drm/i915: Start passing around i915_vma from execbuffer
[deliverable/linux.git] / drivers / gpu / drm / i915 / i915_gem_gtt.h
1 /*
2 * Copyright © 2014 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 * Please try to maintain the following order within this file unless it makes
24 * sense to do otherwise. From top to bottom:
25 * 1. typedefs
26 * 2. #defines, and macros
27 * 3. structure definitions
28 * 4. function prototypes
29 *
30 * Within each section, please try to order by generation in ascending order,
31 * from top to bottom (ie. gen6 on the top, gen8 on the bottom).
32 */
33
34 #ifndef __I915_GEM_GTT_H__
35 #define __I915_GEM_GTT_H__
36
37 #include <linux/io-mapping.h>
38
39 #include "i915_gem_request.h"
40
41 struct drm_i915_file_private;
42
43 typedef uint32_t gen6_pte_t;
44 typedef uint64_t gen8_pte_t;
45 typedef uint64_t gen8_pde_t;
46 typedef uint64_t gen8_ppgtt_pdpe_t;
47 typedef uint64_t gen8_ppgtt_pml4e_t;
48
49 #define ggtt_total_entries(ggtt) ((ggtt)->base.total >> PAGE_SHIFT)
50
51 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
52 #define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
53 #define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
54 #define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
55 #define GEN6_PTE_CACHE_LLC (2 << 1)
56 #define GEN6_PTE_UNCACHED (1 << 1)
57 #define GEN6_PTE_VALID (1 << 0)
58
59 #define I915_PTES(pte_len) (PAGE_SIZE / (pte_len))
60 #define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
61 #define I915_PDES 512
62 #define I915_PDE_MASK (I915_PDES - 1)
63 #define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
64
65 #define GEN6_PTES I915_PTES(sizeof(gen6_pte_t))
66 #define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE)
67 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
68 #define GEN6_PDE_SHIFT 22
69 #define GEN6_PDE_VALID (1 << 0)
70
71 #define GEN7_PTE_CACHE_L3_LLC (3 << 1)
72
73 #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
74 #define BYT_PTE_WRITEABLE (1 << 1)
75
76 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
77 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
78 */
79 #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
80 (((bits) & 0x8) << (11 - 3)))
81 #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
82 #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
83 #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
84 #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
85 #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
86 #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
87 #define HSW_PTE_UNCACHED (0)
88 #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
89 #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
90
91 /* GEN8 legacy style address is defined as a 3 level page table:
92 * 31:30 | 29:21 | 20:12 | 11:0
93 * PDPE | PDE | PTE | offset
94 * The difference as compared to normal x86 3 level page table is the PDPEs are
95 * programmed via register.
96 *
97 * GEN8 48b legacy style address is defined as a 4 level page table:
98 * 47:39 | 38:30 | 29:21 | 20:12 | 11:0
99 * PML4E | PDPE | PDE | PTE | offset
100 */
101 #define GEN8_PML4ES_PER_PML4 512
102 #define GEN8_PML4E_SHIFT 39
103 #define GEN8_PML4E_MASK (GEN8_PML4ES_PER_PML4 - 1)
104 #define GEN8_PDPE_SHIFT 30
105 /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
106 * tables */
107 #define GEN8_PDPE_MASK 0x1ff
108 #define GEN8_PDE_SHIFT 21
109 #define GEN8_PDE_MASK 0x1ff
110 #define GEN8_PTE_SHIFT 12
111 #define GEN8_PTE_MASK 0x1ff
112 #define GEN8_LEGACY_PDPES 4
