Quelle  vmwgfx_cotable.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0 OR MIT
/**************************************************************************
 *
 * Copyright 2014-2023 VMware, Inc., Palo Alto, CA., USA
 *
 * 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.
 *
 **************************************************************************/
/*
 * Treat context OTables as resources to make use of the resource
 * backing MOB eviction mechanism, that is used to read back the COTable
 * whenever the backing MOB is evicted.
 */

#include "vmwgfx_bo.h"
#include "vmwgfx_drv.h"
#include "vmwgfx_mksstat.h"
#include "vmwgfx_resource_priv.h"
#include "vmwgfx_so.h"

#include <drm/ttm/ttm_placement.h>

/**
 * struct vmw_cotable - Context Object Table resource
 *
 * @res: struct vmw_resource we are deriving from.
 * @ctx: non-refcounted pointer to the owning context.
 * @size_read_back: Size of data read back during eviction.
 * @seen_entries: Seen entries in command stream for this cotable.
 * @type: The cotable type.
 * @scrubbed: Whether the cotable has been scrubbed.
 * @resource_list: List of resources in the cotable.
 */
struct vmw_cotable {
 struct vmw_resource res;
 struct vmw_resource *ctx;
 size_t size_read_back;
 int seen_entries;
 u32 type;
 bool scrubbed;
 struct list_head resource_list;
};

/**
 * struct vmw_cotable_info - Static info about cotable types
 *
 * @min_initial_entries: Min number of initial intries at cotable allocation
 * for this cotable type.
 * @size: Size of each entry.
 * @unbind_func: Unbind call-back function.
 */
struct vmw_cotable_info {
 u32 min_initial_entries;
 u32 size;
 void (*unbind_func)(struct vmw_private *, struct list_head *,
       bool);
};


/*
 * Getting the initial size right is difficult because it all depends
 * on what the userspace is doing. The sizes will be aligned up to
 * a PAGE_SIZE so we just want to make sure that for majority of apps
 * the initial number of entries doesn't require an immediate resize.
 * For all cotables except SVGACOTableDXElementLayoutEntry and
 * SVGACOTableDXBlendStateEntry the initial number of entries fits
 * within the PAGE_SIZE. For SVGACOTableDXElementLayoutEntry and
 * SVGACOTableDXBlendStateEntry we want to reserve two pages,
 * because that's what all apps will require initially.
 */
static const struct vmw_cotable_info co_info[] = {
 {1, sizeof(SVGACOTableDXRTViewEntry), &vmw_view_cotable_list_destroy},
 {1, sizeof(SVGACOTableDXDSViewEntry), &vmw_view_cotable_list_destroy},
 {1, sizeof(SVGACOTableDXSRViewEntry), &vmw_view_cotable_list_destroy},
 {PAGE_SIZE/sizeof(SVGACOTableDXElementLayoutEntry) + 1, sizeof(SVGACOTableDXElementLayoutEntry), NULL},
 {PAGE_SIZE/sizeof(SVGACOTableDXBlendStateEntry) + 1, sizeof(SVGACOTableDXBlendStateEntry), NULL},
 {1, sizeof(SVGACOTableDXDepthStencilEntry), NULL},
 {1, sizeof(SVGACOTableDXRasterizerStateEntry), NULL},
 {1, sizeof(SVGACOTableDXSamplerEntry), NULL},
 {1, sizeof(SVGACOTableDXStreamOutputEntry), &vmw_dx_streamoutput_cotable_list_scrub},
 {1, sizeof(SVGACOTableDXQueryEntry), NULL},
 {1, sizeof(SVGACOTableDXShaderEntry), &vmw_dx_shader_cotable_list_scrub},
 {1, sizeof(SVGACOTableDXUAViewEntry), &vmw_view_cotable_list_destroy}
};

