1 /* SPDX-License-Identifier: MIT */
3 * Copyright © 2020 Intel Corporation
5 * Please try to maintain the following order within this file unless it makes
6 * sense to do otherwise. From top to bottom:
8 * 2. #defines, and macros
9 * 3. structure definitions
10 * 4. function prototypes
12 * Within each section, please try to order by generation in ascending order,
13 * from top to bottom (ie. gen6 on the top, gen8 on the bottom).
16 #ifndef __INTEL_GTT_H__
17 #define __INTEL_GTT_H__
19 #include <linux/io-mapping.h>
20 #include <linux/kref.h>
22 #include <linux/pagevec.h>
23 #include <linux/scatterlist.h>
24 #include <linux/workqueue.h>
26 #include <drm/drm_mm.h>
28 #include "gt/intel_reset.h"
29 #include "i915_selftest.h"
30 #include "i915_vma_types.h"
32 #define I915_GFP_ALLOW_FAIL (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)
34 #if IS_ENABLED(CONFIG_DRM_I915_TRACE_GTT)
35 #define DBG(...) trace_printk(__VA_ARGS__)
40 #define NALLOC 3 /* 1 normal, 1 for concurrent threads, 1 for preallocation */
42 #define I915_GTT_PAGE_SIZE_4K BIT_ULL(12)
43 #define I915_GTT_PAGE_SIZE_64K BIT_ULL(16)
44 #define I915_GTT_PAGE_SIZE_2M BIT_ULL(21)
46 #define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K
47 #define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M
49 #define I915_GTT_PAGE_MASK -I915_GTT_PAGE_SIZE
51 #define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE
53 #define I915_FENCE_REG_NONE -1
54 #define I915_MAX_NUM_FENCES 32
55 /* 32 fences + sign bit for FENCE_REG_NONE */
56 #define I915_MAX_NUM_FENCE_BITS 6
58 typedef u32 gen6_pte_t;
59 typedef u64 gen8_pte_t;
61 #define ggtt_total_entries(ggtt) ((ggtt)->vm.total >> PAGE_SHIFT)
63 #define I915_PTES(pte_len) ((unsigned int)(PAGE_SIZE / (pte_len)))
64 #define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1)
66 #define I915_PDE_MASK (I915_PDES - 1)
68 /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */
69 #define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0))
70 #define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
71 #define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr)
72 #define GEN6_PTE_CACHE_LLC (2 << 1)
73 #define GEN6_PTE_UNCACHED (1 << 1)
74 #define GEN6_PTE_VALID REG_BIT(0)
76 #define GEN6_PTES I915_PTES(sizeof(gen6_pte_t))
77 #define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE)
78 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
79 #define GEN6_PDE_SHIFT 22
80 #define GEN6_PDE_VALID REG_BIT(0)
81 #define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT))
83 #define GEN7_PTE_CACHE_L3_LLC (3 << 1)
85 #define BYT_PTE_SNOOPED_BY_CPU_CACHES REG_BIT(2)
86 #define BYT_PTE_WRITEABLE REG_BIT(1)
88 #define GEN12_PPGTT_PTE_LM BIT_ULL(11)
90 #define GEN12_GGTT_PTE_LM BIT_ULL(1)
93 * Cacheability Control is a 4-bit value. The low three bits are stored in bits
94 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE.
96 #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \
97 (((bits) & 0x8) << (11 - 3)))
98 #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2)
99 #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3)
100 #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8)
101 #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb)
102 #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7)
103 #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6)
104 #define HSW_PTE_UNCACHED (0)
105 #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0))
106 #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr)
109 * GEN8 32b style address is defined as a 3 level page table:
110 * 31:30 | 29:21 | 20:12 | 11:0
111 * PDPE | PDE | PTE | offset
112 * The difference as compared to normal x86 3 level page table is the PDPEs are
113 * programmed via register.
