drivers/gpu/drm/i915/i915_vma.h

   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_active.h"
  38 #include "i915_request.h"
  39
  40 enum i915_cache_level;
  41
  42 /**
  43  * A VMA represents a GEM BO that is bound into an address space. Therefore, a
  44  * VMA's presence cannot be guaranteed before binding, or after unbinding the
  45  * object into/from the address space.
  46  *
  47  * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
  48  * will always be <= an objects lifetime. So object refcounting should cover us.
  49  */
  50 struct i915_vma {
  51         struct drm_mm_node node;
  52         struct drm_i915_gem_object *obj;
  53         struct i915_address_space *vm;
  54         const struct i915_vma_ops *ops;
  55         struct drm_i915_fence_reg *fence;
  56         struct reservation_object *resv; /** Alias of obj->resv */
  57         struct sg_table *pages;
  58         void __iomem *iomap;
  59         void *private; /* owned by creator */
  60         u64 size;
  61         u64 display_alignment;
  62         struct i915_page_sizes page_sizes;
  63
  64         u32 fence_size;
  65         u32 fence_alignment;
  66
  67         /**
  68          * Count of the number of times this vma has been opened by different
  69          * handles (but same file) for execbuf, i.e. the number of aliases
  70          * that exist in the ctx->handle_vmas LUT for this vma.
  71          */
  72         unsigned int open_count;
  73         unsigned long flags;
  74         /**
  75          * How many users have pinned this object in GTT space.
  76          *
  77          * This is a tightly bound, fairly small number of users, so we
  78          * stuff inside the flags field so that we can both check for overflow
  79          * and detect a no-op i915_vma_pin() in a single check, while also
  80          * pinning the vma.
  81          *
  82          * The worst case display setup would have the same vma pinned for
  83          * use on each plane on each crtc, while also building the next atomic
  84          * state and holding a pin for the length of the cleanup queue. In the
  85          * future, the flip queue may be increased from 1.
  86          * Estimated worst case: 3 [qlen] * 4 [max crtcs] * 7 [max planes] = 84
  87          *
  88          * For GEM, the number of concurrent users for pwrite/pread is
  89          * unbounded. For execbuffer, it is currently one but will in future
  90          * be extended to allow multiple clients to pin vma concurrently.
  91          *
  92          * We also use suballocated pages, with each suballocation claiming
  93          * its own pin on the shared vma. At present, this is limited to
  94          * exclusive cachelines of a single page, so a maximum of 64 possible
  95          * users.
  96          */
  97 #define I915_VMA_PIN_MASK 0xff
  98 #define I915_VMA_PIN_OVERFLOW   BIT(8)
  99
 100         /** Flags and address space this VMA is bound to */
 101 #define I915_VMA_GLOBAL_BIND    BIT(9)
 102 #define I915_VMA_LOCAL_BIND     BIT(10)
 103 #define I915_VMA_BIND_MASK (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND | I915_VMA_PIN_OVERFLOW)
 104
 105 #define I915_VMA_GGTT           BIT(11)
 106 #define I915_VMA_CAN_FENCE      BIT(12)
 107 #define I915_VMA_CLOSED         BIT(13)
 108 #define I915_VMA_USERFAULT_BIT  14
 109 #define I915_VMA_USERFAULT      BIT(I915_VMA_USERFAULT_BIT)
 110 #define I915_VMA_GGTT_WRITE     BIT(15)
 111
 112         struct i915_active active;
 113         struct i915_active_request last_fence;
 114
 115         /**
 116          * Support different GGTT views into the same object.
 117          * This means there can be multiple VMA mappings per object and per VM.
 118          * i915_ggtt_view_type is used to distinguish between those entries.
 119          * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
 120          * assumed in GEM functions which take no ggtt view parameter.
