tree: git://people.freedesktop.org/~agd5f/linux.git drm-next-5.2-wip head: 5666aea3ea494d4dd96df8f092cab32dbeeac321 commit: ba5896bd6a1a175b21609c6c81dea9813cbf296c [19/42] drm/amdgpu: use HMM callback to replace mmu notifier config: xtensa-allyesconfig (attached as .config) compiler: xtensa-linux-gcc (GCC) 8.1.0 reproduce: wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross git checkout ba5896bd6a1a175b21609c6c81dea9813cbf296c # save the attached .config to linux build tree GCC_VERSION=8.1.0 make.cross ARCH=xtensa

Note: the radeon-alex/drm-next-5.2-wip HEAD 5666aea3ea494d4dd96df8f092cab32dbeeac321 builds fine. It only hurts bisectibility.

All error/warnings (new ones prefixed by >>):

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:87:20: error: field 'mirror' has incomplete type

struct hmm_mirror mirror; ^~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: In function 'amdgpu_mn_destroy':

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:131:2: error: implicit declaration of function 'hmm_mirror_unregister'; did you mean 'drm_dp_aux_unregister'? [-Werror=implicit-function-declaration]

hmm_mirror_unregister(&amn->mirror); ^~~~~~~~~~~~~~~~~~~~~ drm_dp_aux_unregister In file included from include/linux/firmware.h:6, from drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:46: drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: In function 'amdgpu_hmm_mirror_release':

...

include/linux/kernel.h:979:32: error: dereferencing pointer to incomplete type 'struct hmm_mirror'

BUILD_BUG_ON_MSG(!__same_type(*(ptr), ((type *)0)->member) && \ ^~~~~~ include/linux/compiler.h:324:9: note: in definition of macro '__compiletime_assert' if (!(condition)) \ ^~~~~~~~~ include/linux/compiler.h:344:2: note: in expansion of macro '_compiletime_assert' _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__) ^~~~~~~~~~~~~~~~~~~ include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert' #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg) ^~~~~~~~~~~~~~~~~~ include/linux/kernel.h:979:2: note: in expansion of macro 'BUILD_BUG_ON_MSG' BUILD_BUG_ON_MSG(!__same_type(*(ptr), ((type *)0)->member) && \ ^~~~~~~~~~~~~~~~ include/linux/kernel.h:979:20: note: in expansion of macro '__same_type' BUILD_BUG_ON_MSG(!__same_type(*(ptr), ((type *)0)->member) && \ ^~~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:144:26: note: in expansion of macro 'container_of' struct amdgpu_mn *amn = container_of(mirror, struct amdgpu_mn, mirror); ^~~~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: At top level:

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:238:17: warning: 'struct hmm_update' declared inside parameter list will not be visible outside of this definition or declaration

const struct hmm_update *update) ^~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: In function 'amdgpu_mn_sync_pagetables_gfx':

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:241:30: error: dereferencing pointer to incomplete type 'const struct hmm_update'

unsigned long start = update->start; ^~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: At top level: drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:286:17: warning: 'struct hmm_update' declared inside parameter list will not be visible outside of this definition or declaration const struct hmm_update *update) ^~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: In function 'amdgpu_mn_sync_pagetables_hsa': drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:289:30: error: dereferencing pointer to incomplete type 'const struct hmm_update' unsigned long start = update->start; ^~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: At top level:

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:333:30: error: array type has incomplete element type 'struct hmm_mirror_ops'

static struct hmm_mirror_ops amdgpu_hmm_mirror_ops[] = { ^~~~~~~~~~~~~~~~~~~~~

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:334:3: error: array index in non-array initializer

[AMDGPU_MN_TYPE_GFX] = { ^~~~~~~~~~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:334:3: note: (near initialization for 'amdgpu_hmm_mirror_ops')

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:335:3: error: field name not in record or union initializer

.sync_cpu_device_pagetables = amdgpu_mn_sync_pagetables_gfx, ^ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:335:3: note: (near initialization for 'amdgpu_hmm_mirror_ops') drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:336:3: error: field name not in record or union initializer .release = amdgpu_hmm_mirror_release ^ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:336:3: note: (near initialization for 'amdgpu_hmm_mirror_ops') drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:338:3: error: array index in non-array initializer [AMDGPU_MN_TYPE_HSA] = { ^~~~~~~~~~~~~~~~~~ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:338:3: note: (near initialization for 'amdgpu_hmm_mirror_ops') drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:339:3: error: field name not in record or union initializer .sync_cpu_device_pagetables = amdgpu_mn_sync_pagetables_hsa, ^ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:339:3: note: (near initialization for 'amdgpu_hmm_mirror_ops') drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:340:3: error: field name not in record or union initializer .release = amdgpu_hmm_mirror_release ^ drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:340:3: note: (near initialization for 'amdgpu_hmm_mirror_ops') drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c: In function 'amdgpu_mn_get':

...