115 * GEN8 48b style address is defined as a 4 level page table:
116 * 47:39 | 38:30 | 29:21 | 20:12 | 11:0
117 * PML4E | PDPE | PDE | PTE | offset
119 #define GEN8_3LVL_PDPES 4
121 #define PPAT_UNCACHED (_PAGE_PWT | _PAGE_PCD)
122 #define PPAT_CACHED_PDE 0 /* WB LLC */
123 #define PPAT_CACHED _PAGE_PAT /* WB LLCeLLC */
124 #define PPAT_DISPLAY_ELLC _PAGE_PCD /* WT eLLC */
126 #define CHV_PPAT_SNOOP REG_BIT(6)
127 #define GEN8_PPAT_AGE(x) ((x)<<4)
128 #define GEN8_PPAT_LLCeLLC (3<<2)
129 #define GEN8_PPAT_LLCELLC (2<<2)
130 #define GEN8_PPAT_LLC (1<<2)
131 #define GEN8_PPAT_WB (3<<0)
132 #define GEN8_PPAT_WT (2<<0)
133 #define GEN8_PPAT_WC (1<<0)
134 #define GEN8_PPAT_UC (0<<0)
135 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2)
136 #define GEN8_PPAT(i, x) ((u64)(x) << ((i) * 8))
138 #define GEN8_PDE_IPS_64K BIT(11)
139 #define GEN8_PDE_PS_2M BIT(7)
141 enum i915_cache_level;
143 struct drm_i915_gem_object;
144 struct i915_fence_reg;
148 #define for_each_sgt_daddr(__dp, __iter, __sgt) \
149 __for_each_sgt_daddr(__dp, __iter, __sgt, I915_GTT_PAGE_SIZE)
151 struct i915_page_table {
152 struct drm_i915_gem_object *base;
155 struct i915_page_table *stash;
159 struct i915_page_directory {
160 struct i915_page_table pt;
165 #define __px_choose_expr(x, type, expr, other) \
166 __builtin_choose_expr( \
167 __builtin_types_compatible_p(typeof(x), type) || \
168 __builtin_types_compatible_p(typeof(x), const type), \
169 ({ type __x = (type)(x); expr; }), \
172 #define px_base(px) \
173 __px_choose_expr(px, struct drm_i915_gem_object *, __x, \
174 __px_choose_expr(px, struct i915_page_table *, __x->base, \
175 __px_choose_expr(px, struct i915_page_directory *, __x->pt.base, \
178 struct page *__px_page(struct drm_i915_gem_object *p);
179 dma_addr_t __px_dma(struct drm_i915_gem_object *p);
180 #define px_dma(px) (__px_dma(px_base(px)))
182 void *__px_vaddr(struct drm_i915_gem_object *p);
183 #define px_vaddr(px) (__px_vaddr(px_base(px)))
186 __px_choose_expr(px, struct i915_page_table *, __x, \
187 __px_choose_expr(px, struct i915_page_directory *, &__x->pt, \
189 #define px_used(px) (&px_pt(px)->used)
191 struct i915_vm_pt_stash {
192 /* preallocated chains of page tables/directories */
193 struct i915_page_table *pt[2];
196 struct i915_vma_ops {
197 /* Map an object into an address space with the given cache flags. */
198 void (*bind_vma)(struct i915_address_space *vm,
199 struct i915_vm_pt_stash *stash,
200 struct i915_vma *vma,
201 enum i915_cache_level cache_level,
204 * Unmap an object from an address space. This usually consists of
205 * setting the valid PTE entries to a reserved scratch page.
207 void (*unbind_vma)(struct i915_address_space *vm,
208 struct i915_vma *vma);
210 int (*set_pages)(struct i915_vma *vma);
211 void (*clear_pages)(struct i915_vma *vma);
214 struct i915_address_space {
216 struct work_struct release_work;
220 struct drm_i915_private *i915;
222 u64 total; /* size addr space maps (ex. 2GB for ggtt) */
223 u64 reserved; /* size addr space reserved */
225 unsigned int bind_async_flags;
228 * Each active user context has its own address space (in full-ppgtt).
229 * Since the vm may be shared between multiple contexts, we count how
230 * many contexts keep us "open". Once open hits zero, we are closed
231 * and do not allow any new attachments, and proceed to shutdown our
232 * vma and page directories.
236 struct mutex mutex; /* protects vma and our lists */
238 struct kref resv_ref; /* kref to keep the reservation lock alive. */
239 struct dma_resv _resv; /* reservation lock for all pd objects, and buffer pool */
240 #define VM_CLASS_GGTT 0
241 #define VM_CLASS_PPGTT 1
242 #define VM_CLASS_DPT 2
244 struct drm_i915_gem_object *scratch[4];
246 * List of vma currently bound.
