drm/i915/gen8: Add PML4 structure
[deliverable/linux.git] / drivers / gpu / drm / i915 / i915_gem_gtt.h
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 struct drm_i915_file_private;
38
39 typedef uint32_t gen6_pte_t;
40 typedef uint64_t gen8_pte_t;
41 typedef uint64_t gen8_pde_t;
42
43 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
44
45
46 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
47 #define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
48 #define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
49 #define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
50 #define GEN6_PTE_CACHE_LLC (2 << 1)
51 #define GEN6_PTE_UNCACHED (1 << 1)
52 #define GEN6_PTE_VALID (1 << 0)
53
54 #define I915_PTES(pte_len) (PAGE_SIZE / (pte_len))
55 #define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
56 #define I915_PDES 512
57 #define I915_PDE_MASK (I915_PDES - 1)
58 #define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
59
60 #define GEN6_PTES I915_PTES(sizeof(gen6_pte_t))
61 #define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE)
62 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
63 #define GEN6_PDE_SHIFT 22
64 #define GEN6_PDE_VALID (1 << 0)
65
66 #define GEN7_PTE_CACHE_L3_LLC (3 << 1)
67
68 #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
69 #define BYT_PTE_WRITEABLE (1 << 1)
70
71 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits
72 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
73 */
74 #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
75 (((bits) & 0x8) << (11 - 3)))
76 #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
77 #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
78 #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
79 #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
80 #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
81 #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
82 #define HSW_PTE_UNCACHED (0)
83 #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
84 #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
85
86 /* GEN8 legacy style address is defined as a 3 level page table:
87 * 31:30 | 29:21 | 20:12 | 11:0
88 * PDPE | PDE | PTE | offset
89 * The difference as compared to normal x86 3 level page table is the PDPEs are
90 * programmed via register.
91 *
92 * GEN8 48b legacy style address is defined as a 4 level page table:
93 * 47:39 | 38:30 | 29:21 | 20:12 | 11:0
94 * PML4E | PDPE | PDE | PTE | offset
95 */
96 #define GEN8_PML4ES_PER_PML4 512
97 #define GEN8_PML4E_SHIFT 39
98 #define GEN8_PDPE_SHIFT 30
99 /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page
100 * tables */
101 #define GEN8_PDPE_MASK 0x1ff
102 #define GEN8_PDE_SHIFT 21
103 #define GEN8_PDE_MASK 0x1ff
104 #define GEN8_PTE_SHIFT 12
105 #define GEN8_PTE_MASK 0x1ff
106 #define GEN8_LEGACY_PDPES 4
107 #define GEN8_PTES I915_PTES(sizeof(gen8_pte_t))
108
109 #define I915_PDPES_PER_PDP(dev) (USES_FULL_48BIT_PPGTT(dev) ?\
110 GEN8_PML4ES_PER_PML4 : GEN8_LEGACY_PDPES)
111
112 #define PPAT_UNCACHED_INDEX (_PAGE_PWT | _PAGE_PCD)
113 #define PPAT_CACHED_PDE_INDEX 0 /* WB LLC */
114 #define PPAT_CACHED_INDEX _PAGE_PAT /* WB LLCeLLC */
115 #define PPAT_DISPLAY_ELLC_INDEX _PAGE_PCD /* WT eLLC */
116
117 #define CHV_PPAT_SNOOP (1<<6)
118 #define GEN8_PPAT_AGE(x) (x<<4)
119 #define GEN8_PPAT_LLCeLLC (3<<2)
120 #define GEN8_PPAT_LLCELLC (2<<2)
121 #define GEN8_PPAT_LLC (1<<2)
122 #define GEN8_PPAT_WB (3<<0)
123 #define GEN8_PPAT_WT (2<<0)
124 #define GEN8_PPAT_WC (1<<0)
125 #define GEN8_PPAT_UC (0<<0)
126 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
127 #define GEN8_PPAT(i, x) ((uint64_t) (x) << ((i) * 8))
128
129 enum i915_ggtt_view_type {
130 I915_GGTT_VIEW_NORMAL = 0,
131 I915_GGTT_VIEW_ROTATED,
132 I915_GGTT_VIEW_PARTIAL,
133 };
134
135 struct intel_rotation_info {
136 unsigned int height;
137 unsigned int pitch;
138 uint32_t pixel_format;
139 uint64_t fb_modifier;
140 unsigned int width_pages, height_pages;
141 uint64_t size;
142 };
143
144 struct i915_ggtt_view {
145 enum i915_ggtt_view_type type;
146
147 union {
148 struct {
149 unsigned long offset;
150 unsigned int size;
151 } partial;
152 } params;
153
154 struct sg_table *pages;
155
156 union {
157 struct intel_rotation_info rotation_info;
158 };
159 };
160
161 extern const struct i915_ggtt_view i915_ggtt_view_normal;
162 extern const struct i915_ggtt_view i915_ggtt_view_rotated;
163
164 enum i915_cache_level;
165
166 /**
167 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
168 * VMA's presence cannot be guaranteed before binding, or after unbinding the
169 * object into/from the address space.