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:383:6: error: implicit declaration of function 'hmm_mirror_register'; did you mean 'drm_dp_aux_register'? [-Werror=implicit-function-declaration]

r = hmm_mirror_register(&amn->mirror, mm); ^~~~~~~~~~~~~~~~~~~ drm_dp_aux_register At top level: drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c:333:30: warning: 'amdgpu_hmm_mirror_ops' defined but not used [-Wunused-variable] static struct hmm_mirror_ops amdgpu_hmm_mirror_ops[] = { ^~~~~~~~~~~~~~~~~~~~~ cc1: some warnings being treated as errors

vim +/mirror +87 drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c

55 56 /** 57 * struct amdgpu_mn 58 * 59 * @adev: amdgpu device pointer 60 * @mm: process address space 61 * @type: type of MMU notifier 62 * @work: destruction work item 63 * @node: hash table node to find structure by adev and mn 64 * @lock: rw semaphore protecting the notifier nodes 65 * @objects: interval tree containing amdgpu_mn_nodes 66 * @mirror: HMM mirror function support 67 * 68 * Data for each amdgpu device and process address space. 69 */ 70 struct amdgpu_mn { 71 /* constant after initialisation */ 72 struct amdgpu_device *adev; 73 struct mm_struct *mm; 74 enum amdgpu_mn_type type; 75 76 /* only used on destruction */ 77 struct work_struct work; 78 79 /* protected by adev->mn_lock */ 80 struct hlist_node node; 81 82 /* objects protected by lock */ 83 struct rw_semaphore lock; 84 struct rb_root_cached objects; 85 86 /* HMM mirror */

87 struct hmm_mirror mirror;

88 }; 89 90 /** 91 * struct amdgpu_mn_node 92 * 93 * @it: interval node defining start-last of the affected address range 94 * @bos: list of all BOs in the affected address range 95 * 96 * Manages all BOs which are affected of a certain range of address space. 97 */ 98 struct amdgpu_mn_node { 99 struct interval_tree_node it; 100 struct list_head bos; 101 }; 102 103 /** 104 * amdgpu_mn_destroy - destroy the HMM mirror 105 * 106 * @work: previously sheduled work item 107 * 108 * Lazy destroys the notifier from a work item 109 */ 110 static void amdgpu_mn_destroy(struct work_struct *work) 111 { 112 struct amdgpu_mn *amn = container_of(work, struct amdgpu_mn, work); 113 struct amdgpu_device *adev = amn->adev; 114 struct amdgpu_mn_node *node, *next_node; 115 struct amdgpu_bo *bo, *next_bo; 116 117 mutex_lock(&adev->mn_lock); 118 down_write(&amn->lock); 119 hash_del(&amn->node); 120 rbtree_postorder_for_each_entry_safe(node, next_node, 121 &amn->objects.rb_root, it.rb) { 122 list_for_each_entry_safe(bo, next_bo, &node->bos, mn_list) { 123 bo->mn = NULL; 124 list_del_init(&bo->mn_list); 125 } 126 kfree(node); 127 } 128 up_write(&amn->lock); 129 mutex_unlock(&adev->mn_lock); 130

131 hmm_mirror_unregister(&amn->mirror);

132 kfree(amn); 133 } 134 135 /** 136 * amdgpu_hmm_mirror_release - callback to notify about mm destruction 137 * 138 * @mirror: the HMM mirror (mm) this callback is about 139 * 140 * Shedule a work item to lazy destroy HMM mirror. 141 */ 142 static void amdgpu_hmm_mirror_release(struct hmm_mirror *mirror) 143 {

144 struct amdgpu_mn *amn = container_of(mirror, struct amdgpu_mn, mirror);