248 struct list_head bound_list;
253 /* Display page table */
256 /* Some systems support read-only mappings for GGTT and/or PPGTT */
257 bool has_read_only:1;
263 /* Flags used when creating page-table objects for this vm */
264 unsigned long lmem_pt_obj_flags;
266 struct drm_i915_gem_object *
267 (*alloc_pt_dma)(struct i915_address_space *vm, int sz);
269 u64 (*pte_encode)(dma_addr_t addr,
270 enum i915_cache_level level,
271 u32 flags); /* Create a valid PTE */
272 #define PTE_READ_ONLY BIT(0)
273 #define PTE_LM BIT(1)
275 void (*allocate_va_range)(struct i915_address_space *vm,
276 struct i915_vm_pt_stash *stash,
277 u64 start, u64 length);
278 void (*clear_range)(struct i915_address_space *vm,
279 u64 start, u64 length);
280 void (*insert_page)(struct i915_address_space *vm,
283 enum i915_cache_level cache_level,
285 void (*insert_entries)(struct i915_address_space *vm,
286 struct i915_vma *vma,
287 enum i915_cache_level cache_level,
289 void (*cleanup)(struct i915_address_space *vm);
291 void (*foreach)(struct i915_address_space *vm,
292 u64 start, u64 length,
293 void (*fn)(struct i915_address_space *vm,
294 struct i915_page_table *pt,
298 struct i915_vma_ops vma_ops;
300 I915_SELFTEST_DECLARE(struct fault_attr fault_attr);
301 I915_SELFTEST_DECLARE(bool scrub_64K);
305 * The Graphics Translation Table is the way in which GEN hardware translates a
306 * Graphics Virtual Address into a Physical Address. In addition to the normal
307 * collateral associated with any va->pa translations GEN hardware also has a
308 * portion of the GTT which can be mapped by the CPU and remain both coherent
309 * and correct (in cases like swizzling). That region is referred to as GMADR in
313 struct i915_address_space vm;
315 struct io_mapping iomap; /* Mapping to our CPU mappable region */
316 struct resource gmadr; /* GMADR resource */
317 resource_size_t mappable_end; /* End offset that we can CPU map */
319 /** "Graphics Stolen Memory" holds the global PTEs */
321 void (*invalidate)(struct i915_ggtt *ggtt);
323 /** PPGTT used for aliasing the PPGTT with the GTT */
324 struct i915_ppgtt *alias;
330 /** Bit 6 swizzling required for X tiling */
332 /** Bit 6 swizzling required for Y tiling */
337 unsigned int num_fences;
338 struct i915_fence_reg *fence_regs;
339 struct list_head fence_list;
342 * List of all objects in gtt_space, currently mmaped by userspace.
343 * All objects within this list must also be on bound_list.
345 struct list_head userfault_list;
347 /* Manual runtime pm autosuspend delay for user GGTT mmaps */
348 struct intel_wakeref_auto userfault_wakeref;
350 struct mutex error_mutex;
351 struct drm_mm_node error_capture;
352 struct drm_mm_node uc_fw;
356 struct i915_address_space vm;
358 struct i915_page_directory *pd;
361 #define i915_is_ggtt(vm) ((vm)->is_ggtt)
362 #define i915_is_dpt(vm) ((vm)->is_dpt)
363 #define i915_is_ggtt_or_dpt(vm) (i915_is_ggtt(vm) || i915_is_dpt(vm))
366 i915_vm_lock_objects(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww);
369 i915_vm_is_4lvl(const struct i915_address_space *vm)
371 return (vm->total - 1) >> 32;
375 i915_vm_has_scratch_64K(struct i915_address_space *vm)
377 return vm->scratch_order == get_order(I915_GTT_PAGE_SIZE_64K);
381 i915_vm_has_cache_coloring(struct i915_address_space *vm)
383 return i915_is_ggtt(vm) && vm->mm.color_adjust;
386 static inline struct i915_ggtt *
387 i915_vm_to_ggtt(struct i915_address_space *vm)
389 BUILD_BUG_ON(offsetof(struct i915_ggtt, vm));
390 GEM_BUG_ON(!i915_is_ggtt(vm));
391 return container_of(vm, struct i915_ggtt, vm);
394 static inline struct i915_ppgtt *
395 i915_vm_to_ppgtt(struct i915_address_space *vm)
397 BUILD_BUG_ON(offsetof(struct i915_ppgtt, vm));
398 GEM_BUG_ON(i915_is_ggtt_or_dpt(vm));
399 return container_of(vm, struct i915_ppgtt, vm);
402 static inline struct i915_address_space *
403 i915_vm_get(struct i915_address_space *vm)
410 * i915_vm_resv_get - Obtain a reference on the vm's reservation lock
411 * @vm: The vm whose reservation lock we want to share.
