Linux Audio

Check our new training course

Loading...
v4.6
  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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 37struct drm_i915_file_private;
 
 38
 39typedef uint32_t gen6_pte_t;
 40typedef uint64_t gen8_pte_t;
 41typedef uint64_t gen8_pde_t;
 42typedef uint64_t gen8_ppgtt_pdpe_t;
 43typedef uint64_t gen8_ppgtt_pml4e_t;
 44
 45#define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
 46
 47/* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
 48#define GEN6_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0xff0))
 49#define GEN6_PTE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
 50#define GEN6_PDE_ADDR_ENCODE(addr)	GEN6_GTT_ADDR_ENCODE(addr)
 51#define GEN6_PTE_CACHE_LLC		(2 << 1)
 52#define GEN6_PTE_UNCACHED		(1 << 1)
 53#define GEN6_PTE_VALID			(1 << 0)
 54
 55#define I915_PTES(pte_len)		(PAGE_SIZE / (pte_len))
 56#define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
 57#define I915_PDES			512
 58#define I915_PDE_MASK			(I915_PDES - 1)
 59#define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
 60
 61#define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
 62#define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
 63#define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
 64#define GEN6_PDE_SHIFT			22
 65#define GEN6_PDE_VALID			(1 << 0)
 66
 67#define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
 68
 69#define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
 70#define BYT_PTE_WRITEABLE		(1 << 1)
 71
 72/* Cacheability Control is a 4-bit value. The low three bits are stored in bits
 73 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
 74 */
 75#define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
 76					 (((bits) & 0x8) << (11 - 3)))
 77#define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
 78#define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
 79#define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
 80#define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
 81#define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
 82#define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
 83#define HSW_PTE_UNCACHED		(0)
 84#define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
 85#define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
 86
 87/* GEN8 legacy style address is defined as a 3 level page table:
 88 * 31:30 | 29:21 | 20:12 |  11:0
 89 * PDPE  |  PDE  |  PTE  | offset
 90 * The difference as compared to normal x86 3 level page table is the PDPEs are
 91 * programmed via register.
 92 *
 93 * GEN8 48b legacy style address is defined as a 4 level page table:
 
 
 
 
 
 
 
 94 * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
 95 * PML4E | PDPE  |  PDE  |  PTE  | offset
 96 */
 97#define GEN8_PML4ES_PER_PML4		512
 98#define GEN8_PML4E_SHIFT		39
 99#define GEN8_PML4E_MASK			(GEN8_PML4ES_PER_PML4 - 1)
100#define GEN8_PDPE_SHIFT			30
101/* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
102 * tables */
103#define GEN8_PDPE_MASK			0x1ff
104#define GEN8_PDE_SHIFT			21
105#define GEN8_PDE_MASK			0x1ff
106#define GEN8_PTE_SHIFT			12
107#define GEN8_PTE_MASK			0x1ff
108#define GEN8_LEGACY_PDPES		4
109#define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
110
111#define I915_PDPES_PER_PDP(dev) (USES_FULL_48BIT_PPGTT(dev) ?\
112				 GEN8_PML4ES_PER_PML4 : GEN8_LEGACY_PDPES)
113
114#define PPAT_UNCACHED_INDEX		(_PAGE_PWT | _PAGE_PCD)
115#define PPAT_CACHED_PDE_INDEX		0 /* WB LLC */
116#define PPAT_CACHED_INDEX		_PAGE_PAT /* WB LLCeLLC */
117#define PPAT_DISPLAY_ELLC_INDEX		_PAGE_PCD /* WT eLLC */
118
119#define CHV_PPAT_SNOOP			(1<<6)
120#define GEN8_PPAT_AGE(x)		(x<<4)
121#define GEN8_PPAT_LLCeLLC		(3<<2)
122#define GEN8_PPAT_LLCELLC		(2<<2)
123#define GEN8_PPAT_LLC			(1<<2)
124#define GEN8_PPAT_WB			(3<<0)
125#define GEN8_PPAT_WT			(2<<0)
126#define GEN8_PPAT_WC			(1<<0)
127#define GEN8_PPAT_UC			(0<<0)
128#define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
129#define GEN8_PPAT(i, x)			((uint64_t) (x) << ((i) * 8))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
130
131enum i915_ggtt_view_type {
132	I915_GGTT_VIEW_NORMAL = 0,
133	I915_GGTT_VIEW_ROTATED,
134	I915_GGTT_VIEW_PARTIAL,
135};
136
137struct intel_rotation_info {
138	unsigned int height;
139	unsigned int pitch;
140	unsigned int uv_offset;
141	uint32_t pixel_format;
142	uint64_t fb_modifier;
143	unsigned int width_pages, height_pages;
144	uint64_t size;
145	unsigned int width_pages_uv, height_pages_uv;
146	uint64_t size_uv;
147	unsigned int uv_start_page;
148};
 
149
150struct i915_ggtt_view {
151	enum i915_ggtt_view_type type;
152
153	union {
154		struct {
155			u64 offset;
156			unsigned int size;
157		} partial;
158		struct intel_rotation_info rotated;
159	} params;
160
161	struct sg_table *pages;
162};
163
164extern const struct i915_ggtt_view i915_ggtt_view_normal;
165extern const struct i915_ggtt_view i915_ggtt_view_rotated;
166
167enum i915_cache_level;
168
169/**
170 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
171 * VMA's presence cannot be guaranteed before binding, or after unbinding the
172 * object into/from the address space.
173 *
174 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
175 * will always be <= an objects lifetime. So object refcounting should cover us.
176 */
177struct i915_vma {
178	struct drm_mm_node node;
179	struct drm_i915_gem_object *obj;
180	struct i915_address_space *vm;
181
182	/** Flags and address space this VMA is bound to */
183#define GLOBAL_BIND	(1<<0)
184#define LOCAL_BIND	(1<<1)
185	unsigned int bound : 4;
186	bool is_ggtt : 1;
187
188	/**
189	 * Support different GGTT views into the same object.
190	 * This means there can be multiple VMA mappings per object and per VM.
191	 * i915_ggtt_view_type is used to distinguish between those entries.
192	 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
193	 * assumed in GEM functions which take no ggtt view parameter.
194	 */
195	struct i915_ggtt_view ggtt_view;
196
197	/** This object's place on the active/inactive lists */
198	struct list_head vm_link;
199
200	struct list_head obj_link; /* Link in the object's VMA list */
201
202	/** This vma's place in the batchbuffer or on the eviction list */
203	struct list_head exec_list;
204
205	/**
206	 * Used for performing relocations during execbuffer insertion.
207	 */
208	struct hlist_node exec_node;
209	unsigned long exec_handle;
210	struct drm_i915_gem_exec_object2 *exec_entry;
211
212	/**
213	 * How many users have pinned this object in GTT space. The following
214	 * users can each hold at most one reference: pwrite/pread, execbuffer
215	 * (objects are not allowed multiple times for the same batchbuffer),
216	 * and the framebuffer code. When switching/pageflipping, the
217	 * framebuffer code has at most two buffers pinned per crtc.
218	 *
219	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
220	 * bits with absolutely no headroom. So use 4 bits. */
221	unsigned int pin_count:4;
222#define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
223};
224
225struct i915_page_dma {
226	struct page *page;
 
227	union {
228		dma_addr_t daddr;
229
230		/* For gen6/gen7 only. This is the offset in the GGTT
231		 * where the page directory entries for PPGTT begin
232		 */
233		uint32_t ggtt_offset;
234	};
235};
236
237#define px_base(px) (&(px)->base)
238#define px_page(px) (px_base(px)->page)
239#define px_dma(px) (px_base(px)->daddr)
240
241struct i915_page_scratch {
242	struct i915_page_dma base;
243};
244
245struct i915_page_table {
246	struct i915_page_dma base;
247
248	unsigned long *used_ptes;
249};
250
251struct i915_page_directory {
252	struct i915_page_dma base;
253
254	unsigned long *used_pdes;
255	struct i915_page_table *page_table[I915_PDES]; /* PDEs */
 
256};
257
258struct i915_page_directory_pointer {
259	struct i915_page_dma base;
260
261	unsigned long *used_pdpes;
262	struct i915_page_directory **page_directory;
 
263};
264
265struct i915_pml4 {
266	struct i915_page_dma base;
267
268	DECLARE_BITMAP(used_pml4es, GEN8_PML4ES_PER_PML4);
269	struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
270};
271
272struct i915_address_space {
273	struct drm_mm mm;
274	struct drm_device *dev;
 
 
 
 
 
 
 
 
 
 
 
275	struct list_head global_link;
276	u64 start;		/* Start offset always 0 for dri2 */
277	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
 
278
279	bool is_ggtt;
280
281	struct i915_page_scratch *scratch_page;
282	struct i915_page_table *scratch_pt;
283	struct i915_page_directory *scratch_pd;
284	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
285
286	/**
287	 * List of objects currently involved in rendering.
288	 *
289	 * Includes buffers having the contents of their GPU caches
290	 * flushed, not necessarily primitives. last_read_req
291	 * represents when the rendering involved will be completed.
292	 *
293	 * A reference is held on the buffer while on this list.
294	 */
295	struct list_head active_list;
296
297	/**
298	 * LRU list of objects which are not in the ringbuffer and
299	 * are ready to unbind, but are still in the GTT.
300	 *
301	 * last_read_req is NULL while an object is in this list.
302	 *
303	 * A reference is not held on the buffer while on this list,
304	 * as merely being GTT-bound shouldn't prevent its being
305	 * freed, and we'll pull it off the list in the free path.
306	 */
307	struct list_head inactive_list;
308
 
 
 
 
 
 
 
 
 
 
309	/* FIXME: Need a more generic return type */
310	gen6_pte_t (*pte_encode)(dma_addr_t addr,
311				 enum i915_cache_level level,
312				 bool valid, u32 flags); /* Create a valid PTE */
313	/* flags for pte_encode */
314#define PTE_READ_ONLY	(1<<0)
315	int (*allocate_va_range)(struct i915_address_space *vm,
316				 uint64_t start,
317				 uint64_t length);
318	void (*clear_range)(struct i915_address_space *vm,
319			    uint64_t start,
320			    uint64_t length,
321			    bool use_scratch);
 
 
 
322	void (*insert_entries)(struct i915_address_space *vm,
323			       struct sg_table *st,
324			       uint64_t start,
325			       enum i915_cache_level cache_level, u32 flags);
326	void (*cleanup)(struct i915_address_space *vm);
327	/** Unmap an object from an address space. This usually consists of
328	 * setting the valid PTE entries to a reserved scratch page. */
329	void (*unbind_vma)(struct i915_vma *vma);
330	/* Map an object into an address space with the given cache flags. */
331	int (*bind_vma)(struct i915_vma *vma,
332			enum i915_cache_level cache_level,
333			u32 flags);
 
 
 
 
334};
335
336#define i915_is_ggtt(V) ((V)->is_ggtt)
 
 
 
 
 
 
 
 
 
 
 
 
337
338/* The Graphics Translation Table is the way in which GEN hardware translates a
339 * Graphics Virtual Address into a Physical Address. In addition to the normal
340 * collateral associated with any va->pa translations GEN hardware also has a
341 * portion of the GTT which can be mapped by the CPU and remain both coherent
342 * and correct (in cases like swizzling). That region is referred to as GMADR in
343 * the spec.
344 */
345struct i915_gtt {
346	struct i915_address_space base;
347
348	size_t stolen_size;		/* Total size of stolen memory */
349	size_t stolen_usable_size;	/* Total size minus BIOS reserved */
350	size_t stolen_reserved_base;
351	size_t stolen_reserved_size;
352	u64 mappable_end;		/* End offset that we can CPU map */
353	struct io_mapping *mappable;	/* Mapping to our CPU mappable region */
354	phys_addr_t mappable_base;	/* PA of our GMADR */
355
356	/** "Graphics Stolen Memory" holds the global PTEs */
357	void __iomem *gsm;
 
358
359	bool do_idle_maps;
360
361	int mtrr;
362
363	/* global gtt ops */
364	int (*gtt_probe)(struct drm_device *dev, u64 *gtt_total,
365			  size_t *stolen, phys_addr_t *mappable_base,
366			  u64 *mappable_end);
367};
368
369struct i915_hw_ppgtt {
370	struct i915_address_space base;
371	struct kref ref;
372	struct drm_mm_node node;
373	unsigned long pd_dirty_rings;
374	union {
375		struct i915_pml4 pml4;		/* GEN8+ & 48b PPGTT */
376		struct i915_page_directory_pointer pdp;	/* GEN8+ */
377		struct i915_page_directory pd;		/* GEN6-7 */
378	};
379
380	struct drm_i915_file_private *file_priv;
381
382	gen6_pte_t __iomem *pd_addr;
383
384	int (*enable)(struct i915_hw_ppgtt *ppgtt);
385	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
386			 struct drm_i915_gem_request *req);
387	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
388};
389
390/* For each pde iterates over every pde between from start until start + length.
391 * If start, and start+length are not perfectly divisible, the macro will round
392 * down, and up as needed. The macro modifies pde, start, and length. Dev is
393 * only used to differentiate shift values. Temp is temp.  On gen6/7, start = 0,
394 * and length = 2G effectively iterates over every PDE in the system.
395 *
396 * XXX: temp is not actually needed, but it saves doing the ALIGN operation.
397 */
398#define gen6_for_each_pde(pt, pd, start, length, temp, iter) \
399	for (iter = gen6_pde_index(start); \
400	     length > 0 && iter < I915_PDES ? \
401			(pt = (pd)->page_table[iter]), 1 : 0; \
402	     iter++, \
403	     temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT) - start, \
404	     temp = min_t(unsigned, temp, length), \
405	     start += temp, length -= temp)
406
407#define gen6_for_all_pdes(pt, ppgtt, iter)  \
408	for (iter = 0;		\
409	     pt = ppgtt->pd.page_table[iter], iter < I915_PDES;	\
410	     iter++)
411
412static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
413{
414	const uint32_t mask = NUM_PTE(pde_shift) - 1;
415
416	return (address >> PAGE_SHIFT) & mask;
417}
418
419/* Helper to counts the number of PTEs within the given length. This count
420 * does not cross a page table boundary, so the max value would be
421 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
422*/
423static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
424				      uint32_t pde_shift)
425{
426	const uint64_t mask = ~((1ULL << pde_shift) - 1);
427	uint64_t end;
428
429	WARN_ON(length == 0);
430	WARN_ON(offset_in_page(addr|length));
431
432	end = addr + length;
433
434	if ((addr & mask) != (end & mask))
435		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
436
437	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
438}
439
440static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
441{
442	return (addr >> shift) & I915_PDE_MASK;
443}
444
445static inline uint32_t gen6_pte_index(uint32_t addr)
446{
447	return i915_pte_index(addr, GEN6_PDE_SHIFT);
448}
449
450static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
451{
452	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
453}
454
455static inline uint32_t gen6_pde_index(uint32_t addr)
456{
457	return i915_pde_index(addr, GEN6_PDE_SHIFT);
458}
459
 
 
 
 
 
 
 
 
 
460/* Equivalent to the gen6 version, For each pde iterates over every pde
461 * between from start until start + length. On gen8+ it simply iterates
462 * over every page directory entry in a page directory.
463 */
464#define gen8_for_each_pde(pt, pd, start, length, iter)			\
465	for (iter = gen8_pde_index(start);				\
466	     length > 0 && iter < I915_PDES &&				\
467		(pt = (pd)->page_table[iter], true);			\
468	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT);		\
469		    temp = min(temp - start, length);			\
470		    start += temp, length -= temp; }), ++iter)
471
472#define gen8_for_each_pdpe(pd, pdp, start, length, iter)		\
473	for (iter = gen8_pdpe_index(start);				\
474	     length > 0 && iter < I915_PDPES_PER_PDP(dev) &&		\
475		(pd = (pdp)->page_directory[iter], true);		\
476	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT);	\
477		    temp = min(temp - start, length);			\
478		    start += temp, length -= temp; }), ++iter)
479
480#define gen8_for_each_pml4e(pdp, pml4, start, length, iter)		\
481	for (iter = gen8_pml4e_index(start);				\
482	     length > 0 && iter < GEN8_PML4ES_PER_PML4 &&		\
483		(pdp = (pml4)->pdps[iter], true);			\
484	     ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT);	\
485		    temp = min(temp - start, length);			\
486		    start += temp, length -= temp; }), ++iter)
487
488static inline uint32_t gen8_pte_index(uint64_t address)
489{
490	return i915_pte_index(address, GEN8_PDE_SHIFT);
491}
492
493static inline uint32_t gen8_pde_index(uint64_t address)
494{
495	return i915_pde_index(address, GEN8_PDE_SHIFT);
496}
497
498static inline uint32_t gen8_pdpe_index(uint64_t address)
499{
500	return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
501}
502
503static inline uint32_t gen8_pml4e_index(uint64_t address)
504{
505	return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
506}
507
508static inline size_t gen8_pte_count(uint64_t address, uint64_t length)
509{
510	return i915_pte_count(address, length, GEN8_PDE_SHIFT);
511}
512
513static inline dma_addr_t
514i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
515{
516	return test_bit(n, ppgtt->pdp.used_pdpes) ?
517		px_dma(ppgtt->pdp.page_directory[n]) :
518		px_dma(ppgtt->base.scratch_pd);
 
 
 
 
 
519}
520
521int i915_gem_gtt_init(struct drm_device *dev);
522void i915_gem_init_global_gtt(struct drm_device *dev);
523void i915_global_gtt_cleanup(struct drm_device *dev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
524
 
 
525
526int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
527int i915_ppgtt_init_hw(struct drm_device *dev);
528int i915_ppgtt_init_ring(struct drm_i915_gem_request *req);
 
 
 
 
 
 
529void i915_ppgtt_release(struct kref *kref);
530struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
531					struct drm_i915_file_private *fpriv);
 
 
532static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
533{
534	if (ppgtt)
535		kref_get(&ppgtt->ref);
536}
537static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
538{
539	if (ppgtt)
540		kref_put(&ppgtt->ref, i915_ppgtt_release);
541}
542
543void i915_check_and_clear_faults(struct drm_device *dev);
544void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
545void i915_gem_restore_gtt_mappings(struct drm_device *dev);
546
547int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
548void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
549
550static inline bool
551i915_ggtt_view_equal(const struct i915_ggtt_view *a,
552                     const struct i915_ggtt_view *b)
553{
554	if (WARN_ON(!a || !b))
555		return false;
556
557	if (a->type != b->type)
558		return false;
559	if (a->type != I915_GGTT_VIEW_NORMAL)
560		return !memcmp(&a->params, &b->params, sizeof(a->params));
561	return true;
562}
563
564size_t
565i915_ggtt_view_size(struct drm_i915_gem_object *obj,
566		    const struct i915_ggtt_view *view);
 
 
 
 
 
 
 
 
 
 
 
567
568#endif
v4.17
  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
 43#include "i915_request.h"
 44#include "i915_selftest.h"
 45
 46#define I915_GTT_PAGE_SIZE_4K BIT(12)
 47#define I915_GTT_PAGE_SIZE_64K BIT(16)
 48#define I915_GTT_PAGE_SIZE_2M BIT(21)
 49
 50#define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
 51#define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
 52
 53#define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
 54
 55#define I915_FENCE_REG_NONE -1
 56#define I915_MAX_NUM_FENCES 32
 57/* 32 fences + sign bit for FENCE_REG_NONE */
 58#define I915_MAX_NUM_FENCE_BITS 6
 59
 60struct drm_i915_file_private;
 61struct drm_i915_fence_reg;
 62
 63typedef u32 gen6_pte_t;
 64typedef u64 gen8_pte_t;
 65typedef u64 gen8_pde_t;
 66typedef u64 gen8_ppgtt_pdpe_t;
 67typedef u64 gen8_ppgtt_pml4e_t;
 68
 69#define ggtt_total_entries(ggtt) ((ggtt)->base.total >> PAGE_SHIFT)
 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			(1 << 0)
 78
 79#define I915_PTES(pte_len)		((unsigned int)(PAGE_SIZE / (pte_len)))
 80#define I915_PTE_MASK(pte_len)		(I915_PTES(pte_len) - 1)
 81#define I915_PDES			512
 82#define I915_PDE_MASK			(I915_PDES - 1)
 83#define NUM_PTE(pde_shift)     (1 << (pde_shift - PAGE_SHIFT))
 84
 85#define GEN6_PTES			I915_PTES(sizeof(gen6_pte_t))
 86#define GEN6_PD_SIZE		        (I915_PDES * PAGE_SIZE)
 87#define GEN6_PD_ALIGN			(PAGE_SIZE * 16)
 88#define GEN6_PDE_SHIFT			22
 89#define GEN6_PDE_VALID			(1 << 0)
 90
 91#define GEN7_PTE_CACHE_L3_LLC		(3 << 1)
 92
 93#define BYT_PTE_SNOOPED_BY_CPU_CACHES	(1 << 2)
 94#define BYT_PTE_WRITEABLE		(1 << 1)
 95
 96/* Cacheability Control is a 4-bit value. The low three bits are stored in bits
 97 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
 98 */
 99#define HSW_CACHEABILITY_CONTROL(bits)	((((bits) & 0x7) << 1) | \
100					 (((bits) & 0x8) << (11 - 3)))
101#define HSW_WB_LLC_AGE3			HSW_CACHEABILITY_CONTROL(0x2)
102#define HSW_WB_LLC_AGE0			HSW_CACHEABILITY_CONTROL(0x3)
103#define HSW_WB_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x8)
104#define HSW_WB_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0xb)
105#define HSW_WT_ELLC_LLC_AGE3		HSW_CACHEABILITY_CONTROL(0x7)
106#define HSW_WT_ELLC_LLC_AGE0		HSW_CACHEABILITY_CONTROL(0x6)
107#define HSW_PTE_UNCACHED		(0)
108#define HSW_GTT_ADDR_ENCODE(addr)	((addr) | (((addr) >> 28) & 0x7f0))
109#define HSW_PTE_ADDR_ENCODE(addr)	HSW_GTT_ADDR_ENCODE(addr)
110
111/* GEN8 32b style address is defined as a 3 level page table:
112 * 31:30 | 29:21 | 20:12 |  11:0
113 * PDPE  |  PDE  |  PTE  | offset
114 * The difference as compared to normal x86 3 level page table is the PDPEs are
115 * programmed via register.
116 */
117#define GEN8_3LVL_PDPES			4
118#define GEN8_PDE_SHIFT			21
119#define GEN8_PDE_MASK			0x1ff
120#define GEN8_PTE_SHIFT			12
121#define GEN8_PTE_MASK			0x1ff
122#define GEN8_PTES			I915_PTES(sizeof(gen8_pte_t))
123
124/* GEN8 48b style address is defined as a 4 level page table:
125 * 47:39 | 38:30 | 29:21 | 20:12 |  11:0
126 * PML4E | PDPE  |  PDE  |  PTE  | offset
127 */
128#define GEN8_PML4ES_PER_PML4		512
129#define GEN8_PML4E_SHIFT		39
130#define GEN8_PML4E_MASK			(GEN8_PML4ES_PER_PML4 - 1)
131#define GEN8_PDPE_SHIFT			30
132/* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
133 * tables */
134#define GEN8_PDPE_MASK			0x1ff
 
 
 
 
 
 
 
 
 
135
136#define PPAT_UNCACHED			(_PAGE_PWT | _PAGE_PCD)
137#define PPAT_CACHED_PDE			0 /* WB LLC */
138#define PPAT_CACHED			_PAGE_PAT /* WB LLCeLLC */
139#define PPAT_DISPLAY_ELLC		_PAGE_PCD /* WT eLLC */
140
141#define CHV_PPAT_SNOOP			(1<<6)
142#define GEN8_PPAT_AGE(x)		((x)<<4)
143#define GEN8_PPAT_LLCeLLC		(3<<2)
144#define GEN8_PPAT_LLCELLC		(2<<2)
145#define GEN8_PPAT_LLC			(1<<2)
146#define GEN8_PPAT_WB			(3<<0)
147#define GEN8_PPAT_WT			(2<<0)
148#define GEN8_PPAT_WC			(1<<0)
149#define GEN8_PPAT_UC			(0<<0)
150#define GEN8_PPAT_ELLC_OVERRIDE		(0<<2)
151#define GEN8_PPAT(i, x)			((u64)(x) << ((i) * 8))
152
153#define GEN8_PPAT_GET_CA(x) ((x) & 3)
154#define GEN8_PPAT_GET_TC(x) ((x) & (3 << 2))
155#define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4))
156#define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6))
157
158#define GEN8_PDE_IPS_64K BIT(11)
159#define GEN8_PDE_PS_2M   BIT(7)
160
161struct sg_table;
162
163struct intel_rotation_info {
164	struct intel_rotation_plane_info {
165		/* tiles */
166		unsigned int width, height, stride, offset;
167	} plane[2];
168} __packed;
169
170static inline void assert_intel_rotation_info_is_packed(void)
171{
172	BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int));
173}
174
175struct intel_partial_info {
176	u64 offset;
177	unsigned int size;
178} __packed;
179
180static inline void assert_intel_partial_info_is_packed(void)
181{
182	BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int));
183}
184
185enum i915_ggtt_view_type {
186	I915_GGTT_VIEW_NORMAL = 0,
187	I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
188	I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info),
189};
190
191static inline void assert_i915_ggtt_view_type_is_unique(void)
192{
193	/* As we encode the size of each branch inside the union into its type,
194	 * we have to be careful that each branch has a unique size.
195	 */
196	switch ((enum i915_ggtt_view_type)0) {
197	case I915_GGTT_VIEW_NORMAL:
198	case I915_GGTT_VIEW_PARTIAL:
199	case I915_GGTT_VIEW_ROTATED:
200		/* gcc complains if these are identical cases */
201		break;
202	}
203}
204
205struct i915_ggtt_view {
206	enum i915_ggtt_view_type type;
 
207	union {
208		/* Members need to contain no holes/padding */
209		struct intel_partial_info partial;
 
 
210		struct intel_rotation_info rotated;
211	};
 
 
212};
213
 
 
 
214enum i915_cache_level;
215
216struct i915_vma;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
217
218struct i915_page_dma {
219	struct page *page;
220	int order;
221	union {
222		dma_addr_t daddr;
223
224		/* For gen6/gen7 only. This is the offset in the GGTT
225		 * where the page directory entries for PPGTT begin
226		 */
227		u32 ggtt_offset;
228	};
229};
230
231#define px_base(px) (&(px)->base)
232#define px_page(px) (px_base(px)->page)
233#define px_dma(px) (px_base(px)->daddr)
234
 
 
 
 
235struct i915_page_table {
236	struct i915_page_dma base;
237	unsigned int used_ptes;
 
238};
239
240struct i915_page_directory {
241	struct i915_page_dma base;
242
 
243	struct i915_page_table *page_table[I915_PDES]; /* PDEs */
244	unsigned int used_pdes;
245};
246
247struct i915_page_directory_pointer {
248	struct i915_page_dma base;
 
 
249	struct i915_page_directory **page_directory;
250	unsigned int used_pdpes;
251};
252
253struct i915_pml4 {
254	struct i915_page_dma base;
 
 
255	struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
256};
257
258struct i915_address_space {
259	struct drm_mm mm;
260	struct i915_gem_timeline timeline;
261	struct drm_i915_private *i915;
262	struct device *dma;
263	/* Every address space belongs to a struct file - except for the global
264	 * GTT that is owned by the driver (and so @file is set to NULL). In
265	 * principle, no information should leak from one context to another
266	 * (or between files/processes etc) unless explicitly shared by the
267	 * owner. Tracking the owner is important in order to free up per-file
268	 * objects along with the file, to aide resource tracking, and to
269	 * assign blame.
270	 */
271	struct drm_i915_file_private *file;
272	struct list_head global_link;
 
273	u64 total;		/* size addr space maps (ex. 2GB for ggtt) */
274	u64 reserved;		/* size addr space reserved */
275
276	bool closed;
277
278	struct i915_page_dma scratch_page;
279	struct i915_page_table *scratch_pt;
280	struct i915_page_directory *scratch_pd;
281	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
282
283	/**
284	 * List of objects currently involved in rendering.
285	 *
286	 * Includes buffers having the contents of their GPU caches
287	 * flushed, not necessarily primitives. last_read_req
288	 * represents when the rendering involved will be completed.
289	 *
290	 * A reference is held on the buffer while on this list.
291	 */
292	struct list_head active_list;
293
294	/**
295	 * LRU list of objects which are not in the ringbuffer and
296	 * are ready to unbind, but are still in the GTT.
297	 *
298	 * last_read_req is NULL while an object is in this list.
299	 *
300	 * A reference is not held on the buffer while on this list,
301	 * as merely being GTT-bound shouldn't prevent its being
302	 * freed, and we'll pull it off the list in the free path.
303	 */
304	struct list_head inactive_list;
305
306	/**
307	 * List of vma that have been unbound.
308	 *
309	 * A reference is not held on the buffer while on this list.
310	 */
311	struct list_head unbound_list;
312
313	struct pagevec free_pages;
314	bool pt_kmap_wc;
315
316	/* FIXME: Need a more generic return type */
317	gen6_pte_t (*pte_encode)(dma_addr_t addr,
318				 enum i915_cache_level level,
319				 u32 flags); /* Create a valid PTE */
320	/* flags for pte_encode */
321#define PTE_READ_ONLY	(1<<0)
322	int (*allocate_va_range)(struct i915_address_space *vm,
323				 u64 start, u64 length);
 
324	void (*clear_range)(struct i915_address_space *vm,
325			    u64 start, u64 length);
326	void (*insert_page)(struct i915_address_space *vm,
327			    dma_addr_t addr,
328			    u64 offset,
329			    enum i915_cache_level cache_level,
330			    u32 flags);
331	void (*insert_entries)(struct i915_address_space *vm,
332			       struct i915_vma *vma,
333			       enum i915_cache_level cache_level,
334			       u32 flags);
335	void (*cleanup)(struct i915_address_space *vm);
336	/** Unmap an object from an address space. This usually consists of
337	 * setting the valid PTE entries to a reserved scratch page. */
338	void (*unbind_vma)(struct i915_vma *vma);
339	/* Map an object into an address space with the given cache flags. */
340	int (*bind_vma)(struct i915_vma *vma,
341			enum i915_cache_level cache_level,
342			u32 flags);
343	int (*set_pages)(struct i915_vma *vma);
344	void (*clear_pages)(struct i915_vma *vma);
345
346	I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
347};
348
349#define i915_is_ggtt(V) (!(V)->file)
350
351static inline bool
352i915_vm_is_48bit(const struct i915_address_space *vm)
353{
354	return (vm->total - 1) >> 32;
355}
356
357static inline bool
358i915_vm_has_scratch_64K(struct i915_address_space *vm)
359{
360	return vm->scratch_page.order == get_order(I915_GTT_PAGE_SIZE_64K);
361}
362
363/* The Graphics Translation Table is the way in which GEN hardware translates a
364 * Graphics Virtual Address into a Physical Address. In addition to the normal
365 * collateral associated with any va->pa translations GEN hardware also has a
366 * portion of the GTT which can be mapped by the CPU and remain both coherent
367 * and correct (in cases like swizzling). That region is referred to as GMADR in
368 * the spec.
369 */
370struct i915_ggtt {
371	struct i915_address_space base;
372
373	struct io_mapping iomap;	/* Mapping to our CPU mappable region */
374	struct resource gmadr;          /* GMADR resource */
375	resource_size_t mappable_end;	/* End offset that we can CPU map */
 
 
 
 
376
377	/** "Graphics Stolen Memory" holds the global PTEs */
378	void __iomem *gsm;
379	void (*invalidate)(struct drm_i915_private *dev_priv);
380
381	bool do_idle_maps;
382
383	int mtrr;
384
385	struct drm_mm_node error_capture;
 
 
 
386};
387
388struct i915_hw_ppgtt {
389	struct i915_address_space base;
390	struct kref ref;
391	struct drm_mm_node node;
392	unsigned long pd_dirty_rings;
393	union {
394		struct i915_pml4 pml4;		/* GEN8+ & 48b PPGTT */
395		struct i915_page_directory_pointer pdp;	/* GEN8+ */
396		struct i915_page_directory pd;		/* GEN6-7 */
397	};
398
 
 
399	gen6_pte_t __iomem *pd_addr;
400
 
401	int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
402			 struct i915_request *rq);
403	void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
404};
405
406/*
407 * gen6_for_each_pde() iterates over every pde from start until start+length.
408 * If start and start+length are not perfectly divisible, the macro will round
409 * down and up as needed. Start=0 and length=2G effectively iterates over
410 * every PDE in the system. The macro modifies ALL its parameters except 'pd',
411 * so each of the other parameters should preferably be a simple variable, or
412 * at most an lvalue with no side-effects!
413 */
414#define gen6_for_each_pde(pt, pd, start, length, iter)			\
415	for (iter = gen6_pde_index(start);				\
416	     length > 0 && iter < I915_PDES &&				\
417		(pt = (pd)->page_table[iter], true);			\
418	     ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT);		\
419		    temp = min(temp - start, length);			\
420		    start += temp, length -= temp; }), ++iter)
421
422#define gen6_for_all_pdes(pt, pd, iter)					\
423	for (iter = 0;							\
424	     iter < I915_PDES &&					\
425		(pt = (pd)->page_table[iter], true);			\
426	     ++iter)
427
428static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
429{
430	const u32 mask = NUM_PTE(pde_shift) - 1;
431
432	return (address >> PAGE_SHIFT) & mask;
433}
434
435/* Helper to counts the number of PTEs within the given length. This count
436 * does not cross a page table boundary, so the max value would be
437 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
438*/
439static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
 
440{
441	const u64 mask = ~((1ULL << pde_shift) - 1);
442	u64 end;
443
444	WARN_ON(length == 0);
445	WARN_ON(offset_in_page(addr|length));
446
447	end = addr + length;
448
449	if ((addr & mask) != (end & mask))
450		return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
451
452	return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
453}
454
455static inline u32 i915_pde_index(u64 addr, u32 shift)
456{
457	return (addr >> shift) & I915_PDE_MASK;
458}
459
460static inline u32 gen6_pte_index(u32 addr)
461{
462	return i915_pte_index(addr, GEN6_PDE_SHIFT);
463}
464
465static inline u32 gen6_pte_count(u32 addr, u32 length)
466{
467	return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
468}
469
470static inline u32 gen6_pde_index(u32 addr)
471{
472	return i915_pde_index(addr, GEN6_PDE_SHIFT);
473}
474
475static inline unsigned int
476i915_pdpes_per_pdp(const struct i915_address_space *vm)
477{
478	if (i915_vm_is_48bit(vm))
479		return GEN8_PML4ES_PER_PML4;
480
481	return GEN8_3LVL_PDPES;
482}
483
484/* Equivalent to the gen6 version, For each pde iterates over every pde
485 * between from start until start + length. On gen8+ it simply iterates
486 * over every page directory entry in a page directory.
487 */
488#define gen8_for_each_pde(pt, pd, start, length, iter)			\
489	for (iter = gen8_pde_index(start);				\
490	     length > 0 && iter < I915_PDES &&				\
491		(pt = (pd)->page_table[iter], true);			\
492	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT);		\
493		    temp = min(temp - start, length);			\
494		    start += temp, length -= temp; }), ++iter)
495
496#define gen8_for_each_pdpe(pd, pdp, start, length, iter)		\
497	for (iter = gen8_pdpe_index(start);				\
498	     length > 0 && iter < i915_pdpes_per_pdp(vm) &&		\
499		(pd = (pdp)->page_directory[iter], true);		\
500	     ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT);	\
501		    temp = min(temp - start, length);			\
502		    start += temp, length -= temp; }), ++iter)
503
504#define gen8_for_each_pml4e(pdp, pml4, start, length, iter)		\
505	for (iter = gen8_pml4e_index(start);				\
506	     length > 0 && iter < GEN8_PML4ES_PER_PML4 &&		\
507		(pdp = (pml4)->pdps[iter], true);			\
508	     ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT);	\
509		    temp = min(temp - start, length);			\
510		    start += temp, length -= temp; }), ++iter)
511
512static inline u32 gen8_pte_index(u64 address)
513{
514	return i915_pte_index(address, GEN8_PDE_SHIFT);
515}
516
517static inline u32 gen8_pde_index(u64 address)
518{
519	return i915_pde_index(address, GEN8_PDE_SHIFT);
520}
521
522static inline u32 gen8_pdpe_index(u64 address)
523{
524	return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
525}
526
527static inline u32 gen8_pml4e_index(u64 address)
528{
529	return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK;
530}
531
532static inline u64 gen8_pte_count(u64 address, u64 length)
533{
534	return i915_pte_count(address, length, GEN8_PDE_SHIFT);
535}
536
537static inline dma_addr_t
538i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
539{
540	return px_dma(ppgtt->pdp.page_directory[n]);
541}
542
543static inline struct i915_ggtt *
544i915_vm_to_ggtt(struct i915_address_space *vm)
545{
546	GEM_BUG_ON(!i915_is_ggtt(vm));
547	return container_of(vm, struct i915_ggtt, base);
548}
549
550#define INTEL_MAX_PPAT_ENTRIES 8
551#define INTEL_PPAT_PERFECT_MATCH (~0U)
552
553struct intel_ppat;
554
555struct intel_ppat_entry {
556	struct intel_ppat *ppat;
557	struct kref ref;
558	u8 value;
559};
560
561struct intel_ppat {
562	struct intel_ppat_entry entries[INTEL_MAX_PPAT_ENTRIES];
563	DECLARE_BITMAP(used, INTEL_MAX_PPAT_ENTRIES);
564	DECLARE_BITMAP(dirty, INTEL_MAX_PPAT_ENTRIES);
565	unsigned int max_entries;
566	u8 clear_value;
567	/*
568	 * Return a score to show how two PPAT values match,
569	 * a INTEL_PPAT_PERFECT_MATCH indicates a perfect match
570	 */
571	unsigned int (*match)(u8 src, u8 dst);
572	void (*update_hw)(struct drm_i915_private *i915);
573
574	struct drm_i915_private *i915;
575};
576
577const struct intel_ppat_entry *
578intel_ppat_get(struct drm_i915_private *i915, u8 value);
579void intel_ppat_put(const struct intel_ppat_entry *entry);
580
581int i915_gem_init_aliasing_ppgtt(struct drm_i915_private *i915);
582void i915_gem_fini_aliasing_ppgtt(struct drm_i915_private *i915);
583
584int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
585int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
586int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
587void i915_ggtt_enable_guc(struct drm_i915_private *i915);
588void i915_ggtt_disable_guc(struct drm_i915_private *i915);
589int i915_gem_init_ggtt(struct drm_i915_private *dev_priv);
590void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv);
591
592int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv);
593void i915_ppgtt_release(struct kref *kref);
594struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv,
595					struct drm_i915_file_private *fpriv,
596					const char *name);
597void i915_ppgtt_close(struct i915_address_space *vm);
598static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
599{
600	if (ppgtt)
601		kref_get(&ppgtt->ref);
602}
603static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
604{
605	if (ppgtt)
606		kref_put(&ppgtt->ref, i915_ppgtt_release);
607}
608
609void i915_check_and_clear_faults(struct drm_i915_private *dev_priv);
610void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv);
611void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv);
612
613int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
614					    struct sg_table *pages);
615void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
616			       struct sg_table *pages);
617
618int i915_gem_gtt_reserve(struct i915_address_space *vm,
619			 struct drm_mm_node *node,
620			 u64 size, u64 offset, unsigned long color,
621			 unsigned int flags);
622
623int i915_gem_gtt_insert(struct i915_address_space *vm,
624			struct drm_mm_node *node,
625			u64 size, u64 alignment, unsigned long color,
626			u64 start, u64 end, unsigned int flags);
627
628/* Flags used by pin/bind&friends. */
629#define PIN_NONBLOCK		BIT(0)
630#define PIN_MAPPABLE		BIT(1)
631#define PIN_ZONE_4G		BIT(2)
632#define PIN_NONFAULT		BIT(3)
633#define PIN_NOEVICT		BIT(4)
634
635#define PIN_MBZ			BIT(5) /* I915_VMA_PIN_OVERFLOW */
636#define PIN_GLOBAL		BIT(6) /* I915_VMA_GLOBAL_BIND */
637#define PIN_USER		BIT(7) /* I915_VMA_LOCAL_BIND */
638#define PIN_UPDATE		BIT(8)
639
640#define PIN_HIGH		BIT(9)
641#define PIN_OFFSET_BIAS		BIT(10)
642#define PIN_OFFSET_FIXED	BIT(11)
643#define PIN_OFFSET_MASK		(-I915_GTT_PAGE_SIZE)
644
645#endif