linux/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
Deepak Rawat 5e8ec0d919 drm/vmwgfx: Add support for UA view commands
Virtual device now support new commands to manage unordered access
views. Allow them as part of user-space command buffer. This involves
adding UA view cotable, binding tracker info, new view type and command
verifier functions.

v2: fix comment typo
v3: style fixes (don't use deprecated PTR_RET)

Signed-off-by: Deepak Rawat <drawat.floss@gmail.com>
Signed-off-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Thomas Hellström (VMware) <thomas_os@shipmail.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Roland Scheidegger <sroland@vmware.com>
2020-03-23 22:39:34 +01:00

584 lines
18 KiB
C

// SPDX-License-Identifier: GPL-2.0 OR MIT
/**************************************************************************
* Copyright 2014-2015 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.
*
**************************************************************************/
#include "vmwgfx_drv.h"
#include "vmwgfx_resource_priv.h"
#include "vmwgfx_so.h"
#include "vmwgfx_binding.h"
/*
* The currently only reason we need to keep track of views is that if we
* destroy a hardware surface, all views pointing to it must also be destroyed,
* otherwise the device will error.
* So in particuar if a surface is evicted, we must destroy all views pointing
* to it, and all context bindings of that view. Similarly we must restore
* the view bindings, views and surfaces pointed to by the views when a
* context is referenced in the command stream.
*/
/**
* struct vmw_view - view metadata
*
* @res: The struct vmw_resource we derive from
* @ctx: Non-refcounted pointer to the context this view belongs to.
* @srf: Refcounted pointer to the surface pointed to by this view.
* @cotable: Refcounted pointer to the cotable holding this view.
* @srf_head: List head for the surface-to-view list.
* @cotable_head: List head for the cotable-to_view list.
* @view_type: View type.
* @view_id: User-space per context view id. Currently used also as per
* context device view id.
* @cmd_size: Size of the SVGA3D define view command that we've copied from the
* command stream.
* @committed: Whether the view is actually created or pending creation at the
* device level.
* @cmd: The SVGA3D define view command copied from the command stream.
*/
struct vmw_view {
struct rcu_head rcu;
struct vmw_resource res;
struct vmw_resource *ctx; /* Immutable */
struct vmw_resource *srf; /* Immutable */
struct vmw_resource *cotable; /* Immutable */
struct list_head srf_head; /* Protected by binding_mutex */
struct list_head cotable_head; /* Protected by binding_mutex */
unsigned view_type; /* Immutable */
unsigned view_id; /* Immutable */
u32 cmd_size; /* Immutable */
bool committed; /* Protected by binding_mutex */
u32 cmd[1]; /* Immutable */
};
static int vmw_view_create(struct vmw_resource *res);
static int vmw_view_destroy(struct vmw_resource *res);
static void vmw_hw_view_destroy(struct vmw_resource *res);
static void vmw_view_commit_notify(struct vmw_resource *res,
enum vmw_cmdbuf_res_state state);
static const struct vmw_res_func vmw_view_func = {
.res_type = vmw_res_view,
.needs_backup = false,
.may_evict = false,
.type_name = "DX view",
.backup_placement = NULL,
.create = vmw_view_create,
.commit_notify = vmw_view_commit_notify,
};
/**
* struct vmw_view - view define command body stub
*
* @view_id: The device id of the view being defined
* @sid: The surface id of the view being defined
*
* This generic struct is used by the code to change @view_id and @sid of a
* saved view define command.
*/
struct vmw_view_define {
uint32 view_id;
uint32 sid;
};
/**
* vmw_view - Convert a struct vmw_resource to a struct vmw_view
*
* @res: Pointer to the resource to convert.
*
* Returns a pointer to a struct vmw_view.
*/
static struct vmw_view *vmw_view(struct vmw_resource *res)
{
return container_of(res, struct vmw_view, res);
}
/**
* vmw_view_commit_notify - Notify that a view operation has been committed to
* hardware from a user-supplied command stream.
*
* @res: Pointer to the view resource.
* @state: Indicating whether a creation or removal has been committed.
