2 * Copyright © 2014 Intel Corporation
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:
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
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
23 * Please try to maintain the following order within this file unless it makes
24 * sense to do otherwise. From top to bottom:
26 * 2. #defines, and macros
27 * 3. structure definitions
28 * 4. function prototypes
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).
34 #ifndef __I915_GEM_GTT_H__
35 #define __I915_GEM_GTT_H__
37 struct drm_i915_file_private;
39 typedef uint32_t gen6_pte_t;
40 typedef uint64_t gen8_pte_t;
41 typedef uint64_t gen8_pde_t;
43 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
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)
54 #define I915_PTES(pte_len) (PAGE_SIZE / (pte_len))
55 #define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
57 #define I915_PDE_MASK (I915_PDES - 1)
58 #define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
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)
66 #define GEN7_PTE_CACHE_L3_LLC (3 << 1)
68 #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2)
69 #define BYT_PTE_WRITEABLE (1 << 1)
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.
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)
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.
92 #define GEN8_PDPE_SHIFT 30
93 #define GEN8_PDPE_MASK 0x3
94 #define GEN8_PDE_SHIFT 21
95 #define GEN8_PDE_MASK 0x1ff
96 #define GEN8_PTE_SHIFT 12
97 #define GEN8_PTE_MASK 0x1ff
98 #define GEN8_LEGACY_PDPES 4
99 #define GEN8_PTES I915_PTES(sizeof(gen8_pte_t))
101 #define PPAT_UNCACHED_INDEX (_PAGE_PWT | _PAGE_PCD)
102 #define PPAT_CACHED_PDE_INDEX 0 /* WB LLC */
103 #define PPAT_CACHED_INDEX _PAGE_PAT /* WB LLCeLLC */
104 #define PPAT_DISPLAY_ELLC_INDEX _PAGE_PCD /* WT eLLC */
106 #define CHV_PPAT_SNOOP (1<<6)
107 #define GEN8_PPAT_AGE(x) (x<<4)
108 #define GEN8_PPAT_LLCeLLC (3<<2)
109 #define GEN8_PPAT_LLCELLC (2<<2)
110 #define GEN8_PPAT_LLC (1<<2)
111 #define GEN8_PPAT_WB (3<<0)
112 #define GEN8_PPAT_WT (2<<0)
113 #define GEN8_PPAT_WC (1<<0)
114 #define GEN8_PPAT_UC (0<<0)
115 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
116 #define GEN8_PPAT(i, x) ((uint64_t) (x) << ((i) * 8))
118 enum i915_ggtt_view_type {
119 I915_GGTT_VIEW_NORMAL = 0,
120 I915_GGTT_VIEW_ROTATED,
121 I915_GGTT_VIEW_PARTIAL,
124 struct intel_rotation_info {
127 uint32_t pixel_format;
128 uint64_t fb_modifier;
129 unsigned int width_pages, height_pages;
133 struct i915_ggtt_view {
134 enum i915_ggtt_view_type type;
138 unsigned long offset;
143 struct sg_table *pages;
146 struct intel_rotation_info rotation_info;
150 extern const struct i915_ggtt_view i915_ggtt_view_normal;
151 extern const struct i915_ggtt_view i915_ggtt_view_rotated;
153 enum i915_cache_level;
156 * A VMA represents a GEM BO that is bound into an address space. Therefore, a
157 * VMA's presence cannot be guaranteed before binding, or after unbinding the
158 * object into/from the address space.
160 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime
161 * will always be <= an objects lifetime. So object refcounting should cover us.
164 struct drm_mm_node node;
165 struct drm_i915_gem_object *obj;
166 struct i915_address_space *vm;
168 /** Flags and address space this VMA is bound to */
169 #define GLOBAL_BIND (1<<0)
170 #define LOCAL_BIND (1<<1)
171 unsigned int bound : 4;
174 * Support different GGTT views into the same object.
175 * This means there can be multiple VMA mappings per object and per VM.
176 * i915_ggtt_view_type is used to distinguish between those entries.
177 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also
178 * assumed in GEM functions which take no ggtt view parameter.
180 struct i915_ggtt_view ggtt_view;
182 /** This object's place on the active/inactive lists */
183 struct list_head mm_list;
185 struct list_head vma_link; /* Link in the object's VMA list */
187 /** This vma's place in the batchbuffer or on the eviction list */
188 struct list_head exec_list;
191 * Used for performing relocations during execbuffer insertion.
193 struct hlist_node exec_node;
194 unsigned long exec_handle;
195 struct drm_i915_gem_exec_object2 *exec_entry;
198 * How many users have pinned this object in GTT space. The following
199 * users can each hold at most one reference: pwrite/pread, execbuffer
200 * (objects are not allowed multiple times for the same batchbuffer),
201 * and the framebuffer code. When switching/pageflipping, the
202 * framebuffer code has at most two buffers pinned per crtc.
