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 MTL_PPGTT_PTE_PAT3	BIT_ULL(62)
92 #define GEN12_PPGTT_PTE_LM	BIT_ULL(11)
93 #define GEN12_PPGTT_PTE_PAT2	BIT_ULL(7)
94 #define GEN12_PPGTT_PTE_PAT1	BIT_ULL(4)
95 #define GEN12_PPGTT_PTE_PAT0	BIT_ULL(3)
96 
97 #define GEN12_GGTT_PTE_LM		BIT_ULL(1)
98 #define MTL_GGTT_PTE_PAT0		BIT_ULL(52)
99 #define MTL_GGTT_PTE_PAT1		BIT_ULL(53)
100 #define GEN12_GGTT_PTE_ADDR_MASK	GENMASK_ULL(45, 12)
101 #define MTL_GGTT_PTE_PAT_MASK		GENMASK_ULL(53, 52)
102 
103 #define GEN12_PDE_64K BIT(6)
104 #define GEN12_PTE_PS64 BIT(8)
105 
106 /*
107  * Cacheability Control is a 4-bit value. The low three bits are stored in bits
108  * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
109  */
110 #define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
111 					 (((bits) & 0x8) << (11 - 3)))
112 #define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
113 #define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
114 #define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
115 #define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
116 #define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
117 #define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
118 #define HSW_PTE_UNCACHED		(0)
119 #define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
120 #define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
121 
122 /*
123  * GEN8 32b style address is defined as a 3 level page table:
124  * 31:30 | 29:21 | 20:12 |  11:0
125  * PDPE  |  PDE  |  PTE  | offset
126  * The difference as compared to normal x86 3 level page table is the PDPEs are
127  * programmed via register.
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_3LVL_PDPES			4
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			REG_BIT(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_PAGE_PRESENT		BIT_ULL(0)
153 #define GEN8_PAGE_RW			BIT_ULL(1)
154 
155 #define GEN8_PDE_IPS_64K BIT(11)
156 #define GEN8_PDE_PS_2M   BIT(7)
157 
158 #define MTL_PPAT_L4_CACHE_POLICY_MASK	REG_GENMASK(3, 2)
159 #define MTL_PAT_INDEX_COH_MODE_MASK	REG_GENMASK(1, 0)
160 #define MTL_PPAT_L4_3_UC	REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 3)
161 #define MTL_PPAT_L4_1_WT	REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 1)
162 #define MTL_PPAT_L4_0_WB	REG_FIELD_PREP(MTL_PPAT_L4_CACHE_POLICY_MASK, 0)
163 #define MTL_3_COH_2W	REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 3)
164 #define MTL_2_COH_1W	REG_FIELD_PREP(MTL_PAT_INDEX_COH_MODE_MASK, 2)
165 
166 struct drm_i915_gem_object;
167 struct i915_fence_reg;
168 struct i915_vma;
169 struct intel_gt;
170 
171 #define for_each_sgt_daddr(__dp, __iter, __sgt) \
172 	__for_each_sgt_daddr(__dp, __iter, __sgt, I915_GTT_PAGE_SIZE)
173 
174 struct i915_page_table {
175 	struct drm_i915_gem_object *base;
176 	union {
177 		atomic_t used;
178 		struct i915_page_table *stash;
179 	};
180 	bool is_compact;
181 };
182 
183 struct i915_page_directory {
184 	struct i915_page_table pt;
185 	spinlock_t lock;
186 	void **entry;
187 };
188 
189 #define __px_choose_expr(x, type, expr, other) \
190 	__builtin_choose_expr( \
191 	__builtin_types_compatible_p(typeof(x), type) || \
192 	__builtin_types_compatible_p(typeof(x), const type), \
193 	({ type __x = (type)(x); expr; }), \
194 	other)
195 
196 #define px_base(px) \
197 	__px_choose_expr(px, struct drm_i915_gem_object *, __x, \
198 	__px_choose_expr(px, struct i915_page_table *, __x->base, \
199 	__px_choose_expr(px, struct i915_page_directory *, __x->pt.base, \
200 	(void)0)))
201 
202 struct page *__px_page(struct drm_i915_gem_object *p);
203 dma_addr_t __px_dma(struct drm_i915_gem_object *p);
204 #define px_dma(px) (__px_dma(px_base(px)))
205 
206 void *__px_vaddr(struct drm_i915_gem_object *p);
207 #define px_vaddr(px) (__px_vaddr(px_base(px)))
208 
209 #define px_pt(px) \
210 	__px_choose_expr(px, struct i915_page_table *, __x, \
211 	__px_choose_expr(px, struct i915_page_directory *, &__x->pt, \
212 	(void)0))
213 #define px_used(px) (&px_pt(px)->used)
214 
215 struct i915_vm_pt_stash {
216 	/* preallocated chains of page tables/directories */
217 	struct i915_page_table *pt[2];
218 	/*
219 	 * Optionally override the alignment/size of the physical page that
220 	 * contains each PT. If not set defaults back to the usual
221 	 * I915_GTT_PAGE_SIZE_4K. This does not influence the other paging
222 	 * structures. MUST be a power-of-two. ONLY applicable on discrete
223 	 * platforms.
