xref: /openbmc/linux/drivers/gpu/drm/i915/gt/intel_gtt.h (revision 8ab59da2)
1 /* SPDX-License-Identifier: MIT */
2 /*
3  * Copyright © 2020 Intel Corporation
4  *
5  * Please try to maintain the following order within this file unless it makes
6  * sense to do otherwise. From top to bottom:
7  * 1. typedefs
8  * 2. #defines, and macros
9  * 3. structure definitions
10  * 4. function prototypes
11  *
12  * Within each section, please try to order by generation in ascending order,
13  * from top to bottom (ie. gen6 on the top, gen8 on the bottom).
14  */
15 
16 #ifndef __INTEL_GTT_H__
17 #define __INTEL_GTT_H__
18 
19 #include <linux/io-mapping.h>
20 #include <linux/kref.h>
21 #include <linux/mm.h>
22 #include <linux/pagevec.h>
23 #include <linux/scatterlist.h>
24 #include <linux/workqueue.h>
25 
26 #include <drm/drm_mm.h>
27 
28 #include "gt/intel_reset.h"
29 #include "i915_selftest.h"
30 #include "i915_vma_resource.h"
31 #include "i915_vma_types.h"
32 #include "i915_params.h"
33 #include "intel_memory_region.h"
34 
35 #define I915_GFP_ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
36 
37 #if IS_ENABLED(CONFIG_DRM_I915_TRACE_GTT)
38 #define DBG(...) trace_printk(__VA_ARGS__)
39 #else
40 #define DBG(...)
41 #endif
42 
43 #define NALLOC 3 /* 1 normal, 1 for concurrent threads, 1 for preallocation */
44 
45 #define I915_GTT_PAGE_SIZE_4K	BIT_ULL(12)
46 #define I915_GTT_PAGE_SIZE_64K	BIT_ULL(16)
47 #define I915_GTT_PAGE_SIZE_2M	BIT_ULL(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_PAGE_MASK -I915_GTT_PAGE_SIZE
53 
54 #define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
55 
56 #define I915_FENCE_REG_NONE -1
57 #define I915_MAX_NUM_FENCES 32
58 /* 32 fences + sign bit for FENCE_REG_NONE */
59 #define I915_MAX_NUM_FENCE_BITS 6
60 
61 typedef u32 gen6_pte_t;
62 typedef u64 gen8_pte_t;
63 
64 #define ggtt_total_entries(ggtt) ((ggtt)->vm.total >> PAGE_SHIFT)
65 
66 #define I915_PTES(pte_len)		((unsigned int)(PAGE_SIZE / (pte_len)))
67 #define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
68 #define I915_PDES			512
69 #define I915_PDE_MASK			(I915_PDES - 1)
70 
71 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
72 #define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
73 #define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
74 #define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
75 #define GEN6_PTE_CACHE_LLC		(2 << 1)
76 #define GEN6_PTE_UNCACHED		(1 << 1)
77 #define GEN6_PTE_VALID			REG_BIT(0)
78 
79 #define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
80 #define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
81 #define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
82 #define GEN6_PDE_SHIFT			22
83 #define GEN6_PDE_VALID			REG_BIT(0)
84 #define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
85 
86 #define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
87 
88 #define BYT_PTE_SNOOPED_BY_CPU_CACHES	REG_BIT(2)
89 #define BYT_PTE_WRITEABLE		REG_BIT(1)
90 
91 #define GEN12_PPGTT_PTE_LM	BIT_ULL(11)
92 
93 #define GEN12_GGTT_PTE_LM	BIT_ULL(1)
94 
95 #define GEN12_PDE_64K BIT(6)
96 #define GEN12_PTE_PS64 BIT(8)
97 
98 /*
99  * Cacheability Control is a 4-bit value. The low three bits are stored in bits
100  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
101  */
102 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
103 					 (((bits) & 0x8) << (11 - 3)))
104 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
105 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
106 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
107 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
108 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
109 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
110 #define HSW_PTE_UNCACHED		(0)
111 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
112 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
113 
114 /*
115  * GEN8 32b style address is defined as a 3 level page table:
116  * 31:30 | 29:21 | 20:12 |  11:0
117  * PDPE  |  PDE  |  PTE  | offset
118  * The difference as compared to normal x86 3 level page table is the PDPEs are
119  * programmed via register.
