)
+ {
+       struct vmw_user_buffer_object *vmw_user_bo = vmw_user_buffer_object(bo);
+       vmw_bo_unmap(&vmw_user_bo->vbo);
+       ttm_prime_object_kfree(vmw_user_bo, prime);
+ }
+ /**
+  * vmw_bo_init - Initialize a vmw buffer object
+  *
+  * @dev_priv: Pointer to the device private struct
+  * @vmw_bo: Pointer to the struct vmw_buffer_object to initialize.
+  * @size: Buffer object size in bytes.
+  * @placement: Initial placement.
+  * @interruptible: Whether waits should be performed interruptible.
+  * @bo_free: The buffer object destructor.
+  * Returns: Zero on success, negative error code on error.
+  *
+  * Note that on error, the code will free the buffer object.
+  */
+ int vmw_bo_init(struct vmw_private *dev_priv,
+               struct vmw_buffer_object *vmw_bo,
+               size_t size, struct ttm_placement *placement,
+               bool interruptible,
+               void (*bo_free)(struct ttm_buffer_object *bo))
+ {
+       struct ttm_bo_device *bdev = &dev_priv->bdev;
+       size_t acc_size;
+       int ret;
+       bool user = (bo_free == &vmw_user_bo_destroy);
+       WARN_ON_ONCE(!bo_free && (!user && (bo_free != vmw_bo_bo_free)));
+       acc_size = vmw_bo_acc_size(dev_priv, size, user);
+       memset(vmw_bo, 0, sizeof(*vmw_bo));
+       INIT_LIST_HEAD(&vmw_bo->res_list);
+       ret = ttm_bo_init(bdev, &vmw_bo->base, size,
+                         ttm_bo_type_device, placement,
+                         0, interruptible, acc_size,
+                         NULL, NULL, bo_free);
+       return ret;
+ }
+ /**
+  * vmw_user_bo_release - TTM reference base object release callback for
+  * vmw user buffer objects
+  *
+  * @p_base: The TTM base object pointer about to be unreferenced.
+  *
+  * Clears the TTM base object pointer and drops the reference the
+  * base object has on the underlying struct vmw_buffer_object.
+  */
+ static void vmw_user_bo_release(struct ttm_base_object **p_base)
+ {
+       struct vmw_user_buffer_object *vmw_user_bo;
+       struct ttm_base_object *base = *p_base;
+       struct ttm_buffer_object *bo;
+       *p_base = NULL;
+       if (unlikely(base == NULL))
+               return;
+       vmw_user_bo = container_of(base, struct vmw_user_buffer_object,
+                                  prime.base);
+       bo = &vmw_user_bo->vbo.base;
+       ttm_bo_unref(&bo);
+ }
+ /**
+  * vmw_user_bo_ref_obj-release - TTM synccpu reference object release callback
+  * for vmw user buffer objects
+  *
+  * @base: Pointer to the TTM base object
+  * @ref_type: Reference type of the reference reaching zero.
+  *
+  * Called when user-space drops its last synccpu reference on the buffer
+  * object, Either explicitly or as part of a cleanup file close.
+  */
+ static void vmw_user_bo_ref_obj_release(struct ttm_base_object *base,
+                                       enum ttm_ref_type ref_type)
+ {
+       struct vmw_user_buffer_object *user_bo;
+       user_bo = container_of(base, struct vmw_user_buffer_object, prime.base);
+       switch (ref_type) {
+       case TTM_REF_SYNCCPU_WRITE:
+               ttm_bo_synccpu_write_release(&user_bo->vbo.base);
+               break;
+       default:
+               WARN_ONCE(true, "Undefined buffer object reference release.\n");
+       }
+ }
+ /**
+  * vmw_user_bo_alloc - Allocate a user buffer object
+  *
+  * @dev_priv: Pointer to a struct device private.
+  * @tfile: Pointer to a struct ttm_object_file on which to register the user
+  * object.
+  * @size: Size of the buffer object.
+  * @shareable: Boolean whether the buffer is shareable with other open files.