224 	 */
225 	int pt_sz;
226 };
227 
228 struct i915_vma_ops {
229 	/* Map an object into an address space with the given cache flags. */
230 	void (*bind_vma)(struct i915_address_space *vm,
231 			 struct i915_vm_pt_stash *stash,
232 			 struct i915_vma_resource *vma_res,
233 			 unsigned int pat_index,
234 			 u32 flags);
235 	/*
236 	 * Unmap an object from an address space. This usually consists of
237 	 * setting the valid PTE entries to a reserved scratch page.
238 	 */
239 	void (*unbind_vma)(struct i915_address_space *vm,
240 			   struct i915_vma_resource *vma_res);
241 
242 };
243 
244 struct i915_address_space {
245 	struct kref ref;
246 	struct work_struct release_work;
247 
248 	struct drm_mm mm;
249 	struct intel_gt *gt;
250 	struct drm_i915_private *i915;
251 	struct device *dma;
252 	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
253 	u64 reserved;		/* size addr space reserved */
254 	u64 min_alignment[INTEL_MEMORY_STOLEN_LOCAL + 1];
255 
256 	unsigned int bind_async_flags;
257 
258 	struct mutex mutex; /* protects vma and our lists */
259 
260 	struct kref resv_ref; /* kref to keep the reservation lock alive. */
261 	struct dma_resv _resv; /* reservation lock for all pd objects, and buffer pool */
262 #define VM_CLASS_GGTT 0
263 #define VM_CLASS_PPGTT 1
264 #define VM_CLASS_DPT 2
265 
266 	struct drm_i915_gem_object *scratch[4];
267 	/**
268 	 * List of vma currently bound.
269 	 */
270 	struct list_head bound_list;
271 
272 	/**
273 	 * List of vmas not yet bound or evicted.
274 	 */
275 	struct list_head unbound_list;
276 
277 	/* Global GTT */
278 	bool is_ggtt:1;
279 
280 	/* Display page table */
281 	bool is_dpt:1;
282 
283 	/* Some systems support read-only mappings for GGTT and/or PPGTT */
284 	bool has_read_only:1;
285 
286 	/* Skip pte rewrite on unbind for suspend. Protected by @mutex */
287 	bool skip_pte_rewrite:1;
288 
289 	u8 top;
290 	u8 pd_shift;
291 	u8 scratch_order;
292 
293 	/* Flags used when creating page-table objects for this vm */
294 	unsigned long lmem_pt_obj_flags;
295 
296 	/* Interval tree for pending unbind vma resources */
297 	struct rb_root_cached pending_unbind;
298 
299 	struct drm_i915_gem_object *
300 		(*alloc_pt_dma)(struct i915_address_space *vm, int sz);
301 	struct drm_i915_gem_object *
302 		(*alloc_scratch_dma)(struct i915_address_space *vm, int sz);
303 
304 	u64 (*pte_encode)(dma_addr_t addr,
305 			  unsigned int pat_index,
306 			  u32 flags); /* Create a valid PTE */
307 #define PTE_READ_ONLY	BIT(0)
308 #define PTE_LM		BIT(1)
309 
310 	void (*allocate_va_range)(struct i915_address_space *vm,
311 				  struct i915_vm_pt_stash *stash,
312 				  u64 start, u64 length);
