i915_gem_gtt.c revision 1.1.1.4
1 /* $NetBSD: i915_gem_gtt.c,v 1.1.1.4 2021/12/18 20:15:25 riastradh Exp $ */ 2 3 // SPDX-License-Identifier: MIT 4 /* 5 * Copyright 2010 Daniel Vetter 6 * Copyright 2020 Intel Corporation 7 */ 8 9 #include <sys/cdefs.h> 10 __KERNEL_RCSID(0, "$NetBSD: i915_gem_gtt.c,v 1.1.1.4 2021/12/18 20:15:25 riastradh Exp $"); 11 12 #include <linux/slab.h> /* fault-inject.h is not standalone! */ 13 14 #include <linux/fault-inject.h> 15 #include <linux/log2.h> 16 #include <linux/random.h> 17 #include <linux/seq_file.h> 18 #include <linux/stop_machine.h> 19 20 #include <asm/set_memory.h> 21 #include <asm/smp.h> 22 23 #include <drm/i915_drm.h> 24 25 #include "display/intel_frontbuffer.h" 26 #include "gt/intel_gt.h" 27 #include "gt/intel_gt_requests.h" 28 29 #include "i915_drv.h" 30 #include "i915_scatterlist.h" 31 #include "i915_trace.h" 32 #include "i915_vgpu.h" 33 34 int i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj, 35 struct sg_table *pages) 36 { 37 do { 38 if (dma_map_sg_attrs(&obj->base.dev->pdev->dev, 39 pages->sgl, pages->nents, 40 PCI_DMA_BIDIRECTIONAL, 41 DMA_ATTR_NO_WARN)) 42 return 0; 43 44 /* 45 * If the DMA remap fails, one cause can be that we have 46 * too many objects pinned in a small remapping table, 47 * such as swiotlb. Incrementally purge all other objects and 48 * try again - if there are no more pages to remove from 49 * the DMA remapper, i915_gem_shrink will return 0. 50 */ 51 GEM_BUG_ON(obj->mm.pages == pages); 52 } while (i915_gem_shrink(to_i915(obj->base.dev), 53 obj->base.size >> PAGE_SHIFT, NULL, 54 I915_SHRINK_BOUND | 55 I915_SHRINK_UNBOUND)); 56 57 return -ENOSPC; 58 } 59 60 void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj, 61 struct sg_table *pages) 62 { 63 struct drm_i915_private *dev_priv = to_i915(obj->base.dev); 64 struct device *kdev = &dev_priv->drm.pdev->dev; 65 struct i915_ggtt *ggtt = &dev_priv->ggtt; 66 67 if (unlikely(ggtt->do_idle_maps)) { 68 /* XXX This does not prevent more requests being submitted! */ 69 if (intel_gt_retire_requests_timeout(ggtt->vm.gt, 70 -MAX_SCHEDULE_TIMEOUT)) { 71 DRM_ERROR("Failed to wait for idle; VT'd may hang.\n"); 72 /* Wait a bit, in hopes it avoids the hang */ 73 udelay(10); 74 } 75 } 76 77 dma_unmap_sg(kdev, pages->sgl, pages->nents, PCI_DMA_BIDIRECTIONAL); 78 } 79 80 /** 81 * i915_gem_gtt_reserve - reserve a node in an address_space (GTT) 82 * @vm: the &struct i915_address_space 83 * @node: the &struct drm_mm_node (typically i915_vma.mode) 84 * @size: how much space to allocate inside the GTT, 85 * must be #I915_GTT_PAGE_SIZE aligned 86 * @offset: where to insert inside the GTT, 87 * must be #I915_GTT_MIN_ALIGNMENT aligned, and the node 88 * (@offset + @size) must fit within the address space 89 * @color: color to apply to node, if this node is not from a VMA, 90 * color must be #I915_COLOR_UNEVICTABLE 91 * @flags: control search and eviction behaviour 92 * 93 * i915_gem_gtt_reserve() tries to insert the @node at the exact @offset inside 94 * the address space (using @size and @color). If the @node does not fit, it 95 * tries to evict any overlapping nodes from the GTT, including any 96 * neighbouring nodes if the colors do not match (to ensure guard pages between 97 * differing domains). See i915_gem_evict_for_node() for the gory details 98 * on the eviction algorithm. #PIN_NONBLOCK may used to prevent waiting on 99 * evicting active overlapping objects, and any overlapping node that is pinned 100 * or marked as unevictable will also result in failure. 101 * 102 * Returns: 0 on success, -ENOSPC if no suitable hole is found, -EINTR if 103 * asked to wait for eviction and interrupted. 