+  * @handle: Pointer to where the handle value should be assigned.
+  * @p_vbo: Pointer to where the refcounted struct vmw_buffer_object pointer
+  * should be assigned.
+  * Return: Zero on success, negative error code on error.
+  */
+ int vmw_user_bo_alloc(struct vmw_private *dev_priv,
+                     struct ttm_object_file *tfile,
+                     uint32_t size,
+                     bool shareable,
+                     uint32_t *handle,
+                     struct vmw_buffer_object **p_vbo,
+                     struct ttm_base_object **p_base)
+ {
+       struct vmw_user_buffer_object *user_bo;
+       struct ttm_buffer_object *tmp;
+       int ret;
+       user_bo = kzalloc(sizeof(*user_bo), GFP_KERNEL);
+       if (unlikely(!user_bo)) {
+               DRM_ERROR("Failed to allocate a buffer.\n");
+               return -ENOMEM;
+       }
+       ret = vmw_bo_init(dev_priv, &user_bo->vbo, size,
+                         (dev_priv->has_mob) ?
+                         &vmw_sys_placement :
+                         &vmw_vram_sys_placement, true,
+                         &vmw_user_bo_destroy);
+       if (unlikely(ret != 0))
+               return ret;
+       tmp = ttm_bo_reference(&user_bo->vbo.base);
+       ret = ttm_prime_object_init(tfile,
+                                   size,
+                                   &user_bo->prime,
+                                   shareable,
+                                   ttm_buffer_type,
+                                   &vmw_user_bo_release,
+                                   &vmw_user_bo_ref_obj_release);
+       if (unlikely(ret != 0)) {
+               ttm_bo_unref(&tmp);
+               goto out_no_base_object;
+       }
+       *p_vbo = &user_bo->vbo;
+       if (p_base) {
+               *p_base = &user_bo->prime.base;
+               kref_get(&(*p_base)->refcount);
+       }
+       *handle = user_bo->prime.base.hash.key;
+ out_no_base_object:
+       return ret;
+ }
+ /**
+  * vmw_user_bo_verify_access - verify access permissions on this
+  * buffer object.
+  *
+  * @bo: Pointer to the buffer object being accessed
+  * @tfile: Identifying the caller.
+  */
+ int vmw_user_bo_verify_access(struct ttm_buffer_object *bo,
+                             struct ttm_object_file *tfile)
+ {
+       struct vmw_user_buffer_object *vmw_user_bo;
+       if (unlikely(bo->destroy != vmw_user_bo_destroy))
+               return -EPERM;
+       vmw_user_bo = vmw_user_buffer_object(bo);
+       /* Check that the caller has opened the object. */
+       if (likely(ttm_ref_object_exists(tfile, &vmw_user_bo->prime.base)))
+               return 0;
+       DRM_ERROR("Could not grant buffer access.\n");
+       return -EPERM;
+ }
+ /**
+  * vmw_user_bo_synccpu_grab - Grab a struct vmw_user_buffer_object for cpu
+  * access, idling previous GPU operations on the buffer and optionally
+  * blocking it for further command submissions.
+  *
+  * @user_bo: Pointer to the buffer object being grabbed for CPU access
+  * @tfile: Identifying the caller.
+  * @flags: Flags indicating how the grab should be performed.
+  * Return: Zero on success, Negative error code on error. In particular,
+  * -EBUSY will be returned if a dontblock operation is requested and the
+  * buffer object is busy, and -ERESTARTSYS will be returned if a wait is
+  * interrupted by a signal.
+  *
+  * A blocking grab will be automatically released when @tfile is closed.