413 * Return: A pointer to the vm's reservation lock.
415 static inline struct dma_resv *i915_vm_resv_get(struct i915_address_space *vm)
417 kref_get(&vm->resv_ref);
421 void i915_vm_release(struct kref *kref);
423 void i915_vm_resv_release(struct kref *kref);
425 static inline void i915_vm_put(struct i915_address_space *vm)
427 kref_put(&vm->ref, i915_vm_release);
431 * i915_vm_resv_put - Release a reference on the vm's reservation lock
432 * @resv: Pointer to a reservation lock obtained from i915_vm_resv_get()
434 static inline void i915_vm_resv_put(struct i915_address_space *vm)
436 kref_put(&vm->resv_ref, i915_vm_resv_release);
439 static inline struct i915_address_space *
440 i915_vm_open(struct i915_address_space *vm)
442 GEM_BUG_ON(!atomic_read(&vm->open));
443 atomic_inc(&vm->open);
444 return i915_vm_get(vm);
448 i915_vm_tryopen(struct i915_address_space *vm)
450 if (atomic_add_unless(&vm->open, 1, 0))
451 return i915_vm_get(vm);
456 void __i915_vm_close(struct i915_address_space *vm);
459 i915_vm_close(struct i915_address_space *vm)
461 GEM_BUG_ON(!atomic_read(&vm->open));
467 void i915_address_space_init(struct i915_address_space *vm, int subclass);
468 void i915_address_space_fini(struct i915_address_space *vm);
470 static inline u32 i915_pte_index(u64 address, unsigned int pde_shift)
472 const u32 mask = NUM_PTE(pde_shift) - 1;
474 return (address >> PAGE_SHIFT) & mask;
478 * Helper to counts the number of PTEs within the given length. This count
479 * does not cross a page table boundary, so the max value would be
480 * GEN6_PTES for GEN6, and GEN8_PTES for GEN8.
482 static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift)
484 const u64 mask = ~((1ULL << pde_shift) - 1);
487 GEM_BUG_ON(length == 0);
488 GEM_BUG_ON(offset_in_page(addr | length));
492 if ((addr & mask) != (end & mask))
493 return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift);
495 return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift);
498 static inline u32 i915_pde_index(u64 addr, u32 shift)
500 return (addr >> shift) & I915_PDE_MASK;
503 static inline struct i915_page_table *
504 i915_pt_entry(const struct i915_page_directory * const pd,
505 const unsigned short n)
510 static inline struct i915_page_directory *
511 i915_pd_entry(const struct i915_page_directory * const pdp,
512 const unsigned short n)
514 return pdp->entry[n];
517 static inline dma_addr_t
518 i915_page_dir_dma_addr(const struct i915_ppgtt *ppgtt, const unsigned int n)
520 struct i915_page_table *pt = ppgtt->pd->entry[n];
522 return __px_dma(pt ? px_base(pt) : ppgtt->vm.scratch[ppgtt->vm.top]);
525 void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
526 unsigned long lmem_pt_obj_flags);
528 int i915_ggtt_probe_hw(struct drm_i915_private *i915);
529 int i915_ggtt_init_hw(struct drm_i915_private *i915);
530 int i915_ggtt_enable_hw(struct drm_i915_private *i915);
531 void i915_ggtt_enable_guc(struct i915_ggtt *ggtt);
532 void i915_ggtt_disable_guc(struct i915_ggtt *ggtt);
533 int i915_init_ggtt(struct drm_i915_private *i915);
534 void i915_ggtt_driver_release(struct drm_i915_private *i915);
535 void i915_ggtt_driver_late_release(struct drm_i915_private *i915);
537 static inline bool i915_ggtt_has_aperture(const struct i915_ggtt *ggtt)