120  *
121  * GEN8 48b style address is defined as a 4 level page table:
122  * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
123  * PML4E | PDPE  |  PDE  |  PTE  | offset
124  */
125 #define GEN8_3LVL_PDPES			4
126 
127 #define PPAT_UNCACHED			(_PAGE_PWT | _PAGE_PCD)
128 #define PPAT_CACHED_PDE			0 /* WB LLC */
129 #define PPAT_CACHED			_PAGE_PAT /* WB LLCeLLC */
130 #define PPAT_DISPLAY_ELLC		_PAGE_PCD /* WT eLLC */
131 
132 #define CHV_PPAT_SNOOP			REG_BIT(6)
133 #define GEN8_PPAT_AGE(x)		((x)<<4)
134 #define GEN8_PPAT_LLCeLLC		(3<<2)
135 #define GEN8_PPAT_LLCELLC		(2<<2)
136 #define GEN8_PPAT_LLC			(1<<2)
137 #define GEN8_PPAT_WB			(3<<0)
138 #define GEN8_PPAT_WT			(2<<0)
139 #define GEN8_PPAT_WC			(1<<0)
140 #define GEN8_PPAT_UC			(0<<0)
141 #define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
142 #define GEN8_PPAT(i, x)			((u64)(x) << ((i) * 8))
143 
144 #define GEN8_PAGE_PRESENT		BIT_ULL(0)
145 #define GEN8_PAGE_RW			BIT_ULL(1)
146 
147 #define GEN8_PDE_IPS_64K BIT(11)
148 #define GEN8_PDE_PS_2M   BIT(7)
149 
150 enum i915_cache_level;
151 
152 struct drm_i915_gem_object;
153 struct i915_fence_reg;
154 struct i915_vma;
155 struct intel_gt;
156 
157 #define for_each_sgt_daddr(__dp, __iter, __sgt) \
158 	__for_each_sgt_daddr(__dp, __iter, __sgt, I915_GTT_PAGE_SIZE)
159 
160 struct i915_page_table {
161 	struct drm_i915_gem_object *base;
162 	union {
163 		atomic_t used;
164 		struct i915_page_table *stash;
165 	};
166 	bool is_compact;
167 };
168 
169 struct i915_page_directory {
170 	struct i915_page_table pt;
171 	spinlock_t lock;
172 	void **entry;
173 };
174 
175 #define __px_choose_expr(x, type, expr, other) \
176 	__builtin_choose_expr( \
177 	__builtin_types_compatible_p(typeof(x), type) || \
178 	__builtin_types_compatible_p(typeof(x), const type), \
179 	({ type __x = (type)(x); expr; }), \
180 	other)
181 
182 #define px_base(px) \
183 	__px_choose_expr(px, struct drm_i915_gem_object *, __x, \
184 	__px_choose_expr(px, struct i915_page_table *, __x->base, \
185 	__px_choose_expr(px, struct i915_page_directory *, __x->pt.base, \
186 	(void)0)))
187 
188 struct page *__px_page(struct drm_i915_gem_object *p);
189 dma_addr_t __px_dma(struct drm_i915_gem_object *p);
190 #define px_dma(px) (__px_dma(px_base(px)))
191 
192 void *__px_vaddr(struct drm_i915_gem_object *p);
193 #define px_vaddr(px) (__px_vaddr(px_base(px)))
194 
195 #define px_pt(px) \
196 	__px_choose_expr(px, struct i915_page_table *, __x, \
197 	__px_choose_expr(px, struct i915_page_directory *, &__x->pt, \
198 	(void)0))
199 #define px_used(px) (&px_pt(px)->used)
200 
201 struct i915_vm_pt_stash {
202 	/* preallocated chains of page tables/directories */
203 	struct i915_page_table *pt[2];
204 	/*
205 	 * Optionally override the alignment/size of the physical page that
206 	 * contains each PT. If not set defaults back to the usual
207 	 * I915_GTT_PAGE_SIZE_4K. This does not influence the other paging
208 	 * structures. MUST be a power-of-two. ONLY applicable on discrete
209 	 * platforms.