*
*/
static void vmw_view_commit_notify(struct vmw_resource *res,
enum vmw_cmdbuf_res_state state)
{
struct vmw_view *view = vmw_view(res);
struct vmw_private *dev_priv = res->dev_priv;
mutex_lock(&dev_priv->binding_mutex);
if (state == VMW_CMDBUF_RES_ADD) {
struct vmw_surface *srf = vmw_res_to_srf(view->srf);
list_add_tail(&view->srf_head, &srf->view_list);
vmw_cotable_add_resource(view->cotable, &view->cotable_head);
view->committed = true;
res->id = view->view_id;
} else {
list_del_init(&view->cotable_head);
list_del_init(&view->srf_head);
view->committed = false;
res->id = -1;
}
mutex_unlock(&dev_priv->binding_mutex);
}
/**
* vmw_view_create - Create a hardware view.
*
* @res: Pointer to the view resource.
*
* Create a hardware view. Typically used if that view has previously been
* destroyed by an eviction operation.
*/
static int vmw_view_create(struct vmw_resource *res)
{
struct vmw_view *view = vmw_view(res);
struct vmw_surface *srf = vmw_res_to_srf(view->srf);
struct vmw_private *dev_priv = res->dev_priv;
struct {
SVGA3dCmdHeader header;
struct vmw_view_define body;
} *cmd;
mutex_lock(&dev_priv->binding_mutex);
if (!view->committed) {
mutex_unlock(&dev_priv->binding_mutex);
return 0;
}
cmd = VMW_FIFO_RESERVE_DX(res->dev_priv, view->cmd_size, view->ctx->id);
if (!cmd) {
mutex_unlock(&dev_priv->binding_mutex);
return -ENOMEM;
}
memcpy(cmd, &view->cmd, view->cmd_size);
WARN_ON(cmd->body.view_id != view->view_id);
/* Sid may have changed due to surface eviction. */
WARN_ON(view->srf->id == SVGA3D_INVALID_ID);
cmd->body.sid = view->srf->id;
vmw_fifo_commit(res->dev_priv, view->cmd_size);
res->id = view->view_id;
list_add_tail(&view->srf_head, &srf->view_list);
vmw_cotable_add_resource(view->cotable, &view->cotable_head);
mutex_unlock(&dev_priv->binding_mutex);
return 0;
}
/**
* vmw_view_destroy - Destroy a hardware view.
*
* @res: Pointer to the view resource.
*
* Destroy a hardware view. Typically used on unexpected termination of the
* owning process or if the surface the view is pointing to is destroyed.
*/
static int vmw_view_destroy(struct vmw_resource *res)
{
struct vmw_private *dev_priv = res->dev_priv;
struct vmw_view *view = vmw_view(res);
struct {
SVGA3dCmdHeader header;
union vmw_view_destroy body;
} *cmd;
lockdep_assert_held_once(&dev_priv->binding_mutex);
vmw_binding_res_list_scrub(&res->binding_head);
if (!view->committed || res->id == -1)
return 0;
cmd = VMW_FIFO_RESERVE_DX(dev_priv, sizeof(*cmd), view->ctx->id);
if (!cmd)
return -ENOMEM;
cmd->header.id = vmw_view_destroy_cmds[view->view_type];
cmd->header.size = sizeof(cmd->body);
cmd->body.view_id = view->view_id;
vmw_fifo_commit(dev_priv, sizeof(*cmd));
res->id = -1;
list_del_init(&view->cotable_head);
list_del_init(&view->srf_head);
return 0;
}
/**
* vmw_hw_view_destroy - Destroy a hardware view as part of resource cleanup.
*
* @res: Pointer to the view resource.
*
* Destroy a hardware view if it's still present.
*/
static void vmw_hw_view_destroy(struct vmw_resource *res)
{
struct vmw_private *dev_priv = res->dev_priv;
mutex_lock(&dev_priv->binding_mutex);
WARN_ON(vmw_view_destroy(res));
res->id = -1;
mutex_unlock(&dev_priv->binding_mutex);
}
/**
* vmw_view_key - Compute a view key suitable for the cmdbuf resource manager
*
* @user_key: The user-space id used for the view.