313 	void (*clear_range)(struct i915_address_space *vm,
314 			    u64 start, u64 length);
315 	void (*scratch_range)(struct i915_address_space *vm,
316 			      u64 start, u64 length);
317 	void (*insert_page)(struct i915_address_space *vm,
318 			    dma_addr_t addr,
319 			    u64 offset,
320 			    unsigned int pat_index,
321 			    u32 flags);
322 	void (*insert_entries)(struct i915_address_space *vm,
323 			       struct i915_vma_resource *vma_res,
324 			       unsigned int pat_index,
325 			       u32 flags);
326 	void (*raw_insert_page)(struct i915_address_space *vm,
327 				dma_addr_t addr,
328 				u64 offset,
329 				unsigned int pat_index,
330 				u32 flags);
331 	void (*raw_insert_entries)(struct i915_address_space *vm,
332 				   struct i915_vma_resource *vma_res,
333 				   unsigned int pat_index,
334 				   u32 flags);
335 	void (*cleanup)(struct i915_address_space *vm);
336 
337 	void (*foreach)(struct i915_address_space *vm,
338 			u64 start, u64 length,
339 			void (*fn)(struct i915_address_space *vm,
340 				   struct i915_page_table *pt,
341 				   void *data),
342 			void *data);
343 
344 	struct i915_vma_ops vma_ops;
345 
346 	I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
347 	I915_SELFTEST_DECLARE(bool scrub_64K);
348 };
349 
350 /*
351  * The Graphics Translation Table is the way in which GEN hardware translates a
352  * Graphics Virtual Address into a Physical Address. In addition to the normal
353  * collateral associated with any va->pa translations GEN hardware also has a
354  * portion of the GTT which can be mapped by the CPU and remain both coherent
355  * and correct (in cases like swizzling). That region is referred to as GMADR in
356  * the spec.
357  */
358 struct i915_ggtt {
359 	struct i915_address_space vm;
360 
361 	struct io_mapping iomap;	/* Mapping to our CPU mappable region */
362 	struct resource gmadr;          /* GMADR resource */
363 	resource_size_t mappable_end;	/* End offset that we can CPU map */
364 
365 	/** "Graphics Stolen Memory" holds the global PTEs */
366 	void __iomem *gsm;
367 	void (*invalidate)(struct i915_ggtt *ggtt);
368 
369 	/** PPGTT used for aliasing the PPGTT with the GTT */
370 	struct i915_ppgtt *alias;
371 
372 	bool do_idle_maps;
373 
374 	int mtrr;
375 
376 	/** Bit 6 swizzling required for X tiling */
377 	u32 bit_6_swizzle_x;
378 	/** Bit 6 swizzling required for Y tiling */
379 	u32 bit_6_swizzle_y;
380 
381 	u32 pin_bias;
382 
383 	unsigned int num_fences;
384 	struct i915_fence_reg *fence_regs;
385 	struct list_head fence_list;
386 
387 	/**
388 	 * List of all objects in gtt_space, currently mmaped by userspace.
389 	 * All objects within this list must also be on bound_list.