+  */
+ static int vmw_user_bo_synccpu_grab(struct vmw_user_buffer_object *user_bo,
+                                   struct ttm_object_file *tfile,
+                                   uint32_t flags)
+ {
+       struct ttm_buffer_object *bo = &user_bo->vbo.base;
+       bool existed;
+       int ret;
+       if (flags & drm_vmw_synccpu_allow_cs) {
+               bool nonblock = !!(flags & drm_vmw_synccpu_dontblock);
+               long lret;
+               lret = reservation_object_wait_timeout_rcu
+                       (bo->resv, true, true,
+                        nonblock ? 0 : MAX_SCHEDULE_TIMEOUT);
+               if (!lret)
+                       return -EBUSY;
+               else if (lret < 0)
+                       return lret;
+               return 0;
+       }
+       ret = ttm_bo_synccpu_write_grab
+               (bo, !!(flags & drm_vmw_synccpu_dontblock));
+       if (unlikely(ret != 0))
+               return ret;
+       ret = ttm_ref_object_add(tfile, &user_bo->prime.base,
+                                TTM_REF_SYNCCPU_WRITE, &existed, false);
+       if (ret != 0 || existed)
+               ttm_bo_synccpu_write_release(&user_bo->vbo.base);
+       return ret;
+ }
+ /**
+  * vmw_user_bo_synccpu_release - Release a previous grab for CPU access,
+  * and unblock command submission on the buffer if blocked.
+  *
+  * @handle: Handle identifying the buffer object.
+  * @tfile: Identifying the caller.
+  * @flags: Flags indicating the type of release.
+  */
+ static int vmw_user_bo_synccpu_release(uint32_t handle,
+                                          struct ttm_object_file *tfile,
+                                          uint32_t flags)
+ {
+       if (!(flags & drm_vmw_synccpu_allow_cs))
+               return ttm_ref_object_base_unref(tfile, handle,
+                                                TTM_REF_SYNCCPU_WRITE);
+       return 0;
+ }
+ /**
+  * vmw_user_bo_synccpu_ioctl - ioctl function implementing the synccpu
+  * functionality.
+  *
+  * @dev: Identifies the drm device.
+  * @data: Pointer to the ioctl argument.
+  * @file_priv: Identifies the caller.
+  * Return: Zero on success, negative error code on error.
+  *
+  * This function checks the ioctl arguments for validity and calls the
+  * relevant synccpu functions.
+  */
+ int vmw_user_bo_synccpu_ioctl(struct drm_device *dev, void *data,
+                             struct drm_file *file_priv)
+ {
+       struct drm_vmw_synccpu_arg *arg =
+               (struct drm_vmw_synccpu_arg *) data;
+       struct vmw_buffer_object *vbo;
+       struct vmw_user_buffer_object *user_bo;
+       struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+       struct ttm_base_object *buffer_base;
+       int ret;
+       if ((arg->flags & (drm_vmw_synccpu_read | drm_vmw_synccpu_write)) == 0
+           || (arg->flags & ~(drm_vmw_synccpu_read | drm_vmw_synccpu_write |
+                              drm_vmw_synccpu_dontblock |
+                              drm_vmw_synccpu_allow_cs)) != 0) {
+               DRM_ERROR("Illegal synccpu flags.\n");
+               return -EINVAL;
+       }
+       switch (arg->op) {
+       case drm_vmw_synccpu_grab:
+               ret = vmw_user_bo_lookup(tfile, arg->handle, &vbo,
+                                            &buffer_base);
+               if (unlikely(ret != 0))
+                       return ret;
+               user_bo = container_of(vbo, struct vmw_user_buffer_object,
+                                      vbo);
+               ret = vmw_user_bo_synccpu_grab(user_bo, tfile, arg->flags);
+               vmw_bo_unreference(&vbo);
+               ttm_base_object_unref(&buffer_base);
+               if (unlikely(ret != 0 && ret != -ERESTARTSYS &&
+                            ret != -EBUSY)) {
+                       DRM_ERROR("Failed synccpu grab on handle 0x%08x.\n",
+                                 (unsigned int) arg->handle);
+                       return ret;
+               }
+               break;
+       case drm_vmw_synccpu_release:
+               ret = vmw_user_bo_synccpu_release(arg->handle, tfile,
+                                                 arg->flags);
+               if (unlikely(ret != 0)) {
+                       DRM_ERROR("Failed synccpu release on handle 0x%08x.\n",
+                                 (unsigned int) arg->handle);
+                       return ret;
+               }
+               break;
+       default:
+               DRM_ERROR("Invalid synccpu operation.\n");
+               return -EINVAL;
+       }
+       return 0;
+ }
+ /**
+  * vmw_bo_alloc_ioctl - ioctl function implementing the buffer object
+  * allocation functionality.