204 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
205 * bits with absolutely no headroom. So use 4 bits. */
206 unsigned int pin_count:4;
207 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
210 struct i915_page_table {
214 unsigned long *used_ptes;
217 struct i915_page_directory {
218 struct page *page; /* NULL for GEN6-GEN7 */
224 unsigned long *used_pdes;
225 struct i915_page_table *page_table[I915_PDES]; /* PDEs */
228 struct i915_page_directory_pointer {
229 /* struct page *page; */
230 DECLARE_BITMAP(used_pdpes, GEN8_LEGACY_PDPES);
231 struct i915_page_directory *page_directory[GEN8_LEGACY_PDPES];
234 struct i915_address_space {
236 struct drm_device *dev;
237 struct list_head global_link;
238 u64 start; /* Start offset always 0 for dri2 */
239 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
247 * List of objects currently involved in rendering.
249 * Includes buffers having the contents of their GPU caches
250 * flushed, not necessarily primitives. last_read_req
251 * represents when the rendering involved will be completed.
253 * A reference is held on the buffer while on this list.
255 struct list_head active_list;
258 * LRU list of objects which are not in the ringbuffer and
259 * are ready to unbind, but are still in the GTT.
261 * last_read_req is NULL while an object is in this list.
263 * A reference is not held on the buffer while on this list,
264 * as merely being GTT-bound shouldn't prevent its being
265 * freed, and we'll pull it off the list in the free path.
267 struct list_head inactive_list;
269 /* FIXME: Need a more generic return type */
270 gen6_pte_t (*pte_encode)(dma_addr_t addr,
271 enum i915_cache_level level,
272 bool valid, u32 flags); /* Create a valid PTE */
273 /* flags for pte_encode */
274 #define PTE_READ_ONLY (1<<0)
275 int (*allocate_va_range)(struct i915_address_space *vm,
278 void (*clear_range)(struct i915_address_space *vm,
282 void (*insert_entries)(struct i915_address_space *vm,
285 enum i915_cache_level cache_level, u32 flags);
286 void (*cleanup)(struct i915_address_space *vm);
287 /** Unmap an object from an address space. This usually consists of
288 * setting the valid PTE entries to a reserved scratch page. */
289 void (*unbind_vma)(struct i915_vma *vma);
290 /* Map an object into an address space with the given cache flags. */
291 int (*bind_vma)(struct i915_vma *vma,
292 enum i915_cache_level cache_level,
296 /* The Graphics Translation Table is the way in which GEN hardware translates a
297 * Graphics Virtual Address into a Physical Address. In addition to the normal
298 * collateral associated with any va->pa translations GEN hardware also has a
299 * portion of the GTT which can be mapped by the CPU and remain both coherent
300 * and correct (in cases like swizzling). That region is referred to as GMADR in
304 struct i915_address_space base;
306 size_t stolen_size; /* Total size of stolen memory */
307 u64 mappable_end; /* End offset that we can CPU map */
308 struct io_mapping *mappable; /* Mapping to our CPU mappable region */
309 phys_addr_t mappable_base; /* PA of our GMADR */
311 /** "Graphics Stolen Memory" holds the global PTEs */
319 int (*gtt_probe)(struct drm_device *dev, u64 *gtt_total,
320 size_t *stolen, phys_addr_t *mappable_base,
324 struct i915_hw_ppgtt {
325 struct i915_address_space base;
327 struct drm_mm_node node;
328 unsigned long pd_dirty_rings;
330 struct i915_page_directory_pointer pdp;
331 struct i915_page_directory pd;
334 struct i915_page_table *scratch_pt;
335 struct i915_page_directory *scratch_pd;
337 struct drm_i915_file_private *file_priv;
339 gen6_pte_t __iomem *pd_addr;
341 int (*enable)(struct i915_hw_ppgtt *ppgtt);
342 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt,
343 struct drm_i915_gem_request *req);
344 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m);
347 /* For each pde iterates over every pde between from start until start + length.
348 * If start, and start+length are not perfectly divisible, the macro will round
349 * down, and up as needed. The macro modifies pde, start, and length. Dev is
350 * only used to differentiate shift values. Temp is temp. On gen6/7, start = 0,
351 * and length = 2G effectively iterates over every PDE in the system.
353 * XXX: temp is not actually needed, but it saves doing the ALIGN operation.
