10260c420SBen Widawsky /* 20260c420SBen Widawsky * Copyright © 2014 Intel Corporation 30260c420SBen Widawsky * 40260c420SBen Widawsky * Permission is hereby granted, free of charge, to any person obtaining a 50260c420SBen Widawsky * copy of this software and associated documentation files (the "Software"), 60260c420SBen Widawsky * to deal in the Software without restriction, including without limitation 70260c420SBen Widawsky * the rights to use, copy, modify, merge, publish, distribute, sublicense, 80260c420SBen Widawsky * and/or sell copies of the Software, and to permit persons to whom the 90260c420SBen Widawsky * Software is furnished to do so, subject to the following conditions: 100260c420SBen Widawsky * 110260c420SBen Widawsky * The above copyright notice and this permission notice (including the next 120260c420SBen Widawsky * paragraph) shall be included in all copies or substantial portions of the 130260c420SBen Widawsky * Software. 140260c420SBen Widawsky * 150260c420SBen Widawsky * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 160260c420SBen Widawsky * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 170260c420SBen Widawsky * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 180260c420SBen Widawsky * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 190260c420SBen Widawsky * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 200260c420SBen Widawsky * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 210260c420SBen Widawsky * IN THE SOFTWARE. 220260c420SBen Widawsky * 230260c420SBen Widawsky * Please try to maintain the following order within this file unless it makes 240260c420SBen Widawsky * sense to do otherwise. From top to bottom: 250260c420SBen Widawsky * 1. typedefs 260260c420SBen Widawsky * 2. #defines, and macros 270260c420SBen Widawsky * 3. structure definitions 280260c420SBen Widawsky * 4. function prototypes 290260c420SBen Widawsky * 300260c420SBen Widawsky * Within each section, please try to order by generation in ascending order, 310260c420SBen Widawsky * from top to bottom (ie. gen6 on the top, gen8 on the bottom). 320260c420SBen Widawsky */ 330260c420SBen Widawsky 340260c420SBen Widawsky #ifndef __I915_GEM_GTT_H__ 350260c420SBen Widawsky #define __I915_GEM_GTT_H__ 360260c420SBen Widawsky 378ef8561fSChris Wilson #include <linux/io-mapping.h> 38b42fe9caSJoonas Lahtinen #include <linux/mm.h> 398448661dSChris Wilson #include <linux/pagevec.h> 408ef8561fSChris Wilson 41e61e0f51SChris Wilson #include "i915_request.h" 428448661dSChris Wilson #include "i915_selftest.h" 43a89d1f92SChris Wilson #include "i915_timeline.h" 44b0decaf7SChris Wilson 452a9654b2SMatthew Auld #define I915_GTT_PAGE_SIZE_4K BIT(12) 462a9654b2SMatthew Auld #define I915_GTT_PAGE_SIZE_64K BIT(16) 472a9654b2SMatthew Auld #define I915_GTT_PAGE_SIZE_2M BIT(21) 482a9654b2SMatthew Auld 492a9654b2SMatthew Auld #define I915_GTT_PAGE_SIZE I915_GTT_PAGE_SIZE_4K 502a9654b2SMatthew Auld #define I915_GTT_MAX_PAGE_SIZE I915_GTT_PAGE_SIZE_2M 512a9654b2SMatthew Auld 52f51455d4SChris Wilson #define I915_GTT_MIN_ALIGNMENT I915_GTT_PAGE_SIZE 53f51455d4SChris Wilson 5449ef5294SChris Wilson #define I915_FENCE_REG_NONE -1 5549ef5294SChris Wilson #define I915_MAX_NUM_FENCES 32 5649ef5294SChris Wilson /* 32 fences + sign bit for FENCE_REG_NONE */ 5749ef5294SChris Wilson #define I915_MAX_NUM_FENCE_BITS 6 5849ef5294SChris Wilson 594d884705SDaniel Vetter struct drm_i915_file_private; 6049ef5294SChris Wilson struct drm_i915_fence_reg; 6193f2cde2SChris Wilson struct i915_vma; 624d884705SDaniel Vetter 6375c7b0b8SChris Wilson typedef u32 gen6_pte_t; 6475c7b0b8SChris Wilson typedef u64 gen8_pte_t; 6575c7b0b8SChris Wilson typedef u64 gen8_pde_t; 6675c7b0b8SChris Wilson typedef u64 gen8_ppgtt_pdpe_t; 6775c7b0b8SChris Wilson typedef u64 gen8_ppgtt_pml4e_t; 680260c420SBen Widawsky 6982ad6443SChris Wilson #define ggtt_total_entries(ggtt) ((ggtt)->vm.total >> PAGE_SHIFT) 700260c420SBen Widawsky 710260c420SBen Widawsky /* gen6-hsw has bit 11-4 for physical addr bit 39-32 */ 720260c420SBen Widawsky #define GEN6_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0xff0)) 730260c420SBen Widawsky #define GEN6_PTE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr) 740260c420SBen Widawsky #define GEN6_PDE_ADDR_ENCODE(addr) GEN6_GTT_ADDR_ENCODE(addr) 750260c420SBen Widawsky #define GEN6_PTE_CACHE_LLC (2 << 1) 760260c420SBen Widawsky #define GEN6_PTE_UNCACHED (1 << 1) 770260c420SBen Widawsky #define GEN6_PTE_VALID (1 << 0) 780260c420SBen Widawsky 79dd19674bSChris Wilson #define I915_PTES(pte_len) ((unsigned