210 	 */
211 	int pt_sz;
212 };
213 
214 struct i915_vma_ops {
215 	/* Map an object into an address space with the given cache flags. */
216 	void (*bind_vma)(struct i915_address_space *vm,
217 			 struct i915_vm_pt_stash *stash,
218 			 struct i915_vma_resource *vma_res,
219 			 enum i915_cache_level cache_level,
220 			 u32 flags);
221 	/*
222 	 * Unmap an object from an address space. This usually consists of
223 	 * setting the valid PTE entries to a reserved scratch page.
224 	 */
225 	void (*unbind_vma)(struct i915_address_space *vm,
226 			   struct i915_vma_resource *vma_res);
227 
228 };
229 
230 struct i915_address_space {
231 	struct kref ref;
232 	struct work_struct release_work;
233 
234 	struct drm_mm mm;
235 	struct intel_gt *gt;
236 	struct drm_i915_private *i915;
237 	struct device *dma;
238 	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
239 	u64 reserved;		/* size addr space reserved */
240 	u64 min_alignment[INTEL_MEMORY_STOLEN_LOCAL + 1];
241 
242 	unsigned int bind_async_flags;
243 
244 	struct mutex mutex; /* protects vma and our lists */
245 
246 	struct kref resv_ref; /* kref to keep the reservation lock alive. */
247 	struct dma_resv _resv; /* reservation lock for all pd objects, and buffer pool */
248 #define VM_CLASS_GGTT 0
249 #define VM_CLASS_PPGTT 1
250 #define VM_CLASS_DPT 2
251 
252 	struct drm_i915_gem_object *scratch[4];
253 	/**
254 	 * List of vma currently bound.
255 	 */
256 	struct list_head bound_list;
257 
258 	/**
259 	 * List of vmas not yet bound or evicted.
260 	 */
261 	struct list_head unbound_list;
262 
263 	/* Global GTT */
264 	bool is_ggtt:1;
265 
266 	/* Display page table */
267 	bool is_dpt:1;
268 
269 	/* Some systems support read-only mappings for GGTT and/or PPGTT */
270 	bool has_read_only:1;
271 
272 	/* Skip pte rewrite on unbind for suspend. Protected by @mutex */
273 	bool skip_pte_rewrite:1;
274 
275 	u8 top;
276 	u8 pd_shift;
277 	u8 scratch_order;
278 
279 	/* Flags used when creating page-table objects for this vm */
280 	unsigned long lmem_pt_obj_flags;
281 
282 	/* Interval tree for pending unbind vma resources */
283 	struct rb_root_cached pending_unbind;
284 
285 	struct drm_i915_gem_object *
286 		(*alloc_pt_dma)(struct i915_address_space *vm, int sz);
287 	struct drm_i915_gem_object *
288 		(*alloc_scratch_dma)(struct i915_address_space *vm, int sz);
289 
290 	u64 (*pte_encode)(dma_addr_t addr,
291 			  enum i915_cache_level level,
292 			  u32 flags); /* Create a valid PTE */
293 #define PTE_READ_ONLY	BIT(0)
294 #define PTE_LM		BIT(1)
295 
296 	void (*allocate_va_range)(struct i915_address_space *vm,
297 				  struct i915_vm_pt_stash *stash,
298 				  u64 start, u64 length);
299 	void (*clear_range)(struct i915_address_space *vm,
300 			    u64 start, u64 length);
301 	void (*insert_page)(struct i915_address_space *vm,
