linux/drivers/gpu/drm/i915/i915_vma.h
<<
>>
Prefs
   1/*
   2 * Copyright © 2016 Intel Corporation
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice (including the next
  12 * paragraph) shall be included in all copies or substantial portions of the
  13 * Software.
  14 *
  15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21 * IN THE SOFTWARE.
  22 *
  23 */
  24
  25#ifndef __I915_VMA_H__
  26#define __I915_VMA_H__
  27
  28#include <linux/io-mapping.h>
  29#include <linux/rbtree.h>
  30
  31#include <drm/drm_mm.h>
  32
  33#include "i915_gem_gtt.h"
  34#include "i915_gem_fence_reg.h"
  35#include "i915_gem_object.h"
  36
  37#include "i915_request.h"
  38
  39enum i915_cache_level;
  40
  41/**
  42 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
  43 * VMA's presence cannot be guaranteed before binding, or after unbinding the
  44 * object into/from the address space.
  45 *
  46 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
  47 * will always be <= an objects lifetime. So object refcounting should cover us.
  48 */
  49struct i915_vma {
  50        struct drm_mm_node node;
  51        struct drm_i915_gem_object *obj;
  52        struct i915_address_space *vm;
  53        const struct i915_vma_ops *ops;
  54        struct drm_i915_fence_reg *fence;
  55        struct reservation_object *resv; /** Alias of obj->resv */
  56        struct sg_table *pages;
  57        void __iomem *iomap;
  58        void *private; /* owned by creator */
  59        u64 size;
  60        u64 display_alignment;
  61        struct i915_page_sizes page_sizes;
  62
  63        u32 fence_size;
  64        u32 fence_alignment;
  65
  66        /**
  67         * Count of the number of times this vma has been opened by different
  68         * handles (but same file) for execbuf, i.e. the number of aliases
  69         * that exist in the ctx->handle_vmas LUT for this vma.
  70         */
  71        unsigned int open_count;
  72        unsigned long flags;
  73        /**
  74         * How many users have pinned this object in GTT space. The following
  75         * users can each hold at most one reference: pwrite/pread, execbuffer
  76         * (objects are not allowed multiple times for the same batchbuffer),
  77         * and the framebuffer code. When switching/pageflipping, the
  78         * framebuffer code has at most two buffers pinned per crtc.
  79         *
  80         * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
  81         * bits with absolutely no headroom. So use 4 bits.
  82         */
  83#define I915_VMA_PIN_MASK 0xf
  84#define I915_VMA_PIN_OVERFLOW   BIT(5)
  85
  86        /** Flags and address space this VMA is bound to */
  87#define I915_VMA_GLOBAL_BIND    BIT(6)
  88#define I915_VMA_LOCAL_BIND     BIT(7)
  89#define I915_VMA_BIND_MASK (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND | I915_VMA_PIN_OVERFLOW)
  90
  91#define I915_VMA_GGTT           BIT(8)
  92#define I915_VMA_CAN_FENCE      BIT(9)
  93#define I915_VMA_CLOSED         BIT(10)
  94#define I915_VMA_USERFAULT_BIT  11
  95#define I915_VMA_USERFAULT      BIT(I915_VMA_USERFAULT_BIT)
  96#define I915_VMA_GGTT_WRITE     BIT(12)
  97
  98        unsigned int active_count;
  99        struct rb_root active;
 100        struct i915_gem_active last_active;
 101        struct i915_gem_active last_fence;
 102
 103        /**
 104         * Support different GGTT views into the same object.
 105         * This means there can be multiple VMA mappings per object and per VM.
 106         * i915_ggtt_view_type is used to distinguish between those entries.
 107         * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
 108         * assumed in GEM functions which take no ggtt view parameter.