170 *
171 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
172 * will always be <= an objects lifetime. So object refcounting should cover us.
173 */
174 struct i915_vma {
175 struct drm_mm_node node;
176 struct drm_i915_gem_object *obj;
177 struct i915_address_space *vm;
178
179 /** Flags and address space this VMA is bound to */
180 #define GLOBAL_BIND (1<<0)
181 #define LOCAL_BIND (1<<1)
182 unsigned int bound : 4;
183
184 /**
185 * Support different GGTT views into the same object.
186 * This means there can be multiple VMA mappings per object and per VM.
187 * i915_ggtt_view_type is used to distinguish between those entries.
188 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
189 * assumed in GEM functions which take no ggtt view parameter.
190 */
191 struct i915_ggtt_view ggtt_view;
192
193 /** This object's place on the active/inactive lists */
194 struct list_head mm_list;
195
196 struct list_head vma_link; /* Link in the object's VMA list */
197
198 /** This vma's place in the batchbuffer or on the eviction list */
199 struct list_head exec_list;
200
201 /**
202 * Used for performing relocations during execbuffer insertion.
203 */
204 struct hlist_node exec_node;
205 unsigned long exec_handle;
206 struct drm_i915_gem_exec_object2 *exec_entry;
207
208 /**
209 * How many users have pinned this object in GTT space. The following
210 * users can each hold at most one reference: pwrite/pread, execbuffer
211 * (objects are not allowed multiple times for the same batchbuffer),
212 * and the framebuffer code. When switching/pageflipping, the
213 * framebuffer code has at most two buffers pinned per crtc.
214 *
215 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
216 * bits with absolutely no headroom. So use 4 bits. */
217 unsigned int pin_count:4;
218 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
219 };
220
221 struct i915_page_dma {
222 struct page *page;
223 union {
224 dma_addr_t daddr;
225
226 /* For gen6/gen7 only. This is the offset in the GGTT
227 * where the page directory entries for PPGTT begin
228 */
229 uint32_t ggtt_offset;
230 };
231 };
232
233 #define px_base(px) (&(px)->base)
234 #define px_page(px) (px_base(px)->page)
235 #define px_dma(px) (px_base(px)->daddr)
236
237 struct i915_page_scratch {
238 struct i915_page_dma base;
239 };
240
241 struct i915_page_table {
242 struct i915_page_dma base;
243
244 unsigned long *used_ptes;
245 };
246
247 struct i915_page_directory {
248 struct i915_page_dma base;
249
250 unsigned long *used_pdes;
251 struct i915_page_table *page_table[I915_PDES]; /* PDEs */
252 };
253
254 struct i915_page_directory_pointer {
255 struct i915_page_dma base;
256
257 unsigned long *used_pdpes;
258 struct i915_page_directory **page_directory;
259 };
260
261 struct i915_pml4 {
262 struct i915_page_dma base;
263
264 DECLARE_BITMAP(used_pml4es, GEN8_PML4ES_PER_PML4);
265 struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4];
266 };
267
268 struct i915_address_space {
269 struct drm_mm mm;
270 struct drm_device *dev;
271 struct list_head global_link;
272 u64 start; /* Start offset always 0 for dri2 */
273 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
274
275 struct i915_page_scratch *scratch_page;
276 struct i915_page_table *scratch_pt;
277 struct i915_page_directory *scratch_pd;
278
279 /**
280 * List of objects currently involved in rendering.
281 *
282 * Includes buffers having the contents of their GPU caches
283 * flushed, not necessarily primitives. last_read_req
284 * represents when the rendering involved will be completed.
285 *
286 * A reference is held on the buffer while on this list.
287 */
288 struct list_head active_list;
289
290 /**
291 * LRU list of objects which are not in the ringbuffer and
292 * are ready to unbind, but are still in the GTT.
293 *
294 * last_read_req is NULL while an object is in this list.