+  *
+  * @dev: Identifies the drm device.
+  * @data: Pointer to the ioctl argument.
+  * @file_priv: Identifies the caller.
+  * Return: Zero on success, negative error code on error.
+  *
+  * This function checks the ioctl arguments for validity and allocates a
+  * struct vmw_user_buffer_object bo.
+  */
+ int vmw_bo_alloc_ioctl(struct drm_device *dev, void *data,
+                      struct drm_file *file_priv)
+ {
+       struct vmw_private *dev_priv = vmw_priv(dev);
+       union drm_vmw_alloc_dmabuf_arg *arg =
+           (union drm_vmw_alloc_dmabuf_arg *)data;
+       struct drm_vmw_alloc_dmabuf_req *req = &arg->req;
+       struct drm_vmw_dmabuf_rep *rep = &arg->rep;
+       struct vmw_buffer_object *vbo;
+       uint32_t handle;
+       int ret;
+       ret = ttm_read_lock(&dev_priv->reservation_sem, true);
+       if (unlikely(ret != 0))
+               return ret;
+       ret = vmw_user_bo_alloc(dev_priv, vmw_fpriv(file_priv)->tfile,
+                               req->size, false, &handle, &vbo,
+                               NULL);
+       if (unlikely(ret != 0))
+               goto out_no_bo;
+       rep->handle = handle;
+       rep->map_handle = drm_vma_node_offset_addr(&vbo->base.vma_node);
+       rep->cur_gmr_id = handle;
+       rep->cur_gmr_offset = 0;
+       vmw_bo_unreference(&vbo);
+ out_no_bo:
+       ttm_read_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_bo_unref_ioctl - Generic handle close ioctl.
+  *
+  * @dev: Identifies the drm device.
+  * @data: Pointer to the ioctl argument.
+  * @file_priv: Identifies the caller.
+  * Return: Zero on success, negative error code on error.
+  *
+  * This function checks the ioctl arguments for validity and closes a
+  * handle to a TTM base object, optionally freeing the object.
+  */
+ int vmw_bo_unref_ioctl(struct drm_device *dev, void *data,
+                      struct drm_file *file_priv)
+ {
+       struct drm_vmw_unref_dmabuf_arg *arg =
+           (struct drm_vmw_unref_dmabuf_arg *)data;
+       return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
+                                        arg->handle,
+                                        TTM_REF_USAGE);
+ }
+ /**
+  * vmw_user_bo_lookup - Look up a vmw user buffer object from a handle.
+  *
+  * @tfile: The TTM object file the handle is registered with.
+  * @handle: The user buffer object handle
+  * @out: Pointer to a where a pointer to the embedded
+  * struct vmw_buffer_object should be placed.
+  * @p_base: Pointer to where a pointer to the TTM base object should be
+  * placed, or NULL if no such pointer is required.
+  * Return: Zero on success, Negative error code on error.
+  *
+  * Both the output base object pointer and the vmw buffer object pointer
+  * will be refcounted.