 109         */
 110        struct i915_ggtt_view ggtt_view;
 111
 112        /** This object's place on the active/inactive lists */
 113        struct list_head vm_link;
 114
 115        struct list_head obj_link; /* Link in the object's VMA list */
 116        struct rb_node obj_node;
 117        struct hlist_node obj_hash;
 118
 119        /** This vma's place in the execbuf reservation list */
 120        struct list_head exec_link;
 121        struct list_head reloc_link;
 122
 123        /** This vma's place in the eviction list */
 124        struct list_head evict_link;
 125
 126        struct list_head closed_link;
 127
 128        /**
 129         * Used for performing relocations during execbuffer insertion.
 130         */
 131        unsigned int *exec_flags;
 132        struct hlist_node exec_node;
 133        u32 exec_handle;
 134};
 135
 136struct i915_vma *
 137i915_vma_instance(struct drm_i915_gem_object *obj,
 138                  struct i915_address_space *vm,
 139                  const struct i915_ggtt_view *view);
 140
 141void i915_vma_unpin_and_release(struct i915_vma **p_vma, unsigned int flags);
 142#define I915_VMA_RELEASE_MAP BIT(0)
 143
 144static inline bool i915_vma_is_active(struct i915_vma *vma)
 145{
 146        return vma->active_count;
 147}
 148
 149int __must_check i915_vma_move_to_active(struct i915_vma *vma,
 150                                         struct i915_request *rq,
 151                                         unsigned int flags);
 152
 153static inline bool i915_vma_is_ggtt(const struct i915_vma *vma)
 154{
 155        return vma->flags & I915_VMA_GGTT;
 156}
 157
 158static inline bool i915_vma_has_ggtt_write(const struct i915_vma *vma)
 159{
 160        return vma->flags & I915_VMA_GGTT_WRITE;
 161}
 162
 163static inline void i915_vma_set_ggtt_write(struct i915_vma *vma)
 164{
 165        GEM_BUG_ON(!i915_vma_is_ggtt(vma));
 166        vma->flags |= I915_VMA_GGTT_WRITE;
 167}
 168
 169static inline void i915_vma_unset_ggtt_write(struct i915_vma *vma)
 170{
 171        vma->flags &= ~I915_VMA_GGTT_WRITE;
 172}
 173
 174void i915_vma_flush_writes(struct i915_vma *vma);
 175
 176static inline bool i915_vma_is_map_and_fenceable(const struct i915_vma *vma)
 177{
 178        return vma->flags & I915_VMA_CAN_FENCE;
 179}
 180
 181static inline bool i915_vma_is_closed(const struct i915_vma *vma)
 182{
 183        return vma->flags & I915_VMA_CLOSED;
 184}
 185
 186static inline bool i915_vma_set_userfault(struct i915_vma *vma)
 187{
 188        GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
 189        return __test_and_set_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 190}
 191
 192static inline void i915_vma_unset_userfault(struct i915_vma *vma)
 193{
 194        return __clear_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 195}
 196
 197static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
 198{
 199        return test_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 200}
 201
 202static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
 203{
 204        GEM_BUG_ON(!i915_vma_is_ggtt(vma));
 205        GEM_BUG_ON(!vma->node.allocated);
 206        GEM_BUG_ON(upper_32_bits(vma->node.start));
 207        GEM_BUG_ON(upper_32_bits(vma->node.start + vma->node.size - 1));
 208        return lower_32_bits(vma->node.start);
 209}
 210
 211static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
 212{
 213        return i915_vm_to_ggtt(vma->vm)->pin_bias;
 214}
 215
 216static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
 217{
 218        i915_gem_object_get(vma->obj);
 219        return vma;
 220}
 221
 222static inline void i915_vma_put(struct i915_vma *vma)
 223{
 224        i915_gem_object_put(vma->obj);
 225}
 226
 227static __always_inline ptrdiff_t ptrdiff(const void *a, const void *b)
 228{
 229        return a - b;
 230}
 231
 232static inline long
 233i915_vma_compare(struct i915_vma *vma,
 234                 struct i915_address_space *vm,
 235                 const struct i915_ggtt_view *view)
 236{
 237        ptrdiff_t cmp;
 238
 239        GEM_BUG_ON(view && !i915_is_ggtt(vm));
 240
 241        cmp = ptrdiff(vma->vm, vm);
 242        if (cmp)
 243                return cmp;
 244
 245        BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL != 0);
 246        cmp = vma->ggtt_view.type;
 247        if (!view)
 248                return cmp;
 249
 250        cmp -= view->type;
 251        if (cmp)
 252                return cmp;
 253
 254        assert_i915_gem_gtt_types();
 255
 256        /* ggtt_view.type also encodes its size so that we both distinguish
 257         * different views using it as a "type" and also use a compact (no
 258         * accessing of uninitialised padding bytes) memcmp without storing
 259         * an extra parameter or adding more code.
