Commit | Line | Data |
---|---|---|
c030f2e4 | 1 | /* |
2 | * Copyright 2018 Advanced Micro Devices, Inc. | |
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 shall be included in | |
12 | * all copies or substantial portions of the Software. | |
13 | * | |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
20 | * OTHER DEALINGS IN THE SOFTWARE. | |
21 | * | |
22 | * | |
23 | */ | |
24 | #include <linux/debugfs.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/module.h> | |
f867723b | 27 | #include <linux/uaccess.h> |
7c6e68c7 AG |
28 | #include <linux/reboot.h> |
29 | #include <linux/syscalls.h> | |
05adfd80 | 30 | #include <linux/pm_runtime.h> |
dbf3850d | 31 | #include <linux/list_sort.h> |
f867723b | 32 | |
c030f2e4 | 33 | #include "amdgpu.h" |
34 | #include "amdgpu_ras.h" | |
b404ae82 | 35 | #include "amdgpu_atomfirmware.h" |
19744f5f | 36 | #include "amdgpu_xgmi.h" |
4e644fff | 37 | #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" |
9af357bc | 38 | #include "nbio_v4_3.h" |
7692e1ee | 39 | #include "nbio_v7_9.h" |
f50160cf | 40 | #include "atom.h" |
25a2b22e | 41 | #include "amdgpu_reset.h" |
4e2965bd | 42 | #include "amdgpu_psp.h" |
25a2b22e | 43 | |
12b2cab7 MJ |
44 | #ifdef CONFIG_X86_MCE_AMD |
45 | #include <asm/mce.h> | |
c030f2e4 | 46 | |
12b2cab7 MJ |
47 | static bool notifier_registered; |
48 | #endif | |
eb0c3cd4 GC |
49 | static const char *RAS_FS_NAME = "ras"; |
50 | ||
c030f2e4 | 51 | const char *ras_error_string[] = { |
52 | "none", | |
53 | "parity", | |
54 | "single_correctable", | |
55 | "multi_uncorrectable", | |
56 | "poison", | |
57 | }; | |
58 | ||
59 | const char *ras_block_string[] = { | |
60 | "umc", | |
61 | "sdma", | |
62 | "gfx", | |
63 | "mmhub", | |
64 | "athub", | |
65 | "pcie_bif", | |
66 | "hdp", | |
67 | "xgmi_wafl", | |
68 | "df", | |
69 | "smn", | |
70 | "sem", | |
71 | "mp0", | |
72 | "mp1", | |
73 | "fuse", | |
640ae42e | 74 | "mca", |
a3d63c62 MZZ |
75 | "vcn", |
76 | "jpeg", | |
7ed97155 YW |
77 | "ih", |
78 | "mpio", | |
c030f2e4 | 79 | }; |
80 | ||
640ae42e JC |
81 | const char *ras_mca_block_string[] = { |
82 | "mca_mp0", | |
83 | "mca_mp1", | |
84 | "mca_mpio", | |
85 | "mca_iohc", | |
86 | }; | |
87 | ||
d5e8ff5f | 88 | struct amdgpu_ras_block_list { |
89 | /* ras block link */ | |
90 | struct list_head node; | |
91 | ||
92 | struct amdgpu_ras_block_object *ras_obj; | |
93 | }; | |
94 | ||
640ae42e JC |
95 | const char *get_ras_block_str(struct ras_common_if *ras_block) |
96 | { | |
97 | if (!ras_block) | |
98 | return "NULL"; | |
99 | ||
7ed97155 YW |
100 | if (ras_block->block >= AMDGPU_RAS_BLOCK_COUNT || |
101 | ras_block->block >= ARRAY_SIZE(ras_block_string)) | |
640ae42e JC |
102 | return "OUT OF RANGE"; |
103 | ||
104 | if (ras_block->block == AMDGPU_RAS_BLOCK__MCA) | |
105 | return ras_mca_block_string[ras_block->sub_block_index]; | |
106 | ||
107 | return ras_block_string[ras_block->block]; | |
108 | } | |
109 | ||
954ea6aa | 110 | #define ras_block_str(_BLOCK_) \ |
111 | (((_BLOCK_) < ARRAY_SIZE(ras_block_string)) ? ras_block_string[_BLOCK_] : "Out Of Range") | |
8b0fb0e9 | 112 | |
c030f2e4 | 113 | #define ras_err_str(i) (ras_error_string[ffs(i)]) |
c030f2e4 | 114 | |
108c6a63 | 115 | #define RAS_DEFAULT_FLAGS (AMDGPU_RAS_FLAG_INIT_BY_VBIOS) |
116 | ||
7cdc2ee3 TZ |
117 | /* inject address is 52 bits */ |
118 | #define RAS_UMC_INJECT_ADDR_LIMIT (0x1ULL << 52) | |
119 | ||
e4e6a589 LT |
120 | /* typical ECC bad page rate is 1 bad page per 100MB VRAM */ |
121 | #define RAS_BAD_PAGE_COVER (100 * 1024 * 1024ULL) | |
c84d4670 | 122 | |
6c23f3d1 YC |
123 | #define MAX_UMC_POISON_POLLING_TIME_ASYNC 100 //ms |
124 | ||
2cf8e50e YC |
125 | #define AMDGPU_RAS_RETIRE_PAGE_INTERVAL 100 //ms |
126 | ||
52dd95f2 GC |
127 | enum amdgpu_ras_retire_page_reservation { |
128 | AMDGPU_RAS_RETIRE_PAGE_RESERVED, | |
129 | AMDGPU_RAS_RETIRE_PAGE_PENDING, | |
130 | AMDGPU_RAS_RETIRE_PAGE_FAULT, | |
131 | }; | |
7c6e68c7 AG |
132 | |
133 | atomic_t amdgpu_ras_in_intr = ATOMIC_INIT(0); | |
134 | ||
676deb38 DL |
135 | static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, |
136 | uint64_t addr); | |
6e4be987 TZ |
137 | static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, |
138 | uint64_t addr); | |
12b2cab7 | 139 | #ifdef CONFIG_X86_MCE_AMD |
91a1a52d MJ |
140 | static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev); |
141 | struct mce_notifier_adev_list { | |
142 | struct amdgpu_device *devs[MAX_GPU_INSTANCE]; | |
143 | int num_gpu; | |
144 | }; | |
145 | static struct mce_notifier_adev_list mce_adev_list; | |
12b2cab7 | 146 | #endif |
6e4be987 | 147 | |
61380faa JC |
148 | void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready) |
149 | { | |
a9d82d2f | 150 | if (adev && amdgpu_ras_get_context(adev)) |
61380faa JC |
151 | amdgpu_ras_get_context(adev)->error_query_ready = ready; |
152 | } | |
153 | ||
f3167919 | 154 | static bool amdgpu_ras_get_error_query_ready(struct amdgpu_device *adev) |
61380faa | 155 | { |
a9d82d2f | 156 | if (adev && amdgpu_ras_get_context(adev)) |
61380faa JC |
157 | return amdgpu_ras_get_context(adev)->error_query_ready; |
158 | ||
159 | return false; | |
160 | } | |
161 | ||
cbb8f989 JC |
162 | static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t address) |
163 | { | |
5b1270be | 164 | struct ras_err_data err_data; |
cbb8f989 | 165 | struct eeprom_table_record err_rec; |
5b1270be | 166 | int ret; |
cbb8f989 JC |
167 | |
168 | if ((address >= adev->gmc.mc_vram_size) || | |
169 | (address >= RAS_UMC_INJECT_ADDR_LIMIT)) { | |
170 | dev_warn(adev->dev, | |
171 | "RAS WARN: input address 0x%llx is invalid.\n", | |
172 | address); | |
173 | return -EINVAL; | |
174 | } | |
175 | ||
176 | if (amdgpu_ras_check_bad_page(adev, address)) { | |
177 | dev_warn(adev->dev, | |
80b0cd0f | 178 | "RAS WARN: 0x%llx has already been marked as bad page!\n", |
cbb8f989 JC |
179 | address); |
180 | return 0; | |
181 | } | |
182 | ||
5b1270be YW |
