xref: /dragonfly/sys/dev/drm/i915/i915_gem_gtt.h (revision 5ca0a96d)
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 #include <linux/mm.h>
39 #include <linux/pagevec.h>
40 
41 #include "i915_gem_timeline.h"
42 #include "i915_gem_request.h"
43 #include "i915_selftest.h"
44 
45 #define I915_GTT_PAGE_SIZE_4K BIT(12)
46 #define I915_GTT_PAGE_SIZE_64K BIT(16)
47 #define I915_GTT_PAGE_SIZE_2M BIT(21)
48 
49 #define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
50 #define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
51 
52 #define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
53 
54 #define I915_FENCE_REG_NONE -1
55 #define I915_MAX_NUM_FENCES 32
56 /* 32 fences + sign bit for FENCE_REG_NONE */
57 #define I915_MAX_NUM_FENCE_BITS 6
58 
59 struct drm_i915_file_private;
60 struct drm_i915_fence_reg;
61 
62 typedef u32 gen6_pte_t;
63 typedef u64 gen8_pte_t;
64 typedef u64 gen8_pde_t;
65 typedef u64 gen8_ppgtt_pdpe_t;
66 typedef u64 gen8_ppgtt_pml4e_t;
67 
68 #define ggtt_total_entries(ggtt) ((ggtt)->base.total >> PAGE_SHIFT)
69 
70 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
71 #define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
72 #define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
73 #define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
74 #define GEN6_PTE_CACHE_LLC		(2 << 1)
75 #define GEN6_PTE_UNCACHED		(1 << 1)
76 #define GEN6_PTE_VALID			(1 << 0)
77 
78 #define I915_PTES(pte_len)		((unsigned int)(PAGE_SIZE / (pte_len)))
79 #define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
80 #define I915_PDES			512
81 #define I915_PDE_MASK			(I915_PDES - 1)
82 #define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
83 
84 #define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
85 #define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
86 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
87 #define GEN6_PDE_SHIFT			22
88 #define GEN6_PDE_VALID			(1 << 0)
89 
90 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
91 
92 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
93 #define BYT_PTE_WRITEABLE		(1 << 1)
94 
95 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
96  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
97  */
98 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
99 					 (((bits) & 0x8) << (11 - 3)))
100 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
101 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
102 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
103 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
104 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
105 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
106 #define HSW_PTE_UNCACHED		(0)
107 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
108 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
109 
110 /* GEN8 32b style address is defined as a 3 level page table:
111  * 31:30 | 29:21 | 20:12 |  11:0
112  * PDPE  |  PDE  |  PTE  | offset
113  * The difference as compared to normal x86 3 level page table is the PDPEs are
114  * programmed via register.
115  */
116 #define GEN8_3LVL_PDPES			4
117 #define GEN8_PDE_SHIFT			21
118 #define GEN8_PDE_MASK			0x1ff
119 #define GEN8_PTE_SHIFT			12
120 #define GEN8_PTE_MASK			0x1ff
121 #define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
122 
123 /* GEN8 48b style address is defined as a 4 level page table:
124  * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
125  * PML4E | PDPE  |  PDE  |  PTE  | offset
126  */
127 #define GEN8_PML4ES_PER_PML4		512
128 #define GEN8_PML4E_SHIFT		39
129 #define GEN8_PML4E_MASK			(GEN8_PML4ES_PER_PML4 - 1)
130 #define GEN8_PDPE_SHIFT			30
131 /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
132  * tables */
133 #define GEN8_PDPE_MASK			0x1ff
134 
135 #define PPAT_UNCACHED			(_PAGE_PWT | _PAGE_PCD)
136 #define PPAT_CACHED_PDE			0 /* WB LLC */
137 #define PPAT_CACHED			_PAGE_PAT /* WB LLCeLLC */
138 #define PPAT_DISPLAY_ELLC		_PAGE_PCD /* WT eLLC */
139 
140 #define CHV_PPAT_SNOOP			(1<<6)
141 #define GEN8_PPAT_AGE(x)		((x)<<4)
142 #define GEN8_PPAT_LLCeLLC		(3<<2)
143 #define GEN8_PPAT_LLCELLC		(2<<2)
144 #define GEN8_PPAT_LLC			(1<<2)
145 #define GEN8_PPAT_WB			(3<<0)
146 #define GEN8_PPAT_WT			(2<<0)
147 #define GEN8_PPAT_WC			(1<<0)
148 #define GEN8_PPAT_UC			(0<<0)
149 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
150 #define GEN8_PPAT(i, x)			((u64)(x) << ((i) * 8))
151 
152 #define GEN8_PPAT_GET_CA(x) ((x) & 3)
153 #define GEN8_PPAT_GET_TC(x) ((x) & (3 << 2))
154 #define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4))
155 #define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6))
156 
157 #define GEN8_PDE_IPS_64K BIT(11)
158 #define GEN8_PDE_PS_2M   BIT(7)
159 
160 struct sg_table;
161 
162 struct intel_rotation_info {
163 	struct intel_rotation_plane_info {
164 		/* tiles */
165 		unsigned int width, height, stride, offset;
166 	} plane[2];
167 } __packed;
168 
169 static inline void assert_intel_rotation_info_is_packed(void)
170 {
171 	BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int));
172 }
173 
174 struct intel_partial_info {
175 	u64 offset;
176 	unsigned int size;
177 } __packed;
178 
179 static inline void assert_intel_partial_info_is_packed(void)
180 {
181 	BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int));
182 }
183 
184 enum i915_ggtt_view_type {
185 	I915_GGTT_VIEW_NORMAL = 0,
186 	I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
187 	I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info),
188 };
189 
190 static inline void assert_i915_ggtt_view_type_is_unique(void)
191 {
192 	/* As we encode the size of each branch inside the union into its type,
193 	 * we have to be careful that each branch has a unique size.