int)(PAGE_SIZE / (pte_len))) 8007749ef3SMichel Thierry #define I915_PTE_MASK(pte_len) (I915_PTES(pte_len) - 1) 8107749ef3SMichel Thierry #define I915_PDES 512 8207749ef3SMichel Thierry #define I915_PDE_MASK (I915_PDES - 1) 83678d96fbSBen Widawsky #define NUM_PTE(pde_shift) (1 << (pde_shift - PAGE_SHIFT)) 8407749ef3SMichel Thierry 8507749ef3SMichel Thierry #define GEN6_PTES I915_PTES(sizeof(gen6_pte_t)) 8607749ef3SMichel Thierry #define GEN6_PD_SIZE (I915_PDES * PAGE_SIZE) 870260c420SBen Widawsky #define GEN6_PD_ALIGN (PAGE_SIZE * 16) 88678d96fbSBen Widawsky #define GEN6_PDE_SHIFT 22 890260c420SBen Widawsky #define GEN6_PDE_VALID (1 << 0) 900260c420SBen Widawsky 910260c420SBen Widawsky #define GEN7_PTE_CACHE_L3_LLC (3 << 1) 920260c420SBen Widawsky 930260c420SBen Widawsky #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2) 940260c420SBen Widawsky #define BYT_PTE_WRITEABLE (1 << 1) 950260c420SBen Widawsky 960260c420SBen Widawsky /* Cacheability Control is a 4-bit value. The low three bits are stored in bits 970260c420SBen Widawsky * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE. 980260c420SBen Widawsky */ 990260c420SBen Widawsky #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \ 1000260c420SBen Widawsky (((bits) & 0x8) << (11 - 3))) 1010260c420SBen Widawsky #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2) 1020260c420SBen Widawsky #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3) 1030260c420SBen Widawsky #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8) 1040260c420SBen Widawsky #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb) 1050260c420SBen Widawsky #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7) 1060260c420SBen Widawsky #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6) 1070260c420SBen Widawsky #define HSW_PTE_UNCACHED (0) 1080260c420SBen Widawsky #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0)) 1090260c420SBen Widawsky #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr) 1100260c420SBen Widawsky 111e7167769SMika Kuoppala /* GEN8 32b style address is defined as a 3 level page table: 1120260c420SBen Widawsky * 31:30 | 29:21 | 20:12 | 11:0 1130260c420SBen Widawsky * PDPE | PDE | PTE | offset 1140260c420SBen Widawsky * The difference as compared to normal x86 3 level page table is the PDPEs are 1150260c420SBen Widawsky * programmed via register. 116e7167769SMika Kuoppala */ 117e7167769SMika Kuoppala #define GEN8_3LVL_PDPES 4 118e7167769SMika Kuoppala #define GEN8_PDE_SHIFT 21 119e7167769SMika Kuoppala #define GEN8_PDE_MASK 0x1ff 120e7167769SMika Kuoppala #define GEN8_PTE_SHIFT 12 121e7167769SMika Kuoppala #define GEN8_PTE_MASK 0x1ff 122e7167769SMika Kuoppala #define GEN8_PTES I915_PTES(sizeof(gen8_pte_t)) 123e7167769SMika Kuoppala 124e7167769SMika Kuoppala /* GEN8 48b style address is defined as a 4 level page table: 12581ba8aefSMichel Thierry * 47:39 | 38:30 | 29:21 | 20:12 | 11:0 12681ba8aefSMichel Thierry * PML4E | PDPE | PDE | PTE | offset 1270260c420SBen Widawsky */ 12881ba8aefSMichel Thierry #define GEN8_PML4ES_PER_PML4 512 12981ba8aefSMichel Thierry #define GEN8_PML4E_SHIFT 39 130762d9936SMichel Thierry #define GEN8_PML4E_MASK (GEN8_PML4ES_PER_PML4 - 1) 1310260c420SBen Widawsky #define GEN8_PDPE_SHIFT 30 13281ba8aefSMichel Thierry /* NB: GEN8_PDPE_MASK is untrue for 32b platforms, but it has no impact on 32b page 13381ba8aefSMichel Thierry * tables */ 13481ba8aefSMichel Thierry #define GEN8_PDPE_MASK 0x1ff 1350260c420SBen Widawsky 136c095b97cSZhi Wang #define PPAT_UNCACHED (_PAGE_PWT | _PAGE_PCD) 137c095b97cSZhi Wang #define PPAT_CACHED_PDE 0 /* WB LLC */ 138c095b97cSZhi Wang #define PPAT_CACHED _PAGE_PAT /* WB LLCeLLC */ 139c095b97cSZhi Wang #define PPAT_DISPLAY_ELLC _PAGE_PCD /* WT eLLC */ 1400260c420SBen Widawsky 141ee0ce478SVille Syrjälä #define CHV_PPAT_SNOOP (1<<6) 1421790625bSMichal Wajdeczko #define GEN8_PPAT_AGE(x) ((x)<<4) 1430260c420SBen Widawsky #define GEN8_PPAT_LLCeLLC (3<<2) 1440260c420SBen Widawsky #define GEN8_PPAT_LLCELLC (2<<2) 1450260c420SBen Widawsky #define GEN8_PPAT_LLC (1<<2) 1460260c420SBen Widawsky #define GEN8_PPAT_WB (3<<0) 1470260c420SBen Widawsky #define GEN8_PPAT_WT (2<<0) 1480260c420SBen Widawsky #define GEN8_PPAT_WC (1<<0) 1490260c420SBen Widawsky #define GEN8_PPAT_UC (0<<0) 1500260c420SBen Widawsky #define GEN8_PPAT_ELLC_OVERRIDE (0<<2) 15175c7b0b8SChris Wilson #define GEN8_PPAT(i, x) ((u64)(x) << ((i) * 8)) 1520260c420SBen Widawsky 1534395890aSZhi Wang #define GEN8_PPAT_GET_CA(x) ((x) & 3) 1544395890aSZhi Wang #define GEN8_PPAT_GET_TC(x) ((x) & (3 << 