113 #define GEN8_PTES I915_PTES(sizeof(gen8_pte_t))
114
115 #define I915_PDPES_PER_PDP(dev) (USES_FULL_48BIT_PPGTT(dev) ?\
116 GEN8_PML4ES_PER_PML4 : GEN8_LEGACY_PDPES)
117
118 #define PPAT_UNCACHED_INDEX (_PAGE_PWT | _PAGE_PCD)
119 #define PPAT_CACHED_PDE_INDEX 0 /* WB LLC */
120 #define PPAT_CACHED_INDEX _PAGE_PAT /* WB LLCeLLC */
121 #define PPAT_DISPLAY_ELLC_INDEX _PAGE_PCD /* WT eLLC */
122
123 #define CHV_PPAT_SNOOP (1<<6)
124 #define GEN8_PPAT_AGE(x) (x<<4)
125 #define GEN8_PPAT_LLCeLLC (3<<2)
126 #define GEN8_PPAT_LLCELLC (2<<2)
127 #define GEN8_PPAT_LLC (1<<2)
128 #define GEN8_PPAT_WB (3<<0)
129 #define GEN8_PPAT_WT (2<<0)
130 #define GEN8_PPAT_WC (1<<0)
131 #define GEN8_PPAT_UC (0<<0)
132 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
133 #define GEN8_PPAT(i, x) ((uint64_t) (x) << ((i) * 8))
134
135 enum i915_ggtt_view_type {
136 I915_GGTT_VIEW_NORMAL = 0,
137 I915_GGTT_VIEW_ROTATED,
138 I915_GGTT_VIEW_PARTIAL,
139 };
140
141 struct intel_rotation_info {
142 unsigned int uv_offset;
143 uint32_t pixel_format;
144 unsigned int uv_start_page;
145 struct {
146 /* tiles */
147 unsigned int width, height;
148 } plane[2];
149 };
150
151 struct i915_ggtt_view {
152 enum i915_ggtt_view_type type;
153
154 union {
155 struct {
156 u64 offset;
157 unsigned int size;
158 } partial;
159 struct intel_rotation_info rotated;
160 } params;
161
162 struct sg_table *pages;
163 };
164
165 extern const struct i915_ggtt_view i915_ggtt_view_normal;
166 extern const struct i915_ggtt_view i915_ggtt_view_rotated;
167
168 enum i915_cache_level;
169
170 /**
171 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
172 * VMA's presence cannot be guaranteed before binding, or after unbinding the
173 * object into/from the address space.
174 *
175 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
176 * will always be <= an objects lifetime. So object refcounting should cover us.
177 */
178 struct i915_vma {
179 struct drm_mm_node node;
180 struct drm_i915_gem_object *obj;
181 struct i915_address_space *vm;
182 void __iomem *iomap;
183 u64 size;
184
185 unsigned int active;
186 struct i915_gem_active last_read[I915_NUM_ENGINES];
187
188 /** Flags and address space this VMA is bound to */
189 #define GLOBAL_BIND (1<<0)
190 #define LOCAL_BIND (1<<1)
191 unsigned int bound : 4;
192 bool is_ggtt : 1;
193 bool closed : 1;
194
195 /**
196 * Support different GGTT views into the same object.
197 * This means there can be multiple VMA mappings per object and per VM.
198 * i915_ggtt_view_type is used to distinguish between those entries.
199 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
200 * assumed in GEM functions which take no ggtt view parameter.
201 */
202 struct i915_ggtt_view ggtt_view;
203
204 /** This object's place on the active/inactive lists */
205 struct list_head vm_link;
206
207 struct list_head obj_link; /* Link in the object's VMA list */
208
209 /** This vma's place in the batchbuffer or on the eviction list */
210 struct list_head exec_list;
211
212 /**
213 * Used for performing relocations during execbuffer insertion.
214 */
215 struct hlist_node exec_node;
216 unsigned long exec_handle;
217 struct drm_i915_gem_exec_object2 *exec_entry;
218
219 /**
220 * How many users have pinned this object in GTT space. The following
221 * users can each hold at most one reference: pwrite/pread, execbuffer
222 * (objects are not allowed multiple times for the same batchbuffer),
223 * and the framebuffer code. When switching/pageflipping, the
224 * framebuffer code has at most two buffers pinned per crtc.