183 | ret = amdgpu_ras_error_data_init(&err_data); |
184 | if (ret) | |
185 | return ret; | |
186 | ||
cbb8f989 | 187 | memset(&err_rec, 0x0, sizeof(struct eeprom_table_record)); |
cbb8f989 | 188 | err_data.err_addr = &err_rec; |
71344a71 | 189 | amdgpu_umc_fill_error_record(&err_data, address, address, 0, 0); |
cbb8f989 JC |
190 | |
191 | if (amdgpu_bad_page_threshold != 0) { | |
192 | amdgpu_ras_add_bad_pages(adev, err_data.err_addr, | |
193 | err_data.err_addr_cnt); | |
4d33e0f1 | 194 | amdgpu_ras_save_bad_pages(adev, NULL); |
cbb8f989 JC |
195 | } |
196 | ||
5b1270be YW |
197 | amdgpu_ras_error_data_fini(&err_data); |
198 | ||
cbb8f989 JC |
199 | dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL CORRUPT RAS EEPROM\n"); |
200 | dev_warn(adev->dev, "Clear EEPROM:\n"); | |
201 | dev_warn(adev->dev, " echo 1 > /sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n"); | |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
c030f2e4 | 206 | static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf, |
207 | size_t size, loff_t *pos) | |
208 | { | |
209 | struct ras_manager *obj = (struct ras_manager *)file_inode(f)->i_private; | |
210 | struct ras_query_if info = { | |
211 | .head = obj->head, | |
212 | }; | |
213 | ssize_t s; | |
214 | char val[128]; | |
215 | ||
761d86d3 | 216 | if (amdgpu_ras_query_error_status(obj->adev, &info)) |
c030f2e4 | 217 | return -EINVAL; |
218 | ||
2a460963 | 219 | /* Hardware counter will be reset automatically after the query on Vega20 and Arcturus */ |
4e8303cf LL |
220 | if (amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) && |
221 | amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 4)) { | |
2a460963 CL |
222 | if (amdgpu_ras_reset_error_status(obj->adev, info.head.block)) |
223 | dev_warn(obj->adev->dev, "Failed to reset error counter and error status"); | |
224 | } | |
225 | ||
c030f2e4 | 226 | s = snprintf(val, sizeof(val), "%s: %lu\n%s: %lu\n", |
227 | "ue", info.ue_count, | |
228 | "ce", info.ce_count); | |
229 | if (*pos >= s) | |
230 | return 0; | |
231 | ||
232 | s -= *pos; | |
233 | s = min_t(u64, s, size); | |
234 | ||
235 | ||
236 | if (copy_to_user(buf, &val[*pos], s)) | |
237 | return -EINVAL; | |
238 | ||
239 | *pos += s; | |
240 | ||
241 | return s; | |
242 | } | |
243 | ||
c030f2e4 | 244 | static const struct file_operations amdgpu_ras_debugfs_ops = { |
245 | .owner = THIS_MODULE, | |
246 | .read = amdgpu_ras_debugfs_read, | |
190211ab | 247 | .write = NULL, |
c030f2e4 | 248 | .llseek = default_llseek |
249 | }; | |
250 | ||
96ebb307 | 251 | static int amdgpu_ras_find_block_id_by_name(const char *name, int *block_id) |
252 | { | |
253 | int i; | |
254 | ||
255 | for (i = 0; i < ARRAY_SIZE(ras_block_string); i++) { | |
256 | *block_id = i; | |
640ae42e | 257 | if (strcmp(name, ras_block_string[i]) == 0) |
96ebb307 | 258 | return 0; |
259 | } | |
260 | return -EINVAL; | |
261 | } | |
262 | ||
263 | static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f, | |
264 | const char __user *buf, size_t size, | |
265 | loff_t *pos, struct ras_debug_if *data) | |
266 | { | |
267 | ssize_t s = min_t(u64, 64, size); | |
268 | char str[65]; | |
269 | char block_name[33]; | |
270 | char err[9] = "ue"; | |
271 | int op = -1; | |
272 | int block_id; | |
44494f96 | 273 | uint32_t sub_block; |
96ebb307 | 274 | u64 address, value; |
2c22ed0b TZ |
275 | /* default value is 0 if the mask is not set by user */ |
276 | u32 instance_mask = 0; | |
96ebb307 | 277 | |
278 | if (*pos) | |
279 | return -EINVAL; | |
280 | *pos = size; | |
281 | ||
282 | memset(str, 0, sizeof(str)); | |
283 | memset(data, 0, sizeof(*data)); | |
284 | ||
285 | if (copy_from_user(str, buf, s)) | |
286 | return -EINVAL; | |
287 | ||
288 | if (sscanf(str, "disable %32s", block_name) == 1) | |
289 | op = 0; | |
290 | else if (sscanf(str, "enable %32s %8s", block_name, err) == 2) | |
291 | op = 1; | |
292 | else if (sscanf(str, "inject %32s %8s", block_name, err) == 2) | |
293 | op = 2; | |
6df23f4c | 294 | else if (strstr(str, "retire_page") != NULL) |
cbb8f989 | 295 | op = 3; |
b076296b | 296 | else if (str[0] && str[1] && str[2] && str[3]) |
96ebb307 | 297 | /* ascii string, but commands are not matched. */ |
298 | return -EINVAL; | |
299 | ||
300 | if (op != -1) { | |
cbb8f989 | 301 | if (op == 3) { |
546aa546 LT |
302 | if (sscanf(str, "%*s 0x%llx", &address) != 1 && |
303 | sscanf(str, "%*s %llu", &address) != 1) | |
6cb7a1d4 | 304 | return -EINVAL; |
cbb8f989 JC |
305 | |
306 | data->op = op; | |
307 | data->inject.address = address; | |
308 | ||
309 | return 0; | |
310 | } | |
311 | ||
96ebb307 | 312 | if (amdgpu_ras_find_block_id_by_name(block_name, &block_id)) |
313 | return -EINVAL; | |
314 | ||
315 | data->head.block = block_id; | |
fb1e9171 | 316 | /* only ue, ce and poison errors are supported */ |
e1063493 TZ |
317 | if (!memcmp("ue", err, 2)) |
318 | data->head.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; | |
319 | else if (!memcmp("ce", err, 2)) | |
320 | data->head.type = AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE; | |
fb1e9171 CL |
321 | else if (!memcmp("poison", err, 6)) |
322 | data->head.type = AMDGPU_RAS_ERROR__POISON; | |
e1063493 TZ |
323 | else |
324 | return -EINVAL; | |
325 | ||
96ebb307 | 326 | data->op = op; |
327 | ||
328 | if (op == 2) { | |
2c22ed0b TZ |
329 | if (sscanf(str, "%*s %*s %*s 0x%x 0x%llx 0x%llx 0x%x", |
330 | &sub_block, &address, &value, &instance_mask) != 4 && | |
331 | sscanf(str, "%*s %*s %*s %u %llu %llu %u", | |
332 | &sub_block, &address, &value, &instance_mask) != 4 && | |
333 | sscanf(str, "%*s %*s %*s 0x%x 0x%llx 0x%llx", | |
546aa546 LT |
334 | &sub_block, &address, &value) != 3 && |
335 | sscanf(str, "%*s %*s %*s %u %llu %llu", | |
6cb7a1d4 LT |
336 | &sub_block, &address, &value) != 3) |
337 | return -EINVAL; | |
44494f96 | 338 | data->head.sub_block_index = sub_block; |
96ebb307 | 339 | data->inject.address = address; |
340 | data->inject.value = value; | |