295 *
296 * A reference is not held on the buffer while on this list,
297 * as merely being GTT-bound shouldn't prevent its being
298 * freed, and we'll pull it off the list in the free path.
299 */
300 struct list_head inactive_list;
301
302 /* FIXME: Need a more generic return type */
303 gen6_pte_t (*pte_encode)(dma_addr_t addr,
304 enum i915_cache_level level,
305 bool valid, u32 flags); /* Create a valid PTE */
306 /* flags for pte_encode */
307 #define PTE_READ_ONLY (1<<0)
308 int (*allocate_va_range)(struct i915_address_space *vm,
309 uint64_t start,
310 uint64_t length);
311 void (*clear_range)(struct i915_address_space *vm,
312 uint64_t start,
313 uint64_t length,
314 bool use_scratch);
315 void (*insert_entries)(struct i915_address_space *vm,
316 struct sg_table *st,
317 uint64_t start,
318 enum i915_cache_level cache_level, u32 flags);
319 void (*cleanup)(struct i915_address_space *vm);
320 /** Unmap an object from an address space. This usually consists of
321 * setting the valid PTE entries to a reserved scratch page. */
322 void (*unbind_vma)(struct i915_vma *vma);
323 /* Map an object into an address space with the given cache flags. */
324 int (*bind_vma)(struct i915_vma *vma,
325 enum i915_cache_level cache_level,
326 u32 flags);
327 };
328
329 /* The Graphics Translation Table is the way in which GEN hardware translates a
330 * Graphics Virtual Address into a Physical Address. In addition to the normal
331 * collateral associated with any va->pa translations GEN hardware also has a
332 * portion of the GTT which can be mapped by the CPU and remain both coherent
333 * and correct (in cases like swizzling). That region is referred to as GMADR in
334 * the spec.
335 */
336 struct i915_gtt {
337 struct i915_address_space base;
338
339 size_t stolen_size; /* Total size of stolen memory */
340 u64 mappable_end; /* End offset that we can CPU map */
341 struct io_mapping *mappable; /* Mapping to our CPU mappable region */
342 phys_addr_t mappable_base; /* PA of our GMADR */
343
344 /** "Graphics Stolen Memory" holds the global PTEs */
345 void __iomem *gsm;
346
347 bool do_idle_maps;
348
349 int mtrr;
350
351 /* global gtt ops */
352 int (*gtt_probe)(struct drm_device *dev, u64 *gtt_total,
353 size_t *stolen, phys_addr_t *mappable_base,
354 u64 *mappable_end);
355 };
356
357 struct i915_hw_ppgtt {
358 struct i915_address_space base;
359 struct kref ref;
360 struct drm_mm_node node;
361 unsigned long pd_dirty_rings;
362 union {
363 struct i915_pml4 pml4; /* GEN8+ & 48b PPGTT */
364 struct i915_page_directory_pointer pdp; /* GEN8+ */
365 struct i915_page_directory pd; /* GEN6-7 */
366 };
367
368 struct drm_i915_file_private *file_priv;
369
370 gen6_pte_t __iomem *pd_addr;
371
372 int (*enable)(struct i915_hw_ppgtt *ppgtt);
373 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
374 struct drm_i915_gem_request *req);
375 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
376 };
377
378 /* For each pde iterates over every pde between from start until start + length.
379 * If start, and start+length are not perfectly divisible, the macro will round
380 * down, and up as needed. The macro modifies pde, start, and length. Dev is
381 * only used to differentiate shift values. Temp is temp. On gen6/7, start = 0,
382 * and length = 2G effectively iterates over every PDE in the system.
383 *
384 * XXX: temp is not actually needed, but it saves doing the ALIGN operation.
