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 "gt/intel_reset.h"
42 #include "i915_gem_fence_reg.h"
43 #include "i915_request.h"
44 #include "i915_scatterlist.h"
45 #include "i915_selftest.h"
46 #include "i915_timeline.h"
47 
48 #define I915_GTT_PAGE_SIZE_4K	BIT_ULL(12)
49 #define I915_GTT_PAGE_SIZE_64K	BIT_ULL(16)
50 #define I915_GTT_PAGE_SIZE_2M	BIT_ULL(21)
51 
52 #define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
53 #define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
54 
55 #define I915_GTT_PAGE_MASK -I915_GTT_PAGE_SIZE
56 
57 #define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
58 
59 #define I915_FENCE_REG_NONE -1
60 #define I915_MAX_NUM_FENCES 32
61 /* 32 fences + sign bit for FENCE_REG_NONE */
62 #define I915_MAX_NUM_FENCE_BITS 6
63 
64 struct drm_i915_file_private;
65 struct drm_i915_gem_object;
66 struct i915_vma;
67 
68 typedef u32 gen6_pte_t;
69 typedef u64 gen8_pte_t;
70 typedef u64 gen8_pde_t;
71 typedef u64 gen8_ppgtt_pdpe_t;
72 typedef u64 gen8_ppgtt_pml4e_t;
73 
74 #define ggtt_total_entries(ggtt) ((ggtt)->vm.total >> PAGE_SHIFT)
75 
76 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
77 #define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
78 #define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
79 #define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
80 #define GEN6_PTE_CACHE_LLC		(2 << 1)
81 #define GEN6_PTE_UNCACHED		(1 << 1)
82 #define GEN6_PTE_VALID			(1 << 0)
83 
84 #define I915_PTES(pte_len)		((unsigned int)(PAGE_SIZE / (pte_len)))
85 #define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
86 #define I915_PDES			512
87 #define I915_PDE_MASK			(I915_PDES - 1)
88 #define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
89 
90 #define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
91 #define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
92 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
93 #define GEN6_PDE_SHIFT			22
94 #define GEN6_PDE_VALID			(1 << 0)
95 
96 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
97 
98 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
99 #define BYT_PTE_WRITEABLE		(1 << 1)
100 
101 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
102  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
103  */
104 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
105 					 (((bits) & 0x8) << (11 - 3)))
106 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
107 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
108 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
109 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
110 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
111 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
112 #define HSW_PTE_UNCACHED		(0)
113 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
114 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
115 
116 /* GEN8 32b style address is defined as a 3 level page table:
117  * 31:30 | 29:21 | 20:12 |  11:0
118  * PDPE  |  PDE  |  PTE  | offset
119  * The difference as compared to normal x86 3 level page table is the PDPEs are
120  * programmed via register.
121  */
122 #define GEN8_3LVL_PDPES			4
123 #define GEN8_PDE_SHIFT			21
124 #define GEN8_PDE_MASK			0x1ff
125 #define GEN8_PTE_SHIFT			12
126 #define GEN8_PTE_MASK			0x1ff
127 #define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
128 
129 /* GEN8 48b style address is defined as a 4 level page table:
130  * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
131  * PML4E | PDPE  |  PDE  |  PTE  | offset
132  */
133 #define GEN8_PML4ES_PER_PML4		512
134 #define GEN8_PML4E_SHIFT		39
135 #define GEN8_PML4E_MASK			(GEN8_PML4ES_PER_PML4 - 1)
136 #define GEN8_PDPE_SHIFT			30
137 /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
138  * tables */
139 #define GEN8_PDPE_MASK			0x1ff
140 
141 #define PPAT_UNCACHED			(_PAGE_PWT | _PAGE_PCD)
142 #define PPAT_CACHED_PDE			0 /* WB LLC */
143 #define PPAT_CACHED			_PAGE_PAT /* WB LLCeLLC */