2)) 1554395890aSZhi Wang #define GEN8_PPAT_GET_AGE(x) ((x) & (3 << 4)) 1564395890aSZhi Wang #define CHV_PPAT_GET_SNOOP(x) ((x) & (1 << 6)) 1574395890aSZhi Wang 15817a00cf7SMatthew Auld #define GEN8_PDE_IPS_64K BIT(11) 1590a03852eSMatthew Auld #define GEN8_PDE_PS_2M BIT(7) 1600a03852eSMatthew Auld 161b42fe9caSJoonas Lahtinen struct sg_table; 162b42fe9caSJoonas Lahtinen 16350470bb0STvrtko Ursulin struct intel_rotation_info { 1647ff19c56SChris Wilson struct intel_rotation_plane_info { 1651663b9d6SVille Syrjälä /* tiles */ 1666687c906SVille Syrjälä unsigned int width, height, stride, offset; 1671663b9d6SVille Syrjälä } plane[2]; 1688d9046adSChris Wilson } __packed; 1698d9046adSChris Wilson 1708d9046adSChris Wilson static inline void assert_intel_rotation_info_is_packed(void) 1718d9046adSChris Wilson { 1728d9046adSChris Wilson BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 8*sizeof(unsigned int)); 1738d9046adSChris Wilson } 174fe14d5f4STvrtko Ursulin 1757ff19c56SChris Wilson struct intel_partial_info { 1767ff19c56SChris Wilson u64 offset; 1777ff19c56SChris Wilson unsigned int size; 1788d9046adSChris Wilson } __packed; 1798d9046adSChris Wilson 1808d9046adSChris Wilson static inline void assert_intel_partial_info_is_packed(void) 1818d9046adSChris Wilson { 1828d9046adSChris Wilson BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int)); 1838d9046adSChris Wilson } 1847ff19c56SChris Wilson 185992e418dSChris Wilson enum i915_ggtt_view_type { 186992e418dSChris Wilson I915_GGTT_VIEW_NORMAL = 0, 187992e418dSChris Wilson I915_GGTT_VIEW_ROTATED = sizeof(struct intel_rotation_info), 188992e418dSChris Wilson I915_GGTT_VIEW_PARTIAL = sizeof(struct intel_partial_info), 189992e418dSChris Wilson }; 190992e418dSChris Wilson 191992e418dSChris Wilson static inline void assert_i915_ggtt_view_type_is_unique(void) 192992e418dSChris Wilson { 193992e418dSChris Wilson /* As we encode the size of each branch inside the union into its type, 194992e418dSChris Wilson * we have to be careful that each branch has a unique size. 195992e418dSChris Wilson */ 196992e418dSChris Wilson switch ((enum i915_ggtt_view_type)0) { 197992e418dSChris Wilson case I915_GGTT_VIEW_NORMAL: 198992e418dSChris Wilson case I915_GGTT_VIEW_PARTIAL: 199992e418dSChris Wilson case I915_GGTT_VIEW_ROTATED: 200992e418dSChris Wilson /* gcc complains if these are identical cases */ 201992e418dSChris Wilson break; 202992e418dSChris Wilson } 203992e418dSChris Wilson } 204992e418dSChris Wilson 205fe14d5f4STvrtko Ursulin struct i915_ggtt_view { 206fe14d5f4STvrtko Ursulin enum i915_ggtt_view_type type; 2078bd7ef16SJoonas Lahtinen union { 208992e418dSChris Wilson /* Members need to contain no holes/padding */ 2097ff19c56SChris Wilson struct intel_partial_info partial; 2107723f47dSVille Syrjälä struct intel_rotation_info rotated; 2118bab1193SChris Wilson }; 212fe14d5f4STvrtko Ursulin }; 213fe14d5f4STvrtko Ursulin 2140260c420SBen Widawsky enum i915_cache_level; 215fe14d5f4STvrtko Ursulin 216b42fe9caSJoonas Lahtinen struct i915_vma; 217bde13ebdSChris Wilson 21844159ddbSMika Kuoppala struct i915_page_dma { 219d7b3de91SBen Widawsky struct page *page; 220aa095871SMatthew Auld int order; 22144159ddbSMika Kuoppala union { 2227324cc04SBen Widawsky dma_addr_t daddr; 223678d96fbSBen Widawsky 22444159ddbSMika Kuoppala /* For gen6/gen7 only. This is the offset in the GGTT 22544159ddbSMika Kuoppala * where the page directory entries for PPGTT begin 22644159ddbSMika Kuoppala */ 22775c7b0b8SChris Wilson u32 ggtt_offset; 22844159ddbSMika Kuoppala }; 22944159ddbSMika Kuoppala }; 23044159ddbSMika Kuoppala 231567047beSMika Kuoppala #define px_base(px) (&(px)->base) 232567047beSMika Kuoppala #define px_page(px) (px_base(px)->page) 233567047beSMika Kuoppala #define px_dma(px) (px_base(px)->daddr) 234567047beSMika Kuoppala 23544159ddbSMika Kuoppala struct i915_page_table { 23644159ddbSMika Kuoppala struct i915_page_dma base; 237dd19674bSChris Wilson unsigned int used_ptes; 238d7b3de91SBen Widawsky }; 239d7b3de91SBen Widawsky 240ec565b3cSMichel Thierry struct i915_page_directory { 24144159ddbSMika Kuoppala struct i915_page_dma base; 2427324cc04SBen Widawsky 243ec565b3cSMichel Thierry struct i915_page_table *page_table[I915_PDES]; /* PDEs */ 244fe52e37fSChris Wilson unsigned int used_pdes; 245d7b3de91SBen Widawsky }; 246d7b3de91SBen Widawsky 247ec565b3cSMichel Thierry struct i915_page_directory_pointer { 2486ac18502SMichel Thierry struct i915_page_dma base; 2496ac18502SMichel Thierry struct i915_page_directory **page_directory; 250e2b763caSChris Wilson