 121          */
 122         struct i915_ggtt_view ggtt_view;
 123
 124         /** This object's place on the active/inactive lists */
 125         struct list_head vm_link;
 126
 127         struct list_head obj_link; /* Link in the object's VMA list */
 128         struct rb_node obj_node;
 129         struct hlist_node obj_hash;
 130
 131         /** This vma's place in the execbuf reservation list */
 132         struct list_head exec_link;
 133         struct list_head reloc_link;
 134
 135         /** This vma's place in the eviction list */
 136         struct list_head evict_link;
 137
 138         struct list_head closed_link;
 139
 140         /**
 141          * Used for performing relocations during execbuffer insertion.
 142          */
 143         unsigned int *exec_flags;
 144         struct hlist_node exec_node;
 145         u32 exec_handle;
 146 };
 147
 148 struct i915_vma *
 149 i915_vma_instance(struct drm_i915_gem_object *obj,
 150                   struct i915_address_space *vm,
 151                   const struct i915_ggtt_view *view);
 152
 153 void i915_vma_unpin_and_release(struct i915_vma **p_vma, unsigned int flags);
 154 #define I915_VMA_RELEASE_MAP BIT(0)
 155
 156 static inline bool i915_vma_is_active(const struct i915_vma *vma)
 157 {
 158         return !i915_active_is_idle(&vma->active);
 159 }
 160
 161 int __must_check i915_vma_move_to_active(struct i915_vma *vma,
 162                                          struct i915_request *rq,
 163                                          unsigned int flags);
 164
 165 static inline bool i915_vma_is_ggtt(const struct i915_vma *vma)
 166 {
 167         return vma->flags & I915_VMA_GGTT;
 168 }
 169
 170 static inline bool i915_vma_has_ggtt_write(const struct i915_vma *vma)
 171 {
 172         return vma->flags & I915_VMA_GGTT_WRITE;
 173 }
 174
 175 static inline void i915_vma_set_ggtt_write(struct i915_vma *vma)
 176 {
 177         GEM_BUG_ON(!i915_vma_is_ggtt(vma));
 178         vma->flags |= I915_VMA_GGTT_WRITE;
 179 }
 180
 181 static inline void i915_vma_unset_ggtt_write(struct i915_vma *vma)
 182 {
 183         vma->flags &= ~I915_VMA_GGTT_WRITE;
 184 }
 185
 186 void i915_vma_flush_writes(struct i915_vma *vma);
 187
 188 static inline bool i915_vma_is_map_and_fenceable(const struct i915_vma *vma)
 189 {
 190         return vma->flags & I915_VMA_CAN_FENCE;
 191 }
 192
 193 static inline bool i915_vma_is_closed(const struct i915_vma *vma)
 194 {
 195         return vma->flags & I915_VMA_CLOSED;
 196 }
 197
 198 static inline bool i915_vma_set_userfault(struct i915_vma *vma)
 199 {
 200         GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
 201         return __test_and_set_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 202 }
 203
 204 static inline void i915_vma_unset_userfault(struct i915_vma *vma)
 205 {
 206         return __clear_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 207 }
 208
 209 static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
 210 {
 211         return test_bit(I915_VMA_USERFAULT_BIT, &vma->flags);
 212 }
 213
 214 static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
 215 {
 216         GEM_BUG_ON(!i915_vma_is_ggtt(vma));
 217         GEM_BUG_ON(!vma->node.allocated);
 218         GEM_BUG_ON(upper_32_bits(vma->node.start));
 219         GEM_BUG_ON(upper_32_bits(vma->node.start + vma->node.size - 1));
 220         return lower_32_bits(vma->node.start);
 221 }
 222
 223 static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
 224 {
 225         return i915_vm_to_ggtt(vma->vm)->pin_bias;
 226 }
 227
 228 static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
 229 {
 230         i915_gem_object_get(vma->obj);
 231         return vma;
 232 }
 233
 234 static inline void i915_vma_put(struct i915_vma *vma)
 235 {
 236         i915_gem_object_put(vma->obj);
 237 }
 238
 239 static __always_inline ptrdiff_t ptrdiff(const void *a, const void *b)
 240 {
 241         return a - b;
 242 }
 243
 244 static inline long
 245 i915_vma_compare(struct i915_vma *vma,
 246                  struct i915_address_space *vm,
 247                  const struct i915_ggtt_view *view)
 248 {
 249         ptrdiff_t cmp;
 250
 251         GEM_BUG_ON(view && !i915_is_ggtt(vm));
 252
 253         cmp = ptrdiff(vma->vm, vm);
 254         if (cmp)
 255                 return cmp;
 256
 257         BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL != 0);
 258         cmp = vma->ggtt_view.type;
 259         if (!view)
 260                 return cmp;
 261
 262         cmp -= view->type;
 263         if (cmp)
 264                 return cmp;
 265
 266         assert_i915_gem_gtt_types();
 267
 268         /* ggtt_view.type also encodes its size so that we both distinguish
 269          * different views using it as a "type" and also use a compact (no
 270          * accessing of uninitialised padding bytes) memcmp without storing
 271          * an extra parameter or adding more code.