302 			    dma_addr_t addr,
303 			    u64 offset,
304 			    enum i915_cache_level cache_level,
305 			    u32 flags);
306 	void (*insert_entries)(struct i915_address_space *vm,
307 			       struct i915_vma_resource *vma_res,
308 			       enum i915_cache_level cache_level,
309 			       u32 flags);
310 	void (*raw_insert_page)(struct i915_address_space *vm,
311 				dma_addr_t addr,
312 				u64 offset,
313 				enum i915_cache_level cache_level,
314 				u32 flags);
315 	void (*raw_insert_entries)(struct i915_address_space *vm,
316 				   struct i915_vma_resource *vma_res,
317 				   enum i915_cache_level cache_level,
318 				   u32 flags);
319 	void (*cleanup)(struct i915_address_space *vm);
320 
321 	void (*foreach)(struct i915_address_space *vm,
322 			u64 start, u64 length,
323 			void (*fn)(struct i915_address_space *vm,
324 				   struct i915_page_table *pt,
325 				   void *data),
326 			void *data);
327 
328 	struct i915_vma_ops vma_ops;
329 
330 	I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
331 	I915_SELFTEST_DECLARE(bool scrub_64K);
332 };
333 
334 /*
335  * The Graphics Translation Table is the way in which GEN hardware translates a
336  * Graphics Virtual Address into a Physical Address. In addition to the normal
337  * collateral associated with any va->pa translations GEN hardware also has a
338  * portion of the GTT which can be mapped by the CPU and remain both coherent
339  * and correct (in cases like swizzling). That region is referred to as GMADR in
340  * the spec.
341  */
342 struct i915_ggtt {
343 	struct i915_address_space vm;
344 
345 	struct io_mapping iomap;	/* Mapping to our CPU mappable region */
346 	struct resource gmadr;          /* GMADR resource */
347 	resource_size_t mappable_end;	/* End offset that we can CPU map */
348 
349 	/** "Graphics Stolen Memory" holds the global PTEs */
350 	void __iomem *gsm;
351 	void (*invalidate)(struct i915_ggtt *ggtt);
352 
353 	/** PPGTT used for aliasing the PPGTT with the GTT */
354 	struct i915_ppgtt *alias;
355 
356 	bool do_idle_maps;
357 
358 	/**
359 	 * @pte_lost: Are ptes lost on resume?
360 	 *
361 	 * Whether the system was recently restored from hibernate and
362 	 * thus may have lost pte content.
363 	 */
364 	bool pte_lost;
365 
366 	/**
367 	 * @probed_pte: Probed pte value on suspend. Re-checked on resume.
368 	 */
369 	u64 probed_pte;
370 
371 	int mtrr;
372 
373 	/** Bit 6 swizzling required for X tiling */
374 	u32 bit_6_swizzle_x;
375 	/** Bit 6 swizzling required for Y tiling */
376 	u32 bit_6_swizzle_y;
377 
378 	u32 pin_bias;
379 
380 	unsigned int num_fences;
381 	struct i915_fence_reg *fence_regs;
382 	struct list_head fence_list;
383 
384 	/**
385 	 * List of all objects in gtt_space, currently mmaped by userspace.
386 	 * All objects within this list must also be on bound_list.