194 	 */
195 	switch ((enum i915_ggtt_view_type)0) {
196 	case I915_GGTT_VIEW_NORMAL:
197 	case I915_GGTT_VIEW_PARTIAL:
198 	case I915_GGTT_VIEW_ROTATED:
199 		/* gcc complains if these are identical cases */
200 		break;
201 	}
202 }
203 
204 struct i915_ggtt_view {
205 	enum i915_ggtt_view_type type;
206 	union {
207 		/* Members need to contain no holes/padding */
208 		struct intel_partial_info partial;
209 		struct intel_rotation_info rotated;
210 	};
211 };
212 
213 enum i915_cache_level;
214 
215 struct i915_vma;
216 
217 struct i915_page_dma {
218 	struct page *page;
219 	int order;
220 	union {
221 		dma_addr_t daddr;
222 
223 		/* For gen6/gen7 only. This is the offset in the GGTT
224 		 * where the page directory entries for PPGTT begin
225 		 */
226 		u32 ggtt_offset;
227 	};
228 };
229 
230 #define px_base(px) (&(px)->base)
231 #define px_page(px) (px_base(px)->page)
232 #define px_dma(px) (px_base(px)->daddr)
233 
234 struct i915_page_table {
235 	struct i915_page_dma base;
236 	unsigned int used_ptes;
237 };
238 
239 struct i915_page_directory {
240 	struct i915_page_dma base;
241 
242 	struct i915_page_table *page_table[I915_PDES]; /* PDEs */
243 	unsigned int used_pdes;
244 };
245 
246 struct i915_page_directory_pointer {
247 	struct i915_page_dma base;
248 	struct i915_page_directory **page_directory;
249 	unsigned int used_pdpes;
250 };
251 
252 struct i915_pml4 {
253 	struct i915_page_dma base;
254 	struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
255 };
256 
257 struct i915_address_space {
258 	struct drm_mm mm;
259 	struct i915_gem_timeline timeline;
260 	struct drm_i915_private *i915;
261 	struct device *dma;
262 	/* Every address space belongs to a struct file - except for the global
263 	 * GTT that is owned by the driver (and so @file is set to NULL). In
264 	 * principle, no information should leak from one context to another
265 	 * (or between files/processes etc) unless explicitly shared by the
266 	 * owner. Tracking the owner is important in order to free up per-file
267 	 * objects along with the file, to aide resource tracking, and to
268 	 * assign blame.
269 	 */
270 	struct drm_i915_file_private *file;
271 	struct list_head global_link;
272 	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
273 	u64 reserved;		/* size addr space reserved */
274 
275 	bool closed;
276 
277 	struct i915_page_dma scratch_page;
278 	struct i915_page_table *scratch_pt;
279 	struct i915_page_directory *scratch_pd;
280 	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
281 
282 	/**
283 	 * List of objects currently involved in rendering.
284 	 *
285 	 * Includes buffers having the contents of their GPU caches
286 	 * flushed, not necessarily primitives. last_read_req
287 	 * represents when the rendering involved will be completed.
288 	 *
289 	 * A reference is held on the buffer while on this list.
290 	 */
291 	struct list_head active_list;
292 
293 	/**
294 	 * LRU list of objects which are not in the ringbuffer and
295 	 * are ready to unbind, but are still in the GTT.