144 #define PPAT_DISPLAY_ELLC		_PAGE_PCD /* WT eLLC */
145 
146 #define CHV_PPAT_SNOOP			(1<<6)
147 #define GEN8_PPAT_AGE(x)		((x)<<4)
148 #define GEN8_PPAT_LLCeLLC		(3<<2)
149 #define GEN8_PPAT_LLCELLC		(2<<2)
150 #define GEN8_PPAT_LLC			(1<<2)
151 #define GEN8_PPAT_WB			(3<<0)
152 #define GEN8_PPAT_WT			(2<<0)
153 #define GEN8_PPAT_WC			(1<<0)
154 #define GEN8_PPAT_UC			(0<<0)
155 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
156 #define GEN8_PPAT(i, x)			((u64)(x) << ((i) * 8))
157 
158 #define GEN8_PPAT_GET_CA(x) ((x) & 3)
159 #define GEN8_PPAT_GET_TC(x) ((x) & (3 << 2))
160 #define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4))
161 #define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6))
162 
163 #define GEN8_PDE_IPS_64K BIT(11)
164 #define GEN8_PDE_PS_2M   BIT(7)
165 
166 #define for_each_sgt_dma(__dmap, __iter, __sgt) \
167 	__for_each_sgt_dma(__dmap, __iter, __sgt, I915_GTT_PAGE_SIZE)
168 
169 struct intel_remapped_plane_info {
170 	/* in gtt pages */
171 	unsigned int width, height, stride, offset;
172 } __packed;
173 
174 struct intel_remapped_info {
175 	struct intel_remapped_plane_info plane[2];
176 	unsigned int unused_mbz;
177 } __packed;
178 
179 struct intel_rotation_info {
180 	struct intel_remapped_plane_info plane[2];
181 } __packed;
182 
183 struct intel_partial_info {
184 	u64 offset;
185 	unsigned int size;
186 } __packed;
187 
188 enum i915_ggtt_view_type {
189 	I915_GGTT_VIEW_NORMAL = 0,
190 	I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
191 	I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info),
192 	I915_GGTT_VIEW_REMAPPED = sizeof(struct intel_remapped_info),
193 };
194 
195 static inline void assert_i915_gem_gtt_types(void)
196 {
197 	BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int));
198 	BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int));
199 	BUILD_BUG_ON(sizeof(struct intel_remapped_info) != 9*sizeof(unsigned int));
200 
201 	/* Check that rotation/remapped shares offsets for simplicity */
202 	BUILD_BUG_ON(offsetof(struct intel_remapped_info, plane[0]) !=
203 		     offsetof(struct intel_rotation_info, plane[0]));
204 	BUILD_BUG_ON(offsetofend(struct intel_remapped_info, plane[1]) !=
205 		     offsetofend(struct intel_rotation_info, plane[1]));
206 
207 	/* As we encode the size of each branch inside the union into its type,
208 	 * we have to be careful that each branch has a unique size.
209 	 */
210 	switch ((enum i915_ggtt_view_type)0) {
211 	case I915_GGTT_VIEW_NORMAL:
212 	case I915_GGTT_VIEW_PARTIAL:
213 	case I915_GGTT_VIEW_ROTATED:
214 	case I915_GGTT_VIEW_REMAPPED:
215 		/* gcc complains if these are identical cases */
216 		break;
217 	}
218 }
219 
220 struct i915_ggtt_view {
221 	enum i915_ggtt_view_type type;
222 	union {
223 		/* Members need to contain no holes/padding */
224 		struct intel_partial_info partial;
225 		struct intel_rotation_info rotated;
226 		struct intel_remapped_info remapped;
227 	};
228 };
229 
230 enum i915_cache_level;
231 
232 struct i915_vma;
233 
234 struct i915_page_dma {
235 	struct page *page;
236 	union {
237 		dma_addr_t daddr;
238 
239 		/* For gen6/gen7 only. This is the offset in the GGTT
240 		 * where the page directory entries for PPGTT begin
241 		 */
242 		u32 ggtt_offset;
243 	};
244 };
245 
246 #define px_base(px) (&(px)->base)
247 #define px_dma(px) (px_base(px)->daddr)
248 
249 struct i915_page_table {
250 	struct i915_page_dma base;
251 	atomic_t used;
252 };
253 
254 struct i915_page_directory {
255 	struct i915_page_dma base;
256 	atomic_t used;
257 	spinlock_t lock;
258 	void *entry[512];
259 };
260 
261 struct i915_vma_ops {
262 	/* Map an object into an address space with the given cache flags. */
263 	int (*bind_vma)(struct i915_vma *vma,
264 			enum i915_cache_level cache_level,
265 			u32 flags);
266 	/*
267 	 * Unmap an object from an address space. This usually consists of
268 	 * setting the valid PTE entries to a reserved scratch page.