225 *
226 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
227 * bits with absolutely no headroom. So use 4 bits. */
228 unsigned int pin_count:4;
229 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
230 };
231
232 static inline unsigned int i915_vma_get_active(const struct i915_vma *vma)
233 {
234 return vma->active;
235 }
236
237 static inline bool i915_vma_is_active(const struct i915_vma *vma)
238 {
239 return i915_vma_get_active(vma);
240 }
241
242 static inline void i915_vma_set_active(struct i915_vma *vma,
243 unsigned int engine)
244 {
245 vma->active |= BIT(engine);
246 }
247
248 static inline void i915_vma_clear_active(struct i915_vma *vma,
249 unsigned int engine)
250 {
251 vma->active &= ~BIT(engine);
252 }
253
254 static inline bool i915_vma_has_active_engine(const struct i915_vma *vma,
255 unsigned int engine)
256 {
257 return vma->active & BIT(engine);
258 }
259
260 struct i915_page_dma {
261 struct page *page;
262 union {
263 dma_addr_t daddr;
264
265 /* For gen6/gen7 only. This is the offset in the GGTT
266 * where the page directory entries for PPGTT begin
267 */
268 uint32_t ggtt_offset;
269 };
270 };
271
272 #define px_base(px) (&(px)->base)
273 #define px_page(px) (px_base(px)->page)
274 #define px_dma(px) (px_base(px)->daddr)
275
276 struct i915_page_scratch {
277 struct i915_page_dma base;
278 };
279
280 struct i915_page_table {
281 struct i915_page_dma base;
282
283 unsigned long *used_ptes;
284 };
285
286 struct i915_page_directory {
287 struct i915_page_dma base;
288
289 unsigned long *used_pdes;
290 struct i915_page_table *page_table[I915_PDES]; /* PDEs */
291 };
292
293 struct i915_page_directory_pointer {
294 struct i915_page_dma base;
295
296 unsigned long *used_pdpes;
297 struct i915_page_directory **page_directory;
298 };
299
300 struct i915_pml4 {
301 struct i915_page_dma base;
302
303 DECLARE_BITMAP(used_pml4es, GEN8_PML4ES_PER_PML4);
304 struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
305 };
306
307 struct i915_address_space {
308 struct drm_mm mm;
309 struct drm_device *dev;
310 /* Every address space belongs to a struct file - except for the global
311 * GTT that is owned by the driver (and so @file is set to NULL). In
312 * principle, no information should leak from one context to another
313 * (or between files/processes etc) unless explicitly shared by the
314 * owner. Tracking the owner is important in order to free up per-file
315 * objects along with the file, to aide resource tracking, and to
316 * assign blame.
317 */
318 struct drm_i915_file_private *file;
319 struct list_head global_link;
320 u64 start; /* Start offset always 0 for dri2 */
321 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
322
323 bool closed;
324
325 struct i915_page_scratch *scratch_page;
326 struct i915_page_table *scratch_pt;
327 struct i915_page_directory *scratch_pd;
328 struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
329
330 /**
331 * List of objects currently involved in rendering.
332 *
333 * Includes buffers having the contents of their GPU caches
334 * flushed, not necessarily primitives. last_read_req
335 * represents when the rendering involved will be completed.
336 *
337 * A reference is held on the buffer while on this list.
338 */
339 struct list_head active_list;
340
341 /**
342 * LRU list of objects which are not in the ringbuffer and
343 * are ready to unbind, but are still in the GTT.
344 *
345 * last_read_req is NULL while an object is in this list.
346 *
347 * A reference is not held on the buffer while on this list,
348 * as merely being GTT-bound shouldn't prevent its being
349 * freed, and we'll pull it off the list in the free path.
350 */
351 struct list_head inactive_list;
352
353 /**
354 * List of vma that have been unbound.