 260         *
 261         * To ensure that the memcmp is valid for all branches of the union,
 262         * even though the code looks like it is just comparing one branch,
 263         * we assert above that all branches have the same address, and that
 264         * each branch has a unique type/size.
 265         */
 266        BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL >= I915_GGTT_VIEW_PARTIAL);
 267        BUILD_BUG_ON(I915_GGTT_VIEW_PARTIAL >= I915_GGTT_VIEW_ROTATED);
 268        BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
 269                     offsetof(typeof(*view), partial));
 270        return memcmp(&vma->ggtt_view.partial, &view->partial, view->type);
 271}
 272
 273int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level,
 274                  u32 flags);
 275bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long cache_level);
 276bool i915_vma_misplaced(const struct i915_vma *vma,
 277                        u64 size, u64 alignment, u64 flags);
 278void __i915_vma_set_map_and_fenceable(struct i915_vma *vma);
 279void i915_vma_revoke_mmap(struct i915_vma *vma);
 280int __must_check i915_vma_unbind(struct i915_vma *vma);
 281void i915_vma_unlink_ctx(struct i915_vma *vma);
 282void i915_vma_close(struct i915_vma *vma);
 283void i915_vma_reopen(struct i915_vma *vma);
 284void i915_vma_destroy(struct i915_vma *vma);
 285
 286int __i915_vma_do_pin(struct i915_vma *vma,
 287                      u64 size, u64 alignment, u64 flags);
 288static inline int __must_check
 289i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 290{
 291        BUILD_BUG_ON(PIN_MBZ != I915_VMA_PIN_OVERFLOW);
 292        BUILD_BUG_ON(PIN_GLOBAL != I915_VMA_GLOBAL_BIND);
 293        BUILD_BUG_ON(PIN_USER != I915_VMA_LOCAL_BIND);
 294
 295        /* Pin early to prevent the shrinker/eviction logic from destroying
 296         * our vma as we insert and bind.
 297         */
 298        if (likely(((++vma->flags ^ flags) & I915_VMA_BIND_MASK) == 0)) {
 299                GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 300                GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags));
 301                return 0;
 302        }
 303
 304        return __i915_vma_do_pin(vma, size, alignment, flags);
 305}
 306
 307static inline int i915_vma_pin_count(const struct i915_vma *vma)
 308{
 309        return vma->flags & I915_VMA_PIN_MASK;
 310}
 311
 312static inline bool i915_vma_is_pinned(const struct i915_vma *vma)
 313{
 314        return i915_vma_pin_count(vma);
 315}
 316
 317static inline void __i915_vma_pin(struct i915_vma *vma)
 318{
 319        vma->flags++;
 320        GEM_BUG_ON(vma->flags & I915_VMA_PIN_OVERFLOW);
 321}
 322
 323static inline void __i915_vma_unpin(struct i915_vma *vma)
 324{
 325        vma->flags--;
 326}
 327
 328static inline void i915_vma_unpin(struct i915_vma *vma)
 329{
 330        GEM_BUG_ON(!i915_vma_is_pinned(vma));
 331        GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 332        __i915_vma_unpin(vma);
 333}
 334
 335static inline bool i915_vma_is_bound(const struct i915_vma *vma,
 336                                     unsigned int where)
 337{
 338        return vma->flags & where;
 339}
 340
 341/**
 342 * i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
 343 * @vma: VMA to iomap
 344 *
 345 * The passed in VMA has to be pinned in the global GTT mappable region.