387 	 */
388 	struct list_head userfault_list;
389 
390 	struct mutex error_mutex;
391 	struct drm_mm_node error_capture;
392 	struct drm_mm_node uc_fw;
393 };
394 
395 struct i915_ppgtt {
396 	struct i915_address_space vm;
397 
398 	struct i915_page_directory *pd;
399 };
400 
401 #define i915_is_ggtt(vm) ((vm)->is_ggtt)
402 #define i915_is_dpt(vm) ((vm)->is_dpt)
403 #define i915_is_ggtt_or_dpt(vm) (i915_is_ggtt(vm) || i915_is_dpt(vm))
404 
405 bool intel_vm_no_concurrent_access_wa(struct drm_i915_private *i915);
406 
407 int __must_check
408 i915_vm_lock_objects(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww);
409 
410 static inline bool
411 i915_vm_is_4lvl(const struct i915_address_space *vm)
412 {
413 	return (vm->total - 1) >> 32;
414 }
415 
416 static inline bool
417 i915_vm_has_scratch_64K(struct i915_address_space *vm)
418 {
419 	return vm->scratch_order == get_order(I915_GTT_PAGE_SIZE_64K);
420 }
421 
422 static inline u64 i915_vm_min_alignment(struct i915_address_space *vm,
423 					enum intel_memory_type type)
424 {
425 	/* avoid INTEL_MEMORY_MOCK overflow */
426 	if ((int)type >= ARRAY_SIZE(vm->min_alignment))
427 		type = INTEL_MEMORY_SYSTEM;
428 
429 	return vm->min_alignment[type];
430 }
431 
432 static inline u64 i915_vm_obj_min_alignment(struct i915_address_space *vm,
433 					    struct drm_i915_gem_object  *obj)
434 {
435 	struct intel_memory_region *mr = READ_ONCE(obj->mm.region);
436 	enum intel_memory_type type = mr ? mr->type : INTEL_MEMORY_SYSTEM;
437 
438 	return i915_vm_min_alignment(vm, type);
439 }
440 
441 static inline bool
442 i915_vm_has_cache_coloring(struct i915_address_space *vm)
443 {
444 	return i915_is_ggtt(vm) && vm->mm.color_adjust;
445 }
446 
447 static inline struct i915_ggtt *
448 i915_vm_to_ggtt(struct i915_address_space *vm)
449 {
450 	BUILD_BUG_ON(offsetof(struct i915_ggtt, vm));
451 	GEM_BUG_ON(!i915_is_ggtt(vm));
452 	return container_of(vm, struct i915_ggtt, vm);
453 }
454 
455 static inline struct i915_ppgtt *
456 i915_vm_to_ppgtt(struct i915_address_space *vm)
457 {
458 	BUILD_BUG_ON(offsetof(struct i915_ppgtt, vm));
459 	GEM_BUG_ON(i915_is_ggtt_or_dpt(vm));
460 	return container_of(vm, struct i915_ppgtt, vm);
461 }
462 
463 static inline struct i915_address_space *
464 i915_vm_get(struct i915_address_space *vm)
465 {
466 	kref_get(&vm->ref);
467 	return vm;
468 }
469 
470 static inline struct i915_address_space *
471 i915_vm_tryget(struct i915_address_space *vm)
472 {
473 	return kref_get_unless_zero(&vm->ref) ? vm : NULL;
474 }
475 
476 static inline void assert_vm_alive(struct i915_address_space *vm)
477 {
478 	GEM_BUG_ON(!kref_read(&vm->ref));
479 }
480 
481 /**
482  * i915_vm_resv_get - Obtain a reference on the vm's reservation lock
483  * @vm: The vm whose reservation lock we want to share.
484  *
485  * Return: A pointer to the vm's reservation lock.
