xref: /dragonfly/sys/dev/drm/i915/i915_gem_gtt.h (revision 99d38c70)
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 struct drm_i915_file_private;
40 
41 typedef uint32_t gen6_pte_t;
42 typedef uint64_t gen8_pte_t;
43 typedef uint64_t gen8_pde_t;
44 typedef uint64_t gen8_ppgtt_pdpe_t;
45 typedef uint64_t gen8_ppgtt_pml4e_t;
46 
47 #define ggtt_total_entries(ggtt) ((ggtt)->base.total >> PAGE_SHIFT)
48 
49 #define I915_GTT_PAGE_SIZE_4K	(1ULL << 12)
50 #define I915_GTT_PAGE_MASK	-I915_GTT_PAGE_SIZE_4K
51 
52 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
53 #define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
54 #define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
55 #define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
56 #define GEN6_PTE_CACHE_LLC		(2 << 1)
57 #define GEN6_PTE_UNCACHED		(1 << 1)
58 #define GEN6_PTE_VALID			(1 << 0)
59 
60 #define I915_PTES(pte_len)		(PAGE_SIZE / (pte_len))
61 #define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
62 #define I915_PDES			512
63 #define I915_PDE_MASK			(I915_PDES - 1)
64 #define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
65 
66 #define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
67 #define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
68 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
69 #define GEN6_PDE_SHIFT			22
70 #define GEN6_PDE_VALID			(1 << 0)
71 
72 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
73 
74 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
75 #define BYT_PTE_WRITEABLE		(1 << 1)
76 
77 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
78  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
79  */
80 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
81 					 (((bits) & 0x8) << (11 - 3)))
82 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
83 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
84 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
85 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
86 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
87 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
88 #define HSW_PTE_UNCACHED		(0)
89 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
90 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
91 
92 /* GEN8 legacy style address is defined as a 3 level page table:
93  * 31:30 | 29:21 | 20:12 |  11:0
94  * PDPE  |  PDE  |  PTE  | offset
95  * The difference as compared to normal x86 3 level page table is the PDPEs are
96  * programmed via register.
97  *
98  * GEN8 48b legacy style address is defined as a 4 level page table:
99  * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
100  * PML4E | PDPE  |  PDE  |  PTE  | offset
101  */
102 #define GEN8_PML4ES_PER_PML4		512
103 #define GEN8_PML4E_SHIFT		39
104 #define GEN8_PML4E_MASK			(GEN8_PML4ES_PER_PML4 - 1)
105 #define GEN8_PDPE_SHIFT			30
106 /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
107  * tables */
108 #define GEN8_PDPE_MASK			0x1ff
109 #define GEN8_PDE_SHIFT			21
110 #define GEN8_PDE_MASK			0x1ff
111 #define GEN8_PTE_SHIFT			12
112 #define GEN8_PTE_MASK			0x1ff
113 #define GEN8_LEGACY_PDPES		4
114 #define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
115 
116 #define I915_PDPES_PER_PDP(dev) (USES_FULL_48BIT_PPGTT(dev) ?\
117 				 GEN8_PML4ES_PER_PML4 : GEN8_LEGACY_PDPES)
118 
119 #define PPAT_UNCACHED_INDEX		(_PAGE_PWT | _PAGE_PCD)
120 #define PPAT_CACHED_PDE_INDEX		0 /* WB LLC */
121 #define PPAT_CACHED_INDEX		_PAGE_PAT /* WB LLCeLLC */
122 #define PPAT_DISPLAY_ELLC_INDEX		_PAGE_PCD /* WT eLLC */
123 
124 #define CHV_PPAT_SNOOP			(1<<6)
125 #define GEN8_PPAT_AGE(x)		(x<<4)
126 #define GEN8_PPAT_LLCeLLC		(3<<2)
127 #define GEN8_PPAT_LLCELLC		(2<<2)
128 #define GEN8_PPAT_LLC			(1<<2)
129 #define GEN8_PPAT_WB			(3<<0)
130 #define GEN8_PPAT_WT			(2<<0)
131 #define GEN8_PPAT_WC			(1<<0)
132 #define GEN8_PPAT_UC			(0<<0)
133 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
134 #define GEN8_PPAT(i, x)			((uint64_t) (x) << ((i) * 8))
135 
136 enum i915_ggtt_view_type {
137 	I915_GGTT_VIEW_NORMAL = 0,
138 	I915_GGTT_VIEW_ROTATED,
139 	I915_GGTT_VIEW_PARTIAL,
140 };
141 
142 struct intel_rotation_info {
143 	unsigned int uv_offset;
144 	uint32_t pixel_format;
145 	unsigned int uv_start_page;
146 	struct {
147 		/* tiles */
148 		unsigned int width, height;
149 	} plane[2];
150 };
151 
152 struct i915_ggtt_view {
153 	enum i915_ggtt_view_type type;
154 
155 	union {
156 		struct {
157 			u64 offset;
158 			unsigned int size;
159 		} partial;
160 		struct intel_rotation_info rotated;
161 	} params;
162 
163 	struct sg_table *pages;
164 };
165 
166 extern const struct i915_ggtt_view i915_ggtt_view_normal;
167 extern const struct i915_ggtt_view i915_ggtt_view_rotated;
168 
169 enum i915_cache_level;
170 
171 /**
172  * A VMA represents a GEM BO that is bound into an address space. Therefore, a
173  * VMA's presence cannot be guaranteed before binding, or after unbinding the
174  * object into/from the address space.