269 	 */
270 	void (*unbind_vma)(struct i915_vma *vma);
271 
272 	int (*set_pages)(struct i915_vma *vma);
273 	void (*clear_pages)(struct i915_vma *vma);
274 };
275 
276 struct pagestash {
277 	spinlock_t lock;
278 	struct pagevec pvec;
279 };
280 
281 struct i915_address_space {
282 	struct kref ref;
283 
284 	struct drm_mm mm;
285 	struct drm_i915_private *i915;
286 	struct device *dma;
287 	/* Every address space belongs to a struct file - except for the global
288 	 * GTT that is owned by the driver (and so @file is set to NULL). In
289 	 * principle, no information should leak from one context to another
290 	 * (or between files/processes etc) unless explicitly shared by the
291 	 * owner. Tracking the owner is important in order to free up per-file
292 	 * objects along with the file, to aide resource tracking, and to
293 	 * assign blame.
294 	 */
295 	struct drm_i915_file_private *file;
296 	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
297 	u64 reserved;		/* size addr space reserved */
298 
299 	bool closed;
300 
301 	struct mutex mutex; /* protects vma and our lists */
302 #define VM_CLASS_GGTT 0
303 #define VM_CLASS_PPGTT 1
304 
305 	u64 scratch_pte;
306 	int scratch_order;
307 	struct i915_page_dma scratch_page;
308 	struct i915_page_table *scratch_pt;
309 	struct i915_page_directory *scratch_pd;
310 	struct i915_page_directory *scratch_pdp; /* GEN8+ & 48b PPGTT */
311 
312 	/**
313 	 * List of vma currently bound.
314 	 */
315 	struct list_head bound_list;
316 
317 	/**
318 	 * List of vma that are not unbound.
319 	 */
320 	struct list_head unbound_list;
321 
322 	struct pagestash free_pages;
323 
324 	/* Global GTT */
325 	bool is_ggtt:1;
326 
327 	/* Some systems require uncached updates of the page directories */
328 	bool pt_kmap_wc:1;
329 
330 	/* Some systems support read-only mappings for GGTT and/or PPGTT */
331 	bool has_read_only:1;
332 
333 	u64 (*pte_encode)(dma_addr_t addr,
334 			  enum i915_cache_level level,
335 			  u32 flags); /* Create a valid PTE */
336 #define PTE_READ_ONLY	(1<<0)
337 
338 	int (*allocate_va_range)(struct i915_address_space *vm,
339 				 u64 start, u64 length);
340 	void (*clear_range)(struct i915_address_space *vm,
341 			    u64 start, u64 length);
342 	void (*insert_page)(struct i915_address_space *vm,
343 			    dma_addr_t addr,
344 			    u64 offset,
345 			    enum i915_cache_level cache_level,
346 			    u32 flags);
347 	void (*insert_entries)(struct i915_address_space *vm,
348 			       struct i915_vma *vma,
349 			       enum i915_cache_level cache_level,
350 			       u32 flags);
351 	void (*cleanup)(struct i915_address_space *vm);
352 
353 	struct i915_vma_ops vma_ops;
354 
355 	I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
356 	I915_SELFTEST_DECLARE(bool scrub_64K);
357 };
358 
359 #define i915_is_ggtt(vm) ((vm)->is_ggtt)
360 
361 static inline bool
362 i915_vm_is_4lvl(const struct i915_address_space *vm)
363 {
364 	return (vm->total - 1) >> 32;
365 }
366 
367 static inline bool
368 i915_vm_has_scratch_64K(struct i915_address_space *vm)
369 {
370 	return vm->scratch_order == get_order(I915_GTT_PAGE_SIZE_64K);
371 }
372 
373 /* The Graphics Translation Table is the way in which GEN hardware translates a
374  * Graphics Virtual Address into a Physical Address. In addition to the normal
375  * collateral associated with any va->pa translations GEN hardware also has a
376  * portion of the GTT which can be mapped by the CPU and remain both coherent
377  * and correct (in cases like swizzling). That region is referred to as GMADR in
378  * the spec.