390 	 */
391 	struct list_head userfault_list;
392 
393 	struct mutex error_mutex;
394 	struct drm_mm_node error_capture;
395 	struct drm_mm_node uc_fw;
396 
397 	/** List of GTs mapping this GGTT */
398 	struct list_head gt_list;
399 };
400 
401 struct i915_ppgtt {
402 	struct i915_address_space vm;
403 
404 	struct i915_page_directory *pd;
405 };
406 
407 #define i915_is_ggtt(vm) ((vm)->is_ggtt)
408 #define i915_is_dpt(vm) ((vm)->is_dpt)
409 #define i915_is_ggtt_or_dpt(vm) (i915_is_ggtt(vm) || i915_is_dpt(vm))
410 
411 bool intel_vm_no_concurrent_access_wa(struct drm_i915_private *i915);
412 
413 int __must_check
414 i915_vm_lock_objects(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww);
415 
416 static inline bool
i915_vm_is_4lvl(const struct i915_address_space * vm)417 i915_vm_is_4lvl(const struct i915_address_space *vm)
418 {
419 	return (vm->total - 1) >> 32;
420 }
421 
422 static inline bool
i915_vm_has_scratch_64K(struct i915_address_space * vm)423 i915_vm_has_scratch_64K(struct i915_address_space *vm)
424 {
425 	return vm->scratch_order == get_order(I915_GTT_PAGE_SIZE_64K);
426 }
427 
i915_vm_min_alignment(struct i915_address_space * vm,enum intel_memory_type type)428 static inline u64 i915_vm_min_alignment(struct i915_address_space *vm,
429 					enum intel_memory_type type)
430 {
431 	/* avoid INTEL_MEMORY_MOCK overflow */
432 	if ((int)type >= ARRAY_SIZE(vm->min_alignment))
433 		type = INTEL_MEMORY_SYSTEM;
434 
435 	return vm->min_alignment[type];
436 }
437 
i915_vm_obj_min_alignment(struct i915_address_space * vm,struct drm_i915_gem_object * obj)438 static inline u64 i915_vm_obj_min_alignment(struct i915_address_space *vm,
439 					    struct drm_i915_gem_object  *obj)
440 {
441 	struct intel_memory_region *mr = READ_ONCE(obj->mm.region);
442 	enum intel_memory_type type = mr ? mr->type : INTEL_MEMORY_SYSTEM;
443 
444 	return i915_vm_min_alignment(vm, type);
445 }
446 
447 static inline bool
i915_vm_has_cache_coloring(struct i915_address_space * vm)448 i915_vm_has_cache_coloring(struct i915_address_space *vm)
449 {
450 	return i915_is_ggtt(vm) && vm->mm.color_adjust;
451 }
452 
453 static inline struct i915_ggtt *
i915_vm_to_ggtt(struct i915_address_space * vm)454 i915_vm_to_ggtt(struct i915_address_space *vm)
455 {
456 	BUILD_BUG_ON(offsetof(struct i915_ggtt, vm));
457 	GEM_BUG_ON(!i915_is_ggtt(vm));
458 	return container_of(vm, struct i915_ggtt, vm);
459 }
460 
461 static inline struct i915_ppgtt *
i915_vm_to_ppgtt(struct i915_address_space * vm)462 i915_vm_to_ppgtt(struct i915_address_space *vm)
463 {
464 	BUILD_BUG_ON(offsetof(struct i915_ppgtt, vm));
465 	GEM_BUG_ON(i915_is_ggtt_or_dpt(vm));
466 	return container_of(vm, struct i915_ppgtt, vm);
467 }
468 
469 static inline struct i915_address_space *
i915_vm_get(struct i915_address_space * vm)470 i915_vm_get(struct i915_address_space *vm)
471 {
472 	kref_get(&vm->ref);
473 	return vm;
474 }
475 
476 static inline struct i915_address_space *
i915_vm_tryget(struct i915_address_space * vm)477 i915_vm_tryget(struct i915_address_space *vm)
478 {
479 	return kref_get_unless_zero(&vm->ref) ? vm : NULL;
480 }
481 
assert_vm_alive(struct i915_address_space * vm)482 static inline void assert_vm_alive(struct i915_address_space *vm)
483 {
484 	GEM_BUG_ON(!kref_read(&vm->ref));
485 }
486 
487 /**
488  * i915_vm_resv_get - Obtain a reference on the vm's reservation lock
489  * @vm: The vm whose reservation lock we want to share.
490  *
491  * Return: A pointer to the vm's reservation lock.
