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_gtt_pte_t; 40 typedef uint64_t gen8_gtt_pte_t; 41 typedef gen8_gtt_pte_t gen8_ppgtt_pde_t; 42 43 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT) 44 45 #define I915_PPGTT_PT_ENTRIES (PAGE_SIZE / sizeof(gen6_gtt_pte_t)) 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 GEN6_PPGTT_PD_ENTRIES 512 55 #define GEN6_PD_SIZE (GEN6_PPGTT_PD_ENTRIES * PAGE_SIZE) 56 #define GEN6_PD_ALIGN (PAGE_SIZE * 16) 57 #define GEN6_PDE_VALID (1 << 0) 58 59 #define GEN7_PTE_CACHE_L3_LLC (3 << 1) 60 61 #define BYT_PTE_SNOOPED_BY_CPU_CACHES (1 << 2) 62 #define BYT_PTE_WRITEABLE (1 << 1) 63 64 /* Cacheability Control is a 4-bit value. The low three bits are stored in bits 65 * 3:1 of the PTE, while the fourth bit is stored in bit 11 of the PTE. 66 */ 67 #define HSW_CACHEABILITY_CONTROL(bits) ((((bits) & 0x7) << 1) | \ 68 (((bits) & 0x8) << (11 - 3))) 69 #define HSW_WB_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x2) 70 #define HSW_WB_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x3) 71 #define HSW_WB_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x8) 72 #define HSW_WB_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0xb) 73 #define HSW_WT_ELLC_LLC_AGE3 HSW_CACHEABILITY_CONTROL(0x7) 74 #define HSW_WT_ELLC_LLC_AGE0 HSW_CACHEABILITY_CONTROL(0x6) 75 #define HSW_PTE_UNCACHED (0) 76 #define HSW_GTT_ADDR_ENCODE(addr) ((addr) | (((addr) >> 28) & 0x7f0)) 77 #define HSW_PTE_ADDR_ENCODE(addr) HSW_GTT_ADDR_ENCODE(addr) 78 79 /* GEN8 legacy style address is defined as a 3 level page table: 80 * 31:30 | 29:21 | 20:12 | 11:0 81 * PDPE | PDE | PTE | offset 82 * The difference as compared to normal x86 3 level page table is the PDPEs are 83 * programmed via register. 84 */ 85 #define GEN8_PDPE_SHIFT 30 86 #define GEN8_PDPE_MASK 0x3 87 #define GEN8_PDE_SHIFT 21 88 #define GEN8_PDE_MASK 0x1ff 89 #define GEN8_PTE_SHIFT 12 90 #define GEN8_PTE_MASK 0x1ff 91 #define GEN8_LEGACY_PDPS 4 92 #define GEN8_PTES_PER_PAGE (PAGE_SIZE / sizeof(gen8_gtt_pte_t)) 93 #define GEN8_PDES_PER_PAGE (PAGE_SIZE / sizeof(gen8_ppgtt_pde_t)) 94 95 #define PPAT_UNCACHED_INDEX (_PAGE_PWT | _PAGE_PCD) 96 #define PPAT_CACHED_PDE_INDEX 0 /* WB LLC */ 97 #define PPAT_CACHED_INDEX _PAGE_PAT /* WB LLCeLLC */ 98 #define PPAT_DISPLAY_ELLC_INDEX _PAGE_PCD /* WT eLLC */ 99 100 #define CHV_PPAT_SNOOP (1<<6) 101 #define GEN8_PPAT_AGE(x) (x<<4) 102 #define GEN8_PPAT_LLCeLLC (3<<2) 103 #define GEN8_PPAT_LLCELLC (2<<2) 104 #define GEN8_PPAT_LLC (1<<2) 105 #define GEN8_PPAT_WB (3<<0) 106 #define GEN8_PPAT_WT (2<<0) 107 #define GEN8_PPAT_WC (1<<0) 108 #define GEN8_PPAT_UC (0<<0) 109 #define GEN8_PPAT_ELLC_OVERRIDE (0<<2) 110 #define GEN8_PPAT(i, x) ((uint64_t) (x) << ((i) * 8)) 111 112 enum i915_ggtt_view_type { 113 I915_GGTT_VIEW_NORMAL = 0, 114 }; 115 116 struct i915_ggtt_view { 117 enum i915_ggtt_view_type type; 118 119 struct sg_table *pages; 120 }; 121 122 extern const struct i915_ggtt_view i915_ggtt_view_normal; 123 124 enum i915_cache_level; 125 126 /** 127 * A VMA represents a GEM BO that is bound into an address space. Therefore, a 128 * VMA's presence cannot be guaranteed before binding, or after unbinding the 129 * object into/from the address space. 