296 	 *
297 	 * last_read_req is NULL while an object is in this list.
298 	 *
299 	 * A reference is not held on the buffer while on this list,
300 	 * as merely being GTT-bound shouldn't prevent its being
301 	 * freed, and we'll pull it off the list in the free path.
302 	 */
303 	struct list_head inactive_list;
304 
305 	/**
306 	 * List of vma that have been unbound.
307 	 *
308 	 * A reference is not held on the buffer while on this list.
309 	 */
310 	struct list_head unbound_list;
311 
312 	struct pagevec free_pages;
313 	bool pt_kmap_wc;
314 
315 	/* FIXME: Need a more generic return type */
316 	gen6_pte_t (*pte_encode)(dma_addr_t addr,
317 				 enum i915_cache_level level,
318 				 u32 flags); /* Create a valid PTE */
319 	/* flags for pte_encode */
320 #define PTE_READ_ONLY	(1<<0)
321 	int (*allocate_va_range)(struct i915_address_space *vm,
322 				 u64 start, u64 length);
323 	void (*clear_range)(struct i915_address_space *vm,
324 			    u64 start, u64 length);
325 	void (*insert_page)(struct i915_address_space *vm,
326 			    dma_addr_t addr,
327 			    u64 offset,
328 			    enum i915_cache_level cache_level,
329 			    u32 flags);
330 	void (*insert_entries)(struct i915_address_space *vm,
331 			       struct i915_vma *vma,
332 			       enum i915_cache_level cache_level,
333 			       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 	int (*set_pages)(struct i915_vma *vma);
343 	void (*clear_pages)(struct i915_vma *vma);
344 
345 	I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
346 };
347 
348 #define i915_is_ggtt(V) (!(V)->file)
349 
350 static inline bool
351 i915_vm_is_48bit(const struct i915_address_space *vm)
352 {
353 	return (vm->total - 1) >> 32;
354 }
355 
356 static inline bool
357 i915_vm_has_scratch_64K(struct i915_address_space *vm)
358 {
359 	return vm->scratch_page.order == get_order(I915_GTT_PAGE_SIZE_64K);
360 }
361 
362 /* The Graphics Translation Table is the way in which GEN hardware translates a
363  * Graphics Virtual Address into a Physical Address. In addition to the normal
364  * collateral associated with any va->pa translations GEN hardware also has a
365  * portion of the GTT which can be mapped by the CPU and remain both coherent
366  * and correct (in cases like swizzling). That region is referred to as GMADR in
367  * the spec.
368  */
369 struct i915_ggtt {
370 	struct i915_address_space base;
371 	struct io_mapping mappable;	/* Mapping to our CPU mappable region */
372 
373 	phys_addr_t mappable_base;	/* PA of our GMADR */
374 	u64 mappable_end;		/* End offset that we can CPU map */
375 
376 	/* Stolen memory is segmented in hardware with different portions
377 	 * offlimits to certain functions.
378 	 *
379 	 * The drm_mm is initialised to the total accessible range, as found
380 	 * from the PCI config. On Broadwell+, this is further restricted to
381 	 * avoid the first page! The upper end of stolen memory is reserved for
382 	 * hardware functions and similarly removed from the accessible range.
383 	 */
384 	u32 stolen_size;		/* Total size of stolen memory */
385 	u32 stolen_usable_size;	/* Total size minus reserved ranges */
386 	u32 stolen_reserved_base;
387 	u32 stolen_reserved_size;
388 
389 	/** "Graphics Stolen Memory" holds the global PTEs */
390 	void __iomem *gsm;
391 	void (*invalidate)(struct drm_i915_private *dev_priv);
392 
393 	bool do_idle_maps;
394 
395 	int mtrr;
396 
397 	struct drm_mm_node error_capture;
398 };
399 
400 struct i915_hw_ppgtt {
401 	struct i915_address_space base;
402 	struct kref ref;
403 	struct drm_mm_node node;
404 	unsigned long pd_dirty_rings;
405 	union {
406 		struct i915_pml4 pml4;		/* GEN8+ & 48b PPGTT */
407 		struct i915_page_directory_pointer pdp;	/* GEN8+ */
408 		struct i915_page_directory pd;		/* GEN6-7 */
409 	};
410 
411 	gen6_pte_t __iomem *pd_addr;
412 
413 	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
414 			 struct drm_i915_gem_request *req);
415 	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
416 };
417 
418 /*
419  * gen6_for_each_pde() iterates over every pde from start until start+length.