355 *
356 * A reference is not held on the buffer while on this list.
357 */
358 struct list_head unbound_list;
359
360 /* FIXME: Need a more generic return type */
361 gen6_pte_t (*pte_encode)(dma_addr_t addr,
362 enum i915_cache_level level,
363 bool valid, u32 flags); /* Create a valid PTE */
364 /* flags for pte_encode */
365 #define PTE_READ_ONLY (1<<0)
366 int (*allocate_va_range)(struct i915_address_space *vm,
367 uint64_t start,
368 uint64_t length);
369 void (*clear_range)(struct i915_address_space *vm,
370 uint64_t start,
371 uint64_t length,
372 bool use_scratch);
373 void (*insert_page)(struct i915_address_space *vm,
374 dma_addr_t addr,
375 uint64_t offset,
376 enum i915_cache_level cache_level,
377 u32 flags);
378 void (*insert_entries)(struct i915_address_space *vm,
379 struct sg_table *st,
380 uint64_t start,
381 enum i915_cache_level cache_level, u32 flags);
382 void (*cleanup)(struct i915_address_space *vm);
383 /** Unmap an object from an address space. This usually consists of
384 * setting the valid PTE entries to a reserved scratch page. */
385 void (*unbind_vma)(struct i915_vma *vma);
386 /* Map an object into an address space with the given cache flags. */
387 int (*bind_vma)(struct i915_vma *vma,
388 enum i915_cache_level cache_level,
389 u32 flags);
390 };
391
392 #define i915_is_ggtt(V) (!(V)->file)
393
394 /* The Graphics Translation Table is the way in which GEN hardware translates a
395 * Graphics Virtual Address into a Physical Address. In addition to the normal
396 * collateral associated with any va->pa translations GEN hardware also has a
397 * portion of the GTT which can be mapped by the CPU and remain both coherent
398 * and correct (in cases like swizzling). That region is referred to as GMADR in
399 * the spec.
400 */
401 struct i915_ggtt {
402 struct i915_address_space base;
403
404 size_t stolen_size; /* Total size of stolen memory */
405 size_t stolen_usable_size; /* Total size minus BIOS reserved */
406 size_t stolen_reserved_base;
407 size_t stolen_reserved_size;
408 u64 mappable_end; /* End offset that we can CPU map */
409 struct io_mapping *mappable; /* Mapping to our CPU mappable region */
410 phys_addr_t mappable_base; /* PA of our GMADR */
411
412 /** "Graphics Stolen Memory" holds the global PTEs */
413 void __iomem *gsm;
414
415 bool do_idle_maps;
416
417 int mtrr;
418 };
419
420 struct i915_hw_ppgtt {
421 struct i915_address_space base;
422 struct kref ref;
423 struct drm_mm_node node;
424 unsigned long pd_dirty_rings;
425 union {
426 struct i915_pml4 pml4; /* GEN8+ & 48b PPGTT */
427 struct i915_page_directory_pointer pdp; /* GEN8+ */
428 struct i915_page_directory pd; /* GEN6-7 */
429 };
430
431 gen6_pte_t __iomem *pd_addr;
432
433 int (*enable)(struct i915_hw_ppgtt *ppgtt);
434 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
435 struct drm_i915_gem_request *req);
436 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
437 };
438
439 /*
440 * gen6_for_each_pde() iterates over every pde from start until start+length.
441 * If start and start+length are not perfectly divisible, the macro will round
442 * down and up as needed. Start=0 and length=2G effectively iterates over
443 * every PDE in the system. The macro modifies ALL its parameters except 'pd',
444 * so each of the other parameters should preferably be a simple variable, or
445 * at most an lvalue with no side-effects!