355 #define gen6_for_each_pde(pt, pd, start, length, temp, iter) \
356 for (iter = gen6_pde_index(start); \
357 pt = (pd)->page_table[iter], length > 0 && iter < I915_PDES; \
359 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT) - start, \
360 temp = min_t(unsigned, temp, length), \
361 start += temp, length -= temp)
363 #define gen6_for_all_pdes(pt, ppgtt, iter) \
365 pt = ppgtt->pd.page_table[iter], iter < I915_PDES; \
368 static inline uint32_t i915_pte_index(uint64_t address, uint32_t pde_shift)
370 const uint32_t mask = NUM_PTE(pde_shift) - 1;
372 return (address >> PAGE_SHIFT) & mask;
375 /* Helper to counts the number of PTEs within the given length. This count
376 * does not cross a page table boundary, so the max value would be
377 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
379 static inline uint32_t i915_pte_count(uint64_t addr, size_t length,
382 const uint64_t mask = ~((1 << pde_shift) - 1);
385 WARN_ON(length == 0);
386 WARN_ON(offset_in_page(addr|length));
390 if ((addr & mask) != (end & mask))
391 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
393 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
396 static inline uint32_t i915_pde_index(uint64_t addr, uint32_t shift)
398 return (addr >> shift) & I915_PDE_MASK;
401 static inline uint32_t gen6_pte_index(uint32_t addr)
403 return i915_pte_index(addr, GEN6_PDE_SHIFT);
406 static inline size_t gen6_pte_count(uint32_t addr, uint32_t length)
408 return i915_pte_count(addr, length, GEN6_PDE_SHIFT);
411 static inline uint32_t gen6_pde_index(uint32_t addr)
413 return i915_pde_index(addr, GEN6_PDE_SHIFT);
416 /* Equivalent to the gen6 version, For each pde iterates over every pde
417 * between from start until start + length. On gen8+ it simply iterates
418 * over every page directory entry in a page directory.
420 #define gen8_for_each_pde(pt, pd, start, length, temp, iter) \
421 for (iter = gen8_pde_index(start); \
422 pt = (pd)->page_table[iter], length > 0 && iter < I915_PDES; \
424 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT) - start, \
425 temp = min(temp, length), \
426 start += temp, length -= temp)
428 #define gen8_for_each_pdpe(pd, pdp, start, length, temp, iter) \
429 for (iter = gen8_pdpe_index(start); \
430 pd = (pdp)->page_directory[iter], length > 0 && iter < GEN8_LEGACY_PDPES; \
432 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT) - start, \
433 temp = min(temp, length), \
434 start += temp, length -= temp)
436 /* Clamp length to the next page_directory boundary */
437 static inline uint64_t gen8_clamp_pd(uint64_t start, uint64_t length)
439 uint64_t next_pd = ALIGN(start + 1, 1 << GEN8_PDPE_SHIFT);
441 if (next_pd > (start + length))
444 return next_pd - start;
447 static inline uint32_t gen8_pte_index(uint64_t address)
449 return i915_pte_index(address, GEN8_PDE_SHIFT);
452 static inline uint32_t gen8_pde_index(uint64_t address)
454 return i915_pde_index(address, GEN8_PDE_SHIFT);
457 static inline uint32_t gen8_pdpe_index(uint64_t address)
459 return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK;
462 static inline uint32_t gen8_pml4e_index(uint64_t address)
464 WARN_ON(1); /* For 64B */
468 static inline size_t gen8_pte_count(uint64_t address, uint64_t length)
470 return i915_pte_count(address, length, GEN8_PDE_SHIFT);
473 static inline dma_addr_t
474 i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n)
476 return test_bit(n, ppgtt->pdp.used_pdpes) ?
477 ppgtt->pdp.page_directory[n]->daddr :
478 ppgtt->scratch_pd->daddr;
481 int i915_gem_gtt_init(struct drm_device *dev);
482 void i915_gem_init_global_gtt(struct drm_device *dev);
483 void i915_global_gtt_cleanup(struct drm_device *dev);
486 int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt);
487 int i915_ppgtt_init_hw(struct drm_device *dev);
488 int i915_ppgtt_init_ring(struct drm_i915_gem_request *req);
489 void i915_ppgtt_release(struct kref *kref);
490 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev,
491 struct drm_i915_file_private *fpriv);
492 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt)
495 kref_get(&ppgtt->ref);
497 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt)
500 kref_put(&ppgtt->ref, i915_ppgtt_release);
503 void i915_check_and_clear_faults(struct drm_device *dev);
504 void i915_gem_suspend_gtt_mappings(struct drm_device *dev);
505 void i915_gem_restore_gtt_mappings(struct drm_device *dev);
507 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj);
508 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj);
511 i915_ggtt_view_equal(const struct i915_ggtt_view *a,
512 const struct i915_ggtt_view *b)
514 if (WARN_ON(!a || !b))
517 if (a->type != b->type)
519 if (a->type == I915_GGTT_VIEW_PARTIAL)
520 return !memcmp(&a->params, &b->params, sizeof(a->params));
525 i915_ggtt_view_size(struct drm_i915_gem_object *obj,
526 const struct i915_ggtt_view *view);