175  *
176  * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
177  * will always be <= an objects lifetime. So object refcounting should cover us.
178  */
179 struct i915_vma {
180 	struct drm_mm_node node;
181 	struct drm_i915_gem_object *obj;
182 	struct i915_address_space *vm;
183 	void __iomem *iomap;
184 
185 	/** Flags and address space this VMA is bound to */
186 #define GLOBAL_BIND	(1<<0)
187 #define LOCAL_BIND	(1<<1)
188 	unsigned int bound : 4;
189 	bool is_ggtt : 1;
190 
191 	/**
192 	 * Support different GGTT views into the same object.
193 	 * This means there can be multiple VMA mappings per object and per VM.
194 	 * i915_ggtt_view_type is used to distinguish between those entries.
195 	 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
196 	 * assumed in GEM functions which take no ggtt view parameter.
197 	 */
198 	struct i915_ggtt_view ggtt_view;
199 
200 	/** This object's place on the active/inactive lists */
201 	struct list_head vm_link;
202 
203 	struct list_head obj_link; /* Link in the object's VMA list */
204 
205 	/** This vma's place in the batchbuffer or on the eviction list */
206 	struct list_head exec_list;
207 
208 	/**
209 	 * Used for performing relocations during execbuffer insertion.
210 	 */
211 	struct hlist_node exec_node;
212 	unsigned long exec_handle;
213 	struct drm_i915_gem_exec_object2 *exec_entry;
214 
215 	/**
216 	 * How many users have pinned this object in GTT space. The following
217 	 * users can each hold at most one reference: pwrite/pread, execbuffer
218 	 * (objects are not allowed multiple times for the same batchbuffer),
219 	 * and the framebuffer code. When switching/pageflipping, the
220 	 * framebuffer code has at most two buffers pinned per crtc.
221 	 *
222 	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
223 	 * bits with absolutely no headroom. So use 4 bits. */
224 	unsigned int pin_count:4;
225 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
226 };
227 
228 struct i915_page_dma {
229 	struct page *page;
230 	union {
231 		dma_addr_t daddr;
232 
233 		/* For gen6/gen7 only. This is the offset in the GGTT
234 		 * where the page directory entries for PPGTT begin
235 		 */
236 		uint32_t ggtt_offset;
237 	};
238 };
239 
240 #define px_base(px) (&(px)->base)
241 #define px_page(px) (px_base(px)->page)
242 #define px_dma(px) (px_base(px)->daddr)
243 
244 struct i915_page_scratch {
245 	struct i915_page_dma base;
246 };
247 
248 struct i915_page_table {
249 	struct i915_page_dma base;
250 
251 	unsigned long *used_ptes;
252 };
253 
254 struct i915_page_directory {
255 	struct i915_page_dma base;
256 
257 	unsigned long *used_pdes;
258 	struct i915_page_table *page_table[I915_PDES]; /* PDEs */
259 };
260 
261 struct i915_page_directory_pointer {
262 	struct i915_page_dma base;
263 
264 	unsigned long *used_pdpes;
265 	struct i915_page_directory **page_directory;
266 };
267 
268 struct i915_pml4 {
269 	struct i915_page_dma base;
270 
271 	DECLARE_BITMAP(used_pml4es, GEN8_PML4ES_PER_PML4);
272 	struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
273 };
274 
275 struct i915_address_space {
276 	struct drm_mm mm;
277 	struct drm_device *dev;
278 	struct list_head global_link;
279 	u64 start;		/* Start offset always 0 for dri2 */
280 	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
281 
282 	bool is_ggtt;
283 
284 	struct i915_page_scratch *scratch_page;
285 	struct i915_page_table *scratch_pt;
286 	struct i915_page_directory *scratch_pd;
287 	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
288 
289 	/**
290 	 * List of objects currently involved in rendering.