/*
 * Cotables with bindings that we remove must be scrubbed first,
 * otherwise, the device will swap in an invalid context when we remove
 * bindings before scrubbing a cotable...
 */
const SVGACOTableType vmw_cotable_scrub_order[] = {
 SVGA_COTABLE_RTVIEW,
 SVGA_COTABLE_DSVIEW,
 SVGA_COTABLE_SRVIEW,
 SVGA_COTABLE_DXSHADER,
 SVGA_COTABLE_ELEMENTLAYOUT,
 SVGA_COTABLE_BLENDSTATE,
 SVGA_COTABLE_DEPTHSTENCIL,
 SVGA_COTABLE_RASTERIZERSTATE,
 SVGA_COTABLE_SAMPLER,
 SVGA_COTABLE_STREAMOUTPUT,
 SVGA_COTABLE_DXQUERY,
 SVGA_COTABLE_UAVIEW,
};

static int vmw_cotable_bind(struct vmw_resource *res,
       struct ttm_validate_buffer *val_buf);
static int vmw_cotable_unbind(struct vmw_resource *res,
         bool readback,
         struct ttm_validate_buffer *val_buf);
static int vmw_cotable_create(struct vmw_resource *res);
static int vmw_cotable_destroy(struct vmw_resource *res);

static const struct vmw_res_func vmw_cotable_func = {
 .res_type = vmw_res_cotable,
 .needs_guest_memory = true,
 .may_evict = true,
 .prio = 3,
 .dirty_prio = 3,
 .type_name = "context guest backed object tables",
 .domain = VMW_BO_DOMAIN_MOB,
 .busy_domain = VMW_BO_DOMAIN_MOB,
 .create = vmw_cotable_create,
 .destroy = vmw_cotable_destroy,
 .bind = vmw_cotable_bind,
 .unbind = vmw_cotable_unbind,
};

/**
 * vmw_cotable - Convert a struct vmw_resource pointer to a struct
 * vmw_cotable pointer
 *
 * @res: Pointer to the resource.
 */
static struct vmw_cotable *vmw_cotable(struct vmw_resource *res)
{
 return container_of(res, struct vmw_cotable, res);
}

/**
 * vmw_cotable_destroy - Cotable resource destroy callback
 *
 * @res: Pointer to the cotable resource.
 *
 * There is no device cotable destroy command, so this function only
 * makes sure that the resource id is set to invalid.
 */
static int vmw_cotable_destroy(struct vmw_resource *res)
{
 res->id = -1;
 return 0;
}

/**
 * vmw_cotable_unscrub - Undo a cotable unscrub operation
 *
 * @res: Pointer to the cotable resource
 *
 * This function issues commands to (re)bind the cotable to
 * its backing mob, which needs to be validated and reserved at this point.
 * This is identical to bind() except the function interface looks different.
 */
static int vmw_cotable_unscrub(struct vmw_resource *res)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 struct vmw_private *dev_priv = res->dev_priv;
 struct ttm_buffer_object *bo = &res->guest_memory_bo->tbo;
 struct {
  SVGA3dCmdHeader header;
  SVGA3dCmdDXSetCOTable body;
 } *cmd;

 WARN_ON_ONCE(bo->resource->mem_type != VMW_PL_MOB);
 dma_resv_assert_held(bo->base.resv);

 cmd = VMW_CMD_RESERVE(dev_priv, sizeof(*cmd));
 if (!cmd)
  return -ENOMEM;

 WARN_ON(vcotbl->ctx->id == SVGA3D_INVALID_ID);
 WARN_ON(bo->resource->mem_type != VMW_PL_MOB);
 cmd->header.id = SVGA_3D_CMD_DX_SET_COTABLE;
 cmd->header.size = sizeof(cmd->body);
 cmd->body.cid = vcotbl->ctx->id;
 cmd->body.type = vcotbl->type;
 cmd->body.mobid = bo->resource->start;
 cmd->body.validSizeInBytes = vcotbl->size_read_back;

 vmw_cmd_commit_flush(dev_priv, sizeof(*cmd));
 vcotbl->scrubbed = false;