486  */
487 static inline struct dma_resv *i915_vm_resv_get(struct i915_address_space *vm)
488 {
489 	kref_get(&vm->resv_ref);
490 	return &vm->_resv;
491 }
492 
493 void i915_vm_release(struct kref *kref);
494 
495 void i915_vm_resv_release(struct kref *kref);
496 
497 static inline void i915_vm_put(struct i915_address_space *vm)
498 {
499 	kref_put(&vm->ref, i915_vm_release);
500 }
501 
502 /**
503  * i915_vm_resv_put - Release a reference on the vm's reservation lock
504  * @resv: Pointer to a reservation lock obtained from i915_vm_resv_get()
505  */
506 static inline void i915_vm_resv_put(struct i915_address_space *vm)
507 {
508 	kref_put(&vm->resv_ref, i915_vm_resv_release);
509 }
510 
511 void i915_address_space_init(struct i915_address_space *vm, int subclass);
512 void i915_address_space_fini(struct i915_address_space *vm);
513 
514 static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
515 {
516 	const u32 mask = NUM_PTE(pde_shift) - 1;
517 
518 	return (address >> PAGE_SHIFT) & mask;
519 }
520 
521 /*
522  * Helper to counts the number of PTEs within the given length. This count
523  * does not cross a page table boundary, so the max value would be
524  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
525  */
526 static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
527 {
528 	const u64 mask = ~((1ULL << pde_shift) - 1);
529 	u64 end;
530 
531 	GEM_BUG_ON(length == 0);
532 	GEM_BUG_ON(offset_in_page(addr | length));
533 
534 	end = addr + length;
535 
536 	if ((addr & mask) != (end & mask))
537 		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
538 
539 	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
540 }
541 
542 static inline u32 i915_pde_index(u64 addr, u32 shift)
543 {
544 	return (addr >> shift) & I915_PDE_MASK;
545 }
546 
547 static inline struct i915_page_table *
548 i915_pt_entry(const struct i915_page_directory * const pd,
549 	      const unsigned short n)
550 {
551 	return pd->entry[n];
552 }
553 
554 static inline struct i915_page_directory *
555 i915_pd_entry(const struct i915_page_directory * const pdp,
556 	      const unsigned short n)
557 {
558 	return pdp->entry[n];
559 }
560 
561 static inline dma_addr_t
562 i915_page_dir_dma_addr(const struct i915_ppgtt *ppgtt, const unsigned int n)
563 {
564 	struct i915_page_table *pt = ppgtt->pd->entry[n];
565 
566 	return __px_dma(pt ? px_base(pt) : ppgtt->vm.scratch[ppgtt->vm.top]);
567 }
568 
569 void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
570 		unsigned long lmem_pt_obj_flags);
571 void intel_ggtt_bind_vma(struct i915_address_space *vm,
572 			 struct i915_vm_pt_stash *stash,
573 			 struct i915_vma_resource *vma_res,
574 			 enum i915_cache_level cache_level,
575 			 u32 flags);
576 void intel_ggtt_unbind_vma(struct i915_address_space *vm,
577 			   struct i915_vma_resource *vma_res);
578 
579 int i915_ggtt_probe_hw(struct drm_i915_private *i915);
580 int i915_ggtt_init_hw(struct drm_i915_private *i915);
581 int i915_ggtt_enable_hw(struct drm_i915_private *i915);
582 void i915_ggtt_enable_guc(struct i915_ggtt *ggtt);
583 void i915_ggtt_disable_guc(struct i915_ggtt *ggtt);
584 int i915_init_ggtt(struct drm_i915_private *i915);
585 void i915_ggtt_driver_release(struct drm_i915_private *i915);
586 void i915_ggtt_driver_late_release(struct drm_i915_private *i915);
587 
588 static inline bool i915_ggtt_has_aperture(const struct i915_ggtt *ggtt)
589 {
590 	return ggtt->mappable_end > 0;
591 }
592 
593 int i915_ppgtt_init_hw(struct intel_gt *gt);
594 
595 struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,
596 				     unsigned long lmem_pt_obj_flags);
597 
598 void i915_ggtt_suspend_vm(struct i915_address_space *vm);
599 bool i915_ggtt_resume_vm(struct i915_address_space *vm);
600 void i915_ggtt_suspend(struct i915_ggtt *gtt);
601 void i915_ggtt_resume(struct i915_ggtt *ggtt);
602 
603 /**
604  * i915_ggtt_mark_pte_lost - Mark ggtt ptes as lost or clear such a marking
605  * @i915 The device private.
606  * @val whether the ptes should be marked as lost.