* @view_type: The view type.
*
* Destroy a hardware view if it's still present.
*/
static u32 vmw_view_key(u32 user_key, enum vmw_view_type view_type)
{
return user_key | (view_type << 20);
}
/**
* vmw_view_id_ok - Basic view id and type range checks.
*
* @user_key: The user-space id used for the view.
* @view_type: The view type.
*
* Checks that the view id and type (typically provided by user-space) is
* valid.
*/
static bool vmw_view_id_ok(u32 user_key, enum vmw_view_type view_type)
{
return (user_key < SVGA_COTABLE_MAX_IDS &&
view_type < vmw_view_max);
}
/**
* vmw_view_res_free - resource res_free callback for view resources
*
* @res: Pointer to a struct vmw_resource
*
* Frees memory and memory accounting held by a struct vmw_view.
*/
static void vmw_view_res_free(struct vmw_resource *res)
{
struct vmw_view *view = vmw_view(res);
size_t size = offsetof(struct vmw_view, cmd) + view->cmd_size;
struct vmw_private *dev_priv = res->dev_priv;
vmw_resource_unreference(&view->cotable);
vmw_resource_unreference(&view->srf);
kfree_rcu(view, rcu);
ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
}
/**
* vmw_view_add - Create a view resource and stage it for addition
* as a command buffer managed resource.
*
* @man: Pointer to the compat shader manager identifying the shader namespace.
* @ctx: Pointer to a struct vmw_resource identifying the active context.
* @srf: Pointer to a struct vmw_resource identifying the surface the view
* points to.
* @view_type: The view type deduced from the view create command.
* @user_key: The key that is used to identify the shader. The key is
* unique to the view type and to the context.
* @cmd: Pointer to the view create command in the command stream.
* @cmd_size: Size of the view create command in the command stream.
* @list: Caller's list of staged command buffer resource actions.
*/
int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
struct vmw_resource *ctx,
struct vmw_resource *srf,
enum vmw_view_type view_type,
u32 user_key,
const void *cmd,
size_t cmd_size,
struct list_head *list)
{
static const size_t vmw_view_define_sizes[] = {
[vmw_view_sr] = sizeof(SVGA3dCmdDXDefineShaderResourceView),
[vmw_view_rt] = sizeof(SVGA3dCmdDXDefineRenderTargetView),
[vmw_view_ds] = sizeof(SVGA3dCmdDXDefineDepthStencilView),
[vmw_view_ua] = sizeof(SVGA3dCmdDXDefineUAView)
};
struct vmw_private *dev_priv = ctx->dev_priv;
struct vmw_resource *res;
struct vmw_view *view;
struct ttm_operation_ctx ttm_opt_ctx = {
.interruptible = true,
.no_wait_gpu = false
};
size_t size;
int ret;
if (cmd_size != vmw_view_define_sizes[view_type] +
sizeof(SVGA3dCmdHeader)) {
VMW_DEBUG_USER("Illegal view create command size.\n");
return -EINVAL;
}
if (!vmw_view_id_ok(user_key, view_type)) {
VMW_DEBUG_USER("Illegal view add view id.\n");
return -EINVAL;
}
size = offsetof(struct vmw_view, cmd) + cmd_size;
ret = ttm_mem_global_alloc(vmw_mem_glob(dev_priv), size, &ttm_opt_ctx);
if (ret) {
if (ret != -ERESTARTSYS)
DRM_ERROR("Out of graphics memory for view creation\n");
return ret;
}
view = kmalloc(size, GFP_KERNEL);
if (!view) {
ttm_mem_global_free(vmw_mem_glob(dev_priv), size);
return -ENOMEM;
}
res = &view->res;
view->ctx = ctx;
view->srf = vmw_resource_reference(srf);
view->cotable = vmw_resource_reference
(vmw_context_cotable(ctx, vmw_view_cotables[view_type]));
view->view_type = view_type;
view->view_id = user_key;
view->cmd_size = cmd_size;
view->committed = false;
INIT_LIST_HEAD(&view->srf_head);
INIT_LIST_HEAD(&view->cotable_head);
memcpy(&view->cmd, cmd, cmd_size);
ret = vmw_resource_init(dev_priv, res, true,
vmw_view_res_free, &vmw_view_func);
if (ret)
goto out_resource_init;
ret = vmw_cmdbuf_res_add(man, vmw_cmdbuf_res_view,
vmw_view_key(user_key, view_type),
res, list);
if (ret)
goto out_resource_init;
res->id = view->view_id;
res->hw_destroy = vmw_hw_view_destroy;
out_resource_init:
vmw_resource_unreference(&res);
return ret;
}
/**
* vmw_view_remove - Stage a view for removal.