 return 0;
}

/**
 * vmw_cotable_bind - Undo a cotable unscrub operation
 *
 * @res: Pointer to the cotable resource
 * @val_buf: Pointer to a struct ttm_validate_buffer prepared by the caller
 * for convenience / fencing.
 *
 * This function issues commands to (re)bind the cotable to
 * its backing mob, which needs to be validated and reserved at this point.
 */
static int vmw_cotable_bind(struct vmw_resource *res,
       struct ttm_validate_buffer *val_buf)
{
 /*
  * The create() callback may have changed @res->backup without
  * the caller noticing, and with val_buf->bo still pointing to
  * the old backup buffer. Although hackish, and not used currently,
  * take the opportunity to correct the value here so that it's not
  * misused in the future.
 */
 val_buf->bo = &res->guest_memory_bo->tbo;

 return vmw_cotable_unscrub(res);
}

/**
 * vmw_cotable_scrub - Scrub the cotable from the device.
 *
 * @res: Pointer to the cotable resource.
 * @readback: Whether initiate a readback of the cotable data to the backup
 * buffer.
 *
 * In some situations (context swapouts) it might be desirable to make the
 * device forget about the cotable without performing a full unbind. A full
 * unbind requires reserved backup buffers and it might not be possible to
 * reserve them due to locking order violation issues. The vmw_cotable_scrub
 * function implements a partial unbind() without that requirement but with the
 * following restrictions.
 * 1) Before the cotable is again used by the GPU, vmw_cotable_unscrub() must
 *    be called.
 * 2) Before the cotable backing buffer is used by the CPU, or during the
 *    resource destruction, vmw_cotable_unbind() must be called.
 */
int vmw_cotable_scrub(struct vmw_resource *res, bool readback)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 struct vmw_private *dev_priv = res->dev_priv;
 size_t submit_size;

 struct {
  SVGA3dCmdHeader header;
  SVGA3dCmdDXReadbackCOTable body;
 } *cmd0;
 struct {
  SVGA3dCmdHeader header;
  SVGA3dCmdDXSetCOTable body;
 } *cmd1;

 if (vcotbl->scrubbed)
  return 0;

 if (co_info[vcotbl->type].unbind_func)
  co_info[vcotbl->type].unbind_func(dev_priv,
        &vcotbl->resource_list,
        readback);
 submit_size = sizeof(*cmd1);
 if (readback)
  submit_size += sizeof(*cmd0);

 cmd1 = VMW_CMD_RESERVE(dev_priv, submit_size);
 if (!cmd1)
  return -ENOMEM;

 vcotbl->size_read_back = 0;
 if (readback) {
  cmd0 = (void *) cmd1;
  cmd0->header.id = SVGA_3D_CMD_DX_READBACK_COTABLE;
  cmd0->header.size = sizeof(cmd0->body);
  cmd0->body.cid = vcotbl->ctx->id;
  cmd0->body.type = vcotbl->type;
  cmd1 = (void *) &cmd0[1];
  vcotbl->size_read_back = res->guest_memory_size;
 }
 cmd1->header.id = SVGA_3D_CMD_DX_SET_COTABLE;
 cmd1->header.size = sizeof(cmd1->body);
 cmd1->body.cid = vcotbl->ctx->id;
 cmd1->body.type = vcotbl->type;
 cmd1->body.mobid = SVGA3D_INVALID_ID;
 cmd1->body.validSizeInBytes = 0;
 vmw_cmd_commit_flush(dev_priv, submit_size);
 vcotbl->scrubbed = true;

 /* Trigger a create() on next validate. */
 res->id = -1;

 return 0;
}

/**
 * vmw_cotable_unbind - Cotable resource unbind callback
 *
 * @res: Pointer to the cotable resource.
 * @readback: Whether to read back cotable data to the backup buffer.
 * @val_buf: Pointer to a struct ttm_validate_buffer prepared by the caller
 * for convenience / fencing.
 *
 * Unbinds the cotable from the device and fences the backup buffer.
 */
static int vmw_cotable_unbind(struct vmw_resource *res,
         bool readback,
         struct ttm_validate_buffer *val_buf)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 struct vmw_private *dev_priv = res->dev_priv;
 struct ttm_buffer_object *bo = val_buf->bo;
 struct vmw_fence_obj *fence;

 if (!vmw_resource_mob_attached(res))
  return 0;