291 	 *
292 	 * Includes buffers having the contents of their GPU caches
293 	 * flushed, not necessarily primitives. last_read_req
294 	 * represents when the rendering involved will be completed.
295 	 *
296 	 * A reference is held on the buffer while on this list.
297 	 */
298 	struct list_head active_list;
299 
300 	/**
301 	 * LRU list of objects which are not in the ringbuffer and
302 	 * are ready to unbind, but are still in the GTT.
303 	 *
304 	 * last_read_req is NULL while an object is in this list.
305 	 *
306 	 * A reference is not held on the buffer while on this list,
307 	 * as merely being GTT-bound shouldn't prevent its being
308 	 * freed, and we'll pull it off the list in the free path.
309 	 */
310 	struct list_head inactive_list;
311 
312 	/* FIXME: Need a more generic return type */
313 	gen6_pte_t (*pte_encode)(dma_addr_t addr,
314 				 enum i915_cache_level level,
315 				 bool valid, u32 flags); /* Create a valid PTE */
316 	/* flags for pte_encode */
317 #define PTE_READ_ONLY	(1<<0)
318 	int (*allocate_va_range)(struct i915_address_space *vm,
319 				 uint64_t start,
320 				 uint64_t length);
321 	void (*clear_range)(struct i915_address_space *vm,
322 			    uint64_t start,
323 			    uint64_t length,
324 			    bool use_scratch);
325 	void (*insert_page)(struct i915_address_space *vm,
326 			    dma_addr_t addr,
327 			    uint64_t offset,
328 			    enum i915_cache_level cache_level,
329 			    u32 flags);
330 	void (*insert_entries)(struct i915_address_space *vm,
331 			       struct sg_table *st,
332 			       uint64_t start,
333 			       enum i915_cache_level cache_level, u32 flags);
334 	void (*cleanup)(struct i915_address_space *vm);
335 	/** Unmap an object from an address space. This usually consists of
336 	 * setting the valid PTE entries to a reserved scratch page. */
337 	void (*unbind_vma)(struct i915_vma *vma);
338 	/* Map an object into an address space with the given cache flags. */
339 	int (*bind_vma)(struct i915_vma *vma,
340 			enum i915_cache_level cache_level,
341 			u32 flags);
342 };
343 
344 #define i915_is_ggtt(V) ((V)->is_ggtt)
345 
346 /* The Graphics Translation Table is the way in which GEN hardware translates a
347  * Graphics Virtual Address into a Physical Address. In addition to the normal
348  * collateral associated with any va->pa translations GEN hardware also has a
349  * portion of the GTT which can be mapped by the CPU and remain both coherent
350  * and correct (in cases like swizzling). That region is referred to as GMADR in
351  * the spec.