385 */
386 #define gen6_for_each_pde(pt, pd, start, length, temp, iter) \
387 for (iter = gen6_pde_index(start); \
388 pt = (pd)->page_table[iter], length > 0 && iter < I915_PDES; \
389 iter++, \
390 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT) - start, \
391 temp = min_t(unsigned, temp, length), \
392 start += temp, length -= temp)
393
394 #define gen6_for_all_pdes(pt, ppgtt, iter) \
395 for (iter = 0; \
396 pt = ppgtt->pd.page_table[iter], iter < I915_PDES; \
397 iter++)
398
399 static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
400 {
401 const uint32_t mask = NUM_PTE(pde_shift) - 1;
402
403 return (address >> PAGE_SHIFT) & mask;
404 }
405
406 /* Helper to counts the number of PTEs within the given length. This count
407 * does not cross a page table boundary, so the max value would be
408 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
409 */
410 static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
411 uint32_t pde_shift)
412 {
413 const uint64_t mask = ~((1 << pde_shift) - 1);
414 uint64_t end;
415
416 WARN_ON(length == 0);
417 WARN_ON(offset_in_page(addr|length));
418
419 end = addr + length;
420
421 if ((addr & mask) != (end & mask))
422 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
423
424 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
425 }
426
427 static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
428 {
429 return (addr >> shift) & I915_PDE_MASK;
430 }
431
432 static inline uint32_t gen6_pte_index(uint32_t addr)
433 {
434 return i915_pte_index(addr, GEN6_PDE_SHIFT);
435 }
436
437 static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
438 {
439 return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
440 }
441
442 static inline uint32_t gen6_pde_index(uint32_t addr)
443 {
444 return i915_pde_index(addr, GEN6_PDE_SHIFT);
445 }
446
447 /* Equivalent to the gen6 version, For each pde iterates over every pde
448 * between from start until start + length. On gen8+ it simply iterates
449 * over every page directory entry in a page directory.
450 */
451 #define gen8_for_each_pde(pt, pd, start, length, temp, iter) \
452 for (iter = gen8_pde_index(start); \
453 pt = (pd)->page_table[iter], length > 0 && iter < I915_PDES; \
454 iter++, \
455 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT) - start, \
456 temp = min(temp, length), \
457 start += temp, length -= temp)
458
459 #define gen8_for_each_pdpe(pd, pdp, start, length, temp, iter) \
460 for (iter = gen8_pdpe_index(start); \
461 pd = (pdp)->page_directory[iter], \
462 length > 0 && (iter < I915_PDPES_PER_PDP(dev)); \
463 iter++, \
464 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT) - start, \
465 temp = min(temp, length), \
466 start += temp, length -= temp)
467
468 static inline uint32_t gen8_pte_index(uint64_t address)
469 {
470 return i915_pte_index(address, GEN8_PDE_SHIFT);
471 }
472
473 static inline uint32_t gen8_pde_index(uint64_t address)
474 {
475 return i915_pde_index(address, GEN8_PDE_SHIFT);
476 }
477
478 static inline uint32_t gen8_pdpe_index(uint64_t address)
479 {
480 return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
481 }
482
483 static inline uint32_t gen8_pml4e_index(uint64_t address)
484 {
485 WARN_ON(1); /* For 64B */
486 return 0;
487 }
488
489 static inline size_t gen8_pte_count(uint64_t address, uint64_t length)
490 {
491 return i915_pte_count(address, length, GEN8_PDE_SHIFT);
492 }
493
494 static inline dma_addr_t
495 i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
496 {
497 return test_bit(n, ppgtt->pdp.used_pdpes) ?
498 px_dma(ppgtt->pdp.page_directory[n]) :
499 px_dma(ppgtt->base.scratch_pd);
500 }
501
502 int i915_gem_gtt_init(struct drm_device *dev);
503 void i915_gem_init_global_gtt(struct drm_device *dev);
504 void i915_global_gtt_cleanup(struct drm_device *dev);
505
506
507 int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
508 int i915_ppgtt_init_hw(struct drm_device *dev);
509 int i915_ppgtt_init_ring(struct drm_i915_gem_request *req);
510 void i915_ppgtt_release(struct kref *kref);
511 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
512 struct drm_i915_file_private *fpriv);
513 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
514 {
515 if (ppgtt)
516 kref_get(&ppgtt->ref);
517 }
518 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
519 {
520 if (ppgtt)
521 kref_put(&ppgtt->ref, i915_ppgtt_release);
522 }
523
524 void i915_check_and_clear_faults(struct drm_device *dev);
525 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
526 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
527
528 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
529 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
530
531 static inline bool
532 i915_ggtt_view_equal(const struct i915_ggtt_view *a,
533 const struct i915_ggtt_view *b)
534 {
535 if (WARN_ON(!a || !b))
536 return false;
537
538 if (a->type != b->type)
539 return false;
540 if (a->type == I915_GGTT_VIEW_PARTIAL)
541 return !memcmp(&a->params, &b->params, sizeof(a->params));
542 return true;
543 }
544
545 size_t
546 i915_ggtt_view_size(struct drm_i915_gem_object *obj,
547 const struct i915_ggtt_view *view);
548
549 #endif
This page took 0.04214 seconds and 6 git commands to generate.