 WARN_ON_ONCE(bo->resource->mem_type != VMW_PL_MOB);
 dma_resv_assert_held(bo->base.resv);

 mutex_lock(&dev_priv->binding_mutex);
 if (!vcotbl->scrubbed)
  vmw_dx_context_scrub_cotables(vcotbl->ctx, readback);
 mutex_unlock(&dev_priv->binding_mutex);
 (void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
 vmw_bo_fence_single(bo, fence);
 if (likely(fence != NULL))
  vmw_fence_obj_unreference(&fence);

 return 0;
}

/**
 * vmw_cotable_readback - Read back a cotable without unbinding.
 *
 * @res: The cotable resource.
 *
 * Reads back a cotable to its backing mob without scrubbing the MOB from
 * the cotable. The MOB is fenced for subsequent CPU access.
 */
static int vmw_cotable_readback(struct vmw_resource *res)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 struct vmw_private *dev_priv = res->dev_priv;

 struct {
  SVGA3dCmdHeader header;
  SVGA3dCmdDXReadbackCOTable body;
 } *cmd;
 struct vmw_fence_obj *fence;

 if (!vcotbl->scrubbed) {
  cmd = VMW_CMD_RESERVE(dev_priv, sizeof(*cmd));
  if (!cmd)
   return -ENOMEM;

  cmd->header.id = SVGA_3D_CMD_DX_READBACK_COTABLE;
  cmd->header.size = sizeof(cmd->body);
  cmd->body.cid = vcotbl->ctx->id;
  cmd->body.type = vcotbl->type;
  vcotbl->size_read_back = res->guest_memory_size;
  vmw_cmd_commit(dev_priv, sizeof(*cmd));
 }

 (void) vmw_execbuf_fence_commands(NULL, dev_priv, &fence, NULL);
 vmw_bo_fence_single(&res->guest_memory_bo->tbo, fence);
 vmw_fence_obj_unreference(&fence);

 return 0;
}

/**
 * vmw_cotable_resize - Resize a cotable.
 *
 * @res: The cotable resource.
 * @new_size: The new size.
 *
 * Resizes a cotable and binds the new backup buffer.
 * On failure the cotable is left intact.
 * Important! This function may not fail once the MOB switch has been
 * committed to hardware. That would put the device context in an
 * invalid state which we can't currently recover from.
 */
static int vmw_cotable_resize(struct vmw_resource *res, size_t new_size)
{
 struct ttm_operation_ctx ctx = { false, false };
 struct vmw_private *dev_priv = res->dev_priv;
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 struct vmw_bo *buf, *old_buf = res->guest_memory_bo;
 struct ttm_buffer_object *bo, *old_bo = &res->guest_memory_bo->tbo;
 size_t old_size = res->guest_memory_size;
 size_t old_size_read_back = vcotbl->size_read_back;
 size_t cur_size_read_back;
 struct ttm_bo_kmap_obj old_map, new_map;
 int ret;
 size_t i;
 struct vmw_bo_params bo_params = {
  .domain = VMW_BO_DOMAIN_MOB,
  .busy_domain = VMW_BO_DOMAIN_MOB,
  .bo_type = ttm_bo_type_device,
  .size = new_size,
  .pin = true
 };

 MKS_STAT_TIME_DECL(MKSSTAT_KERN_COTABLE_RESIZE);
 MKS_STAT_TIME_PUSH(MKSSTAT_KERN_COTABLE_RESIZE);

 ret = vmw_cotable_readback(res);
 if (ret)
  goto out_done;

 cur_size_read_back = vcotbl->size_read_back;
 vcotbl->size_read_back = old_size_read_back;

