1 /* SPDX-License-Identifier: GPL-2.0 OR MIT */ 2 3 /* 4 * Copyright (c) 2006-2009 VMware, Inc., Palo Alto, CA., USA 5 * Copyright 2020 Advanced Micro Devices, Inc. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the "Software"), 9 * to deal in the Software without restriction, including without limitation 10 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 11 * and/or sell copies of the Software, and to permit persons to whom the 12 * Software is furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 23 * OTHER DEALINGS IN THE SOFTWARE. 24 * 25 * Authors: Christian König 26 */ 27 28 #define pr_fmt(fmt) "[TTM DEVICE] " fmt 29 30 #include <linux/debugfs.h> 31 #include <linux/mm.h> 32 33 #include <drm/ttm/ttm_bo.h> 34 #include <drm/ttm/ttm_device.h> 35 #include <drm/ttm/ttm_tt.h> 36 #include <drm/ttm/ttm_placement.h> 37 38 #include "ttm_module.h" 39 40 /* 41 * ttm_global_mutex - protecting the global state 42 */ 43 static DEFINE_MUTEX(ttm_global_mutex); 44 static unsigned ttm_glob_use_count; 45 struct ttm_global ttm_glob; 46 EXPORT_SYMBOL(ttm_glob); 47 48 struct dentry *ttm_debugfs_root; 49 50 static void ttm_global_release(void) 51 { 52 struct ttm_global *glob = &ttm_glob; 53 54 mutex_lock(&ttm_global_mutex); 55 if (--ttm_glob_use_count > 0) 56 goto out; 57 58 ttm_pool_mgr_fini(); 59 debugfs_remove(ttm_debugfs_root); 60 61 __free_page(glob->dummy_read_page); 62 memset(glob, 0, sizeof(*glob)); 63 out: 64 mutex_unlock(&ttm_global_mutex); 65 } 66 67 static int ttm_global_init(void) 68 { 69 struct ttm_global *glob = &ttm_glob; 70 unsigned long num_pages, num_dma32; 71 struct sysinfo si; 72 int ret = 0; 73 74 mutex_lock(&ttm_global_mutex); 75 if (++ttm_glob_use_count > 1) 76 goto out; 77 78 si_meminfo(&si); 79 80 ttm_debugfs_root = debugfs_create_dir("ttm", NULL); 81 if (IS_ERR(ttm_debugfs_root)) { 82 ttm_debugfs_root = NULL; 83 } 84 85 /* Limit the number of pages in the pool to about 50% of the total 86 * system memory. 87 */ 88 num_pages = ((u64)si.totalram * si.mem_unit) >> PAGE_SHIFT; 89 num_pages /= 2; 90 91 /* But for DMA32 we limit ourself to only use 2GiB maximum. */ 92 num_dma32 = (u64)(si.totalram - si.totalhigh) * si.mem_unit 93 >> PAGE_SHIFT; 94 num_dma32 = min(num_dma32, 2UL << (30 - PAGE_SHIFT)); 95 96 ttm_pool_mgr_init(num_pages); 97 ttm_tt_mgr_init(num_pages, num_dma32); 98 99 glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32 | 100 __GFP_NOWARN); 101 102 /* Retry without GFP_DMA32 for platforms DMA32 is not available */ 103 if (unlikely(glob->dummy_read_page == NULL)) { 104 glob->dummy_read_page = alloc_page(__GFP_ZERO); 105 if (unlikely(glob->dummy_read_page == NULL)) { 106 ret = -ENOMEM; 107 goto out; 108 } 109 pr_warn("Using GFP_DMA32 fallback for dummy_read_page\n"); 110 } 111 112 INIT_LIST_HEAD(&glob->device_list); 113 atomic_set(&glob->bo_count, 0); 114 115 debugfs_create_atomic_t("buffer_objects", 0444, ttm_debugfs_root, 116 &glob->bo_count); 117 out: 118 if (ret && ttm_debugfs_root) 119 debugfs_remove(ttm_debugfs_root); 120 if (ret) 121 --ttm_glob_use_count; 122 mutex_unlock(&ttm_global_mutex); 123 return ret; 124 } 125 126 /* 127 * A buffer object shrink method that tries to swap out the first 128 * buffer object on the global::swap_lru list. 129 */ 130 int ttm_global_swapout(struct ttm_operation_ctx *ctx, gfp_t gfp_flags) 131 { 132 struct ttm_global *glob = &ttm_glob; 133 struct ttm_device *bdev; 134 int ret = 0; 135 136 mutex_lock(&ttm_global_mutex); 137 list_for_each_entry(bdev, &glob->device_list, device_list) { 138 ret = ttm_device_swapout(bdev, ctx, gfp_flags); 139 if (ret > 0) { 140 list_move_tail(&bdev->device_list, &glob->device_list); 141 break; 142 } 143 } 144 mutex_unlock(&ttm_global_mutex); 145 return ret; 146 } 147 148 int ttm_device_swapout(struct ttm_device *bdev, struct ttm_operation_ctx *ctx, 149 gfp_t gfp_flags) 150 { 151 struct ttm_resource_manager *man; 152 unsigned i; 153 s64 lret; 154 155 for (i = TTM_PL_SYSTEM; i < TTM_NUM_MEM_TYPES; ++i) { 156 man = ttm_manager_type(bdev, i); 157 if (!man || !man->use_tt) 158 continue; 159 160 lret = ttm_bo_swapout(bdev, ctx, man, gfp_flags, 1); 161 /* Can be both positive (num_pages) and negative (error) */ 162 if (lret) 163 return lret; 164 } 165 return 0; 166 } 167 EXPORT_SYMBOL(ttm_device_swapout); 168 169 /** 170 * ttm_device_init 171 * 172 * @bdev: A pointer to a struct ttm_device to initialize. 