379  */
380 struct i915_ggtt {
381 	struct i915_address_space vm;
382 
383 	struct io_mapping iomap;	/* Mapping to our CPU mappable region */
384 	struct resource gmadr;          /* GMADR resource */
385 	resource_size_t mappable_end;	/* End offset that we can CPU map */
386 
387 	/** "Graphics Stolen Memory" holds the global PTEs */
388 	void __iomem *gsm;
389 	void (*invalidate)(struct drm_i915_private *dev_priv);
390 
391 	bool do_idle_maps;
392 
393 	int mtrr;
394 
395 	u32 pin_bias;
396 
397 	unsigned int num_fences;
398 	struct i915_fence_reg fence_regs[I915_MAX_NUM_FENCES];
399 	struct list_head fence_list;
400 
401 	/** List of all objects in gtt_space, currently mmaped by userspace.
402 	 * All objects within this list must also be on bound_list.
403 	 */
404 	struct list_head userfault_list;
405 
406 	/* Manual runtime pm autosuspend delay for user GGTT mmaps */
407 	struct intel_wakeref_auto userfault_wakeref;
408 
409 	struct drm_mm_node error_capture;
410 	struct drm_mm_node uc_fw;
411 };
412 
413 struct i915_ppgtt {
414 	struct i915_address_space vm;
415 
416 	intel_engine_mask_t pd_dirty_engines;
417 	struct i915_page_directory *pd;
418 };
419 
420 struct gen6_ppgtt {
421 	struct i915_ppgtt base;
422 
423 	struct i915_vma *vma;
424 	gen6_pte_t __iomem *pd_addr;
425 
426 	unsigned int pin_count;
427 	bool scan_for_unused_pt;
428 
429 	struct gen6_ppgtt_cleanup_work *work;
430 };
431 
432 #define __to_gen6_ppgtt(base) container_of(base, struct gen6_ppgtt, base)
433 
434 static inline struct gen6_ppgtt *to_gen6_ppgtt(struct i915_ppgtt *base)
435 {
436 	BUILD_BUG_ON(offsetof(struct gen6_ppgtt, base));
437 	return __to_gen6_ppgtt(base);
438 }
439 
440 /*
441  * gen6_for_each_pde() iterates over every pde from start until start+length.
442  * If start and start+length are not perfectly divisible, the macro will round
443  * down and up as needed. Start=0 and length=2G effectively iterates over
444  * every PDE in the system. The macro modifies ALL its parameters except 'pd',
445  * so each of the other parameters should preferably be a simple variable, or
446  * at most an lvalue with no side-effects!
447  */
448 #define gen6_for_each_pde(pt, pd, start, length, iter)			\
449 	for (iter = gen6_pde_index(start);				\
450 	     length > 0 && iter < I915_PDES &&				\
451 		     (pt = i915_pt_entry(pd, iter), true);		\
452 	     ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT);		\
453 		    temp = min(temp - start, length);			\
454 		    start += temp, length -= temp; }), ++iter)
455 
456 #define gen6_for_all_pdes(pt, pd, iter)					\
457 	for (iter = 0;							\
458 	     iter < I915_PDES &&					\
459 		     (pt = i915_pt_entry(pd, iter), true);		\
460 	     ++iter)
461 
462 static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
463 {
464 	const u32 mask = NUM_PTE(pde_shift) - 1;
465 
466 	return (address >> PAGE_SHIFT) & mask;
467 }
468 
469 /* Helper to counts the number of PTEs within the given length. This count
470  * does not cross a page table boundary, so the max value would be
471  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
472 */
473 static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
474 {
475 	const u64 mask = ~((1ULL << pde_shift) - 1);
476 	u64 end;
477 
478 	GEM_BUG_ON(length == 0);
479 	GEM_BUG_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 u32 i915_pde_index(u64 addr, u32 shift)
490 {
491 	return (addr >> shift) & I915_PDE_MASK;
492 }
493 
494 static inline u32 gen6_pte_index(u32 addr)
495 {
496 	return i915_pte_index(addr, GEN6_PDE_SHIFT);
497 }