492  */
i915_vm_resv_get(struct i915_address_space * vm)493 static inline struct dma_resv *i915_vm_resv_get(struct i915_address_space *vm)
494 {
495 	kref_get(&vm->resv_ref);
496 	return &vm->_resv;
497 }
498 
499 void i915_vm_release(struct kref *kref);
500 
501 void i915_vm_resv_release(struct kref *kref);
502 
i915_vm_put(struct i915_address_space * vm)503 static inline void i915_vm_put(struct i915_address_space *vm)
504 {
505 	kref_put(&vm->ref, i915_vm_release);
506 }
507 
508 /**
509  * i915_vm_resv_put - Release a reference on the vm's reservation lock
510  * @vm: The vm whose reservation lock reference we want to release
511  */
i915_vm_resv_put(struct i915_address_space * vm)512 static inline void i915_vm_resv_put(struct i915_address_space *vm)
513 {
514 	kref_put(&vm->resv_ref, i915_vm_resv_release);
515 }
516 
517 void i915_address_space_init(struct i915_address_space *vm, int subclass);
518 void i915_address_space_fini(struct i915_address_space *vm);
519 
i915_pte_index(u64 address,unsigned int pde_shift)520 static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
521 {
522 	const u32 mask = NUM_PTE(pde_shift) - 1;
523 
524 	return (address >> PAGE_SHIFT) & mask;
525 }
526 
527 /*
528  * Helper to counts the number of PTEs within the given length. This count
529  * does not cross a page table boundary, so the max value would be
530  * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
531  */
i915_pte_count(u64 addr,u64 length,unsigned int pde_shift)532 static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
533 {
534 	const u64 mask = ~((1ULL << pde_shift) - 1);
535 	u64 end;
536 
537 	GEM_BUG_ON(length == 0);
538 	GEM_BUG_ON(offset_in_page(addr | length));
539 
540 	end = addr + length;
541 
542 	if ((addr & mask) != (end & mask))
543 		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
544 
545 	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
546 }
547 
i915_pde_index(u64 addr,u32 shift)548 static inline u32 i915_pde_index(u64 addr, u32 shift)
549 {
550 	return (addr >> shift) & I915_PDE_MASK;
551 }
552 
553 static inline struct i915_page_table *
i915_pt_entry(const struct i915_page_directory * const pd,const unsigned short n)554 i915_pt_entry(const struct i915_page_directory * const pd,
555 	      const unsigned short n)
556 {
557 	return pd->entry[n];
558 }
559 
560 static inline struct i915_page_directory *
i915_pd_entry(const struct i915_page_directory * const pdp,const unsigned short n)561 i915_pd_entry(const struct i915_page_directory * const pdp,
562 	      const unsigned short n)
563 {
564 	return pdp->entry[n];
565 }
566 
567 static inline dma_addr_t
i915_page_dir_dma_addr(const struct i915_ppgtt * ppgtt,const unsigned int n)568 i915_page_dir_dma_addr(const struct i915_ppgtt *ppgtt, const unsigned int n)
569 {
570 	struct i915_page_table *pt = ppgtt->pd->entry[n];
571 
572 	return __px_dma(pt ? px_base(pt) : ppgtt->vm.scratch[ppgtt->vm.top]);
573 }
574 
575 void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
576 		unsigned long lmem_pt_obj_flags);
577 void intel_ggtt_bind_vma(struct i915_address_space *vm,
578 			 struct i915_vm_pt_stash *stash,
579 			 struct i915_vma_resource *vma_res,
580 			 unsigned int pat_index,
581 			 u32 flags);
582 void intel_ggtt_unbind_vma(struct i915_address_space *vm,
583 			   struct i915_vma_resource *vma_res);
584 
585 int i915_ggtt_probe_hw(struct drm_i915_private *i915);
586 int i915_ggtt_init_hw(struct drm_i915_private *i915);
587 int i915_ggtt_enable_hw(struct drm_i915_private *i915);
588 int i915_init_ggtt(struct drm_i915_private *i915);
589 void i915_ggtt_driver_release(struct drm_i915_private *i915);
590 void i915_ggtt_driver_late_release(struct drm_i915_private *i915);
591 struct i915_ggtt *i915_ggtt_create(struct drm_i915_private *i915);
592 