unsigned int used_pdpes; 251d7b3de91SBen Widawsky }; 252d7b3de91SBen Widawsky 25381ba8aefSMichel Thierry struct i915_pml4 { 25481ba8aefSMichel Thierry struct i915_page_dma base; 25581ba8aefSMichel Thierry struct i915_page_directory_pointer *pdps[GEN8_PML4ES_PER_PML4]; 25681ba8aefSMichel Thierry }; 25781ba8aefSMichel Thierry 25893f2cde2SChris Wilson struct i915_vma_ops { 25993f2cde2SChris Wilson /* Map an object into an address space with the given cache flags. */ 26093f2cde2SChris Wilson int (*bind_vma)(struct i915_vma *vma, 26193f2cde2SChris Wilson enum i915_cache_level cache_level, 26293f2cde2SChris Wilson u32 flags); 26393f2cde2SChris Wilson /* 26493f2cde2SChris Wilson * Unmap an object from an address space. This usually consists of 26593f2cde2SChris Wilson * setting the valid PTE entries to a reserved scratch page. 26693f2cde2SChris Wilson */ 26793f2cde2SChris Wilson void (*unbind_vma)(struct i915_vma *vma); 26893f2cde2SChris Wilson 26993f2cde2SChris Wilson int (*set_pages)(struct i915_vma *vma); 27093f2cde2SChris Wilson void (*clear_pages)(struct i915_vma *vma); 27193f2cde2SChris Wilson }; 27293f2cde2SChris Wilson 2730260c420SBen Widawsky struct i915_address_space { 2740260c420SBen Widawsky struct drm_mm mm; 27549d73912SChris Wilson struct drm_i915_private *i915; 2768448661dSChris Wilson struct device *dma; 2772bfa996eSChris Wilson /* Every address space belongs to a struct file - except for the global 2782bfa996eSChris Wilson * GTT that is owned by the driver (and so @file is set to NULL). In 2792bfa996eSChris Wilson * principle, no information should leak from one context to another 2802bfa996eSChris Wilson * (or between files/processes etc) unless explicitly shared by the 2812bfa996eSChris Wilson * owner. Tracking the owner is important in order to free up per-file 2822bfa996eSChris Wilson * objects along with the file, to aide resource tracking, and to 2832bfa996eSChris Wilson * assign blame. 2842bfa996eSChris Wilson */ 2852bfa996eSChris Wilson struct drm_i915_file_private *file; 2860260c420SBen Widawsky struct list_head global_link; 287c44ef60eSMika Kuoppala u64 total; /* size addr space maps (ex. 2GB for ggtt) */ 288ff8f7975SWeinan Li u64 reserved; /* size addr space reserved */ 2890260c420SBen Widawsky 29050e046b6SChris Wilson bool closed; 29150e046b6SChris Wilson 2928bcdd0f7SChris Wilson struct i915_page_dma scratch_page; 29379ab9370SMika Kuoppala struct i915_page_table *scratch_pt; 29479ab9370SMika Kuoppala struct i915_page_directory *scratch_pd; 29569ab76fdSMichel Thierry struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */ 2960260c420SBen Widawsky 2970260c420SBen Widawsky /** 2980260c420SBen Widawsky * List of objects currently involved in rendering. 2990260c420SBen Widawsky * 3000260c420SBen Widawsky * Includes buffers having the contents of their GPU caches 30197b2a6a1SJohn Harrison * flushed, not necessarily primitives. last_read_req 3020260c420SBen Widawsky * represents when the rendering involved will be completed. 3030260c420SBen Widawsky * 3040260c420SBen Widawsky * A reference is held on the buffer while on this list. 3050260c420SBen Widawsky */ 3060260c420SBen Widawsky struct list_head active_list; 3070260c420SBen Widawsky 3080260c420SBen Widawsky /** 3090260c420SBen Widawsky * LRU list of objects which are not in the ringbuffer and 3100260c420SBen Widawsky * are ready to unbind, but are still in the GTT. 3110260c420SBen Widawsky * 31297b2a6a1SJohn Harrison * last_read_req is NULL while an object is in this list. 3130260c420SBen Widawsky * 3140260c420SBen Widawsky * A reference is not held on the buffer while on this list, 3150260c420SBen Widawsky * as merely being GTT-bound shouldn't prevent its being 3160260c420SBen Widawsky * freed, and we'll pull it off the list in the free path. 3170260c420SBen Widawsky */ 3180260c420SBen Widawsky struct list_head inactive_list; 3190260c420SBen Widawsky 32050e046b6SChris Wilson /** 32150e046b6SChris Wilson * List of vma that have been unbound. 