498 
499 static inline u32 gen6_pte_count(u32 addr, u32 length)
500 {
501 	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
502 }
503 
504 static inline u32 gen6_pde_index(u32 addr)
505 {
506 	return i915_pde_index(addr, GEN6_PDE_SHIFT);
507 }
508 
509 static inline unsigned int
510 i915_pdpes_per_pdp(const struct i915_address_space *vm)
511 {
512 	if (i915_vm_is_4lvl(vm))
513 		return GEN8_PML4ES_PER_PML4;
514 
515 	return GEN8_3LVL_PDPES;
516 }
517 
518 static inline struct i915_page_table *
519 i915_pt_entry(const struct i915_page_directory * const pd,
520 	      const unsigned short n)
521 {
522 	return pd->entry[n];
523 }
524 
525 static inline struct i915_page_directory *
526 i915_pd_entry(const struct i915_page_directory * const pdp,
527 	      const unsigned short n)
528 {
529 	return pdp->entry[n];
530 }
531 
532 static inline struct i915_page_directory *
533 i915_pdp_entry(const struct i915_page_directory * const pml4,
534 	       const unsigned short n)
535 {
536 	return pml4->entry[n];
537 }
538 
539 /* Equivalent to the gen6 version, For each pde iterates over every pde
540  * between from start until start + length. On gen8+ it simply iterates
541  * over every page directory entry in a page directory.
542  */
543 #define gen8_for_each_pde(pt, pd, start, length, iter)			\
544 	for (iter = gen8_pde_index(start);				\
545 	     length > 0 && iter < I915_PDES &&				\
546 		     (pt = i915_pt_entry(pd, iter), true);		\
547 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT);		\
548 		    temp = min(temp - start, length);			\
549 		    start += temp, length -= temp; }), ++iter)
550 
551 #define gen8_for_each_pdpe(pd, pdp, start, length, iter)		\
552 	for (iter = gen8_pdpe_index(start);				\
553 	     length > 0 && iter < i915_pdpes_per_pdp(vm) &&		\
554 		     (pd = i915_pd_entry(pdp, iter), true);		\
555 	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT);	\
556 		    temp = min(temp - start, length);			\
557 		    start += temp, length -= temp; }), ++iter)
558 
559 #define gen8_for_each_pml4e(pdp, pml4, start, length, iter)		\
560 	for (iter = gen8_pml4e_index(start);				\
561 	     length > 0 && iter < GEN8_PML4ES_PER_PML4 &&		\
562 		     (pdp = i915_pdp_entry(pml4, iter), true);		\
563 	     ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT);	\
564 		    temp = min(temp - start, length);			\
565 		    start += temp, length -= temp; }), ++iter)
566 
567 static inline u32 gen8_pte_index(u64 address)
568 {
569 	return i915_pte_index(address, GEN8_PDE_SHIFT);
570 }
571 
572 static inline u32 gen8_pde_index(u64 address)
573 {
574 	return i915_pde_index(address, GEN8_PDE_SHIFT);
575 }
576 
577 static inline u32 gen8_pdpe_index(u64 address)
578 {
579 	return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
580 }
581 
582 static inline u32 gen8_pml4e_index(u64 address)
583 {
584 	return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
585 }
586 
587 static inline u64 gen8_pte_count(u64 address, u64 length)
588 {
589 	return i915_pte_count(address, length, GEN8_PDE_SHIFT);
590 }
591 
592 static inline dma_addr_t
593 i915_page_dir_dma_addr(const struct i915_ppgtt *ppgtt, const unsigned int n)
594 {
595 	struct i915_page_directory *pd;
596 
597 	pd = i915_pdp_entry(ppgtt->pd, n);
598 	return px_dma(pd);
599 }
600 
601 static inline struct i915_ggtt *
602 i915_vm_to_ggtt(struct i915_address_space *vm)
603 {
604 	BUILD_BUG_ON(offsetof(struct i915_ggtt, vm));
605 	GEM_BUG_ON(!i915_is_ggtt(vm));
606 	return container_of(vm, struct i915_ggtt, vm);
607 }
608 
609 static inline struct i915_ppgtt *