*
* @man: Pointer to the view manager identifying the shader namespace.
* @user_key: The key that is used to identify the view. The key is
* unique to the view type.
* @view_type: View type
* @list: Caller's list of staged command buffer resource actions.
* @res_p: If the resource is in an already committed state, points to the
* struct vmw_resource on successful return. The pointer will be
* non ref-counted.
*/
int vmw_view_remove(struct vmw_cmdbuf_res_manager *man,
u32 user_key, enum vmw_view_type view_type,
struct list_head *list,
struct vmw_resource **res_p)
{
if (!vmw_view_id_ok(user_key, view_type)) {
VMW_DEBUG_USER("Illegal view remove view id.\n");
return -EINVAL;
}
return vmw_cmdbuf_res_remove(man, vmw_cmdbuf_res_view,
vmw_view_key(user_key, view_type),
list, res_p);
}
/**
* vmw_view_cotable_list_destroy - Evict all views belonging to a cotable.
*
* @dev_priv: Pointer to a device private struct.
* @list: List of views belonging to a cotable.
* @readback: Unused. Needed for function interface only.
*
* This function evicts all views belonging to a cotable.
* It must be called with the binding_mutex held, and the caller must hold
* a reference to the view resource. This is typically called before the
* cotable is paged out.
*/
void vmw_view_cotable_list_destroy(struct vmw_private *dev_priv,
struct list_head *list,
bool readback)
{
struct vmw_view *entry, *next;
lockdep_assert_held_once(&dev_priv->binding_mutex);
list_for_each_entry_safe(entry, next, list, cotable_head)
WARN_ON(vmw_view_destroy(&entry->res));
}
/**
* vmw_view_surface_list_destroy - Evict all views pointing to a surface
*
* @dev_priv: Pointer to a device private struct.
* @list: List of views pointing to a surface.
*
* This function evicts all views pointing to a surface. This is typically
* called before the surface is evicted.
*/
void vmw_view_surface_list_destroy(struct vmw_private *dev_priv,
struct list_head *list)
{
struct vmw_view *entry, *next;
lockdep_assert_held_once(&dev_priv->binding_mutex);
list_for_each_entry_safe(entry, next, list, srf_head)
WARN_ON(vmw_view_destroy(&entry->res));
}
/**
* vmw_view_srf - Return a non-refcounted pointer to the surface a view is
* pointing to.
*
* @res: pointer to a view resource.
*
* Note that the view itself is holding a reference, so as long
* the view resource is alive, the surface resource will be.
*/
struct vmw_resource *vmw_view_srf(struct vmw_resource *res)
{
return vmw_view(res)->srf;
}
/**
* vmw_view_lookup - Look up a view.
*
* @man: The context's cmdbuf ref manager.
* @view_type: The view type.
* @user_key: The view user id.
*
* returns a refcounted pointer to a view or an error pointer if not found.
*/
struct vmw_resource *vmw_view_lookup(struct vmw_cmdbuf_res_manager *man,
enum vmw_view_type view_type,
u32 user_key)
{
return vmw_cmdbuf_res_lookup(man, vmw_cmdbuf_res_view,
vmw_view_key(user_key, view_type));
}
/**
* vmw_view_dirtying - Return whether a view type is dirtying its resource
* @res: Pointer to the view
*
* Each time a resource is put on the validation list as the result of a
* view pointing to it, we need to determine whether that resource will
* be dirtied (written to by the GPU) as a result of the corresponding
* GPU operation. Currently only rendertarget-, depth-stencil and unordered
* access views are capable of dirtying its resource.