446 */
447 #define gen6_for_each_pde(pt, pd, start, length, iter) \
448 for (iter = gen6_pde_index(start); \
449 length > 0 && iter < I915_PDES && \
450 (pt = (pd)->page_table[iter], true); \
451 ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT); \
452 temp = min(temp - start, length); \
453 start += temp, length -= temp; }), ++iter)
454
455 #define gen6_for_all_pdes(pt, pd, iter) \
456 for (iter = 0; \
457 iter < I915_PDES && \
458 (pt = (pd)->page_table[iter], true); \
459 ++iter)
460
461 static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
462 {
463 const uint32_t mask = NUM_PTE(pde_shift) - 1;
464
465 return (address >> PAGE_SHIFT) & mask;
466 }
467
468 /* Helper to counts the number of PTEs within the given length. This count
469 * does not cross a page table boundary, so the max value would be
470 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
471 */
472 static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
473 uint32_t pde_shift)
474 {
475 const uint64_t mask = ~((1ULL << pde_shift) - 1);
476 uint64_t end;
477
478 WARN_ON(length == 0);
479 WARN_ON(offset_in_page(addr|length));
480
481 end = addr + length;
482
483 if ((addr & mask) != (end & mask))
484 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
485
486 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
487 }
488
489 static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
490 {
491 return (addr >> shift) & I915_PDE_MASK;
492 }
493
494 static inline uint32_t gen6_pte_index(uint32_t addr)
495 {
496 return i915_pte_index(addr, GEN6_PDE_SHIFT);
497 }
498
499 static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
500 {
501 return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
502 }
503
504 static inline uint32_t gen6_pde_index(uint32_t addr)
505 {
506 return i915_pde_index(addr, GEN6_PDE_SHIFT);
507 }
508
509 /* Equivalent to the gen6 version, For each pde iterates over every pde
510 * between from start until start + length. On gen8+ it simply iterates
511 * over every page directory entry in a page directory.
512 */
513 #define gen8_for_each_pde(pt, pd, start, length, iter) \
514 for (iter = gen8_pde_index(start); \
515 length > 0 && iter < I915_PDES && \
516 (pt = (pd)->page_table[iter], true); \
517 ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT); \
518 temp = min(temp - start, length); \
519 start += temp, length -= temp; }), ++iter)
520
521 #define gen8_for_each_pdpe(pd, pdp, start, length, iter) \
522 for (iter = gen8_pdpe_index(start); \
523 length > 0 && iter < I915_PDPES_PER_PDP(dev) && \
524 (pd = (pdp)->page_directory[iter], true); \
525 ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT); \
526 temp = min(temp - start, length); \
527 start += temp, length -= temp; }), ++iter)
528
529 #define gen8_for_each_pml4e(pdp, pml4, start, length, iter) \
530 for (iter = gen8_pml4e_index(start); \
531 length > 0 && iter < GEN8_PML4ES_PER_PML4 && \
532 (pdp = (pml4)->pdps[iter], true); \
533 ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT); \
534 temp = min(temp - start, length); \
535 start += temp, length -= temp; }), ++iter)
536
537 static inline uint32_t gen8_pte_index(uint64_t address)
538 {
539 return i915_pte_index(address, GEN8_PDE_SHIFT);
540 }
541
542 static inline uint32_t gen8_pde_index(uint64_t address)
543 {
544 return i915_pde_index(address, GEN8_PDE_SHIFT);
545 }
546
547 static inline uint32_t gen8_pdpe_index(uint64_t address)
548 {
549 return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
550 }
551
552 static inline uint32_t gen8_pml4e_index(uint64_t address)
553 {
554 return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
555 }
556
557 static inline size_t gen8_pte_count(uint64_t address, uint64_t length)
558 {
559 return i915_pte_count(address, length, GEN8_PDE_SHIFT);
560 }
561
562 static inline dma_addr_t
563 i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
564 {
565 return test_bit(n, ppgtt->pdp.used_pdpes) ?