610 i915_vm_to_ppgtt(struct i915_address_space *vm)
611 {
612 	BUILD_BUG_ON(offsetof(struct i915_ppgtt, vm));
613 	GEM_BUG_ON(i915_is_ggtt(vm));
614 	return container_of(vm, struct i915_ppgtt, vm);
615 }
616 
617 #define INTEL_MAX_PPAT_ENTRIES 8
618 #define INTEL_PPAT_PERFECT_MATCH (~0U)
619 
620 struct intel_ppat;
621 
622 struct intel_ppat_entry {
623 	struct intel_ppat *ppat;
624 	struct kref ref;
625 	u8 value;
626 };
627 
628 struct intel_ppat {
629 	struct intel_ppat_entry entries[INTEL_MAX_PPAT_ENTRIES];
630 	DECLARE_BITMAP(used, INTEL_MAX_PPAT_ENTRIES);
631 	DECLARE_BITMAP(dirty, INTEL_MAX_PPAT_ENTRIES);
632 	unsigned int max_entries;
633 	u8 clear_value;
634 	/*
635 	 * Return a score to show how two PPAT values match,
636 	 * a INTEL_PPAT_PERFECT_MATCH indicates a perfect match
637 	 */
638 	unsigned int (*match)(u8 src, u8 dst);
639 	void (*update_hw)(struct drm_i915_private *i915);
640 
641 	struct drm_i915_private *i915;
642 };
643 
644 const struct intel_ppat_entry *
645 intel_ppat_get(struct drm_i915_private *i915, u8 value);
646 void intel_ppat_put(const struct intel_ppat_entry *entry);
647 
648 int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
649 int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
650 int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
651 void i915_ggtt_enable_guc(struct drm_i915_private *i915);
652 void i915_ggtt_disable_guc(struct drm_i915_private *i915);
653 int i915_gem_init_ggtt(struct drm_i915_private *dev_priv);
654 void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv);
655 
656 int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv);
657 
658 struct i915_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv);
659 
660 static inline struct i915_address_space *
661 i915_vm_get(struct i915_address_space *vm)
662 {
663 	kref_get(&vm->ref);
664 	return vm;
665 }
666 
667 void i915_vm_release(struct kref *kref);
668 
669 static inline void i915_vm_put(struct i915_address_space *vm)
670 {
671 	kref_put(&vm->ref, i915_vm_release);
672 }
673 
674 int gen6_ppgtt_pin(struct i915_ppgtt *base);
675 void gen6_ppgtt_unpin(struct i915_ppgtt *base);
676 void gen6_ppgtt_unpin_all(struct i915_ppgtt *base);
677 
678 void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv);
679 void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv);
680 
681 int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
682 					    struct sg_table *pages);
683 void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
684 			       struct sg_table *pages);
685 
686 int i915_gem_gtt_reserve(struct i915_address_space *vm,
687 			 struct drm_mm_node *node,
688 			 u64 size, u64 offset, unsigned long color,
689 			 unsigned int flags);
690 
691 int i915_gem_gtt_insert(struct i915_address_space *vm,
692 			struct drm_mm_node *node,
693 			u64 size, u64 alignment, unsigned long color,
694 			u64 start, u64 end, unsigned int flags);
695 
696 /* Flags used by pin/bind&friends. */
697 #define PIN_NONBLOCK		BIT_ULL(0)
698 #define PIN_NONFAULT		BIT_ULL(1)
699 #define PIN_NOEVICT		BIT_ULL(2)
700 #define PIN_MAPPABLE		BIT_ULL(3)
701 #define PIN_ZONE_4G		BIT_ULL(4)
702 #define PIN_HIGH		BIT_ULL(5)
703 #define PIN_OFFSET_BIAS		BIT_ULL(6)
704 #define PIN_OFFSET_FIXED	BIT_ULL(7)
705 
706 #define PIN_MBZ			BIT_ULL(8) /* I915_VMA_PIN_OVERFLOW */
707 #define PIN_GLOBAL		BIT_ULL(9) /* I915_VMA_GLOBAL_BIND */
708 #define PIN_USER		BIT_ULL(10) /* I915_VMA_LOCAL_BIND */
709 #define PIN_UPDATE		BIT_ULL(11)
710 
711 #define PIN_OFFSET_MASK		(-I915_GTT_PAGE_SIZE)
712 
713 #endif
714