32250e046b6SChris Wilson * 32350e046b6SChris Wilson * A reference is not held on the buffer while on this list. 32450e046b6SChris Wilson */ 32550e046b6SChris Wilson struct list_head unbound_list; 32650e046b6SChris Wilson 3278448661dSChris Wilson struct pagevec free_pages; 3288448661dSChris Wilson bool pt_kmap_wc; 3298448661dSChris Wilson 3300260c420SBen Widawsky /* FIXME: Need a more generic return type */ 33107749ef3SMichel Thierry gen6_pte_t (*pte_encode)(dma_addr_t addr, 3320260c420SBen Widawsky enum i915_cache_level level, 3334fb84d99SMichał Winiarski u32 flags); /* Create a valid PTE */ 334f329f5f6SDaniel Vetter /* flags for pte_encode */ 335f329f5f6SDaniel Vetter #define PTE_READ_ONLY (1<<0) 336678d96fbSBen Widawsky int (*allocate_va_range)(struct i915_address_space *vm, 33775c7b0b8SChris Wilson u64 start, u64 length); 3380260c420SBen Widawsky void (*clear_range)(struct i915_address_space *vm, 33975c7b0b8SChris Wilson u64 start, u64 length); 340d6473f56SChris Wilson void (*insert_page)(struct i915_address_space *vm, 341d6473f56SChris Wilson dma_addr_t addr, 34275c7b0b8SChris Wilson u64 offset, 343d6473f56SChris Wilson enum i915_cache_level cache_level, 344d6473f56SChris Wilson u32 flags); 3450260c420SBen Widawsky void (*insert_entries)(struct i915_address_space *vm, 3464a234c5fSMatthew Auld struct i915_vma *vma, 34775c7b0b8SChris Wilson enum i915_cache_level cache_level, 34875c7b0b8SChris Wilson u32 flags); 3490260c420SBen Widawsky void (*cleanup)(struct i915_address_space *vm); 35093f2cde2SChris Wilson 35193f2cde2SChris Wilson struct i915_vma_ops vma_ops; 3528448661dSChris Wilson 3538448661dSChris Wilson I915_SELFTEST_DECLARE(struct fault_attr fault_attr); 354f79401b4SMatthew Auld I915_SELFTEST_DECLARE(bool scrub_64K); 3550260c420SBen Widawsky }; 3560260c420SBen Widawsky 3572bfa996eSChris Wilson #define i915_is_ggtt(V) (!(V)->file) 358596c5923SChris Wilson 3593e490042SMika Kuoppala static inline bool 3603e490042SMika Kuoppala i915_vm_is_48bit(const struct i915_address_space *vm) 3613e490042SMika Kuoppala { 3623e490042SMika Kuoppala return (vm->total - 1) >> 32; 3633e490042SMika Kuoppala } 3643e490042SMika Kuoppala 36517a00cf7SMatthew Auld static inline bool 36617a00cf7SMatthew Auld i915_vm_has_scratch_64K(struct i915_address_space *vm) 36717a00cf7SMatthew Auld { 36817a00cf7SMatthew Auld return vm->scratch_page.order == get_order(I915_GTT_PAGE_SIZE_64K); 36917a00cf7SMatthew Auld } 37017a00cf7SMatthew Auld 3710260c420SBen Widawsky /* The Graphics Translation Table is the way in which GEN hardware translates a 3720260c420SBen Widawsky * Graphics Virtual Address into a Physical Address. In addition to the normal 3730260c420SBen Widawsky * collateral associated with any va->pa translations GEN hardware also has a 3740260c420SBen Widawsky * portion of the GTT which can be mapped by the CPU and remain both coherent 3750260c420SBen Widawsky * and correct (in cases like swizzling). That region is referred to as GMADR in 3760260c420SBen Widawsky * the spec. 3770260c420SBen Widawsky */ 37862106b4fSJoonas Lahtinen struct i915_ggtt { 37982ad6443SChris Wilson struct i915_address_space vm; 3800260c420SBen Widawsky 38173ebd503SMatthew Auld struct io_mapping iomap; /* Mapping to our CPU mappable region */ 38273ebd503SMatthew Auld struct resource gmadr; /* GMADR resource */ 383b7128ef1SMatthew Auld resource_size_t mappable_end; /* End offset that we can CPU map */ 384edd1f2feSChris Wilson 3850260c420SBen Widawsky /** "Graphics Stolen Memory" holds the global PTEs */ 3860260c420SBen Widawsky void __iomem *gsm; 3877c3f86b6SChris Wilson void (*invalidate)(struct drm_i915_private *dev_priv); 3880260c420SBen Widawsky 3890260c420SBen Widawsky bool do_idle_maps; 3900260c420SBen Widawsky 3910260c420SBen Widawsky int mtrr; 39295374d75SChris Wilson 39395374d75SChris Wilson struct drm_mm_node error_capture; 3940260c420SBen Widawsky }; 3950260c420SBen Widawsky 3960260c420SBen Widawsky struct i915_hw_ppgtt { 39782ad6443SChris Wilson struct i915_address_space vm; 3980260c420SBen Widawsky struct kref ref; 39935ac40d8SChris Wilson 400563222a7SBen Widawsky unsigned long pd_dirty_rings; 4010260c420SBen Widawsky union { 40281ba8aefSMichel Thierry struct i915_pml4 pml4; /* GEN8+ & 48b PPGTT */ 40381ba8aefSMichel Thierry struct i915_page_directory_pointer pdp; /* GEN8+ */ 40481ba8aefSMichel Thierry struct i915_page_directory pd; /* GEN6-7 */ 405d7b3de91SBen Widawsky }; 4060260c420SBen Widawsky 4070260c420SBen Widawsky void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m); 4080260c420SBen Widawsky }; 4090260c420SBen Widawsky 41035ac40d8SChris Wilson struct gen6_hw_ppgtt { 41135ac40d8SChris Wilson struct i915_hw_ppgtt base; 41235ac40d8SChris Wilson 413e9e7dc41SChris Wilson struct i915_vma *vma; 41435ac40d8SChris Wilson gen6_pte_t __iomem *pd_addr; 415986dbac4SChris Wilson gen6_pte_t scratch_pte; 416a2bbf714SChris Wilson 417a2bbf714SChris Wilson unsigned int pin_count; 4184a192c7eSChris Wilson bool scan_for_unused_pt; 41935ac40d8SChris Wilson }; 42035ac40d8SChris Wilson 42135ac40d8SChris Wilson #define __to_gen6_ppgtt(base) container_of(base, struct gen6_hw_ppgtt, base) 42235ac40d8SChris Wilson 42335ac40d8SChris Wilson static inline struct gen6_hw_ppgtt *to_gen6_ppgtt(struct i915_hw_ppgtt *base) 42435ac40d8SChris Wilson { 42535ac40d8SChris Wilson BUILD_BUG_ON(offsetof(struct gen6_hw_ppgtt, base)); 42635ac40d8SChris Wilson return __to_gen6_ppgtt(base); 42735ac40d8SChris Wilson } 42835ac40d8SChris Wilson 429731f74c5SDave Gordon /* 430731f74c5SDave Gordon * gen6_for_each_pde() iterates over every pde from start until start+length. 