 346 * An extra pinning of the VMA is acquired for the return iomapping,
 347 * the caller must call i915_vma_unpin_iomap to relinquish the pinning
 348 * after the iomapping is no longer required.
 349 *
 350 * Callers must hold the struct_mutex.
 351 *
 352 * Returns a valid iomapped pointer or ERR_PTR.
 353 */
 354void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
 355#define IO_ERR_PTR(x) ((void __iomem *)ERR_PTR(x))
 356
 357/**
 358 * i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
 359 * @vma: VMA to unpin
 360 *
 361 * Unpins the previously iomapped VMA from i915_vma_pin_iomap().
 362 *
 363 * Callers must hold the struct_mutex. This function is only valid to be
 364 * called on a VMA previously iomapped by the caller with i915_vma_pin_iomap().
 365 */
 366void i915_vma_unpin_iomap(struct i915_vma *vma);
 367
 368static inline struct page *i915_vma_first_page(struct i915_vma *vma)
 369{
 370        GEM_BUG_ON(!vma->pages);
 371        return sg_page(vma->pages->sgl);
 372}
 373
 374/**
 375 * i915_vma_pin_fence - pin fencing state
 376 * @vma: vma to pin fencing for
 377 *
 378 * This pins the fencing state (whether tiled or untiled) to make sure the
 379 * vma (and its object) is ready to be used as a scanout target. Fencing
 380 * status must be synchronize first by calling i915_vma_get_fence():
 381 *
 382 * The resulting fence pin reference must be released again with
 383 * i915_vma_unpin_fence().
 384 *
 385 * Returns:
 386 *
 387 * True if the vma has a fence, false otherwise.
 388 */
 389int i915_vma_pin_fence(struct i915_vma *vma);
 390int __must_check i915_vma_put_fence(struct i915_vma *vma);
 391
 392static inline void __i915_vma_unpin_fence(struct i915_vma *vma)
 393{
 394        GEM_BUG_ON(vma->fence->pin_count <= 0);
 395        vma->fence->pin_count--;
 396}
 397
 398/**
 399 * i915_vma_unpin_fence - unpin fencing state
 400 * @vma: vma to unpin fencing for
 401 *
 402 * This releases the fence pin reference acquired through
 403 * i915_vma_pin_fence. It will handle both objects with and without an
 404 * attached fence correctly, callers do not need to distinguish this.
 405 */
 406static inline void
 407i915_vma_unpin_fence(struct i915_vma *vma)
 408{
 409        /* lockdep_assert_held(&vma->vm->i915->drm.struct_mutex); */
 410        if (vma->fence)
 411                __i915_vma_unpin_fence(vma);
 412}
 413
 414void i915_vma_parked(struct drm_i915_private *i915);
 415
 416#define for_each_until(cond) if (cond) break; else
 417
 418/**
 419 * for_each_ggtt_vma - Iterate over the GGTT VMA belonging to an object.
 420 * @V: the #i915_vma iterator
 421 * @OBJ: the #drm_i915_gem_object
 422 *
 423 * GGTT VMA are placed at the being of the object's vma_list, see
 424 * vma_create(), so we can stop our walk as soon as we see a ppgtt VMA,
 425 * or the list is empty ofc.
 426 */
 427#define for_each_ggtt_vma(V, OBJ) \
 428        list_for_each_entry(V, &(OBJ)->vma_list, obj_link)              \
 429                for_each_until(!i915_vma_is_ggtt(V))
 430
 431#endif
 432