2c22ed0b | 341 | data->inject.instance_mask = instance_mask; |
96ebb307 | 342 | } |
343 | } else { | |
73aa8e1a | 344 | if (size < sizeof(*data)) |
96ebb307 | 345 | return -EINVAL; |
346 | ||
347 | if (copy_from_user(data, buf, sizeof(*data))) | |
348 | return -EINVAL; | |
349 | } | |
350 | ||
351 | return 0; | |
352 | } | |
7c6e68c7 | 353 | |
f464c5dd TZ |
354 | static void amdgpu_ras_instance_mask_check(struct amdgpu_device *adev, |
355 | struct ras_debug_if *data) | |
356 | { | |
357 | int num_xcc = adev->gfx.xcc_mask ? NUM_XCC(adev->gfx.xcc_mask) : 1; | |
358 | uint32_t mask, inst_mask = data->inject.instance_mask; | |
359 | ||
360 | /* no need to set instance mask if there is only one instance */ | |
361 | if (num_xcc <= 1 && inst_mask) { | |
362 | data->inject.instance_mask = 0; | |
363 | dev_dbg(adev->dev, | |
364 | "RAS inject mask(0x%x) isn't supported and force it to 0.\n", | |
365 | inst_mask); | |
366 | ||
367 | return; | |
368 | } | |
369 | ||
370 | switch (data->head.block) { | |
371 | case AMDGPU_RAS_BLOCK__GFX: | |
372 | mask = GENMASK(num_xcc - 1, 0); | |
373 | break; | |
374 | case AMDGPU_RAS_BLOCK__SDMA: | |
375 | mask = GENMASK(adev->sdma.num_instances - 1, 0); | |
376 | break; | |
e3959cb5 SY |
377 | case AMDGPU_RAS_BLOCK__VCN: |
378 | case AMDGPU_RAS_BLOCK__JPEG: | |
379 | mask = GENMASK(adev->vcn.num_vcn_inst - 1, 0); | |
380 | break; | |
f464c5dd | 381 | default: |
e3959cb5 | 382 | mask = inst_mask; |
f464c5dd TZ |
383 | break; |
384 | } | |
385 | ||
386 | /* remove invalid bits in instance mask */ | |
387 | data->inject.instance_mask &= mask; | |
388 | if (inst_mask != data->inject.instance_mask) | |
389 | dev_dbg(adev->dev, | |
390 | "Adjust RAS inject mask 0x%x to 0x%x\n", | |
391 | inst_mask, data->inject.instance_mask); | |
392 | } | |
393 | ||
74abc221 TSD |
394 | /** |
395 | * DOC: AMDGPU RAS debugfs control interface | |
36ea1bd2 | 396 | * |
737c375b | 397 | * The control interface accepts struct ras_debug_if which has two members. |
36ea1bd2 | 398 | * |
399 | * First member: ras_debug_if::head or ras_debug_if::inject. | |
96ebb307 | 400 | * |
401 | * head is used to indicate which IP block will be under control. | |
36ea1bd2 | 402 | * |
403 | * head has four members, they are block, type, sub_block_index, name. | |
404 | * block: which IP will be under control. | |
405 | * type: what kind of error will be enabled/disabled/injected. | |
406 | * sub_block_index: some IPs have subcomponets. say, GFX, sDMA. | |
407 | * name: the name of IP. | |
408 | * | |
2c22ed0b | 409 | * inject has three more members than head, they are address, value and mask. |
36ea1bd2 | 410 | * As their names indicate, inject operation will write the |
411 | * value to the address. | |
412 | * | |
ef177d11 | 413 | * The second member: struct ras_debug_if::op. |
c688a06b | 414 | * It has three kinds of operations. |
879e723d AZ |
415 | * |
416 | * - 0: disable RAS on the block. Take ::head as its data. | |
417 | * - 1: enable RAS on the block. Take ::head as its data. | |
418 | * - 2: inject errors on the block. Take ::inject as its data. | |
36ea1bd2 | 419 | * |
96ebb307 | 420 | * How to use the interface? |
ef177d11 | 421 | * |
737c375b | 422 | * In a program |
ef177d11 | 423 | * |
737c375b LT |
424 | * Copy the struct ras_debug_if in your code and initialize it. |
425 | * Write the struct to the control interface. | |
ef177d11 | 426 | * |
737c375b | 427 | * From shell |
96ebb307 | 428 | * |
879e723d AZ |
429 | * .. code-block:: bash |
430 | * | |
737c375b LT |
431 | * echo "disable <block>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl |
432 | * echo "enable <block> <error>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl | |
2c22ed0b | 433 | * echo "inject <block> <error> <sub-block> <address> <value> <mask>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl |
879e723d | 434 | * |
737c375b | 435 | * Where N, is the card which you want to affect. |
ef177d11 | 436 | * |
737c375b LT |
437 | * "disable" requires only the block. |
438 | * "enable" requires the block and error type. | |
439 | * "inject" requires the block, error type, address, and value. | |
c666bbf0 | 440 | * |
737c375b | 441 | * The block is one of: umc, sdma, gfx, etc. |
879e723d | 442 | * see ras_block_string[] for details |
c666bbf0 | 443 | * |
fb1e9171 | 444 | * The error type is one of: ue, ce and poison where, |
737c375b LT |
445 | * ue is multi-uncorrectable |
446 | * ce is single-correctable | |
fb1e9171 | 447 | * poison is poison |
c666bbf0 | 448 | * |
737c375b LT |
449 | * The sub-block is a the sub-block index, pass 0 if there is no sub-block. |
450 | * The address and value are hexadecimal numbers, leading 0x is optional. | |
2c22ed0b | 451 | * The mask means instance mask, is optional, default value is 0x1. |
879e723d | 452 | * |
737c375b | 453 | * For instance, |
879e723d AZ |
454 | * |
455 | * .. code-block:: bash | |
96ebb307 | 456 | * |
44494f96 | 457 | * echo inject umc ue 0x0 0x0 0x0 > /sys/kernel/debug/dri/0/ras/ras_ctrl |
2c22ed0b | 458 | * echo inject umc ce 0 0 0 3 > /sys/kernel/debug/dri/0/ras/ras_ctrl |
96ebb307 | 459 | * echo disable umc > /sys/kernel/debug/dri/0/ras/ras_ctrl |
460 | * | |
737c375b | 461 | * How to check the result of the operation? |
36ea1bd2 | 462 | * |
737c375b | 463 | * To check disable/enable, see "ras" features at, |
36ea1bd2 | 464 | * /sys/class/drm/card[0/1/2...]/device/ras/features |
465 | * | |
737c375b LT |
466 | * To check inject, see the corresponding error count at, |
467 | * /sys/class/drm/card[0/1/2...]/device/ras/[gfx|sdma|umc|...]_err_count | |
36ea1bd2 | 468 | * |
879e723d | 469 | * .. note:: |
ef177d11 | 470 | * Operations are only allowed on blocks which are supported. |
737c375b | 471 | * Check the "ras" mask at /sys/module/amdgpu/parameters/ras_mask |
ef177d11 AD |
472 | * to see which blocks support RAS on a particular asic. |
473 | * | |
36ea1bd2 | 474 | */ |
cf696091 LT |
475 | static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, |