431731f74c5SDave Gordon * If start and start+length are not perfectly divisible, the macro will round 432731f74c5SDave Gordon * down and up as needed. Start=0 and length=2G effectively iterates over 433731f74c5SDave Gordon * every PDE in the system. The macro modifies ALL its parameters except 'pd', 434731f74c5SDave Gordon * so each of the other parameters should preferably be a simple variable, or 435731f74c5SDave Gordon * at most an lvalue with no side-effects! 436678d96fbSBen Widawsky */ 437731f74c5SDave Gordon #define gen6_for_each_pde(pt, pd, start, length, iter) \ 438fdc454c1SMichel Thierry for (iter = gen6_pde_index(start); \ 439731f74c5SDave Gordon length > 0 && iter < I915_PDES && \ 440731f74c5SDave Gordon (pt = (pd)->page_table[iter], true); \ 441731f74c5SDave Gordon ({ u32 temp = ALIGN(start+1, 1 << GEN6_PDE_SHIFT); \ 442731f74c5SDave Gordon temp = min(temp - start, length); \ 443731f74c5SDave Gordon start += temp, length -= temp; }), ++iter) 444678d96fbSBen Widawsky 445731f74c5SDave Gordon #define gen6_for_all_pdes(pt, pd, iter) \ 44609942c65SMichel Thierry for (iter = 0; \ 447731f74c5SDave Gordon iter < I915_PDES && \ 448731f74c5SDave Gordon (pt = (pd)->page_table[iter], true); \ 449731f74c5SDave Gordon ++iter) 45009942c65SMichel Thierry 45175c7b0b8SChris Wilson static inline u32 i915_pte_index(u64 address, unsigned int pde_shift) 452678d96fbSBen Widawsky { 45375c7b0b8SChris Wilson const u32 mask = NUM_PTE(pde_shift) - 1; 454678d96fbSBen Widawsky 455678d96fbSBen Widawsky return (address >> PAGE_SHIFT) & mask; 456678d96fbSBen Widawsky } 457678d96fbSBen Widawsky 458678d96fbSBen Widawsky /* Helper to counts the number of PTEs within the given length. This count 459678d96fbSBen Widawsky * does not cross a page table boundary, so the max value would be 460678d96fbSBen Widawsky * GEN6_PTES for GEN6, and GEN8_PTES for GEN8. 461678d96fbSBen Widawsky */ 46275c7b0b8SChris Wilson static inline u32 i915_pte_count(u64 addr, u64 length, unsigned int pde_shift) 463678d96fbSBen Widawsky { 46475c7b0b8SChris Wilson const u64 mask = ~((1ULL << pde_shift) - 1); 46575c7b0b8SChris Wilson u64 end; 466678d96fbSBen Widawsky 467678d96fbSBen Widawsky WARN_ON(length == 0); 468678d96fbSBen Widawsky WARN_ON(offset_in_page(addr|length)); 469678d96fbSBen Widawsky 470678d96fbSBen Widawsky end = addr + length; 471678d96fbSBen Widawsky 472678d96fbSBen Widawsky if ((addr & mask) != (end & mask)) 473678d96fbSBen Widawsky return NUM_PTE(pde_shift) - i915_pte_index(addr, pde_shift); 474678d96fbSBen Widawsky 475678d96fbSBen Widawsky return i915_pte_index(end, pde_shift) - i915_pte_index(addr, pde_shift); 476678d96fbSBen Widawsky } 477678d96fbSBen Widawsky 47875c7b0b8SChris Wilson static inline u32 i915_pde_index(u64 addr, u32 shift) 479678d96fbSBen Widawsky { 480678d96fbSBen Widawsky return (addr >> shift) & I915_PDE_MASK; 481678d96fbSBen Widawsky } 482678d96fbSBen Widawsky 48375c7b0b8SChris Wilson static inline u32 gen6_pte_index(u32 addr) 484678d96fbSBen Widawsky { 485678d96fbSBen Widawsky return i915_pte_index(addr, GEN6_PDE_SHIFT); 486678d96fbSBen Widawsky } 487678d96fbSBen Widawsky 48875c7b0b8SChris Wilson static inline u32 gen6_pte_count(u32 addr, u32 length) 489678d96fbSBen Widawsky { 490678d96fbSBen Widawsky return i915_pte_count(addr, length, GEN6_PDE_SHIFT); 491678d96fbSBen Widawsky } 492678d96fbSBen Widawsky 49375c7b0b8SChris Wilson static inline u32 gen6_pde_index(u32 addr) 494678d96fbSBen Widawsky { 495678d96fbSBen Widawsky return i915_pde_index(addr, GEN6_PDE_SHIFT); 496678d96fbSBen Widawsky } 497678d96fbSBen Widawsky 4983e490042SMika Kuoppala static inline unsigned int 4993e490042SMika Kuoppala i915_pdpes_per_pdp(const struct i915_address_space *vm) 5003e490042SMika Kuoppala { 5013e490042SMika Kuoppala if (i915_vm_is_48bit(vm)) 5023e490042SMika Kuoppala return GEN8_PML4ES_PER_PML4; 5033e490042SMika Kuoppala 504e7167769SMika Kuoppala return GEN8_3LVL_PDPES; 5053e490042SMika Kuoppala } 5063e490042SMika Kuoppala 5079271d959SMichel Thierry /* Equivalent to the gen6 version, For each pde iterates over every pde 5089271d959SMichel Thierry * between from start until start + length. On gen8+ it simply iterates 5099271d959SMichel Thierry * over every page directory entry in a page directory. 