352  */
353 struct i915_ggtt {
354 	struct i915_address_space base;
355 
356 	size_t stolen_size;		/* Total size of stolen memory */
357 	size_t stolen_usable_size;	/* Total size minus BIOS reserved */
358 	size_t stolen_reserved_base;
359 	size_t stolen_reserved_size;
360 	size_t size;			/* Total size of Global GTT */
361 	u64 mappable_end;		/* End offset that we can CPU map */
362 	struct io_mapping *mappable;	/* Mapping to our CPU mappable region */
363 	phys_addr_t mappable_base;	/* PA of our GMADR */
364 
365 	/** "Graphics Stolen Memory" holds the global PTEs */
366 	void __iomem *gsm;
367 
368 	bool do_idle_maps;
369 
370 	int mtrr;
371 
372 	int (*probe)(struct i915_ggtt *ggtt);
373 };
374 
375 struct i915_hw_ppgtt {
376 	struct i915_address_space base;
377 	struct kref ref;
378 	struct drm_mm_node node;
379 	unsigned long pd_dirty_rings;
380 	union {
381 		struct i915_pml4 pml4;		/* GEN8+ & 48b PPGTT */
382 		struct i915_page_directory_pointer pdp;	/* GEN8+ */
383 		struct i915_page_directory pd;		/* GEN6-7 */
384 	};
385 
386 	struct drm_i915_file_private *file_priv;
387 
388 	gen6_pte_t __iomem *pd_addr;
389 
390 	int (*enable)(struct i915_hw_ppgtt *ppgtt);
391 	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
392 			 struct drm_i915_gem_request *req);
393 	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
394 };
395 
396 /*
397  * gen6_for_each_pde() iterates over every pde from start until start+length.
398  * If start and start+length are not perfectly divisible, the macro will round
399  * down and up as needed. Start=0 and length=2G effectively iterates over
400  * every PDE in the system. The macro modifies ALL its parameters except 'pd',
401  * so each of the other parameters should preferably be a simple variable, or
402  * at most an lvalue with no side-effects!
403  */
404 #define gen6_for_each_pde(pt, pd, start, length, iter)			\
405 	for (iter = gen6_pde_index(start);				\
406 	     length > 0 && iter < I915_PDES &&				\
407 		(pt = (pd)->page_table[iter], true);			\
408 	     ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT);		\
409 		    temp = min(temp - start, length);			\
410 		    start += temp, length -= temp; }), ++iter)
411 
412 #define gen6_for_all_pdes(pt, pd, iter)					\
413 	for (iter = 0;							\
414 	     iter < I915_PDES &&					\
415 		(pt = (pd)->page_table[iter], true);			\
416 	     ++iter)
417 
418 static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
419 {
420 	const uint32_t mask = NUM_PTE(pde_shift) - 1;
421 
422 	return (address >> PAGE_SHIFT) & mask;
423 }
424 
425 /* Helper to counts the number of PTEs within the given length. This count
426  * does not cross a page table boundary, so the max value would be
427  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
428 */
429 static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
430 				      uint32_t pde_shift)
431 {
432 	const uint64_t mask = ~((1ULL << pde_shift) - 1);
433 	uint64_t end;
434 
435 	WARN_ON(length == 0);
436 	WARN_ON(offset_in_page(addr|length));
437 
438 	end = addr + length;
439 
440 	if ((addr & mask) != (end & mask))
441 		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
442 
443 	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
444 }
445 
446 static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
447 {
448 	return (addr >> shift) & I915_PDE_MASK;
449 }
450 
451 static inline uint32_t gen6_pte_index(uint32_t addr)
452 {
453 	return i915_pte_index(addr, GEN6_PDE_SHIFT);
454 }
455 
456 static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
457 {
458 	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
459 }
460 
461 static inline uint32_t gen6_pde_index(uint32_t addr)
462 {
463 	return i915_pde_index(addr, GEN6_PDE_SHIFT);
464 }
465 
466 /* Equivalent to the gen6 version, For each pde iterates over every pde
467  * between from start until start + length. On gen8+ it simply iterates
468  * over every page directory entry in a page directory.