566 px_dma(ppgtt->pdp.page_directory[n]) :
567 px_dma(ppgtt->base.scratch_pd);
568 }
569
570 int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
571 int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
572 int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
573 int i915_gem_init_ggtt(struct drm_i915_private *dev_priv);
574 void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv);
575
576 int i915_ppgtt_init_hw(struct drm_device *dev);
577 void i915_ppgtt_release(struct kref *kref);
578 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv,
579 struct drm_i915_file_private *fpriv);
580 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
581 {
582 if (ppgtt)
583 kref_get(&ppgtt->ref);
584 }
585 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
586 {
587 if (ppgtt)
588 kref_put(&ppgtt->ref, i915_ppgtt_release);
589 }
590
591 void i915_check_and_clear_faults(struct drm_i915_private *dev_priv);
592 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
593 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
594
595 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
596 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
597
598 static inline bool
599 i915_ggtt_view_equal(const struct i915_ggtt_view *a,
600 const struct i915_ggtt_view *b)
601 {
602 if (WARN_ON(!a || !b))
603 return false;
604
605 if (a->type != b->type)
606 return false;
607 if (a->type != I915_GGTT_VIEW_NORMAL)
608 return !memcmp(&a->params, &b->params, sizeof(a->params));
609 return true;
610 }
611
612 int __must_check
613 i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags);
614 /* Flags used by pin/bind&friends. */
615 #define PIN_MAPPABLE BIT(0)
616 #define PIN_NONBLOCK BIT(1)
617 #define PIN_GLOBAL BIT(2)
618 #define PIN_OFFSET_BIAS BIT(3)
619 #define PIN_USER BIT(4)
620 #define PIN_UPDATE BIT(5)
621 #define PIN_ZONE_4G BIT(6)
622 #define PIN_HIGH BIT(7)
623 #define PIN_OFFSET_FIXED BIT(8)
624 #define PIN_OFFSET_MASK (~4095)
625
626 static inline int i915_vma_pin_count(const struct i915_vma *vma)
627 {
628 return vma->pin_count;
629 }
630
631 static inline bool i915_vma_is_pinned(const struct i915_vma *vma)
632 {
633 return i915_vma_pin_count(vma);
634 }
635
636 static inline void __i915_vma_pin(struct i915_vma *vma)
637 {
638 vma->pin_count++;
639 GEM_BUG_ON(!i915_vma_is_pinned(vma));
640 }
641
642 static inline void __i915_vma_unpin(struct i915_vma *vma)
643 {
644 GEM_BUG_ON(!i915_vma_is_pinned(vma));
645 vma->pin_count--;
646 }
647
648 static inline void i915_vma_unpin(struct i915_vma *vma)
649 {
650 GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
651 __i915_vma_unpin(vma);
652 }
653
654 /**
655 * i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
656 * @vma: VMA to iomap
657 *
658 * The passed in VMA has to be pinned in the global GTT mappable region.
659 * An extra pinning of the VMA is acquired for the return iomapping,
660 * the caller must call i915_vma_unpin_iomap to relinquish the pinning
661 * after the iomapping is no longer required.
662 *
663 * Callers must hold the struct_mutex.
664 *
665 * Returns a valid iomapped pointer or ERR_PTR.
666 */
667 void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
668 #define IO_ERR_PTR(x) ((void __iomem *)ERR_PTR(x))
669
670 /**
671 * i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
672 * @vma: VMA to unpin
673 *
674 * Unpins the previously iomapped VMA from i915_vma_pin_iomap().
675 *
676 * Callers must hold the struct_mutex. This function is only valid to be
677 * called on a VMA previously iomapped by the caller with i915_vma_pin_iomap().
678 */
679 static inline void i915_vma_unpin_iomap(struct i915_vma *vma)
680 {
681 lockdep_assert_held(&vma->vm->dev->struct_mutex);
682 GEM_BUG_ON(vma->iomap == NULL);
683 i915_vma_unpin(vma);
684 }
685
686 #endif
This page took 0.045737 seconds and 6 git commands to generate.