145 146 INIT_WORK(&amn->work, amdgpu_mn_destroy); 147 schedule_work(&amn->work); 148 } 149 150 /** 151 * amdgpu_mn_lock - take the write side lock for this notifier 152 * 153 * @mn: our notifier 154 */ 155 void amdgpu_mn_lock(struct amdgpu_mn *mn) 156 { 157 if (mn) 158 down_write(&mn->lock); 159 } 160 161 /** 162 * amdgpu_mn_unlock - drop the write side lock for this notifier 163 * 164 * @mn: our notifier 165 */ 166 void amdgpu_mn_unlock(struct amdgpu_mn *mn) 167 { 168 if (mn) 169 up_write(&mn->lock); 170 } 171 172 /** 173 * amdgpu_mn_read_lock - take the read side lock for this notifier 174 * 175 * @amn: our notifier 176 */ 177 static int amdgpu_mn_read_lock(struct amdgpu_mn *amn, bool blockable) 178 { 179 if (blockable) 180 down_read(&amn->lock); 181 else if (!down_read_trylock(&amn->lock)) 182 return -EAGAIN; 183 184 return 0; 185 } 186 187 /** 188 * amdgpu_mn_read_unlock - drop the read side lock for this notifier 189 * 190 * @amn: our notifier 191 */ 192 static void amdgpu_mn_read_unlock(struct amdgpu_mn *amn) 193 { 194 up_read(&amn->lock); 195 } 196 197 /** 198 * amdgpu_mn_invalidate_node - unmap all BOs of a node 199 * 200 * @node: the node with the BOs to unmap 201 * @start: start of address range affected 202 * @end: end of address range affected 203 * 204 * Block for operations on BOs to finish and mark pages as accessed and 205 * potentially dirty. 206 */ 207 static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node, 208 unsigned long start, 209 unsigned long end) 210 { 211 struct amdgpu_bo *bo; 212 long r; 213 214 list_for_each_entry(bo, &node->bos, mn_list) { 215 216 if (!amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, start, end)) 217 continue; 218 219 r = reservation_object_wait_timeout_rcu(bo->tbo.resv, 220 true, false, MAX_SCHEDULE_TIMEOUT); 221 if (r <= 0) 222 DRM_ERROR("(%ld) failed to wait for user bo\n", r); 223 224 amdgpu_ttm_tt_mark_user_pages(bo->tbo.ttm); 225 } 226 } 227 228 /** 229 * amdgpu_mn_sync_pagetables_gfx - callback to notify about mm change 230 * 231 * @mirror: the hmm_mirror (mm) is about to update 232 * @update: the update start, end address 233 * 234 * Block for operations on BOs to finish and mark pages as accessed and 235 * potentially dirty. 236 */ 237 static int amdgpu_mn_sync_pagetables_gfx(struct hmm_mirror *mirror,

238 const struct hmm_update *update)

239 { 240 struct amdgpu_mn *amn = container_of(mirror, struct amdgpu_mn, mirror);

241 unsigned long start = update->start;

242 unsigned long end = update->end; 243 bool blockable = update->blockable; 244 struct interval_tree_node *it; 245 246 /* notification is exclusive, but interval is inclusive */ 247 end -= 1; 248 249 /* TODO we should be able to split locking for interval tree and 250 * amdgpu_mn_invalidate_node 251 */ 252 if (amdgpu_mn_read_lock(amn, blockable)) 253 return -EAGAIN; 254 255 it = interval_tree_iter_first(&amn->objects, start, end); 256 while (it) { 257 struct amdgpu_mn_node *node; 258 259 if (!blockable) { 260 amdgpu_mn_read_unlock(amn); 261 return -EAGAIN; 262 } 263 264 node = container_of(it, struct amdgpu_mn_node, it); 265 it = interval_tree_iter_next(it, start, end); 266 267 amdgpu_mn_invalidate_node(node, start, end); 268 } 269 270 amdgpu_mn_read_unlock(amn); 271 272 return 0; 273 } 274 275 /** 276 * amdgpu_mn_sync_pagetables_hsa - callback to notify about mm change 277 * 278 * @mirror: the hmm_mirror (mm) is about to update 279 * @update: the update start, end address 280 * 281 * We temporarily evict all BOs between start and end. This 282 * necessitates evicting all user-mode queues of the process. The BOs 283 * are restorted in amdgpu_mn_invalidate_range_end_hsa. 284 */ 285 static int amdgpu_mn_sync_pagetables_hsa(struct hmm_mirror *mirror,

286 const struct hmm_update *update)

287 { 288 struct amdgpu_mn *amn = container_of(mirror, struct amdgpu_mn, mirror); 289 unsigned long start = update->start; 290 unsigned long end = update->end; 291 bool blockable = update->blockable; 292 struct interval_tree_node *it; 293 294 /* notification is exclusive, but interval is inclusive */ 295 end -= 1; 296 297 if (amdgpu_mn_read_lock(amn, blockable)) 298 return -EAGAIN; 299 300 it = interval_tree_iter_first(&amn->objects, start, end); 301 while (it) { 302 struct amdgpu_mn_node *node; 303 struct amdgpu_bo *bo; 304 305 if (!blockable) { 306 amdgpu_mn_read_unlock(amn); 307 return -EAGAIN; 308 } 309 310 node = container_of(it, struct amdgpu_mn_node, it); 311 it = interval_tree_iter_next(it, start, end); 312 313 list_for_each_entry(bo, &node->bos, mn_list) { 314 struct kgd_mem *mem = bo->kfd_bo; 315 316 if (amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, 317 start, end)) 318 amdgpu_amdkfd_evict_userptr(mem, amn->mm); 319 } 320 } 321 322 amdgpu_mn_read_unlock(amn); 323 324 return 0; 325 } 326

--- 0-DAY kernel test infrastructure Open Source Technology Center https://lists.01.org/pipermail/kbuild-all Intel Corporation