+  */
+ int vmw_user_bo_lookup(struct ttm_object_file *tfile,
+                      uint32_t handle, struct vmw_buffer_object **out,
+                      struct ttm_base_object **p_base)
+ {
+       struct vmw_user_buffer_object *vmw_user_bo;
+       struct ttm_base_object *base;
+       base = ttm_base_object_lookup(tfile, handle);
+       if (unlikely(base == NULL)) {
+               DRM_ERROR("Invalid buffer object handle 0x%08lx.\n",
+                         (unsigned long)handle);
+               return -ESRCH;
+       }
+       if (unlikely(ttm_base_object_type(base) != ttm_buffer_type)) {
+               ttm_base_object_unref(&base);
+               DRM_ERROR("Invalid buffer object handle 0x%08lx.\n",
+                         (unsigned long)handle);
+               return -EINVAL;
+       }
+       vmw_user_bo = container_of(base, struct vmw_user_buffer_object,
+                                  prime.base);
+       (void)ttm_bo_reference(&vmw_user_bo->vbo.base);
+       if (p_base)
+               *p_base = base;
+       else
+               ttm_base_object_unref(&base);
+       *out = &vmw_user_bo->vbo;
+       return 0;
+ }
+ /**
+  * vmw_user_bo_reference - Open a handle to a vmw user buffer object.
+  *
+  * @tfile: The TTM object file to register the handle with.
+  * @vbo: The embedded vmw buffer object.
+  * @handle: Pointer to where the new handle should be placed.
+  * Return: Zero on success, Negative error code on error.
+  */
+ int vmw_user_bo_reference(struct ttm_object_file *tfile,
+                         struct vmw_buffer_object *vbo,
+                         uint32_t *handle)
+ {
+       struct vmw_user_buffer_object *user_bo;
+       if (vbo->base.destroy != vmw_user_bo_destroy)
+               return -EINVAL;
+       user_bo = container_of(vbo, struct vmw_user_buffer_object, vbo);
+       *handle = user_bo->prime.base.hash.key;
+       return ttm_ref_object_add(tfile, &user_bo->prime.base,
+                                 TTM_REF_USAGE, NULL, false);
+ }
+ /**
+  * vmw_bo_fence_single - Utility function to fence a single TTM buffer
+  *                       object without unreserving it.
+  *
+  * @bo:             Pointer to the struct ttm_buffer_object to fence.
+  * @fence:          Pointer to the fence. If NULL, this function will
+  *                  insert a fence into the command stream..
+  *
+  * Contrary to the ttm_eu version of this function, it takes only
+  * a single buffer object instead of a list, and it also doesn't
+  * unreserve the buffer object, which needs to be done separately.
+  */
+ void vmw_bo_fence_single(struct ttm_buffer_object *bo,
+                        struct vmw_fence_obj *fence)
+ {
+       struct ttm_bo_device *bdev = bo->bdev;
+       struct vmw_private *dev_priv =
+               container_of(bdev, struct vmw_private, bdev);
+       if (fence == NULL) {
+               vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
+               reservation_object_add_excl_fence(bo->resv, &fence->base);
+               dma_fence_put(&fence->base);
+       } else
+               reservation_object_add_excl_fence(bo->resv, &fence->base);
+ }
+ /**
+  * vmw_dumb_create - Create a dumb kms buffer
+  *
+  * @file_priv: Pointer to a struct drm_file identifying the caller.
+  * @dev: Pointer to the drm device.
+  * @args: Pointer to a struct drm_mode_create_dumb structure
+  * Return: Zero on success, negative error code on failure.
+  *
+  * This is a driver callback for the core drm create_dumb functionality.
+  * Note that this is very similar to the vmw_bo_alloc ioctl, except
+  * that the arguments have a different format.
+  */
+ int vmw_dumb_create(struct drm_file *file_priv,
+                   struct drm_device *dev,
+                   struct drm_mode_create_dumb *args)
+ {
+       struct vmw_private *dev_priv = vmw_priv(dev);
+       struct vmw_buffer_object *vbo;
+       int ret;
+       args->pitch = args->width * ((args->bpp + 7) / 8);
+       args->size = args->pitch * args->height;
+       ret = ttm_read_lock(&dev_priv->reservation_sem, true);
+       if (unlikely(ret != 0))
+               return ret;
+       ret = vmw_user_bo_alloc(dev_priv, vmw_fpriv(file_priv)->tfile,
+                                   args->size, false, &args->handle,
+                                   &vbo, NULL);
+       if (unlikely(ret != 0))
+               goto out_no_bo;
+       vmw_bo_unreference(&vbo);
+ out_no_bo:
+       ttm_read_unlock(&dev_priv->reservation_sem);
+       return ret;
+ }
+ /**
+  * vmw_dumb_map_offset - Return the address space offset of a dumb buffer
+  *
+  * @file_priv: Pointer to a struct drm_file identifying the caller.