 /*
  * While device is processing, Allocate and reserve a buffer object
  * for the new COTable. Initially pin the buffer object to make sure
  * we can use tryreserve without failure.
 */
 ret = vmw_bo_create(dev_priv, &bo_params, &buf);
 if (ret) {
  DRM_ERROR("Failed initializing new cotable MOB.\n");
  goto out_done;
 }

 bo = &buf->tbo;
 WARN_ON_ONCE(ttm_bo_reserve(bo, false, true, NULL));

 ret = ttm_bo_wait(old_bo, false, false);
 if (unlikely(ret != 0)) {
  DRM_ERROR("Failed waiting for cotable unbind.\n");
  goto out_wait;
 }

 /*
  * Do a page by page copy of COTables. This eliminates slow vmap()s.
  * This should really be a TTM utility.
 */
 for (i = 0; i < PFN_UP(old_bo->resource->size); ++i) {
  bool dummy;

  ret = ttm_bo_kmap(old_bo, i, 1, &old_map);
  if (unlikely(ret != 0)) {
   DRM_ERROR("Failed mapping old COTable on resize.\n");
   goto out_wait;
  }
  ret = ttm_bo_kmap(bo, i, 1, &new_map);
  if (unlikely(ret != 0)) {
   DRM_ERROR("Failed mapping new COTable on resize.\n");
   goto out_map_new;
  }
  memcpy(ttm_kmap_obj_virtual(&new_map, &dummy),
         ttm_kmap_obj_virtual(&old_map, &dummy),
         PAGE_SIZE);
  ttm_bo_kunmap(&new_map);
  ttm_bo_kunmap(&old_map);
 }

 /* Unpin new buffer, and switch backup buffers. */
 vmw_bo_placement_set(buf,
        VMW_BO_DOMAIN_MOB,
        VMW_BO_DOMAIN_MOB);
 ret = ttm_bo_validate(bo, &buf->placement, &ctx);
 if (unlikely(ret != 0)) {
  DRM_ERROR("Failed validating new COTable backup buffer.\n");
  goto out_wait;
 }

 vmw_resource_mob_detach(res);
 res->guest_memory_bo = buf;
 res->guest_memory_size = new_size;
 vcotbl->size_read_back = cur_size_read_back;

 /*
  * Now tell the device to switch. If this fails, then we need to
  * revert the full resize.
 */
 ret = vmw_cotable_unscrub(res);
 if (ret) {
  DRM_ERROR("Failed switching COTable backup buffer.\n");
  res->guest_memory_bo = old_buf;
  res->guest_memory_size = old_size;
  vcotbl->size_read_back = old_size_read_back;
  vmw_resource_mob_attach(res);
  goto out_wait;
 }

 vmw_resource_mob_attach(res);
 /* Let go of the old mob. */
 vmw_user_bo_unref(&old_buf);
 res->id = vcotbl->type;

 ret = dma_resv_reserve_fences(bo->base.resv, 1);
 if (unlikely(ret))
  goto out_wait;

 /* Release the pin acquired in vmw_bo_create */
 ttm_bo_unpin(bo);

 MKS_STAT_TIME_POP(MKSSTAT_KERN_COTABLE_RESIZE);

 return 0;

out_map_new:
 ttm_bo_kunmap(&old_map);
out_wait:
 ttm_bo_unpin(bo);
 ttm_bo_unreserve(bo);
 vmw_user_bo_unref(&buf);

out_done:
 MKS_STAT_TIME_POP(MKSSTAT_KERN_COTABLE_RESIZE);

 return ret;
}

/**
 * vmw_cotable_create - Cotable resource create callback
 *
 * @res: Pointer to a cotable resource.
 *
 * There is no separate create command for cotables, so this callback, which
 * is called before bind() in the validation sequence is instead used for two
 * things.
 * 1) Unscrub the cotable if it is scrubbed and still attached to a backup
 *    buffer.
 * 2) Resize the cotable if needed.
 */
static int vmw_cotable_create(struct vmw_resource *res)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);
 size_t new_size = res->guest_memory_size;
 size_t needed_size;
 int ret;