476 | const char __user *buf, | |
477 | size_t size, loff_t *pos) | |
36ea1bd2 | 478 | { |
479 | struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private; | |
480 | struct ras_debug_if data; | |
481 | int ret = 0; | |
482 | ||
61380faa | 483 | if (!amdgpu_ras_get_error_query_ready(adev)) { |
6952e99c GC |
484 | dev_warn(adev->dev, "RAS WARN: error injection " |
485 | "currently inaccessible\n"); | |
43c4d576 JC |
486 | return size; |
487 | } | |
488 | ||
96ebb307 | 489 | ret = amdgpu_ras_debugfs_ctrl_parse_data(f, buf, size, pos, &data); |
490 | if (ret) | |
cf696091 | 491 | return ret; |
36ea1bd2 | 492 | |
80b0cd0f | 493 | if (data.op == 3) { |
cbb8f989 | 494 | ret = amdgpu_reserve_page_direct(adev, data.inject.address); |
80b0cd0f | 495 | if (!ret) |
cbb8f989 JC |
496 | return size; |
497 | else | |
498 | return ret; | |
499 | } | |
500 | ||
36ea1bd2 | 501 | if (!amdgpu_ras_is_supported(adev, data.head.block)) |
502 | return -EINVAL; | |
503 | ||
504 | switch (data.op) { | |
505 | case 0: | |
506 | ret = amdgpu_ras_feature_enable(adev, &data.head, 0); | |
507 | break; | |
508 | case 1: | |
509 | ret = amdgpu_ras_feature_enable(adev, &data.head, 1); | |
510 | break; | |
511 | case 2: | |
43aedbf4 SY |
512 | if ((data.inject.address >= adev->gmc.mc_vram_size && |
513 | adev->gmc.mc_vram_size) || | |
7cdc2ee3 | 514 | (data.inject.address >= RAS_UMC_INJECT_ADDR_LIMIT)) { |
b0d4783a GC |
515 | dev_warn(adev->dev, "RAS WARN: input address " |
516 | "0x%llx is invalid.", | |
517 | data.inject.address); | |
7cdc2ee3 TZ |
518 | ret = -EINVAL; |
519 | break; | |
520 | } | |
521 | ||
6e4be987 TZ |
522 | /* umc ce/ue error injection for a bad page is not allowed */ |
523 | if ((data.head.block == AMDGPU_RAS_BLOCK__UMC) && | |
524 | amdgpu_ras_check_bad_page(adev, data.inject.address)) { | |
c65b0805 LT |
525 | dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has " |
526 | "already been marked as bad!\n", | |
527 | data.inject.address); | |
6e4be987 TZ |
528 | break; |
529 | } | |
530 | ||
f464c5dd TZ |
531 | amdgpu_ras_instance_mask_check(adev, &data); |
532 | ||
7cdc2ee3 | 533 | /* data.inject.address is offset instead of absolute gpu address */ |
36ea1bd2 | 534 | ret = amdgpu_ras_error_inject(adev, &data.inject); |
535 | break; | |
96ebb307 | 536 | default: |
537 | ret = -EINVAL; | |
538 | break; | |
374bf7bd | 539 | } |
36ea1bd2 | 540 | |
541 | if (ret) | |
79c04621 | 542 | return ret; |
36ea1bd2 | 543 | |
544 | return size; | |
545 | } | |
546 | ||
084fe13b AG |
547 | /** |
548 | * DOC: AMDGPU RAS debugfs EEPROM table reset interface | |
549 | * | |
f77c7109 | 550 | * Some boards contain an EEPROM which is used to persistently store a list of |
ef177d11 | 551 | * bad pages which experiences ECC errors in vram. This interface provides |
f77c7109 AD |
552 | * a way to reset the EEPROM, e.g., after testing error injection. |
553 | * | |
554 | * Usage: | |
555 | * | |
556 | * .. code-block:: bash | |
557 | * | |
558 | * echo 1 > ../ras/ras_eeprom_reset | |
559 | * | |
560 | * will reset EEPROM table to 0 entries. | |
561 | * | |
084fe13b | 562 | */ |
cf696091 LT |
563 | static ssize_t amdgpu_ras_debugfs_eeprom_write(struct file *f, |
564 | const char __user *buf, | |
565 | size_t size, loff_t *pos) | |
084fe13b | 566 | { |
bf0b91b7 GC |
567 | struct amdgpu_device *adev = |
568 | (struct amdgpu_device *)file_inode(f)->i_private; | |
084fe13b AG |
569 | int ret; |
570 | ||
bf0b91b7 | 571 | ret = amdgpu_ras_eeprom_reset_table( |
cf696091 | 572 | &(amdgpu_ras_get_context(adev)->eeprom_control)); |
084fe13b | 573 | |
63d4c081 | 574 | if (!ret) { |
cf696091 LT |
575 | /* Something was written to EEPROM. |
576 | */ | |
bf0b91b7 GC |
577 | amdgpu_ras_get_context(adev)->flags = RAS_DEFAULT_FLAGS; |
578 | return size; | |
579 | } else { | |
cf696091 | 580 | return ret; |
bf0b91b7 | 581 | } |
084fe13b AG |
582 | } |
583 | ||
36ea1bd2 | 584 | static const struct file_operations amdgpu_ras_debugfs_ctrl_ops = { |
585 | .owner = THIS_MODULE, | |
586 | .read = NULL, | |
587 | .write = amdgpu_ras_debugfs_ctrl_write, | |
588 | .llseek = default_llseek | |
589 | }; | |
590 | ||
084fe13b AG |
591 | static const struct file_operations amdgpu_ras_debugfs_eeprom_ops = { |
592 | .owner = THIS_MODULE, | |
593 | .read = NULL, | |
594 | .write = amdgpu_ras_debugfs_eeprom_write, | |
595 | .llseek = default_llseek | |
596 | }; | |
597 | ||
f77c7109 AD |
598 | /** |
599 | * DOC: AMDGPU RAS sysfs Error Count Interface | |
600 | * | |
ef177d11 | 601 | * It allows the user to read the error count for each IP block on the gpu through |
f77c7109 AD |
602 | * /sys/class/drm/card[0/1/2...]/device/ras/[gfx/sdma/...]_err_count |
603 | * | |
604 | * It outputs the multiple lines which report the uncorrected (ue) and corrected | |
605 | * (ce) error counts. | |
606 | * | |
607 | * The format of one line is below, | |
608 | * | |
609 | * [ce|ue]: count | |
610 | * | |
611 | * Example: | |
612 | * | |
613 | * .. code-block:: bash | |
614 | * | |
615 | * ue: 0 | |
616 | * ce: 1 | |
617 | * | |
618 | */ | |
c030f2e4 | 619 | static ssize_t amdgpu_ras_sysfs_read(struct device *dev, |
620 | struct device_attribute *attr, char *buf) | |
621 | { | |
622 | struct ras_manager *obj = container_of(attr, struct ras_manager, sysfs_attr); | |
623 | struct ras_query_if info = { | |
624 | .head = obj->head, | |
625 | }; | |
626 | ||
61380faa | 627 | if (!amdgpu_ras_get_error_query_ready(obj->adev)) |
36000c7a | 628 | return sysfs_emit(buf, "Query currently inaccessible\n"); |
43c4d576 | 629 | |
761d86d3 | 630 | if (amdgpu_ras_query_error_status(obj->adev, &info)) |
c030f2e4 | 631 | return -EINVAL; |
632 | ||
4e8303cf LL |
633 | if (amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) && |
634 | amdgpu_ip_version(obj->adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 4)) { | |
1f0d8e37 | 635 | if (amdgpu_ras_reset_error_status(obj->adev, info.head.block)) |