 272          *
 273          * To ensure that the memcmp is valid for all branches of the union,
 274          * even though the code looks like it is just comparing one branch,
 275          * we assert above that all branches have the same address, and that
 276          * each branch has a unique type/size.
 277          */
 278         BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL >= I915_GGTT_VIEW_PARTIAL);
 279         BUILD_BUG_ON(I915_GGTT_VIEW_PARTIAL >= I915_GGTT_VIEW_ROTATED);
 280         BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
 281                      offsetof(typeof(*view), partial));
 282         return memcmp(&vma->ggtt_view.partial, &view->partial, view->type);
 283 }
 284
 285 int i915_vma_bind(struct i915_vma *vma, enum i915_cache_level cache_level,
 286                   u32 flags);
 287 bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long cache_level);
 288 bool i915_vma_misplaced(const struct i915_vma *vma,
 289                         u64 size, u64 alignment, u64 flags);
 290 void __i915_vma_set_map_and_fenceable(struct i915_vma *vma);
 291 void i915_vma_revoke_mmap(struct i915_vma *vma);
 292 int __must_check i915_vma_unbind(struct i915_vma *vma);
 293 void i915_vma_unlink_ctx(struct i915_vma *vma);
 294 void i915_vma_close(struct i915_vma *vma);
 295 void i915_vma_reopen(struct i915_vma *vma);
 296 void i915_vma_destroy(struct i915_vma *vma);
 297
 298 int __i915_vma_do_pin(struct i915_vma *vma,
 299                       u64 size, u64 alignment, u64 flags);
 300 static inline int __must_check
 301 i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
 302 {
 303         BUILD_BUG_ON(PIN_MBZ != I915_VMA_PIN_OVERFLOW);
 304         BUILD_BUG_ON(PIN_GLOBAL != I915_VMA_GLOBAL_BIND);
 305         BUILD_BUG_ON(PIN_USER != I915_VMA_LOCAL_BIND);
 306
 307         /* Pin early to prevent the shrinker/eviction logic from destroying
 308          * our vma as we insert and bind.
 309          */
 310         if (likely(((++vma->flags ^ flags) & I915_VMA_BIND_MASK) == 0)) {
 311                 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 312                 GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags));
 313                 return 0;
 314         }
 315
 316         return __i915_vma_do_pin(vma, size, alignment, flags);
 317 }
 318
 319 static inline int i915_vma_pin_count(const struct i915_vma *vma)
 320 {
 321         return vma->flags & I915_VMA_PIN_MASK;
 322 }
 323
 324 static inline bool i915_vma_is_pinned(const struct i915_vma *vma)
 325 {
 326         return i915_vma_pin_count(vma);
 327 }
 328
 329 static inline void __i915_vma_pin(struct i915_vma *vma)
 330 {
 331         vma->flags++;
 332         GEM_BUG_ON(vma->flags & I915_VMA_PIN_OVERFLOW);
 333 }
 334
 335 static inline void __i915_vma_unpin(struct i915_vma *vma)
 336 {
 337         vma->flags--;
 338 }
 339
 340 static inline void i915_vma_unpin(struct i915_vma *vma)
 341 {
 342         GEM_BUG_ON(!i915_vma_is_pinned(vma));
 343         GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 344         __i915_vma_unpin(vma);
 345 }
 346
 347 static inline bool i915_vma_is_bound(const struct i915_vma *vma,
 348                                      unsigned int where)
 349 {
 350         return vma->flags & where;
 351 }
 352
 353 /**
 354  * i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
 355  * @vma: VMA to iomap
 356  *
 357  * The passed in VMA has to be pinned in the global GTT mappable region.