420  * If start and start+length are not perfectly divisible, the macro will round
421  * down and up as needed. Start=0 and length=2G effectively iterates over
422  * every PDE in the system. The macro modifies ALL its parameters except 'pd',
423  * so each of the other parameters should preferably be a simple variable, or
424  * at most an lvalue with no side-effects!
425  */
426 #define gen6_for_each_pde(pt, pd, start, length, iter)			\
427 	for (iter = gen6_pde_index(start);				\
428 	     length > 0 && iter < I915_PDES &&				\
429 		(pt = (pd)->page_table[iter], true);			\
430 	     ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT);		\
431 		    temp = min(temp - start, length);			\
432 		    start += temp, length -= temp; }), ++iter)
433 
434 #define gen6_for_all_pdes(pt, pd, iter)					\
435 	for (iter = 0;							\
436 	     iter < I915_PDES &&					\
437 		(pt = (pd)->page_table[iter], true);			\
438 	     ++iter)
439 
440 static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
441 {
442 	const u32 mask = NUM_PTE(pde_shift) - 1;
443 
444 	return (address >> PAGE_SHIFT) & mask;
445 }
446 
447 /* Helper to counts the number of PTEs within the given length. This count
448  * does not cross a page table boundary, so the max value would be
449  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
450 */
451 static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
452 {
453 	const u64 mask = ~((1ULL << pde_shift) - 1);
454 	u64 end;
455 
456 	WARN_ON(length == 0);
457 	WARN_ON(offset_in_page(addr|length));
458 
459 	end = addr + length;
460 
461 	if ((addr & mask) != (end & mask))
462 		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
463 
464 	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
465 }
466 
467 static inline u32 i915_pde_index(u64 addr, u32 shift)
468 {
469 	return (addr >> shift) & I915_PDE_MASK;
470 }
471 
472 static inline u32 gen6_pte_index(u32 addr)
473 {
474 	return i915_pte_index(addr, GEN6_PDE_SHIFT);
475 }
476 
477 static inline u32 gen6_pte_count(u32 addr, u32 length)
478 {
479 	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
480 }
481 
482 static inline u32 gen6_pde_index(u32 addr)
483 {
484 	return i915_pde_index(addr, GEN6_PDE_SHIFT);
485 }
486 
487 static inline unsigned int
488 i915_pdpes_per_pdp(const struct i915_address_space *vm)
489 {
490 	if (i915_vm_is_48bit(vm))
491 		return GEN8_PML4ES_PER_PML4;
492 
493 	return GEN8_3LVL_PDPES;
494 }
495 
496 /* Equivalent to the gen6 version, For each pde iterates over every pde
497  * between from start until start + length. On gen8+ it simply iterates
498  * over every page directory entry in a page directory.
499  */
500 #define gen8_for_each_pde(pt, pd, start, length, iter)			\
501 	for (iter = gen8_pde_index(start);				\
502 	     length > 0 && iter < I915_PDES &&				\
503 		(pt = (pd)->page_table[iter], true);			\
504 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT);		\
505 		    temp = min(temp - start, length);			\
506 		    start += temp, length -= temp; }), ++iter)
507 
508 #define gen8_for_each_pdpe(pd, pdp, start, length, iter)		\
509 	for (iter = gen8_pdpe_index(start);				\
510 	     length > 0 && iter < i915_pdpes_per_pdp(vm) &&		\
511 		(pd = (pdp)->page_directory[iter], true);		\
512 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT);	\
513 		    temp = min(temp - start, length);			\
514 		    start += temp, length -= temp; }), ++iter)
515 
516 #define gen8_for_each_pml4e(pdp, pml4, start, length, iter)		\
517 	for (iter = gen8_pml4e_index(start);				\
518 	     length > 0 && iter < GEN8_PML4ES_PER_PML4 &&		\
519 		(pdp = (pml4)->pdps[iter], true);			\
520 	     ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT);	\
521 		    temp = min(temp - start, length);			\
522 		    start += temp, length -= temp; }), ++iter)
523 
524 static inline u32 gen8_pte_index(u64 address)
525 {
526 	return i915_pte_index(address, GEN8_PDE_SHIFT);
527 }
528 
529 static inline u32 gen8_pde_index(u64 address)
530 {
531 	return i915_pde_index(address, GEN8_PDE_SHIFT);
532 }
533 
534 static inline u32 gen8_pdpe_index(u64 address)
535 {