104 */ 105 int i915_gem_gtt_reserve(struct i915_address_space *vm, 106 struct drm_mm_node *node, 107 u64 size, u64 offset, unsigned long color, 108 unsigned int flags) 109 { 110 int err; 111 112 GEM_BUG_ON(!size); 113 GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); 114 GEM_BUG_ON(!IS_ALIGNED(offset, I915_GTT_MIN_ALIGNMENT)); 115 GEM_BUG_ON(range_overflows(offset, size, vm->total)); 116 GEM_BUG_ON(vm == &vm->i915->ggtt.alias->vm); 117 GEM_BUG_ON(drm_mm_node_allocated(node)); 118 119 node->size = size; 120 node->start = offset; 121 node->color = color; 122 123 err = drm_mm_reserve_node(&vm->mm, node); 124 if (err != -ENOSPC) 125 return err; 126 127 if (flags & PIN_NOEVICT) 128 return -ENOSPC; 129 130 err = i915_gem_evict_for_node(vm, node, flags); 131 if (err == 0) 132 err = drm_mm_reserve_node(&vm->mm, node); 133 134 return err; 135 } 136 137 static u64 random_offset(u64 start, u64 end, u64 len, u64 align) 138 { 139 u64 range, addr; 140 141 GEM_BUG_ON(range_overflows(start, len, end)); 142 GEM_BUG_ON(round_up(start, align) > round_down(end - len, align)); 143 144 range = round_down(end - len, align) - round_up(start, align); 145 if (range) { 146 if (sizeof(unsigned long) == sizeof(u64)) { 147 addr = get_random_long(); 148 } else { 149 addr = get_random_int(); 150 if (range > U32_MAX) { 151 addr <<= 32; 152 addr |= get_random_int(); 153 } 154 } 155 div64_u64_rem(addr, range, &addr); 156 start += addr; 157 } 158 159 return round_up(start, align); 160 } 161 162 /** 163 * i915_gem_gtt_insert - insert a node into an address_space (GTT) 164 * @vm: the &struct i915_address_space 165 * @node: the &struct drm_mm_node (typically i915_vma.node) 166 * @size: how much space to allocate inside the GTT, 167 * must be #I915_GTT_PAGE_SIZE aligned 168 * @alignment: required alignment of starting offset, may be 0 but 169 * if specified, this must be a power-of-two and at least 170 * #I915_GTT_MIN_ALIGNMENT 171 * @color: color to apply to node 172 * @start: start of any range restriction inside GTT (0 for all), 173 * must be #I915_GTT_PAGE_SIZE aligned 174 * @end: end of any range restriction inside GTT (U64_MAX for all), 175 * must be #I915_GTT_PAGE_SIZE aligned if not U64_MAX 176 * @flags: control search and eviction behaviour 177 * 178 * i915_gem_gtt_insert() first searches for an available hole into which 179 * is can insert the node. The hole address is aligned to @alignment and 180 * its @size must then fit entirely within the [@start, @end] bounds. The 181 * nodes on either side of the hole must match @color, or else a guard page 182 * will be inserted between the two nodes (or the node evicted). If no 183 * suitable hole is found, first a victim is randomly selected and tested 184 * for eviction, otherwise then the LRU list of objects within the GTT 185 * is scanned to find the first set of replacement nodes to create the hole. 186 * Those old overlapping nodes are evicted from the GTT (and so must be 187 * rebound before any future use). Any node that is currently pinned cannot 188 * be evicted (see i915_vma_pin()). Similar if the node's VMA is currently 189 * active and #PIN_NONBLOCK is specified, that node is also skipped when 190 * searching for an eviction candidate. See i915_gem_evict_something() for 191 * the gory details on the eviction algorithm. 192 * 193 * Returns: 0 on success, -ENOSPC if no suitable hole is found, -EINTR if 194 * asked to wait for eviction and interrupted. 