539 return ggtt->mappable_end > 0;
542 int i915_ppgtt_init_hw(struct intel_gt *gt);
544 struct i915_ppgtt *i915_ppgtt_create(struct intel_gt *gt,
545 unsigned long lmem_pt_obj_flags);
547 void i915_ggtt_suspend(struct i915_ggtt *gtt);
548 void i915_ggtt_resume(struct i915_ggtt *ggtt);
551 fill_page_dma(struct drm_i915_gem_object *p, const u64 val, unsigned int count);
553 #define fill_px(px, v) fill_page_dma(px_base(px), (v), PAGE_SIZE / sizeof(u64))
554 #define fill32_px(px, v) do { \
555 u64 v__ = lower_32_bits(v); \
556 fill_px((px), v__ << 32 | v__); \
559 int setup_scratch_page(struct i915_address_space *vm);
560 void free_scratch(struct i915_address_space *vm);
562 struct drm_i915_gem_object *alloc_pt_dma(struct i915_address_space *vm, int sz);
563 struct drm_i915_gem_object *alloc_pt_lmem(struct i915_address_space *vm, int sz);
564 struct i915_page_table *alloc_pt(struct i915_address_space *vm);
565 struct i915_page_directory *alloc_pd(struct i915_address_space *vm);
566 struct i915_page_directory *__alloc_pd(int npde);
568 int map_pt_dma(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
569 int map_pt_dma_locked(struct i915_address_space *vm, struct drm_i915_gem_object *obj);
571 void free_px(struct i915_address_space *vm,
572 struct i915_page_table *pt, int lvl);
573 #define free_pt(vm, px) free_px(vm, px, 0)
574 #define free_pd(vm, px) free_px(vm, px_pt(px), 1)
577 __set_pd_entry(struct i915_page_directory * const pd,
578 const unsigned short idx,
579 struct i915_page_table *pt,
580 u64 (*encode)(const dma_addr_t, const enum i915_cache_level));
582 #define set_pd_entry(pd, idx, to) \
583 __set_pd_entry((pd), (idx), px_pt(to), gen8_pde_encode)
586 clear_pd_entry(struct i915_page_directory * const pd,
587 const unsigned short idx,
588 const struct drm_i915_gem_object * const scratch);
591 release_pd_entry(struct i915_page_directory * const pd,
592 const unsigned short idx,
593 struct i915_page_table * const pt,
594 const struct drm_i915_gem_object * const scratch);
595 void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);
597 int ggtt_set_pages(struct i915_vma *vma);
598 int ppgtt_set_pages(struct i915_vma *vma);
599 void clear_pages(struct i915_vma *vma);
601 void ppgtt_bind_vma(struct i915_address_space *vm,
602 struct i915_vm_pt_stash *stash,
603 struct i915_vma *vma,
604 enum i915_cache_level cache_level,
606 void ppgtt_unbind_vma(struct i915_address_space *vm,
607 struct i915_vma *vma);
609 void gtt_write_workarounds(struct intel_gt *gt);
611 void setup_private_pat(struct intel_uncore *uncore);
613 int i915_vm_alloc_pt_stash(struct i915_address_space *vm,
614 struct i915_vm_pt_stash *stash,
616 int i915_vm_map_pt_stash(struct i915_address_space *vm,
617 struct i915_vm_pt_stash *stash);
618 void i915_vm_free_pt_stash(struct i915_address_space *vm,
619 struct i915_vm_pt_stash *stash);
622 __vm_create_scratch_for_read(struct i915_address_space *vm, unsigned long size);
625 __vm_create_scratch_for_read_pinned(struct i915_address_space *vm, unsigned long size);
627 static inline struct sgt_dma {
628 struct scatterlist *sg;
630 } sgt_dma(struct i915_vma *vma) {
631 struct scatterlist *sg = vma->pages->sgl;
632 dma_addr_t addr = sg_dma_address(sg);
634 return (struct sgt_dma){ sg, addr, addr + sg_dma_len(sg) };