5109271d959SMichel Thierry */ 511e8ebd8e2SDave Gordon #define gen8_for_each_pde(pt, pd, start, length, iter) \ 5129271d959SMichel Thierry for (iter = gen8_pde_index(start); \ 513e8ebd8e2SDave Gordon length > 0 && iter < I915_PDES && \ 514e8ebd8e2SDave Gordon (pt = (pd)->page_table[iter], true); \ 515e8ebd8e2SDave Gordon ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDE_SHIFT); \ 516e8ebd8e2SDave Gordon temp = min(temp - start, length); \ 517e8ebd8e2SDave Gordon start += temp, length -= temp; }), ++iter) 5189271d959SMichel Thierry 519e8ebd8e2SDave Gordon #define gen8_for_each_pdpe(pd, pdp, start, length, iter) \ 5209271d959SMichel Thierry for (iter = gen8_pdpe_index(start); \ 5213e490042SMika Kuoppala length > 0 && iter < i915_pdpes_per_pdp(vm) && \ 522e8ebd8e2SDave Gordon (pd = (pdp)->page_directory[iter], true); \ 523e8ebd8e2SDave Gordon ({ u64 temp = ALIGN(start+1, 1 << GEN8_PDPE_SHIFT); \ 524e8ebd8e2SDave Gordon temp = min(temp - start, length); \ 525e8ebd8e2SDave Gordon start += temp, length -= temp; }), ++iter) 5269271d959SMichel Thierry 527e8ebd8e2SDave Gordon #define gen8_for_each_pml4e(pdp, pml4, start, length, iter) \ 528762d9936SMichel Thierry for (iter = gen8_pml4e_index(start); \ 529e8ebd8e2SDave Gordon length > 0 && iter < GEN8_PML4ES_PER_PML4 && \ 530e8ebd8e2SDave Gordon (pdp = (pml4)->pdps[iter], true); \ 531e8ebd8e2SDave Gordon ({ u64 temp = ALIGN(start+1, 1ULL << GEN8_PML4E_SHIFT); \ 532e8ebd8e2SDave Gordon temp = min(temp - start, length); \ 533e8ebd8e2SDave Gordon start += temp, length -= temp; }), ++iter) 534762d9936SMichel Thierry 53575c7b0b8SChris Wilson static inline u32 gen8_pte_index(u64 address) 5369271d959SMichel Thierry { 5379271d959SMichel Thierry return i915_pte_index(address, GEN8_PDE_SHIFT); 5389271d959SMichel Thierry } 5399271d959SMichel Thierry 54075c7b0b8SChris Wilson static inline u32 gen8_pde_index(u64 address) 5419271d959SMichel Thierry { 5429271d959SMichel Thierry return i915_pde_index(address, GEN8_PDE_SHIFT); 5439271d959SMichel Thierry } 5449271d959SMichel Thierry 54575c7b0b8SChris Wilson static inline u32 gen8_pdpe_index(u64 address) 5469271d959SMichel Thierry { 5479271d959SMichel Thierry return (address >> GEN8_PDPE_SHIFT) & GEN8_PDPE_MASK; 5489271d959SMichel Thierry } 5499271d959SMichel Thierry 55075c7b0b8SChris Wilson static inline u32 gen8_pml4e_index(u64 address) 5519271d959SMichel Thierry { 552762d9936SMichel Thierry return (address >> GEN8_PML4E_SHIFT) & GEN8_PML4E_MASK; 5539271d959SMichel Thierry } 5549271d959SMichel Thierry 55575c7b0b8SChris Wilson static inline u64 gen8_pte_count(u64 address, u64 length) 55633c8819fSMichel Thierry { 55733c8819fSMichel Thierry return i915_pte_count(address, length, GEN8_PDE_SHIFT); 55833c8819fSMichel Thierry } 55933c8819fSMichel Thierry 560d852c7bfSMika Kuoppala static inline dma_addr_t 561d852c7bfSMika Kuoppala i915_page_dir_dma_addr(const struct i915_hw_ppgtt *ppgtt, const unsigned n) 562d852c7bfSMika Kuoppala { 563fe52e37fSChris Wilson return px_dma(ppgtt->pdp.page_directory[n]); 564d852c7bfSMika Kuoppala } 565d852c7bfSMika Kuoppala 566b42fe9caSJoonas Lahtinen static inline struct i915_ggtt * 567b42fe9caSJoonas Lahtinen i915_vm_to_ggtt(struct i915_address_space *vm) 568b42fe9caSJoonas Lahtinen { 569b42fe9caSJoonas Lahtinen GEM_BUG_ON(!i915_is_ggtt(vm)); 57082ad6443SChris Wilson return container_of(vm, struct i915_ggtt, vm); 571b42fe9caSJoonas Lahtinen } 572b42fe9caSJoonas Lahtinen 5734395890aSZhi Wang #define INTEL_MAX_PPAT_ENTRIES 8 5744395890aSZhi Wang #define INTEL_PPAT_PERFECT_MATCH (~0U) 5754395890aSZhi Wang 5764395890aSZhi Wang struct intel_ppat; 5774395890aSZhi Wang 5784395890aSZhi Wang struct intel_ppat_entry { 5794395890aSZhi Wang struct intel_ppat *ppat; 5804395890aSZhi Wang struct kref ref; 5814395890aSZhi Wang u8 value; 5824395890aSZhi Wang }; 5834395890aSZhi Wang 5844395890aSZhi Wang struct intel_ppat { 5854395890aSZhi Wang struct intel_ppat_entry entries[INTEL_MAX_PPAT_ENTRIES]; 5864395890aSZhi Wang DECLARE_BITMAP(used, INTEL_MAX_PPAT_ENTRIES); 5874395890aSZhi Wang DECLARE_BITMAP(dirty, INTEL_MAX_PPAT_ENTRIES); 5884395890aSZhi Wang unsigned int max_entries; 5894395890aSZhi Wang u8 clear_value; 5904395890aSZhi Wang /* 5914395890aSZhi Wang * Return a score to show how two PPAT values match, 5924395890aSZhi Wang * a INTEL_PPAT_PERFECT_MATCH indicates a perfect match 5934395890aSZhi Wang */ 