i915_ggtt_has_aperture(const struct i915_ggtt * ggtt)593 static inline bool i915_ggtt_has_aperture(const struct i915_ggtt *ggtt)
594 {
595 	return ggtt->mappable_end > 0;
596 }
597 
598 int i915_ppgtt_init_hw(struct intel_gt *gt);
599 
600 struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,
601 				     unsigned long lmem_pt_obj_flags);
602 
603 void i915_ggtt_suspend_vm(struct i915_address_space *vm);
604 bool i915_ggtt_resume_vm(struct i915_address_space *vm);
605 void i915_ggtt_suspend(struct i915_ggtt *gtt);
606 void i915_ggtt_resume(struct i915_ggtt *ggtt);
607 
608 void
609 fill_page_dma(struct drm_i915_gem_object *p, const u64 val, unsigned int count);
610 
611 #define fill_px(px, v) fill_page_dma(px_base(px), (v), PAGE_SIZE / sizeof(u64))
612 #define fill32_px(px, v) do {						\
613 	u64 v__ = lower_32_bits(v);					\
614 	fill_px((px), v__ << 32 | v__);					\
615 } while (0)
616 
617 int setup_scratch_page(struct i915_address_space *vm);
618 void free_scratch(struct i915_address_space *vm);
619 
620 struct drm_i915_gem_object *alloc_pt_dma(struct i915_address_space *vm, int sz);
621 struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz);
622 struct i915_page_table *alloc_pt(struct i915_address_space *vm, int sz);
623 struct i915_page_directory *alloc_pd(struct i915_address_space *vm);
624 struct i915_page_directory *__alloc_pd(int npde);
625 
626 int map_pt_dma(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
627 int map_pt_dma_locked(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
628 
629 void free_px(struct i915_address_space *vm,
630 	     struct i915_page_table *pt, int lvl);
631 #define free_pt(vm, px) free_px(vm, px, 0)
632 #define free_pd(vm, px) free_px(vm, px_pt(px), 1)
633 
634 void
635 __set_pd_entry(struct i915_page_directory * const pd,
636 	       const unsigned short idx,
637 	       struct i915_page_table *pt,
638 	       u64 (*encode)(const dma_addr_t, const enum i915_cache_level));
639 
640 #define set_pd_entry(pd, idx, to) \
641 	__set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)
642 
643 void
644 clear_pd_entry(struct i915_page_directory * const pd,
645 	       const unsigned short idx,
646 	       const struct drm_i915_gem_object * const scratch);
647 
648 bool
649 release_pd_entry(struct i915_page_directory * const pd,
650 		 const unsigned short idx,
651 		 struct i915_page_table * const pt,
652 		 const struct drm_i915_gem_object * const scratch);
653 void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);
654 
655 void ppgtt_bind_vma(struct i915_address_space *vm,
656 		    struct i915_vm_pt_stash *stash,
657 		    struct i915_vma_resource *vma_res,
658 		    unsigned int pat_index,
659 		    u32 flags);
660 void ppgtt_unbind_vma(struct i915_address_space *vm,
661 		      struct i915_vma_resource *vma_res);
662 
663 void gtt_write_workarounds(struct intel_gt *gt);
664 
665 void setup_private_pat(struct intel_gt *gt);
666 
667 int i915_vm_alloc_pt_stash(struct i915_address_space *vm,
668 			   struct i915_vm_pt_stash *stash,
669 			   u64 size);
670 int i915_vm_map_pt_stash(struct i915_address_space *vm,
671 			 struct i915_vm_pt_stash *stash);
672 void i915_vm_free_pt_stash(struct i915_address_space *vm,
673 			   struct i915_vm_pt_stash *stash);
674 
675 struct i915_vma *
676 __vm_create_scratch_for_read(struct i915_address_space *vm, unsigned long size);
677 
678 struct i915_vma *
679 __vm_create_scratch_for_read_pinned(struct i915_address_space *vm, unsigned long size);
680 
681 static inline struct sgt_dma {
682 	struct scatterlist *sg;
683 	dma_addr_t dma, max;
sgt_dma(struct i915_vma_resource * vma_res)684 } sgt_dma(struct i915_vma_resource *vma_res) {
685 	struct scatterlist *sg = vma_res->bi.pages->sgl;
686 	dma_addr_t addr = sg_dma_address(sg);
687 
688 	return (struct sgt_dma){ sg, addr, addr + sg_dma_len(sg) };
689 }
690 
691 #endif
692