130 * 131 * To make things as simple as possible (ie. no refcounting), a VMA's lifetime 132 * will always be <= an objects lifetime. So object refcounting should cover us. 133 */ 134 struct i915_vma { 135 struct drm_mm_node node; 136 struct drm_i915_gem_object *obj; 137 struct i915_address_space *vm; 138 139 /** Flags and address space this VMA is bound to */ 140 #define GLOBAL_BIND (1<<0) 141 #define LOCAL_BIND (1<<1) 142 #define PTE_READ_ONLY (1<<2) 143 unsigned int bound : 4; 144 145 /** 146 * Support different GGTT views into the same object. 147 * This means there can be multiple VMA mappings per object and per VM. 148 * i915_ggtt_view_type is used to distinguish between those entries. 149 * The default one of zero (I915_GGTT_VIEW_NORMAL) is default and also 150 * assumed in GEM functions which take no ggtt view parameter. 151 */ 152 struct i915_ggtt_view ggtt_view; 153 154 /** This object's place on the active/inactive lists */ 155 struct list_head mm_list; 156 157 struct list_head vma_link; /* Link in the object's VMA list */ 158 159 /** This vma's place in the batchbuffer or on the eviction list */ 160 struct list_head exec_list; 161 162 /** 163 * Used for performing relocations during execbuffer insertion. 164 */ 165 struct hlist_node exec_node; 166 unsigned long exec_handle; 167 struct drm_i915_gem_exec_object2 *exec_entry; 168 169 /** 170 * How many users have pinned this object in GTT space. The following 171 * users can each hold at most one reference: pwrite/pread, execbuffer 172 * (objects are not allowed multiple times for the same batchbuffer), 173 * and the framebuffer code. When switching/pageflipping, the 174 * framebuffer code has at most two buffers pinned per crtc. 175 * 176 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3 177 * bits with absolutely no headroom. So use 4 bits. */ 178 unsigned int pin_count:4; 179 #define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf 180 181 /** Unmap an object from an address space. This usually consists of 182 * setting the valid PTE entries to a reserved scratch page. */ 183 void (*unbind_vma)(struct i915_vma *vma); 184 /* Map an object into an address space with the given cache flags. */ 185 void (*bind_vma)(struct i915_vma *vma, 186 enum i915_cache_level cache_level, 187 u32 flags); 188 }; 189 190 struct i915_address_space { 191 struct drm_mm mm; 192 struct drm_device *dev; 193 struct list_head global_link; 194 unsigned long start; /* Start offset always 0 for dri2 */ 195 size_t total; /* size addr space maps (ex. 2GB for ggtt) */ 196 197 struct { 198 dma_addr_t addr; 199 struct page *page; 200 } scratch; 201 202 /** 203 * List of objects currently involved in rendering. 204 * 205 * Includes buffers having the contents of their GPU caches 206 * flushed, not necessarily primitives. last_read_req 207 * represents when the rendering involved will be completed. 208 * 209 * A reference is held on the buffer while on this list. 210 */ 211 struct list_head active_list; 212 213 /** 214 * LRU list of objects which are not in the ringbuffer and 215 * are ready to unbind, but are still in the GTT. 216 * 217 * last_read_req is NULL while an object is in this list. 218 * 219 * A reference is not held on the buffer while on this list, 220 * as merely being GTT-bound shouldn't prevent its being 221 * freed, and we'll pull it off the list in the free path. 222 */ 223 struct list_head inactive_list; 224 225 /* FIXME: Need a more generic return type */ 226 gen6_gtt_pte_t (*pte_encode)(dma_addr_t addr, 227 enum i915_cache_level level, 228 bool valid, u32 flags); /* Create a valid PTE */ 229 void (*clear_range)(struct i915_address_space *vm, 230 uint64_t start, 231 uint64_t length, 232 bool use_scratch); 233 void (*insert_entries)(struct i915_address_space *vm, 234 struct sg_table *st, 235 uint64_t start, 236 enum i915_cache_level cache_level, u32 flags); 237 void (*cleanup)(struct i915_address_space *vm); 238 }; 239 240 /* The Graphics Translation Table is the way in which GEN hardware translates a 241 * Graphics Virtual Address into a Physical Address. In addition to the normal 242 * collateral associated with any va->pa translations GEN hardware also has a 243 * portion of the GTT which can be mapped by the CPU and remain both coherent 244 * and correct (in cases like swizzling). That region is referred to as GMADR in 245 * the spec. 246 */ 247 struct i915_gtt { 248 struct i915_address_space base; 249 size_t stolen_size; /* Total size of stolen memory */ 250 251 unsigned long mappable_end; /* End offset that we can CPU map */ 252 struct io_mapping *mappable; /* Mapping to our CPU mappable region */ 253 phys_addr_t mappable_base; /* PA of our GMADR */ 254 255 /** "Graphics Stolen Memory" holds the global PTEs */ 256 void __iomem *gsm; 257 258 bool do_idle_maps; 259 260 int mtrr; 261 262 /* global gtt ops */ 263 int (*gtt_probe)(struct drm_device *dev, size_t *gtt_total, 264 size_t *stolen, phys_addr_t *mappable_base, 265 unsigned long *mappable_end); 266 }; 267 268 struct i915_hw_ppgtt { 269 struct i915_address_space base; 270 struct kref ref; 271 struct drm_mm_node node; 272 unsigned num_pd_entries; 273 unsigned num_pd_pages; /* gen8+ */ 274 union { 275 struct page **pt_pages; 276 struct page **gen8_pt_pages[GEN8_LEGACY_PDPS]; 277 }; 278 struct page *pd_pages; 279 union { 280 uint32_t pd_offset; 281 dma_addr_t pd_dma_addr[GEN8_LEGACY_PDPS]; 282 }; 283 union { 284 dma_addr_t *pt_dma_addr; 285 dma_addr_t *gen8_pt_dma_addr[4]; 286 }; 287 288 struct drm_i915_file_private *file_priv; 289 290 int (*enable)(struct i915_hw_ppgtt *ppgtt); 291 int (*switch_mm)(struct i915_hw_ppgtt *ppgtt, 292 struct intel_engine_cs *ring); 293 void (*debug_dump)(struct i915_hw_ppgtt *ppgtt, struct seq_file *m); 294 }; 295 296 int i915_gem_gtt_init(struct drm_device *dev); 297 void i915_gem_init_global_gtt(struct drm_device *dev); 298 void i915_global_gtt_cleanup(struct drm_device *dev); 299 300 301 int i915_ppgtt_init(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt); 302 int i915_ppgtt_init_hw(struct drm_device *dev); 303 void i915_ppgtt_release(struct kref *kref); 304 struct i915_hw_ppgtt *i915_ppgtt_create(struct drm_device *dev, 305 struct drm_i915_file_private *fpriv); 306 static inline void i915_ppgtt_get(struct i915_hw_ppgtt *ppgtt) 307 { 308 if (ppgtt) 309 kref_get(&ppgtt->ref); 310 } 311 static inline void i915_ppgtt_put(struct i915_hw_ppgtt *ppgtt) 312 { 313 if (ppgtt) 314 kref_put(&ppgtt->ref, i915_ppgtt_release); 315 } 316 317 void i915_check_and_clear_faults(struct drm_device *dev); 318 void i915_gem_suspend_gtt_mappings(struct drm_device *dev); 319 void i915_gem_restore_gtt_mappings(struct drm_device *dev); 320 321 int __must_check i915_gem_gtt_prepare_object(struct drm_i915_gem_object *obj); 322 void i915_gem_gtt_finish_object(struct drm_i915_gem_object *obj); 323 324 #endif 325