 358  * An extra pinning of the VMA is acquired for the return iomapping,
 359  * the caller must call i915_vma_unpin_iomap to relinquish the pinning
 360  * after the iomapping is no longer required.
 361  *
 362  * Callers must hold the struct_mutex.
 363  *
 364  * Returns a valid iomapped pointer or ERR_PTR.
 365  */
 366 void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
 367 #define IO_ERR_PTR(x) ((void __iomem *)ERR_PTR(x))
 368
 369 /**
 370  * i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
 371  * @vma: VMA to unpin
 372  *
 373  * Unpins the previously iomapped VMA from i915_vma_pin_iomap().
 374  *
 375  * Callers must hold the struct_mutex. This function is only valid to be
 376  * called on a VMA previously iomapped by the caller with i915_vma_pin_iomap().
 377  */
 378 void i915_vma_unpin_iomap(struct i915_vma *vma);
 379
 380 static inline struct page *i915_vma_first_page(struct i915_vma *vma)
 381 {
 382         GEM_BUG_ON(!vma->pages);
 383         return sg_page(vma->pages->sgl);
 384 }
 385
 386 /**
 387  * i915_vma_pin_fence - pin fencing state
 388  * @vma: vma to pin fencing for
 389  *
 390  * This pins the fencing state (whether tiled or untiled) to make sure the
 391  * vma (and its object) is ready to be used as a scanout target. Fencing
 392  * status must be synchronize first by calling i915_vma_get_fence():
 393  *
 394  * The resulting fence pin reference must be released again with
 395  * i915_vma_unpin_fence().
 396  *
 397  * Returns:
 398  *
 399  * True if the vma has a fence, false otherwise.
 400  */
 401 int i915_vma_pin_fence(struct i915_vma *vma);
 402 int __must_check i915_vma_put_fence(struct i915_vma *vma);
 403
 404 static inline void __i915_vma_unpin_fence(struct i915_vma *vma)
 405 {
 406         GEM_BUG_ON(vma->fence->pin_count <= 0);
 407         vma->fence->pin_count--;
 408 }
 409
 410 /**
 411  * i915_vma_unpin_fence - unpin fencing state
 412  * @vma: vma to unpin fencing for
 413  *
 414  * This releases the fence pin reference acquired through
 415  * i915_vma_pin_fence. It will handle both objects with and without an
 416  * attached fence correctly, callers do not need to distinguish this.
 417  */
 418 static inline void
 419 i915_vma_unpin_fence(struct i915_vma *vma)
 420 {
 421         /* lockdep_assert_held(&vma->vm->i915->drm.struct_mutex); */
 422         if (vma->fence)
 423                 __i915_vma_unpin_fence(vma);
 424 }
 425
 426 void i915_vma_parked(struct drm_i915_private *i915);
 427
 428 #define for_each_until(cond) if (cond) break; else
 429
 430 /**
 431  * for_each_ggtt_vma - Iterate over the GGTT VMA belonging to an object.
 432  * @V: the #i915_vma iterator
 433  * @OBJ: the #drm_i915_gem_object
 434  *
 435  * GGTT VMA are placed at the being of the object's vma_list, see
 436  * vma_create(), so we can stop our walk as soon as we see a ppgtt VMA,
 437  * or the list is empty ofc.
 438  */
 439 #define for_each_ggtt_vma(V, OBJ) \
 440         list_for_each_entry(V, &(OBJ)->vma.list, obj_link)              \
 441                 for_each_until(!i915_vma_is_ggtt(V))
 442
 443 struct i915_vma *i915_vma_alloc(void);
 444 void i915_vma_free(struct i915_vma *vma);
 445
 446 #endif