607  *
608  * In some cases pte content is retained across suspend, but typically lost
609  * across hibernate. Typically they should be marked as lost on
610  * hibernation restore and such marking cleared on suspend.
611  */
612 void i915_ggtt_mark_pte_lost(struct drm_i915_private *i915, bool val);
613 
614 void
615 fill_page_dma(struct drm_i915_gem_object *p, const u64 val, unsigned int count);
616 
617 #define fill_px(px, v) fill_page_dma(px_base(px), (v), PAGE_SIZE / sizeof(u64))
618 #define fill32_px(px, v) do {						\
619 	u64 v__ = lower_32_bits(v);					\
620 	fill_px((px), v__ << 32 | v__);					\
621 } while (0)
622 
623 int setup_scratch_page(struct i915_address_space *vm);
624 void free_scratch(struct i915_address_space *vm);
625 
626 struct drm_i915_gem_object *alloc_pt_dma(struct i915_address_space *vm, int sz);
627 struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz);
628 struct i915_page_table *alloc_pt(struct i915_address_space *vm, int sz);
629 struct i915_page_directory *alloc_pd(struct i915_address_space *vm);
630 struct i915_page_directory *__alloc_pd(int npde);
631 
632 int map_pt_dma(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
633 int map_pt_dma_locked(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
634 
635 void free_px(struct i915_address_space *vm,
636 	     struct i915_page_table *pt, int lvl);
637 #define free_pt(vm, px) free_px(vm, px, 0)
638 #define free_pd(vm, px) free_px(vm, px_pt(px), 1)
639 
640 void
641 __set_pd_entry(struct i915_page_directory * const pd,
642 	       const unsigned short idx,
643 	       struct i915_page_table *pt,
644 	       u64 (*encode)(const dma_addr_t, const enum i915_cache_level));
645 
646 #define set_pd_entry(pd, idx, to) \
647 	__set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)
648 
649 void
650 clear_pd_entry(struct i915_page_directory * const pd,
651 	       const unsigned short idx,
652 	       const struct drm_i915_gem_object * const scratch);
653 
654 bool
655 release_pd_entry(struct i915_page_directory * const pd,
656 		 const unsigned short idx,
657 		 struct i915_page_table * const pt,
658 		 const struct drm_i915_gem_object * const scratch);
659 void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);
660 
661 void ppgtt_bind_vma(struct i915_address_space *vm,
662 		    struct i915_vm_pt_stash *stash,
663 		    struct i915_vma_resource *vma_res,
664 		    enum i915_cache_level cache_level,
665 		    u32 flags);
666 void ppgtt_unbind_vma(struct i915_address_space *vm,
667 		      struct i915_vma_resource *vma_res);
668 
669 void gtt_write_workarounds(struct intel_gt *gt);
670 
671 void setup_private_pat(struct intel_gt *gt);
672 
673 int i915_vm_alloc_pt_stash(struct i915_address_space *vm,
674 			   struct i915_vm_pt_stash *stash,
675 			   u64 size);
676 int i915_vm_map_pt_stash(struct i915_address_space *vm,
677 			 struct i915_vm_pt_stash *stash);
678 void i915_vm_free_pt_stash(struct i915_address_space *vm,
679 			   struct i915_vm_pt_stash *stash);
680 
681 struct i915_vma *
682 __vm_create_scratch_for_read(struct i915_address_space *vm, unsigned long size);
683 
684 struct i915_vma *
685 __vm_create_scratch_for_read_pinned(struct i915_address_space *vm, unsigned long size);
686 
687 static inline struct sgt_dma {
688 	struct scatterlist *sg;
689 	dma_addr_t dma, max;
690 } sgt_dma(struct i915_vma_resource *vma_res) {
691 	struct scatterlist *sg = vma_res->bi.pages->sgl;
692 	dma_addr_t addr = sg_dma_address(sg);
693 
694 	return (struct sgt_dma){ sg, addr, addr + sg_dma_len(sg) };
695 }
696 
697 #endif
698