 /* Check whether we need to resize the cotable */
 needed_size = (vcotbl->seen_entries + 1) * co_info[vcotbl->type].size;
 while (needed_size > new_size)
  new_size *= 2;

 if (likely(new_size <= res->guest_memory_size)) {
  if (vcotbl->scrubbed && vmw_resource_mob_attached(res)) {
   ret = vmw_cotable_unscrub(res);
   if (ret)
    return ret;
  }
  res->id = vcotbl->type;
  return 0;
 }

 return vmw_cotable_resize(res, new_size);
}

/**
 * vmw_hw_cotable_destroy - Cotable hw_destroy callback
 *
 * @res: Pointer to a cotable resource.
 *
 * The final (part of resource destruction) destroy callback.
 */
static void vmw_hw_cotable_destroy(struct vmw_resource *res)
{
 (void) vmw_cotable_destroy(res);
}

/**
 * vmw_cotable_free - Cotable resource destructor
 *
 * @res: Pointer to a cotable resource.
 */
static void vmw_cotable_free(struct vmw_resource *res)
{
 kfree(res);
}

/**
 * vmw_cotable_alloc - Create a cotable resource
 *
 * @dev_priv: Pointer to a device private struct.
 * @ctx: Pointer to the context resource.
 * The cotable resource will not add a refcount.
 * @type: The cotable type.
 */
struct vmw_resource *vmw_cotable_alloc(struct vmw_private *dev_priv,
           struct vmw_resource *ctx,
           u32 type)
{
 struct vmw_cotable *vcotbl;
 int ret;
 u32 num_entries;

 vcotbl = kzalloc(sizeof(*vcotbl), GFP_KERNEL);
 if (unlikely(!vcotbl)) {
  ret = -ENOMEM;
  goto out_no_alloc;
 }

 ret = vmw_resource_init(dev_priv, &vcotbl->res, true,
    vmw_cotable_free, &vmw_cotable_func);
 if (unlikely(ret != 0))
  goto out_no_init;

 INIT_LIST_HEAD(&vcotbl->resource_list);
 vcotbl->res.id = type;
 vcotbl->res.guest_memory_size = PAGE_SIZE;
 num_entries = PAGE_SIZE / co_info[type].size;
 if (num_entries < co_info[type].min_initial_entries) {
  vcotbl->res.guest_memory_size = co_info[type].min_initial_entries *
   co_info[type].size;
  vcotbl->res.guest_memory_size = PFN_ALIGN(vcotbl->res.guest_memory_size);
 }

 vcotbl->scrubbed = true;
 vcotbl->seen_entries = -1;
 vcotbl->type = type;
 vcotbl->ctx = ctx;

 vcotbl->res.hw_destroy = vmw_hw_cotable_destroy;

 return &vcotbl->res;

out_no_init:
 kfree(vcotbl);
out_no_alloc:
 return ERR_PTR(ret);
}

/**
 * vmw_cotable_notify - Notify the cotable about an item creation
 *
 * @res: Pointer to a cotable resource.
 * @id: Item id.
 */
int vmw_cotable_notify(struct vmw_resource *res, int id)
{
 struct vmw_cotable *vcotbl = vmw_cotable(res);

 if (id < 0 || id >= SVGA_COTABLE_MAX_IDS) {
  DRM_ERROR("Illegal COTable id. Type is %u. Id is %d\n",
     (unsigned) vcotbl->type, id);
  return -EINVAL;
 }

 if (vcotbl->seen_entries < id) {
  /* Trigger a call to create() on next validate */
  res->id = -1;
  vcotbl->seen_entries = id;
 }

 return 0;
}

/**
 * vmw_cotable_add_resource - add a view to the cotable's list of active views.
 *
 * @res: pointer struct vmw_resource representing the cotable.
 * @head: pointer to the struct list_head member of the resource, dedicated
 * to the cotable active resource list.
 */
void vmw_cotable_add_resource(struct vmw_resource *res, struct list_head *head)
{
 struct vmw_cotable *vcotbl =
  container_of(res, struct vmw_cotable, res);

 list_add_tail(head, &vcotbl->resource_list);
}