2a460963 | 636 | dev_warn(obj->adev->dev, "Failed to reset error counter and error status"); |
1f0d8e37 MJ |
637 | } |
638 | ||
2c7a1560 SY |
639 | if (info.head.block == AMDGPU_RAS_BLOCK__UMC) |
640 | return sysfs_emit(buf, "%s: %lu\n%s: %lu\n%s: %lu\n", "ue", info.ue_count, | |
641 | "ce", info.ce_count, "de", info.de_count); | |
642 | else | |
643 | return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count, | |
644 | "ce", info.ce_count); | |
c030f2e4 | 645 | } |
646 | ||
647 | /* obj begin */ | |
648 | ||
649 | #define get_obj(obj) do { (obj)->use++; } while (0) | |
650 | #define alive_obj(obj) ((obj)->use) | |
651 | ||
652 | static inline void put_obj(struct ras_manager *obj) | |
653 | { | |
ec3e0a91 | 654 | if (obj && (--obj->use == 0)) { |
c030f2e4 | 655 | list_del(&obj->node); |
ec3e0a91 YW |
656 | amdgpu_ras_error_data_fini(&obj->err_data); |
657 | } | |
658 | ||
f0872686 | 659 | if (obj && (obj->use < 0)) |
640ae42e | 660 | DRM_ERROR("RAS ERROR: Unbalance obj(%s) use\n", get_ras_block_str(&obj->head)); |
c030f2e4 | 661 | } |
662 | ||
663 | /* make one obj and return it. */ | |
664 | static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev, | |
665 | struct ras_common_if *head) | |
666 | { | |
667 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
668 | struct ras_manager *obj; | |
669 | ||
8ab0d6f0 | 670 | if (!adev->ras_enabled || !con) |
c030f2e4 | 671 | return NULL; |
672 | ||
673 | if (head->block >= AMDGPU_RAS_BLOCK_COUNT) | |
674 | return NULL; | |
675 | ||
640ae42e JC |
676 | if (head->block == AMDGPU_RAS_BLOCK__MCA) { |
677 | if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST) | |
678 | return NULL; | |
679 | ||
680 | obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index]; | |
681 | } else | |
682 | obj = &con->objs[head->block]; | |
683 | ||
c030f2e4 | 684 | /* already exist. return obj? */ |
685 | if (alive_obj(obj)) | |
686 | return NULL; | |
687 | ||
ec3e0a91 YW |
688 | if (amdgpu_ras_error_data_init(&obj->err_data)) |
689 | return NULL; | |
690 | ||
c030f2e4 | 691 | obj->head = *head; |
692 | obj->adev = adev; | |
693 | list_add(&obj->node, &con->head); | |
694 | get_obj(obj); | |
695 | ||
696 | return obj; | |
697 | } | |
698 | ||
699 | /* return an obj equal to head, or the first when head is NULL */ | |
f2a79be1 | 700 | struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev, |
c030f2e4 | 701 | struct ras_common_if *head) |
702 | { | |
703 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
704 | struct ras_manager *obj; | |
705 | int i; | |
706 | ||
8ab0d6f0 | 707 | if (!adev->ras_enabled || !con) |
c030f2e4 | 708 | return NULL; |
709 | ||
710 | if (head) { | |
711 | if (head->block >= AMDGPU_RAS_BLOCK_COUNT) | |
712 | return NULL; | |
713 | ||
640ae42e JC |
714 | if (head->block == AMDGPU_RAS_BLOCK__MCA) { |
715 | if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST) | |
716 | return NULL; | |
717 | ||
718 | obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index]; | |
719 | } else | |
720 | obj = &con->objs[head->block]; | |
c030f2e4 | 721 | |
640ae42e | 722 | if (alive_obj(obj)) |
c030f2e4 | 723 | return obj; |
c030f2e4 | 724 | } else { |
640ae42e | 725 | for (i = 0; i < AMDGPU_RAS_BLOCK_COUNT + AMDGPU_RAS_MCA_BLOCK_COUNT; i++) { |
c030f2e4 | 726 | obj = &con->objs[i]; |
640ae42e | 727 | if (alive_obj(obj)) |
c030f2e4 | 728 | return obj; |
c030f2e4 | 729 | } |
730 | } | |
731 | ||
732 | return NULL; | |
733 | } | |
734 | /* obj end */ | |
735 | ||
736 | /* feature ctl begin */ | |
737 | static int amdgpu_ras_is_feature_allowed(struct amdgpu_device *adev, | |
e509965e | 738 | struct ras_common_if *head) |
c030f2e4 | 739 | { |
8ab0d6f0 | 740 | return adev->ras_hw_enabled & BIT(head->block); |
c030f2e4 | 741 | } |
742 | ||
743 | static int amdgpu_ras_is_feature_enabled(struct amdgpu_device *adev, | |
744 | struct ras_common_if *head) | |
745 | { | |
746 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
747 | ||
748 | return con->features & BIT(head->block); | |
749 | } | |
750 | ||
751 | /* | |
752 | * if obj is not created, then create one. | |
753 | * set feature enable flag. | |
754 | */ | |
755 | static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev, | |
756 | struct ras_common_if *head, int enable) | |
757 | { | |
758 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
759 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |
760 | ||
5caf466a | 761 | /* If hardware does not support ras, then do not create obj. |
762 | * But if hardware support ras, we can create the obj. | |
763 | * Ras framework checks con->hw_supported to see if it need do | |
764 | * corresponding initialization. | |
765 | * IP checks con->support to see if it need disable ras. | |
766 | */ | |
c030f2e4 | 767 | if (!amdgpu_ras_is_feature_allowed(adev, head)) |
768 | return 0; | |
c030f2e4 | 769 | |
770 | if (enable) { | |
771 | if (!obj) { | |
772 | obj = amdgpu_ras_create_obj(adev, head); | |
773 | if (!obj) | |
774 | return -EINVAL; | |
775 | } else { | |
776 | /* In case we create obj somewhere else */ | |
777 | get_obj(obj); | |
778 | } | |
779 | con->features |= BIT(head->block); | |
780 | } else { | |
781 | if (obj && amdgpu_ras_is_feature_enabled(adev, head)) { | |
19d0dfda | 782 | con->features &= ~BIT(head->block); |
c030f2e4 | 783 | put_obj(obj); |
784 | } | |
785 | } | |
786 | ||
787 | return 0; | |
788 | } | |
789 | ||
790 | /* wrapper of psp_ras_enable_features */ | |
791 | int amdgpu_ras_feature_enable(struct amdgpu_device *adev, | |
792 | struct ras_common_if *head, bool enable) | |
793 | { | |
794 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
7fcffecf | 795 | union ta_ras_cmd_input *info; |
bf7aa8be | 796 | int ret; |
c030f2e4 | 797 | |
798 | if (!con) | |
799 | return -EINVAL; | |
800 | ||
ec70578c HZ |
801 | /* For non-gfx ip, do not enable ras feature if it is not allowed */ |
802 | /* For gfx ip, regardless of feature support status, */ | |
803 | /* Force issue enable or disable ras feature commands */ | |