+  * @dev: Pointer to the drm device.
+  * @handle: Handle identifying the dumb buffer.
+  * @offset: The address space offset returned.
+  * Return: Zero on success, negative error code on failure.
+  *
+  * This is a driver callback for the core drm dumb_map_offset functionality.
+  */
+ int vmw_dumb_map_offset(struct drm_file *file_priv,
+                       struct drm_device *dev, uint32_t handle,
+                       uint64_t *offset)
+ {
+       struct ttm_object_file *tfile = vmw_fpriv(file_priv)->tfile;
+       struct vmw_buffer_object *out_buf;
+       int ret;
+       ret = vmw_user_bo_lookup(tfile, handle, &out_buf, NULL);
+       if (ret != 0)
+               return -EINVAL;
+       *offset = drm_vma_node_offset_addr(&out_buf->base.vma_node);
+       vmw_bo_unreference(&out_buf);
+       return 0;
+ }
+ /**
+  * vmw_dumb_destroy - Destroy a dumb boffer
+  *
+  * @file_priv: Pointer to a struct drm_file identifying the caller.
+  * @dev: Pointer to the drm device.
+  * @handle: Handle identifying the dumb buffer.
+  * Return: Zero on success, negative error code on failure.
+  *
+  * This is a driver callback for the core drm dumb_destroy functionality.
+  */
+ int vmw_dumb_destroy(struct drm_file *file_priv,
+                    struct drm_device *dev,
+                    uint32_t handle)
+ {
+       return ttm_ref_object_base_unref(vmw_fpriv(file_priv)->tfile,
+                                        handle, TTM_REF_USAGE);
+ }
+ /**
+  * vmw_bo_swap_notify - swapout notify callback.
+  *
+  * @bo: The buffer object to be swapped out.
+  */
+ void vmw_bo_swap_notify(struct ttm_buffer_object *bo)
+ {
+       /* Is @bo embedded in a struct vmw_buffer_object? */
+       if (bo->destroy != vmw_bo_bo_free &&
+           bo->destroy != vmw_user_bo_destroy)
+               return;
+       /* Kill any cached kernel maps before swapout */
+       vmw_bo_unmap(vmw_buffer_object(bo));
+ }
+ /**
+  * vmw_bo_move_notify - TTM move_notify_callback
+  *
+  * @bo: The TTM buffer object about to move.
+  * @mem: The struct ttm_mem_reg indicating to what memory
+  *       region the move is taking place.
+  *
+  * Detaches cached maps and device bindings that require that the
+  * buffer doesn't move.
+  */
+ void vmw_bo_move_notify(struct ttm_buffer_object *bo,
+                       struct ttm_mem_reg *mem)
+ {
+       struct vmw_buffer_object *vbo;
+       if (mem == NULL)
+               return;
+       /* Make sure @bo is embedded in a struct vmw_buffer_object? */
+       if (bo->destroy != vmw_bo_bo_free &&
+           bo->destroy != vmw_user_bo_destroy)
+               return;
+       vbo = container_of(bo, struct vmw_buffer_object, base);
+       /*
+        * Kill any cached kernel maps before move to or from VRAM.
+        * With other types of moves, the underlying pages stay the same,
+        * and the map can be kept.
+        */
+       if (mem->mem_type == TTM_PL_VRAM || bo->mem.mem_type == TTM_PL_VRAM)
+               vmw_bo_unmap(vbo);
+       /*
+        * If we're moving a backup MOB out of MOB placement, then make sure we
+        * read back all resource content first, and unbind the MOB from
+        * the resource.
+        */
+       if (mem->mem_type != VMW_PL_MOB && bo->mem.mem_type == VMW_PL_MOB)
+               vmw_resource_unbind_list(vbo);
+ }
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge