Loading...
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
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/kref.h>
39#include <linux/mm.h>
40#include <linux/pagevec.h>
41#include <linux/workqueue.h>
42
43#include <drm/drm_mm.h>
44
45#include "gt/intel_reset.h"
46#include "i915_gem_fence_reg.h"
47#include "i915_request.h"
48#include "i915_scatterlist.h"
49#include "i915_selftest.h"
50#include "gt/intel_timeline.h"
51
52#define I915_GTT_PAGE_SIZE_4K BIT_ULL(12)
53#define I915_GTT_PAGE_SIZE_64K BIT_ULL(16)
54#define I915_GTT_PAGE_SIZE_2M BIT_ULL(21)
55
56#define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
57#define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
58
59#define I915_GTT_PAGE_MASK -I915_GTT_PAGE_SIZE
60
61#define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
62
63#define I915_FENCE_REG_NONE -1
64#define I915_MAX_NUM_FENCES 32
65/* 32 fences + sign bit for FENCE_REG_NONE */
66#define I915_MAX_NUM_FENCE_BITS 6
67
68struct drm_i915_file_private;
69struct drm_i915_gem_object;
70struct i915_vma;
71struct intel_gt;
72
73typedef u32 gen6_pte_t;
74typedef u64 gen8_pte_t;
75
76#define ggtt_total_entries(ggtt) ((ggtt)->vm.total >> PAGE_SHIFT)
77
78/* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
79#define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
80#define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
81#define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
82#define GEN6_PTE_CACHE_LLC (2 << 1)
83#define GEN6_PTE_UNCACHED (1 << 1)
84#define GEN6_PTE_VALID (1 << 0)
85
86#define I915_PTES(pte_len) ((unsigned int)(PAGE_SIZE / (pte_len)))
87#define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
88#define I915_PDES 512
89#define I915_PDE_MASK (I915_PDES - 1)
90#define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
91
92#define GEN6_PTES I915_PTES(sizeof(gen6_pte_t))
93#define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE)
94#define GEN6_PD_ALIGN (PAGE_SIZE * 16)
95#define GEN6_PDE_SHIFT 22
96#define GEN6_PDE_VALID (1 << 0)
97
98#define GEN7_PTE_CACHE_L3_LLC (3 << 1)
99
100#define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
101#define BYT_PTE_WRITEABLE (1 << 1)
102
103/* Cacheability Control is a 4-bit value. The low three bits are stored in bits
104 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
105 */
106#define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
107 (((bits) & 0x8) << (11 - 3)))
108#define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
109#define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
110#define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
111#define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
112#define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
113#define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
114#define HSW_PTE_UNCACHED (0)
115#define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
116#define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
117
118/*
119 * GEN8 32b style address is defined as a 3 level page table:
120 * 31:30 | 29:21 | 20:12 | 11:0
121 * PDPE | PDE | PTE | offset
122 * The difference as compared to normal x86 3 level page table is the PDPEs are
123 * programmed via register.
124 *
125 * GEN8 48b style address is defined as a 4 level page table:
126 * 47:39 | 38:30 | 29:21 | 20:12 | 11:0
127 * PML4E | PDPE | PDE | PTE | offset
128 */
129#define GEN8_3LVL_PDPES 4
130
131#define PPAT_UNCACHED (_PAGE_PWT | _PAGE_PCD)
132#define PPAT_CACHED_PDE 0 /* WB LLC */
133#define PPAT_CACHED _PAGE_PAT /* WB LLCeLLC */
134#define PPAT_DISPLAY_ELLC _PAGE_PCD /* WT eLLC */
135
136#define CHV_PPAT_SNOOP (1<<6)
137#define GEN8_PPAT_AGE(x) ((x)<<4)
138#define GEN8_PPAT_LLCeLLC (3<<2)
139#define GEN8_PPAT_LLCELLC (2<<2)
140#define GEN8_PPAT_LLC (1<<2)
141#define GEN8_PPAT_WB (3<<0)
142#define GEN8_PPAT_WT (2<<0)
143#define GEN8_PPAT_WC (1<<0)
144#define GEN8_PPAT_UC (0<<0)
145#define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
146#define GEN8_PPAT(i, x) ((u64)(x) << ((i) * 8))
147
148#define GEN8_PDE_IPS_64K BIT(11)
149#define GEN8_PDE_PS_2M BIT(7)
150
151#define for_each_sgt_dma(__dmap, __iter, __sgt) \
152 __for_each_sgt_dma(__dmap, __iter, __sgt, I915_GTT_PAGE_SIZE)
153
154struct intel_remapped_plane_info {
155 /* in gtt pages */
156 unsigned int width, height, stride, offset;
157} __packed;
158
159struct intel_remapped_info {
160 struct intel_remapped_plane_info plane[2];
161 unsigned int unused_mbz;
162} __packed;
163
164struct intel_rotation_info {
165 struct intel_remapped_plane_info plane[2];
166} __packed;
167
168struct intel_partial_info {
169 u64 offset;
170 unsigned int size;
171} __packed;
172
173enum i915_ggtt_view_type {
174 I915_GGTT_VIEW_NORMAL = 0,
175 I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
176 I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info),
177 I915_GGTT_VIEW_REMAPPED = sizeof(struct intel_remapped_info),
178};
179
180static inline void assert_i915_gem_gtt_types(void)
181{
182 BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int));
183 BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int));
184 BUILD_BUG_ON(sizeof(struct intel_remapped_info) != 9*sizeof(unsigned int));
185
186 /* Check that rotation/remapped shares offsets for simplicity */
187 BUILD_BUG_ON(offsetof(struct intel_remapped_info, plane[0]) !=
188 offsetof(struct intel_rotation_info, plane[0]));
189 BUILD_BUG_ON(offsetofend(struct intel_remapped_info, plane[1]) !=
190 offsetofend(struct intel_rotation_info, plane[1]));
191
192 /* As we encode the size of each branch inside the union into its type,
193 * we have to be careful that each branch has a unique size.
194 */
195 switch ((enum i915_ggtt_view_type)0) {
196 case I915_GGTT_VIEW_NORMAL:
197 case I915_GGTT_VIEW_PARTIAL:
198 case I915_GGTT_VIEW_ROTATED:
199 case I915_GGTT_VIEW_REMAPPED:
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 struct intel_remapped_info remapped;
212 };
213};
214
215enum i915_cache_level;
216
217struct i915_vma;
218
219struct i915_page_dma {
220 struct page *page;
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
231struct i915_page_scratch {
232 struct i915_page_dma base;
233 u64 encode;
234};
235
236struct i915_page_table {
237 struct i915_page_dma base;
238 atomic_t used;
239};
240
241struct i915_page_directory {
242 struct i915_page_table pt;
243 spinlock_t lock;
244 void *entry[512];
245};
246
247#define __px_choose_expr(x, type, expr, other) \
248 __builtin_choose_expr( \
249 __builtin_types_compatible_p(typeof(x), type) || \
250 __builtin_types_compatible_p(typeof(x), const type), \
251 ({ type __x = (type)(x); expr; }), \
252 other)
253
254#define px_base(px) \
255 __px_choose_expr(px, struct i915_page_dma *, __x, \
256 __px_choose_expr(px, struct i915_page_scratch *, &__x->base, \
257 __px_choose_expr(px, struct i915_page_table *, &__x->base, \
258 __px_choose_expr(px, struct i915_page_directory *, &__x->pt.base, \
259 (void)0))))
260#define px_dma(px) (px_base(px)->daddr)
261
262#define px_pt(px) \
263 __px_choose_expr(px, struct i915_page_table *, __x, \
264 __px_choose_expr(px, struct i915_page_directory *, &__x->pt, \
265 (void)0))
266#define px_used(px) (&px_pt(px)->used)
267
268struct i915_vma_ops {
269 /* Map an object into an address space with the given cache flags. */
270 int (*bind_vma)(struct i915_vma *vma,
271 enum i915_cache_level cache_level,
272 u32 flags);
273 /*
274 * Unmap an object from an address space. This usually consists of
275 * setting the valid PTE entries to a reserved scratch page.
276 */
277 void (*unbind_vma)(struct i915_vma *vma);
278
279 int (*set_pages)(struct i915_vma *vma);
280 void (*clear_pages)(struct i915_vma *vma);
281};
282
283struct pagestash {
284 spinlock_t lock;
285 struct pagevec pvec;
286};
287
288struct i915_address_space {
289 struct kref ref;
290 struct rcu_work rcu;
291
292 struct drm_mm mm;
293 struct intel_gt *gt;
294 struct drm_i915_private *i915;
295 struct device *dma;
296 /* Every address space belongs to a struct file - except for the global
297 * GTT that is owned by the driver (and so @file is set to NULL). In
298 * principle, no information should leak from one context to another
299 * (or between files/processes etc) unless explicitly shared by the
300 * owner. Tracking the owner is important in order to free up per-file
301 * objects along with the file, to aide resource tracking, and to
302 * assign blame.
303 */
304 struct drm_i915_file_private *file;
305 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
306 u64 reserved; /* size addr space reserved */
307
308 bool closed;
309
310 struct mutex mutex; /* protects vma and our lists */
311#define VM_CLASS_GGTT 0
312#define VM_CLASS_PPGTT 1
313
314 struct i915_page_scratch scratch[4];
315 unsigned int scratch_order;
316 unsigned int top;
317
318 /**
319 * List of vma currently bound.
320 */
321 struct list_head bound_list;
322
323 /**
324 * List of vma that are not unbound.
325 */
326 struct list_head unbound_list;
327
328 struct pagestash free_pages;
329
330 /* Global GTT */
331 bool is_ggtt:1;
332
333 /* Some systems require uncached updates of the page directories */
334 bool pt_kmap_wc:1;
335
336 /* Some systems support read-only mappings for GGTT and/or PPGTT */
337 bool has_read_only:1;
338
339 u64 (*pte_encode)(dma_addr_t addr,
340 enum i915_cache_level level,
341 u32 flags); /* Create a valid PTE */
342#define PTE_READ_ONLY (1<<0)
343
344 int (*allocate_va_range)(struct i915_address_space *vm,
345 u64 start, u64 length);
346 void (*clear_range)(struct i915_address_space *vm,
347 u64 start, u64 length);
348 void (*insert_page)(struct i915_address_space *vm,
349 dma_addr_t addr,
350 u64 offset,
351 enum i915_cache_level cache_level,
352 u32 flags);
353 void (*insert_entries)(struct i915_address_space *vm,
354 struct i915_vma *vma,
355 enum i915_cache_level cache_level,
356 u32 flags);
357 void (*cleanup)(struct i915_address_space *vm);
358
359 struct i915_vma_ops vma_ops;
360
361 I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
362 I915_SELFTEST_DECLARE(bool scrub_64K);
363};
364
365#define i915_is_ggtt(vm) ((vm)->is_ggtt)
366
367static inline bool
368i915_vm_is_4lvl(const struct i915_address_space *vm)
369{
370 return (vm->total - 1) >> 32;
371}
372
373static inline bool
374i915_vm_has_scratch_64K(struct i915_address_space *vm)
375{
376 return vm->scratch_order == get_order(I915_GTT_PAGE_SIZE_64K);
377}
378
379/* The Graphics Translation Table is the way in which GEN hardware translates a
380 * Graphics Virtual Address into a Physical Address. In addition to the normal
381 * collateral associated with any va->pa translations GEN hardware also has a
382 * portion of the GTT which can be mapped by the CPU and remain both coherent
383 * and correct (in cases like swizzling). That region is referred to as GMADR in
384 * the spec.
385 */
386struct i915_ggtt {
387 struct i915_address_space vm;
388
389 struct io_mapping iomap; /* Mapping to our CPU mappable region */
390 struct resource gmadr; /* GMADR resource */
391 resource_size_t mappable_end; /* End offset that we can CPU map */
392
393 /** "Graphics Stolen Memory" holds the global PTEs */
394 void __iomem *gsm;
395 void (*invalidate)(struct i915_ggtt *ggtt);
396
397 /** PPGTT used for aliasing the PPGTT with the GTT */
398 struct i915_ppgtt *alias;
399
400 bool do_idle_maps;
401
402 int mtrr;
403
404 u32 pin_bias;
405
406 unsigned int num_fences;
407 struct i915_fence_reg fence_regs[I915_MAX_NUM_FENCES];
408 struct list_head fence_list;
409
410 /** List of all objects in gtt_space, currently mmaped by userspace.
411 * All objects within this list must also be on bound_list.
412 */
413 struct list_head userfault_list;
414
415 /* Manual runtime pm autosuspend delay for user GGTT mmaps */
416 struct intel_wakeref_auto userfault_wakeref;
417
418 struct drm_mm_node error_capture;
419 struct drm_mm_node uc_fw;
420};
421
422struct i915_ppgtt {
423 struct i915_address_space vm;
424
425 intel_engine_mask_t pd_dirty_engines;
426 struct i915_page_directory *pd;
427};
428
429struct gen6_ppgtt {
430 struct i915_ppgtt base;
431
432 struct i915_vma *vma;
433 gen6_pte_t __iomem *pd_addr;
434
435 unsigned int pin_count;
436 bool scan_for_unused_pt;
437};
438
439#define __to_gen6_ppgtt(base) container_of(base, struct gen6_ppgtt, base)
440
441static inline struct gen6_ppgtt *to_gen6_ppgtt(struct i915_ppgtt *base)
442{
443 BUILD_BUG_ON(offsetof(struct gen6_ppgtt, base));
444 return __to_gen6_ppgtt(base);
445}
446
447/*
448 * gen6_for_each_pde() iterates over every pde from start until start+length.
449 * If start and start+length are not perfectly divisible, the macro will round
450 * down and up as needed. Start=0 and length=2G effectively iterates over
451 * every PDE in the system. The macro modifies ALL its parameters except 'pd',
452 * so each of the other parameters should preferably be a simple variable, or
453 * at most an lvalue with no side-effects!
454 */
455#define gen6_for_each_pde(pt, pd, start, length, iter) \
456 for (iter = gen6_pde_index(start); \
457 length > 0 && iter < I915_PDES && \
458 (pt = i915_pt_entry(pd, iter), true); \
459 ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT); \
460 temp = min(temp - start, length); \
461 start += temp, length -= temp; }), ++iter)
462
463#define gen6_for_all_pdes(pt, pd, iter) \
464 for (iter = 0; \
465 iter < I915_PDES && \
466 (pt = i915_pt_entry(pd, iter), true); \
467 ++iter)
468
469static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
470{
471 const u32 mask = NUM_PTE(pde_shift) - 1;
472
473 return (address >> PAGE_SHIFT) & mask;
474}
475
476/* Helper to counts the number of PTEs within the given length. This count
477 * does not cross a page table boundary, so the max value would be
478 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
479*/
480static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
481{
482 const u64 mask = ~((1ULL << pde_shift) - 1);
483 u64 end;
484
485 GEM_BUG_ON(length == 0);
486 GEM_BUG_ON(offset_in_page(addr | length));
487
488 end = addr + length;
489
490 if ((addr & mask) != (end & mask))
491 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
492
493 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
494}
495
496static inline u32 i915_pde_index(u64 addr, u32 shift)
497{
498 return (addr >> shift) & I915_PDE_MASK;
499}
500
501static inline u32 gen6_pte_index(u32 addr)
502{
503 return i915_pte_index(addr, GEN6_PDE_SHIFT);
504}
505
506static inline u32 gen6_pte_count(u32 addr, u32 length)
507{
508 return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
509}
510
511static inline u32 gen6_pde_index(u32 addr)
512{
513 return i915_pde_index(addr, GEN6_PDE_SHIFT);
514}
515
516static inline struct i915_page_table *
517i915_pt_entry(const struct i915_page_directory * const pd,
518 const unsigned short n)
519{
520 return pd->entry[n];
521}
522
523static inline struct i915_page_directory *
524i915_pd_entry(const struct i915_page_directory * const pdp,
525 const unsigned short n)
526{
527 return pdp->entry[n];
528}
529
530static inline dma_addr_t
531i915_page_dir_dma_addr(const struct i915_ppgtt *ppgtt, const unsigned int n)
532{
533 struct i915_page_dma *pt = ppgtt->pd->entry[n];
534
535 return px_dma(pt ?: px_base(&ppgtt->vm.scratch[ppgtt->vm.top]));
536}
537
538static inline struct i915_ggtt *
539i915_vm_to_ggtt(struct i915_address_space *vm)
540{
541 BUILD_BUG_ON(offsetof(struct i915_ggtt, vm));
542 GEM_BUG_ON(!i915_is_ggtt(vm));
543 return container_of(vm, struct i915_ggtt, vm);
544}
545
546static inline struct i915_ppgtt *
547i915_vm_to_ppgtt(struct i915_address_space *vm)
548{
549 BUILD_BUG_ON(offsetof(struct i915_ppgtt, vm));
550 GEM_BUG_ON(i915_is_ggtt(vm));
551 return container_of(vm, struct i915_ppgtt, vm);
552}
553
554int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv);
555int i915_ggtt_init_hw(struct drm_i915_private *dev_priv);
556int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv);
557void i915_ggtt_enable_guc(struct i915_ggtt *ggtt);
558void i915_ggtt_disable_guc(struct i915_ggtt *ggtt);
559int i915_init_ggtt(struct drm_i915_private *dev_priv);
560void i915_ggtt_driver_release(struct drm_i915_private *dev_priv);
561
562int i915_ppgtt_init_hw(struct intel_gt *gt);
563
564struct i915_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv);
565
566static inline struct i915_address_space *
567i915_vm_get(struct i915_address_space *vm)
568{
569 kref_get(&vm->ref);
570 return vm;
571}
572
573void i915_vm_release(struct kref *kref);
574
575static inline void i915_vm_put(struct i915_address_space *vm)
576{
577 kref_put(&vm->ref, i915_vm_release);
578}
579
580int gen6_ppgtt_pin(struct i915_ppgtt *base);
581void gen6_ppgtt_unpin(struct i915_ppgtt *base);
582void gen6_ppgtt_unpin_all(struct i915_ppgtt *base);
583
584void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv);
585void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv);
586
587int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
588 struct sg_table *pages);
589void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
590 struct sg_table *pages);
591
592int i915_gem_gtt_reserve(struct i915_address_space *vm,
593 struct drm_mm_node *node,
594 u64 size, u64 offset, unsigned long color,
595 unsigned int flags);
596
597int i915_gem_gtt_insert(struct i915_address_space *vm,
598 struct drm_mm_node *node,
599 u64 size, u64 alignment, unsigned long color,
600 u64 start, u64 end, unsigned int flags);
601
602/* Flags used by pin/bind&friends. */
603#define PIN_NOEVICT BIT_ULL(0)
604#define PIN_NOSEARCH BIT_ULL(1)
605#define PIN_NONBLOCK BIT_ULL(2)
606#define PIN_MAPPABLE BIT_ULL(3)
607#define PIN_ZONE_4G BIT_ULL(4)
608#define PIN_HIGH BIT_ULL(5)
609#define PIN_OFFSET_BIAS BIT_ULL(6)
610#define PIN_OFFSET_FIXED BIT_ULL(7)
611
612#define PIN_MBZ BIT_ULL(8) /* I915_VMA_PIN_OVERFLOW */
613#define PIN_GLOBAL BIT_ULL(9) /* I915_VMA_GLOBAL_BIND */
614#define PIN_USER BIT_ULL(10) /* I915_VMA_LOCAL_BIND */
615#define PIN_UPDATE BIT_ULL(11)
616
617#define PIN_OFFSET_MASK (-I915_GTT_PAGE_SIZE)
618
619#endif