536 	return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
537 }
538 
539 static inline u32 gen8_pml4e_index(u64 address)
540 {
541 	return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
542 }
543 
544 static inline u64 gen8_pte_count(u64 address, u64 length)
545 {
546 	return i915_pte_count(address, length, GEN8_PDE_SHIFT);
547 }
548 
549 static inline dma_addr_t
550 i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
551 {
552 	return px_dma(ppgtt->pdp.page_directory[n]);
553 }
554 
555 static inline struct i915_ggtt *
556 i915_vm_to_ggtt(struct i915_address_space *vm)
557 {
558 	GEM_BUG_ON(!i915_is_ggtt(vm));
559 	return container_of(vm, struct i915_ggtt, base);
560 }
561 
562 #define INTEL_MAX_PPAT_ENTRIES 8
563 #define INTEL_PPAT_PERFECT_MATCH (~0U)
564 
565 struct intel_ppat;
566 
567 struct intel_ppat_entry {
568 	struct intel_ppat *ppat;
569 	struct kref ref;
570 	u8 value;
571 };
572 
573 struct intel_ppat {
574 	struct intel_ppat_entry entries[INTEL_MAX_PPAT_ENTRIES];
575 	DECLARE_BITMAP(used, INTEL_MAX_PPAT_ENTRIES);
576 	DECLARE_BITMAP(dirty, INTEL_MAX_PPAT_ENTRIES);
577 	unsigned int max_entries;
578 	u8 clear_value;
579 	/*
580 	 * Return a score to show how two PPAT values match,
581 	 * a INTEL_PPAT_PERFECT_MATCH indicates a perfect match
582 	 */
583 	unsigned int (*match)(u8 src, u8 dst);
584 	void (*update_hw)(struct drm_i915_private *i915);
585 
586 	struct drm_i915_private *i915;
587 };
588 
589 const struct intel_ppat_entry *
590 intel_ppat_get(struct drm_i915_private *i915, u8 value);
591 void intel_ppat_put(const struct intel_ppat_entry *entry);
592 
593 int i915_gem_init_aliasing_ppgtt(struct drm_i915_private *i915);
594 void i915_gem_fini_aliasing_ppgtt(struct drm_i915_private *i915);
595 
596 int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
597 int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
598 int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
599 void i915_ggtt_enable_guc(struct drm_i915_private *i915);
600 void i915_ggtt_disable_guc(struct drm_i915_private *i915);
601 int i915_gem_init_ggtt(struct drm_i915_private *dev_priv);
602 void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv);
603 
604 int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv);
605 void i915_ppgtt_release(struct kref *kref);
606 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv,
607 					struct drm_i915_file_private *fpriv,
608 					const char *name);
609 void i915_ppgtt_close(struct i915_address_space *vm);
610 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
611 {
612 	if (ppgtt)
613 		kref_get(&ppgtt->ref);
614 }
615 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
616 {
617 	if (ppgtt)
618 		kref_put(&ppgtt->ref, i915_ppgtt_release);
619 }
620 
621 void i915_check_and_clear_faults(struct drm_i915_private *dev_priv);
622 void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv);
623 void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv);
624 
625 int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
626 					    struct sg_table *pages);
627 void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
628 			       struct sg_table *pages);
629 
630 int i915_gem_gtt_reserve(struct i915_address_space *vm,
631 			 struct drm_mm_node *node,
632 			 u64 size, u64 offset, unsigned long color,
633 			 unsigned int flags);
634 
635 int i915_gem_gtt_insert(struct i915_address_space *vm,
636 			struct drm_mm_node *node,
637 			u64 size, u64 alignment, unsigned long color,
638 			u64 start, u64 end, unsigned int flags);
639 
640 /* Flags used by pin/bind&friends. */
641 #define PIN_NONBLOCK		BIT(0)
642 #define PIN_MAPPABLE		BIT(1)
643 #define PIN_ZONE_4G		BIT(2)
644 #define PIN_NONFAULT		BIT(3)
645 #define PIN_NOEVICT		BIT(4)
646 
647 #define PIN_MBZ			BIT(5) /* I915_VMA_PIN_OVERFLOW */
648 #define PIN_GLOBAL		BIT(6) /* I915_VMA_GLOBAL_BIND */
649 #define PIN_USER		BIT(7) /* I915_VMA_LOCAL_BIND */
650 #define PIN_UPDATE		BIT(8)
651 
652 #define PIN_HIGH		BIT(9)
653 #define PIN_OFFSET_BIAS		BIT(10)
654 #define PIN_OFFSET_FIXED	BIT(11)
655 #define PIN_OFFSET_MASK		(-I915_GTT_PAGE_SIZE)
656 
657 #endif
658