*
* Return: Whether the view type of @res dirties the resource it points to.
*/
u32 vmw_view_dirtying(struct vmw_resource *res)
{
static u32 view_is_dirtying[vmw_view_max] = {
[vmw_view_rt] = VMW_RES_DIRTY_SET,
[vmw_view_ds] = VMW_RES_DIRTY_SET,
[vmw_view_ua] = VMW_RES_DIRTY_SET,
};
/* Update this function as we add more view types */
BUILD_BUG_ON(vmw_view_max != 4);
return view_is_dirtying[vmw_view(res)->view_type];
}
const u32 vmw_view_destroy_cmds[] = {
[vmw_view_sr] = SVGA_3D_CMD_DX_DESTROY_SHADERRESOURCE_VIEW,
[vmw_view_rt] = SVGA_3D_CMD_DX_DESTROY_RENDERTARGET_VIEW,
[vmw_view_ds] = SVGA_3D_CMD_DX_DESTROY_DEPTHSTENCIL_VIEW,
[vmw_view_ua] = SVGA_3D_CMD_DX_DESTROY_UA_VIEW,
};
const SVGACOTableType vmw_view_cotables[] = {
[vmw_view_sr] = SVGA_COTABLE_SRVIEW,
[vmw_view_rt] = SVGA_COTABLE_RTVIEW,
[vmw_view_ds] = SVGA_COTABLE_DSVIEW,
[vmw_view_ua] = SVGA_COTABLE_UAVIEW,
};
const SVGACOTableType vmw_so_cotables[] = {
[vmw_so_el] = SVGA_COTABLE_ELEMENTLAYOUT,
[vmw_so_bs] = SVGA_COTABLE_BLENDSTATE,
[vmw_so_ds] = SVGA_COTABLE_DEPTHSTENCIL,
[vmw_so_rs] = SVGA_COTABLE_RASTERIZERSTATE,
[vmw_so_ss] = SVGA_COTABLE_SAMPLER,
[vmw_so_so] = SVGA_COTABLE_STREAMOUTPUT
};
/* To remove unused function warning */
static void vmw_so_build_asserts(void) __attribute__((used));
/*
* This function is unused at run-time, and only used to dump various build
* asserts important for code optimization assumptions.
*/
static void vmw_so_build_asserts(void)
{
/* Assert that our vmw_view_cmd_to_type() function is correct. */
BUILD_BUG_ON(SVGA_3D_CMD_DX_DESTROY_SHADERRESOURCE_VIEW !=
SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW + 1);
BUILD_BUG_ON(SVGA_3D_CMD_DX_DEFINE_RENDERTARGET_VIEW !=
SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW + 2);
BUILD_BUG_ON(SVGA_3D_CMD_DX_DESTROY_RENDERTARGET_VIEW !=
SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW + 3);
BUILD_BUG_ON(SVGA_3D_CMD_DX_DEFINE_DEPTHSTENCIL_VIEW !=
SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW + 4);
BUILD_BUG_ON(SVGA_3D_CMD_DX_DESTROY_DEPTHSTENCIL_VIEW !=
SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW + 5);
/* Assert that our "one body fits all" assumption is valid */
BUILD_BUG_ON(sizeof(union vmw_view_destroy) != sizeof(u32));
/* Assert that the view key space can hold all view ids. */
BUILD_BUG_ON(SVGA_COTABLE_MAX_IDS >= ((1 << 20) - 1));
/*
* Assert that the offset of sid in all view define commands
* is what we assume it to be.
*/
BUILD_BUG_ON(offsetof(struct vmw_view_define, sid) !=
offsetof(SVGA3dCmdDXDefineShaderResourceView, sid));
BUILD_BUG_ON(offsetof(struct vmw_view_define, sid) !=
offsetof(SVGA3dCmdDXDefineRenderTargetView, sid));
BUILD_BUG_ON(offsetof(struct vmw_view_define, sid) !=
offsetof(SVGA3dCmdDXDefineDepthStencilView, sid));
}