469  */
470 #define gen8_for_each_pde(pt, pd, start, length, iter)			\
471 	for (iter = gen8_pde_index(start);				\
472 	     length > 0 && iter < I915_PDES &&				\
473 		(pt = (pd)->page_table[iter], true);			\
474 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT);		\
475 		    temp = min(temp - start, length);			\
476 		    start += temp, length -= temp; }), ++iter)
477 
478 #define gen8_for_each_pdpe(pd, pdp, start, length, iter)		\
479 	for (iter = gen8_pdpe_index(start);				\
480 	     length > 0 && iter < I915_PDPES_PER_PDP(dev) &&		\
481 		(pd = (pdp)->page_directory[iter], true);		\
482 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT);	\
483 		    temp = min(temp - start, length);			\
484 		    start += temp, length -= temp; }), ++iter)
485 
486 #define gen8_for_each_pml4e(pdp, pml4, start, length, iter)		\
487 	for (iter = gen8_pml4e_index(start);				\
488 	     length > 0 && iter < GEN8_PML4ES_PER_PML4 &&		\
489 		(pdp = (pml4)->pdps[iter], true);			\
490 	     ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT);	\
491 		    temp = min(temp - start, length);			\
492 		    start += temp, length -= temp; }), ++iter)
493 
494 static inline uint32_t gen8_pte_index(uint64_t address)
495 {
496 	return i915_pte_index(address, GEN8_PDE_SHIFT);
497 }
498 
499 static inline uint32_t gen8_pde_index(uint64_t address)
500 {
501 	return i915_pde_index(address, GEN8_PDE_SHIFT);
502 }
503 
504 static inline uint32_t gen8_pdpe_index(uint64_t address)
505 {
506 	return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
507 }
508 
509 static inline uint32_t gen8_pml4e_index(uint64_t address)
510 {
511 	return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
512 }
513 
514 static inline size_t gen8_pte_count(uint64_t address, uint64_t length)
515 {
516 	return i915_pte_count(address, length, GEN8_PDE_SHIFT);
517 }
518 
519 static inline dma_addr_t
520 i915_page_dir_dma_addr(struct i915_hw_ppgtt *ppgtt, const unsigned n)
521 {
522 	return test_bit(n, ppgtt->pdp.used_pdpes) ?
523 		px_dma(ppgtt->pdp.page_directory[n]) :
524 		px_dma(ppgtt->base.scratch_pd);
525 }
526 
527 int i915_ggtt_init_hw(struct drm_device *dev);
528 int i915_ggtt_enable_hw(struct drm_device *dev);
529 void i915_gem_init_ggtt(struct drm_device *dev);
530 void i915_ggtt_cleanup_hw(struct drm_device *dev);
531 
532 int i915_ppgtt_init_hw(struct drm_device *dev);
533 void i915_ppgtt_release(struct kref *kref);
534 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
535 					struct drm_i915_file_private *fpriv);
536 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
537 {
538 	if (ppgtt)
539 		kref_get(&ppgtt->ref);
540 }
541 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
542 {
543 	if (ppgtt)
544 		kref_put(&ppgtt->ref, i915_ppgtt_release);
545 }
546 
547 void i915_check_and_clear_faults(struct drm_i915_private *dev_priv);
548 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
549 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
550 
551 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
552 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
553 
554 static inline bool
555 i915_ggtt_view_equal(const struct i915_ggtt_view *a,
556                      const struct i915_ggtt_view *b)
557 {
558 	if (WARN_ON(!a || !b))
559 		return false;
560 
561 	if (a->type != b->type)
562 		return false;
563 	if (a->type != I915_GGTT_VIEW_NORMAL)
564 		return !memcmp(&a->params, &b->params, sizeof(a->params));
565 	return true;
566 }
567 
568 size_t
569 i915_ggtt_view_size(struct drm_i915_gem_object *obj,
570 		    const struct i915_ggtt_view *view);
571 
572 /**
573  * i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
574  * @vma: VMA to iomap
575  *
576  * The passed in VMA has to be pinned in the global GTT mappable region.
577  * An extra pinning of the VMA is acquired for the return iomapping,
578  * the caller must call i915_vma_unpin_iomap to relinquish the pinning
579  * after the iomapping is no longer required.
580  *
581  * Callers must hold the struct_mutex.
582  *
583  * Returns a valid iomapped pointer or ERR_PTR.
584  */
585 void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
586 #define IO_ERR_PTR(x) ((void __iomem *)ERR_PTR(x))
587 
588 /**
589  * i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
590  * @vma: VMA to unpin
591  *
592  * Unpins the previously iomapped VMA from i915_vma_pin_iomap().
593  *
594  * Callers must hold the struct_mutex. This function is only valid to be
595  * called on a VMA previously iomapped by the caller with i915_vma_pin_iomap().
596  */
597 static inline void i915_vma_unpin_iomap(struct i915_vma *vma)
598 {
599 	lockdep_assert_held(&vma->vm->dev->struct_mutex);
600 	GEM_BUG_ON(vma->pin_count == 0);
601 	GEM_BUG_ON(vma->iomap == NULL);
602 	vma->pin_count--;
603 }
604 
605 #endif
606