804 | if (head->block != AMDGPU_RAS_BLOCK__GFX && | |
6fc9d92c | 805 | !amdgpu_ras_is_feature_allowed(adev, head)) |
bf7aa8be | 806 | return 0; |
6fc9d92c HZ |
807 | |
808 | /* Only enable gfx ras feature from host side */ | |
809 | if (head->block == AMDGPU_RAS_BLOCK__GFX && | |
810 | !amdgpu_sriov_vf(adev) && | |
811 | !amdgpu_ras_intr_triggered()) { | |
26093ce1 SY |
812 | info = kzalloc(sizeof(union ta_ras_cmd_input), GFP_KERNEL); |
813 | if (!info) | |
814 | return -ENOMEM; | |
815 | ||
816 | if (!enable) { | |
817 | info->disable_features = (struct ta_ras_disable_features_input) { | |
818 | .block_id = amdgpu_ras_block_to_ta(head->block), | |
819 | .error_type = amdgpu_ras_error_to_ta(head->type), | |
820 | }; | |
821 | } else { | |
822 | info->enable_features = (struct ta_ras_enable_features_input) { | |
823 | .block_id = amdgpu_ras_block_to_ta(head->block), | |
824 | .error_type = amdgpu_ras_error_to_ta(head->type), | |
825 | }; | |
826 | } | |
c030f2e4 | 827 | |
7fcffecf | 828 | ret = psp_ras_enable_features(&adev->psp, info, enable); |
bff77e86 | 829 | if (ret) { |
e4348849 | 830 | dev_err(adev->dev, "ras %s %s failed poison:%d ret:%d\n", |
011907fd | 831 | enable ? "enable":"disable", |
640ae42e | 832 | get_ras_block_str(head), |
e4348849 | 833 | amdgpu_ras_is_poison_mode_supported(adev), ret); |
5838f74c | 834 | kfree(info); |
bf7aa8be | 835 | return ret; |
bff77e86 | 836 | } |
bf7aa8be HZ |
837 | |
838 | kfree(info); | |
c030f2e4 | 839 | } |
840 | ||
841 | /* setup the obj */ | |
842 | __amdgpu_ras_feature_enable(adev, head, enable); | |
bf7aa8be HZ |
843 | |
844 | return 0; | |
c030f2e4 | 845 | } |
846 | ||
77de502b | 847 | /* Only used in device probe stage and called only once. */ |
848 | int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev, | |
849 | struct ras_common_if *head, bool enable) | |
850 | { | |
851 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
852 | int ret; | |
853 | ||
854 | if (!con) | |
855 | return -EINVAL; | |
856 | ||
857 | if (con->flags & AMDGPU_RAS_FLAG_INIT_BY_VBIOS) { | |
7af23ebe | 858 | if (enable) { |
859 | /* There is no harm to issue a ras TA cmd regardless of | |
860 | * the currecnt ras state. | |
861 | * If current state == target state, it will do nothing | |
862 | * But sometimes it requests driver to reset and repost | |
863 | * with error code -EAGAIN. | |
864 | */ | |
865 | ret = amdgpu_ras_feature_enable(adev, head, 1); | |
866 | /* With old ras TA, we might fail to enable ras. | |
867 | * Log it and just setup the object. | |
868 | * TODO need remove this WA in the future. | |
869 | */ | |
870 | if (ret == -EINVAL) { | |
871 | ret = __amdgpu_ras_feature_enable(adev, head, 1); | |
872 | if (!ret) | |
6952e99c GC |
873 | dev_info(adev->dev, |
874 | "RAS INFO: %s setup object\n", | |
640ae42e | 875 | get_ras_block_str(head)); |
7af23ebe | 876 | } |
877 | } else { | |
878 | /* setup the object then issue a ras TA disable cmd.*/ | |
879 | ret = __amdgpu_ras_feature_enable(adev, head, 1); | |
880 | if (ret) | |
881 | return ret; | |
77de502b | 882 | |
970fd197 SY |
883 | /* gfx block ras dsiable cmd must send to ras-ta */ |
884 | if (head->block == AMDGPU_RAS_BLOCK__GFX) | |
885 | con->features |= BIT(head->block); | |
886 | ||
77de502b | 887 | ret = amdgpu_ras_feature_enable(adev, head, 0); |
19d0dfda SY |
888 | |
889 | /* clean gfx block ras features flag */ | |
8ab0d6f0 | 890 | if (adev->ras_enabled && head->block == AMDGPU_RAS_BLOCK__GFX) |
19d0dfda | 891 | con->features &= ~BIT(head->block); |
7af23ebe | 892 | } |
77de502b | 893 | } else |
894 | ret = amdgpu_ras_feature_enable(adev, head, enable); | |
895 | ||
896 | return ret; | |
897 | } | |
898 | ||
c030f2e4 | 899 | static int amdgpu_ras_disable_all_features(struct amdgpu_device *adev, |
900 | bool bypass) | |
901 | { | |
902 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
903 | struct ras_manager *obj, *tmp; | |
904 | ||
905 | list_for_each_entry_safe(obj, tmp, &con->head, node) { | |
906 | /* bypass psp. | |
907 | * aka just release the obj and corresponding flags | |
908 | */ | |
909 | if (bypass) { | |
910 | if (__amdgpu_ras_feature_enable(adev, &obj->head, 0)) | |
911 | break; | |
912 | } else { | |
913 | if (amdgpu_ras_feature_enable(adev, &obj->head, 0)) | |
914 | break; | |
915 | } | |
289d513b | 916 | } |
c030f2e4 | 917 | |
918 | return con->features; | |
919 | } | |
920 | ||
921 | static int amdgpu_ras_enable_all_features(struct amdgpu_device *adev, | |
922 | bool bypass) | |
923 | { | |
924 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
c030f2e4 | 925 | int i; |
640ae42e | 926 | const enum amdgpu_ras_error_type default_ras_type = AMDGPU_RAS_ERROR__NONE; |
c030f2e4 | 927 | |
640ae42e | 928 | for (i = 0; i < AMDGPU_RAS_BLOCK_COUNT; i++) { |
c030f2e4 | 929 | struct ras_common_if head = { |
930 | .block = i, | |
191051a1 | 931 | .type = default_ras_type, |
c030f2e4 | 932 | .sub_block_index = 0, |
933 | }; | |
640ae42e JC |
934 | |
935 | if (i == AMDGPU_RAS_BLOCK__MCA) | |
936 | continue; | |
937 | ||
938 | if (bypass) { | |
939 | /* | |
940 | * bypass psp. vbios enable ras for us. | |
941 | * so just create the obj | |
942 | */ | |
943 | if (__amdgpu_ras_feature_enable(adev, &head, 1)) | |
944 | break; | |
945 | } else { | |
946 | if (amdgpu_ras_feature_enable(adev, &head, 1)) | |
947 | break; | |
948 | } | |
949 | } | |
950 | ||
951 | for (i = 0; i < AMDGPU_RAS_MCA_BLOCK_COUNT; i++) { | |
952 | struct ras_common_if head = { | |
953 | .block = AMDGPU_RAS_BLOCK__MCA, | |
954 | .type = default_ras_type, | |
955 | .sub_block_index = i, | |
956 | }; | |
957 | ||
c030f2e4 | 958 | if (bypass) { |
959 | /* | |
960 | * bypass psp. vbios enable ras for us. | |
961 | * so just create the obj | |
962 | */ | |
963 | if (__amdgpu_ras_feature_enable(adev, &head, 1)) | |
964 | break; | |
965 | } else { | |
966 | if (amdgpu_ras_feature_enable(adev, &head, 1)) | |
967 | break; | |
968 | } | |
289d513b | 969 | } |
c030f2e4 | 970 | |
971 | return con->features; | |