195 */ 196 int i915_gem_gtt_insert(struct i915_address_space *vm, 197 struct drm_mm_node *node, 198 u64 size, u64 alignment, unsigned long color, 199 u64 start, u64 end, unsigned int flags) 200 { 201 enum drm_mm_insert_mode mode; 202 u64 offset; 203 int err; 204 205 lockdep_assert_held(&vm->mutex); 206 207 GEM_BUG_ON(!size); 208 GEM_BUG_ON(!IS_ALIGNED(size, I915_GTT_PAGE_SIZE)); 209 GEM_BUG_ON(alignment && !is_power_of_2(alignment)); 210 GEM_BUG_ON(alignment && !IS_ALIGNED(alignment, I915_GTT_MIN_ALIGNMENT)); 211 GEM_BUG_ON(start >= end); 212 GEM_BUG_ON(start > 0 && !IS_ALIGNED(start, I915_GTT_PAGE_SIZE)); 213 GEM_BUG_ON(end < U64_MAX && !IS_ALIGNED(end, I915_GTT_PAGE_SIZE)); 214 GEM_BUG_ON(vm == &vm->i915->ggtt.alias->vm); 215 GEM_BUG_ON(drm_mm_node_allocated(node)); 216 217 if (unlikely(range_overflows(start, size, end))) 218 return -ENOSPC; 219 220 if (unlikely(round_up(start, alignment) > round_down(end - size, alignment))) 221 return -ENOSPC; 222 223 mode = DRM_MM_INSERT_BEST; 224 if (flags & PIN_HIGH) 225 mode = DRM_MM_INSERT_HIGHEST; 226 if (flags & PIN_MAPPABLE) 227 mode = DRM_MM_INSERT_LOW; 228 229 /* We only allocate in PAGE_SIZE/GTT_PAGE_SIZE (4096) chunks, 230 * so we know that we always have a minimum alignment of 4096. 231 * The drm_mm range manager is optimised to return results 232 * with zero alignment, so where possible use the optimal 233 * path. 234 */ 235 BUILD_BUG_ON(I915_GTT_MIN_ALIGNMENT > I915_GTT_PAGE_SIZE); 236 if (alignment <= I915_GTT_MIN_ALIGNMENT) 237 alignment = 0; 238 239 err = drm_mm_insert_node_in_range(&vm->mm, node, 240 size, alignment, color, 241 start, end, mode); 242 if (err != -ENOSPC) 243 return err; 244 245 if (mode & DRM_MM_INSERT_ONCE) { 246 err = drm_mm_insert_node_in_range(&vm->mm, node, 247 size, alignment, color, 248 start, end, 249 DRM_MM_INSERT_BEST); 250 if (err != -ENOSPC) 251 return err; 252 } 253 254 if (flags & PIN_NOEVICT) 255 return -ENOSPC; 256 257 /* 258 * No free space, pick a slot at random. 259 * 260 * There is a pathological case here using a GTT shared between 261 * mmap and GPU (i.e. ggtt/aliasing_ppgtt but not full-ppgtt): 262 * 263 * |<-- 256 MiB aperture -->||<-- 1792 MiB unmappable -->| 264 * (64k objects) (448k objects) 265 * 266 * Now imagine that the eviction LRU is ordered top-down (just because 267 * pathology meets real life), and that we need to evict an object to 268 * make room inside the aperture. The eviction scan then has to walk 269 * the 448k list before it finds one within range. And now imagine that 270 * it has to search for a new hole between every byte inside the memcpy, 271 * for several simultaneous clients. 272 * 273 * On a full-ppgtt system, if we have run out of available space, there 274 * will be lots and lots of objects in the eviction list! Again, 275 * searching that LRU list may be slow if we are also applying any 276 * range restrictions (e.g. restriction to low 4GiB) and so, for 277 * simplicity and similarilty between different GTT, try the single 278 * random replacement first. 279 */ 280 offset = random_offset(start, end, 281 size, alignment ?: I915_GTT_MIN_ALIGNMENT); 282 err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags); 283 if (err != -ENOSPC) 284 return err; 285 286 if (flags & PIN_NOSEARCH) 287 return -ENOSPC; 288 289 /* Randomly selected placement is pinned, do a search */ 290 err = i915_gem_evict_something(vm, size, alignment, color, 291 start, end, flags); 292 if (err) 293 return err; 294 295 return drm_mm_insert_node_in_range(&vm->mm, node, 296 size, alignment, color, 297 start, end, DRM_MM_INSERT_EVICT); 298 } 299 300 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) 301 #include "selftests/i915_gem_gtt.c" 302 #endif 303
Indexes created Mon Oct 27 08:10:08 GMT 2025