173 * @funcs: Function table for the device. 174 * @dev: The core kernel device pointer for DMA mappings and allocations. 175 * @mapping: The address space to use for this bo. 176 * @vma_manager: A pointer to a vma manager. 177 * @use_dma_alloc: If coherent DMA allocation API should be used. 178 * @use_dma32: If we should use GFP_DMA32 for device memory allocations. 179 * 180 * Initializes a struct ttm_device: 181 * Returns: 182 * !0: Failure. 183 */ 184 int ttm_device_init(struct ttm_device *bdev, const struct ttm_device_funcs *funcs, 185 struct device *dev, struct address_space *mapping, 186 struct drm_vma_offset_manager *vma_manager, 187 bool use_dma_alloc, bool use_dma32) 188 { 189 struct ttm_global *glob = &ttm_glob; 190 int ret, nid; 191 192 if (WARN_ON(vma_manager == NULL)) 193 return -EINVAL; 194 195 ret = ttm_global_init(); 196 if (ret) 197 return ret; 198 199 bdev->wq = alloc_workqueue("ttm", 200 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 16); 201 if (!bdev->wq) { 202 ttm_global_release(); 203 return -ENOMEM; 204 } 205 206 bdev->funcs = funcs; 207 208 ttm_sys_man_init(bdev); 209 210 if (dev) 211 nid = dev_to_node(dev); 212 else 213 nid = NUMA_NO_NODE; 214 215 ttm_pool_init(&bdev->pool, dev, nid, use_dma_alloc, use_dma32); 216 217 bdev->vma_manager = vma_manager; 218 spin_lock_init(&bdev->lru_lock); 219 INIT_LIST_HEAD(&bdev->pinned); 220 bdev->dev_mapping = mapping; 221 mutex_lock(&ttm_global_mutex); 222 list_add_tail(&bdev->device_list, &glob->device_list); 223 mutex_unlock(&ttm_global_mutex); 224 225 return 0; 226 } 227 EXPORT_SYMBOL(ttm_device_init); 228 229 void ttm_device_fini(struct ttm_device *bdev) 230 { 231 struct ttm_resource_manager *man; 232 unsigned i; 233 234 mutex_lock(&ttm_global_mutex); 235 list_del(&bdev->device_list); 236 mutex_unlock(&ttm_global_mutex); 237 238 drain_workqueue(bdev->wq); 239 destroy_workqueue(bdev->wq); 240 241 man = ttm_manager_type(bdev, TTM_PL_SYSTEM); 242 ttm_resource_manager_set_used(man, false); 243 ttm_set_driver_manager(bdev, TTM_PL_SYSTEM, NULL); 244 245 spin_lock(&bdev->lru_lock); 246 for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) 247 if (list_empty(&man->lru[0])) 248 pr_debug("Swap list %d was clean\n", i); 249 spin_unlock(&bdev->lru_lock); 250 251 ttm_pool_fini(&bdev->pool); 252 ttm_global_release(); 253 } 254 EXPORT_SYMBOL(ttm_device_fini); 255 256 static void ttm_device_clear_lru_dma_mappings(struct ttm_device *bdev, 257 struct list_head *list) 258 { 259 struct ttm_resource *res; 260 261 spin_lock(&bdev->lru_lock); 262 while ((res = ttm_lru_first_res_or_null(list))) { 263 struct ttm_buffer_object *bo = res->bo; 264 265 /* Take ref against racing releases once lru_lock is unlocked */ 266 if (!ttm_bo_get_unless_zero(bo)) 267 continue; 268 269 list_del_init(&bo->resource->lru.link); 270 spin_unlock(&bdev->lru_lock); 271 272 if (bo->ttm) 273 ttm_tt_unpopulate(bo->bdev, bo->ttm); 274 275 ttm_bo_put(bo); 276 spin_lock(&bdev->lru_lock); 277 } 278 spin_unlock(&bdev->lru_lock); 279 } 280 281 void ttm_device_clear_dma_mappings(struct ttm_device *bdev) 282 { 283 struct ttm_resource_manager *man; 284 unsigned int i, j; 285 286 ttm_device_clear_lru_dma_mappings(bdev, &bdev->pinned); 287 288 for (i = TTM_PL_SYSTEM; i < TTM_NUM_MEM_TYPES; ++i) { 289 man = ttm_manager_type(bdev, i); 290 if (!man || !man->use_tt) 291 continue; 292 293 for (j = 0; j < TTM_MAX_BO_PRIORITY; ++j) 294 ttm_device_clear_lru_dma_mappings(bdev, &man->lru[j]); 295 } 296 } 297 EXPORT_SYMBOL(ttm_device_clear_dma_mappings); 298