972 | } | |
973 | /* feature ctl end */ | |
974 | ||
e3d833f4 | 975 | static int amdgpu_ras_block_match_default(struct amdgpu_ras_block_object *block_obj, |
976 | enum amdgpu_ras_block block) | |
6492e1b0 | 977 | { |
b6efdb02 | 978 | if (!block_obj) |
6492e1b0 | 979 | return -EINVAL; |
980 | ||
bdb3489c | 981 | if (block_obj->ras_comm.block == block) |
6492e1b0 | 982 | return 0; |
640ae42e | 983 | |
6492e1b0 | 984 | return -EINVAL; |
985 | } | |
986 | ||
b6efdb02 | 987 | static struct amdgpu_ras_block_object *amdgpu_ras_get_ras_block(struct amdgpu_device *adev, |
6492e1b0 | 988 | enum amdgpu_ras_block block, uint32_t sub_block_index) |
640ae42e | 989 | { |
d5e8ff5f | 990 | struct amdgpu_ras_block_list *node, *tmp; |
991 | struct amdgpu_ras_block_object *obj; | |
6492e1b0 | 992 | |
993 | if (block >= AMDGPU_RAS_BLOCK__LAST) | |
994 | return NULL; | |
995 | ||
d5e8ff5f | 996 | list_for_each_entry_safe(node, tmp, &adev->ras_list, node) { |
997 | if (!node->ras_obj) { | |
998 | dev_warn(adev->dev, "Warning: abnormal ras list node.\n"); | |
999 | continue; | |
1000 | } | |
1001 | ||
1002 | obj = node->ras_obj; | |
6492e1b0 | 1003 | if (obj->ras_block_match) { |
1004 | if (obj->ras_block_match(obj, block, sub_block_index) == 0) | |
1005 | return obj; | |
1006 | } else { | |
1007 | if (amdgpu_ras_block_match_default(obj, block) == 0) | |
1008 | return obj; | |
1009 | } | |
640ae42e | 1010 | } |
6492e1b0 | 1011 | |
1012 | return NULL; | |
640ae42e JC |
1013 | } |
1014 | ||
fdcb279d SY |
1015 | static void amdgpu_ras_get_ecc_info(struct amdgpu_device *adev, struct ras_err_data *err_data) |
1016 | { | |
1017 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |
1018 | int ret = 0; | |
1019 | ||
1020 | /* | |
1021 | * choosing right query method according to | |
1022 | * whether smu support query error information | |
1023 | */ | |
bc143d8b | 1024 | ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(ras->umc_ecc)); |
fdcb279d | 1025 | if (ret == -EOPNOTSUPP) { |
efe17d5a | 1026 | if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && |
1027 | adev->umc.ras->ras_block.hw_ops->query_ras_error_count) | |
1028 | adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev, err_data); | |
fdcb279d SY |
1029 | |
1030 | /* umc query_ras_error_address is also responsible for clearing | |
1031 | * error status | |
1032 | */ | |
efe17d5a | 1033 | if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && |
1034 | adev->umc.ras->ras_block.hw_ops->query_ras_error_address) | |
1035 | adev->umc.ras->ras_block.hw_ops->query_ras_error_address(adev, err_data); | |
fdcb279d | 1036 | } else if (!ret) { |
efe17d5a | 1037 | if (adev->umc.ras && |
1038 | adev->umc.ras->ecc_info_query_ras_error_count) | |
1039 | adev->umc.ras->ecc_info_query_ras_error_count(adev, err_data); | |
fdcb279d | 1040 | |
efe17d5a | 1041 | if (adev->umc.ras && |
1042 | adev->umc.ras->ecc_info_query_ras_error_address) | |
1043 | adev->umc.ras->ecc_info_query_ras_error_address(adev, err_data); | |
fdcb279d SY |
1044 | } |
1045 | } | |
1046 | ||
5b1270be | 1047 | static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, |
ec3e0a91 | 1048 | struct ras_manager *ras_mgr, |
5b1270be | 1049 | struct ras_err_data *err_data, |
9dc57c2a | 1050 | struct ras_query_context *qctx, |
ec3e0a91 | 1051 | const char *blk_name, |
46e2231c CL |
1052 | bool is_ue, |
1053 | bool is_de) | |
5b1270be | 1054 | { |
5b1270be YW |
1055 | struct amdgpu_smuio_mcm_config_info *mcm_info; |
1056 | struct ras_err_node *err_node; | |
1057 | struct ras_err_info *err_info; | |
9dc57c2a | 1058 | u64 event_id = qctx->event_id; |
5b1270be | 1059 | |
ec3e0a91 YW |
1060 | if (is_ue) { |
1061 | for_each_ras_error(err_node, err_data) { | |
1062 | err_info = &err_node->err_info; | |
1063 | mcm_info = &err_info->mcm_info; | |
1064 | if (err_info->ue_count) { | |
9dc57c2a YW |
1065 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1066 | "%lld new uncorrectable hardware errors detected in %s block\n", | |
1067 | mcm_info->socket_id, | |
1068 | mcm_info->die_id, | |
1069 | err_info->ue_count, | |
1070 | blk_name); | |
ec3e0a91 YW |
1071 | } |
1072 | } | |
1073 | ||
1074 | for_each_ras_error(err_node, &ras_mgr->err_data) { | |
1075 | err_info = &err_node->err_info; | |
1076 | mcm_info = &err_info->mcm_info; | |
9dc57c2a YW |
1077 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1078 | "%lld uncorrectable hardware errors detected in total in %s block\n", | |
1079 | mcm_info->socket_id, mcm_info->die_id, err_info->ue_count, blk_name); | |
ec3e0a91 YW |
1080 | } |
1081 | ||
1082 | } else { | |
46e2231c CL |
1083 | if (is_de) { |
1084 | for_each_ras_error(err_node, err_data) { | |
1085 | err_info = &err_node->err_info; | |
1086 | mcm_info = &err_info->mcm_info; | |
1087 | if (err_info->de_count) { | |
9dc57c2a YW |
1088 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1089 | "%lld new deferred hardware errors detected in %s block\n", | |
1090 | mcm_info->socket_id, | |
1091 | mcm_info->die_id, | |
1092 | err_info->de_count, | |
1093 | blk_name); | |
46e2231c CL |
1094 | } |
1095 | } | |
1096 | ||
1097 | for_each_ras_error(err_node, &ras_mgr->err_data) { | |
1098 | err_info = &err_node->err_info; | |
1099 | mcm_info = &err_info->mcm_info; | |
9dc57c2a YW |
1100 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1101 | "%lld deferred hardware errors detected in total in %s block\n", | |
1102 | mcm_info->socket_id, mcm_info->die_id, | |
1103 | err_info->de_count, blk_name); | |
46e2231c CL |
1104 | } |
1105 | } else { | |
1106 | for_each_ras_error(err_node, err_data) { | |
1107 | err_info = &err_node->err_info; | |
1108 | mcm_info = &err_info->mcm_info; | |
1109 | if (err_info->ce_count) { | |
9dc57c2a YW |
1110 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1111 | "%lld new correctable hardware errors detected in %s block\n", | |
1112 | mcm_info->socket_id, | |
1113 | mcm_info->die_id, | |
1114 | err_info->ce_count, | |
1115 | blk_name); | |