5944395890aSZhi Wang unsigned int (*match)(u8 src, u8 dst); 5954395890aSZhi Wang void (*update_hw)(struct drm_i915_private *i915); 5964395890aSZhi Wang 5974395890aSZhi Wang struct drm_i915_private *i915; 5984395890aSZhi Wang }; 5994395890aSZhi Wang 6004395890aSZhi Wang const struct intel_ppat_entry * 6014395890aSZhi Wang intel_ppat_get(struct drm_i915_private *i915, u8 value); 6024395890aSZhi Wang void intel_ppat_put(const struct intel_ppat_entry *entry); 6034395890aSZhi Wang 6046cde9a02SChris Wilson int i915_gem_init_aliasing_ppgtt(struct drm_i915_private *i915); 6056cde9a02SChris Wilson void i915_gem_fini_aliasing_ppgtt(struct drm_i915_private *i915); 6066cde9a02SChris Wilson 60797d6d7abSChris Wilson int i915_ggtt_probe_hw(struct drm_i915_private *dev_priv); 60897d6d7abSChris Wilson int i915_ggtt_init_hw(struct drm_i915_private *dev_priv); 60997d6d7abSChris Wilson int i915_ggtt_enable_hw(struct drm_i915_private *dev_priv); 6107c3f86b6SChris Wilson void i915_ggtt_enable_guc(struct drm_i915_private *i915); 6117c3f86b6SChris Wilson void i915_ggtt_disable_guc(struct drm_i915_private *i915); 612f6b9d5caSChris Wilson int i915_gem_init_ggtt(struct drm_i915_private *dev_priv); 61397d6d7abSChris Wilson void i915_ggtt_cleanup_hw(struct drm_i915_private *dev_priv); 614ee960be7SDaniel Vetter 615c6be607aSTvrtko Ursulin int i915_ppgtt_init_hw(struct drm_i915_private *dev_priv); 616ee960be7SDaniel Vetter void i915_ppgtt_release(struct kref *kref); 6172bfa996eSChris Wilson struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_i915_private *dev_priv, 61880b204bcSChris Wilson struct drm_i915_file_private *fpriv, 61980b204bcSChris Wilson const char *name); 6200c7eeda1SChris Wilson void i915_ppgtt_close(struct i915_address_space *vm); 621ee960be7SDaniel Vetter static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt) 622ee960be7SDaniel Vetter { 623ee960be7SDaniel Vetter if (ppgtt) 624ee960be7SDaniel Vetter kref_get(&ppgtt->ref); 625ee960be7SDaniel Vetter } 626ee960be7SDaniel Vetter static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt) 627ee960be7SDaniel Vetter { 628ee960be7SDaniel Vetter if (ppgtt) 629ee960be7SDaniel Vetter kref_put(&ppgtt->ref, i915_ppgtt_release); 630ee960be7SDaniel Vetter } 6310260c420SBen Widawsky 632a2bbf714SChris Wilson int gen6_ppgtt_pin(struct i915_hw_ppgtt *base); 633a2bbf714SChris Wilson void gen6_ppgtt_unpin(struct i915_hw_ppgtt *base); 634a2bbf714SChris Wilson 635dc97997aSChris Wilson void i915_check_and_clear_faults(struct drm_i915_private *dev_priv); 636275a991cSTvrtko Ursulin void i915_gem_suspend_gtt_mappings(struct drm_i915_private *dev_priv); 637275a991cSTvrtko Ursulin void i915_gem_restore_gtt_mappings(struct drm_i915_private *dev_priv); 6380260c420SBen Widawsky 63903ac84f1SChris Wilson int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj, 64003ac84f1SChris Wilson struct sg_table *pages); 64103ac84f1SChris Wilson void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj, 64203ac84f1SChris Wilson struct sg_table *pages); 6430260c420SBen Widawsky 644625d988aSChris Wilson int i915_gem_gtt_reserve(struct i915_address_space *vm, 645625d988aSChris Wilson struct drm_mm_node *node, 646625d988aSChris Wilson u64 size, u64 offset, unsigned long color, 647625d988aSChris Wilson unsigned int flags); 648625d988aSChris Wilson 649e007b19dSChris Wilson int i915_gem_gtt_insert(struct i915_address_space *vm, 650e007b19dSChris Wilson struct drm_mm_node *node, 651e007b19dSChris Wilson u64 size, u64 alignment, unsigned long color, 652e007b19dSChris Wilson u64 start, u64 end, unsigned int flags); 653e007b19dSChris Wilson 65459bfa124SChris Wilson /* Flags used by pin/bind&friends. */ 655305bc234SChris Wilson #define PIN_NONBLOCK BIT(0) 656305bc234SChris Wilson #define PIN_MAPPABLE BIT(1) 657305bc234SChris Wilson #define PIN_ZONE_4G BIT(2) 65882118877SChris Wilson #define PIN_NONFAULT BIT(3) 659616d9ceeSChris Wilson #define PIN_NOEVICT BIT(4) 660305bc234SChris Wilson 661305bc234SChris Wilson #define PIN_MBZ BIT(5) /* I915_VMA_PIN_OVERFLOW */ 662305bc234SChris Wilson #define PIN_GLOBAL BIT(6) /* I915_VMA_GLOBAL_BIND */ 663305bc234SChris Wilson #define PIN_USER BIT(7) /* I915_VMA_LOCAL_BIND */ 664305bc234SChris Wilson #define PIN_UPDATE BIT(8) 665305bc234SChris Wilson 666305bc234SChris Wilson #define PIN_HIGH BIT(9) 667305bc234SChris Wilson #define PIN_OFFSET_BIAS BIT(10) 668305bc234SChris Wilson #define PIN_OFFSET_FIXED BIT(11) 669f51455d4SChris Wilson #define PIN_OFFSET_MASK (-I915_GTT_PAGE_SIZE) 67059bfa124SChris Wilson 6710260c420SBen Widawsky #endif 672