46e2231c | 1116 | } |
ec3e0a91 | 1117 | } |
ec3e0a91 | 1118 | |
46e2231c CL |
1119 | for_each_ras_error(err_node, &ras_mgr->err_data) { |
1120 | err_info = &err_node->err_info; | |
1121 | mcm_info = &err_info->mcm_info; | |
9dc57c2a YW |
1122 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d, " |
1123 | "%lld correctable hardware errors detected in total in %s block\n", | |
1124 | mcm_info->socket_id, mcm_info->die_id, | |
1125 | err_info->ce_count, blk_name); | |
46e2231c | 1126 | } |
5b1270be YW |
1127 | } |
1128 | } | |
1129 | } | |
1130 | ||
ec3e0a91 YW |
1131 | static inline bool err_data_has_source_info(struct ras_err_data *data) |
1132 | { | |
1133 | return !list_empty(&data->err_node_list); | |
1134 | } | |
1135 | ||
5b1270be YW |
1136 | static void amdgpu_ras_error_generate_report(struct amdgpu_device *adev, |
1137 | struct ras_query_if *query_if, | |
9dc57c2a YW |
1138 | struct ras_err_data *err_data, |
1139 | struct ras_query_context *qctx) | |
5b1270be YW |
1140 | { |
1141 | struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, &query_if->head); | |
1142 | const char *blk_name = get_ras_block_str(&query_if->head); | |
9dc57c2a | 1143 | u64 event_id = qctx->event_id; |
5b1270be YW |
1144 | |
1145 | if (err_data->ce_count) { | |
ec3e0a91 | 1146 | if (err_data_has_source_info(err_data)) { |
9dc57c2a | 1147 | amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, |
46e2231c | 1148 | blk_name, false, false); |
5b1270be YW |
1149 | } else if (!adev->aid_mask && |
1150 | adev->smuio.funcs && | |
1151 | adev->smuio.funcs->get_socket_id && | |
1152 | adev->smuio.funcs->get_die_id) { | |
9dc57c2a YW |
1153 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d " |
1154 | "%ld correctable hardware errors " | |
1155 | "detected in %s block\n", | |
1156 | adev->smuio.funcs->get_socket_id(adev), | |
1157 | adev->smuio.funcs->get_die_id(adev), | |
1158 | ras_mgr->err_data.ce_count, | |
1159 | blk_name); | |
5b1270be | 1160 | } else { |
9dc57c2a YW |
1161 | RAS_EVENT_LOG(adev, event_id, "%ld correctable hardware errors " |
1162 | "detected in %s block\n", | |
1163 | ras_mgr->err_data.ce_count, | |
1164 | blk_name); | |
5b1270be YW |
1165 | } |
1166 | } | |
1167 | ||
1168 | if (err_data->ue_count) { | |
ec3e0a91 | 1169 | if (err_data_has_source_info(err_data)) { |
9dc57c2a | 1170 | amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, |
46e2231c | 1171 | blk_name, true, false); |
5b1270be YW |
1172 | } else if (!adev->aid_mask && |
1173 | adev->smuio.funcs && | |
1174 | adev->smuio.funcs->get_socket_id && | |
1175 | adev->smuio.funcs->get_die_id) { | |
9dc57c2a YW |
1176 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d " |
1177 | "%ld uncorrectable hardware errors " | |
1178 | "detected in %s block\n", | |
1179 | adev->smuio.funcs->get_socket_id(adev), | |
1180 | adev->smuio.funcs->get_die_id(adev), | |
1181 | ras_mgr->err_data.ue_count, | |
1182 | blk_name); | |
5b1270be | 1183 | } else { |
9dc57c2a YW |
1184 | RAS_EVENT_LOG(adev, event_id, "%ld uncorrectable hardware errors " |
1185 | "detected in %s block\n", | |
1186 | ras_mgr->err_data.ue_count, | |
1187 | blk_name); | |
5b1270be YW |
1188 | } |
1189 | } | |
1190 | ||
46e2231c CL |
1191 | if (err_data->de_count) { |
1192 | if (err_data_has_source_info(err_data)) { | |
9dc57c2a | 1193 | amdgpu_ras_error_print_error_data(adev, ras_mgr, err_data, qctx, |
46e2231c CL |
1194 | blk_name, false, true); |
1195 | } else if (!adev->aid_mask && | |
1196 | adev->smuio.funcs && | |
1197 | adev->smuio.funcs->get_socket_id && | |
1198 | adev->smuio.funcs->get_die_id) { | |
9dc57c2a YW |
1199 | RAS_EVENT_LOG(adev, event_id, "socket: %d, die: %d " |
1200 | "%ld deferred hardware errors " | |
1201 | "detected in %s block\n", | |
1202 | adev->smuio.funcs->get_socket_id(adev), | |
1203 | adev->smuio.funcs->get_die_id(adev), | |
1204 | ras_mgr->err_data.de_count, | |
1205 | blk_name); | |
46e2231c | 1206 | } else { |
9dc57c2a YW |
1207 | RAS_EVENT_LOG(adev, event_id, "%ld deferred hardware errors " |
1208 | "detected in %s block\n", | |
1209 | ras_mgr->err_data.de_count, | |
1210 | blk_name); | |
46e2231c CL |
1211 | } |
1212 | } | |
5b1270be YW |
1213 | } |
1214 | ||
ec3e0a91 YW |
1215 | static void amdgpu_rasmgr_error_data_statistic_update(struct ras_manager *obj, struct ras_err_data *err_data) |
1216 | { | |
1217 | struct ras_err_node *err_node; | |
1218 | struct ras_err_info *err_info; | |
1219 | ||
1220 | if (err_data_has_source_info(err_data)) { | |
1221 | for_each_ras_error(err_node, err_data) { | |
1222 | err_info = &err_node->err_info; | |
46e2231c CL |
1223 | amdgpu_ras_error_statistic_de_count(&obj->err_data, |
1224 | &err_info->mcm_info, NULL, err_info->de_count); | |
9f91e983 YC |
1225 | amdgpu_ras_error_statistic_ce_count(&obj->err_data, |
1226 | &err_info->mcm_info, NULL, err_info->ce_count); | |
1227 | amdgpu_ras_error_statistic_ue_count(&obj->err_data, | |
1228 | &err_info->mcm_info, NULL, err_info->ue_count); | |
ec3e0a91 YW |
1229 | } |
1230 | } else { | |
1231 | /* for legacy asic path which doesn't has error source info */ | |
1232 | obj->err_data.ue_count += err_data->ue_count; | |
1233 | obj->err_data.ce_count += err_data->ce_count; | |
46e2231c | 1234 | obj->err_data.de_count += err_data->de_count; |
ec3e0a91 YW |
1235 | } |
1236 | } | |
1237 | ||
04c4fcd2 YW |
1238 | static struct ras_manager *get_ras_manager(struct amdgpu_device *adev, enum amdgpu_ras_block blk) |
1239 | { | |
1240 | struct ras_common_if head; | |
1241 | ||
1242 | memset(&head, 0, sizeof(head)); | |
1243 | head.block = blk; | |
1244 | ||
1245 | return amdgpu_ras_find_obj(adev, &head); | |
1246 | } | |
1247 | ||
1248 | int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk, | |
1249 | const struct aca_info *aca_info, void *data) | |
1250 | { | |
1251 | struct ras_manager *obj; | |
1252 | ||
6f3b6913 YC |
1253 | /* in resume phase, no need to create aca fs node */ |
1254 | if (adev->in_suspend || amdgpu_in_reset(adev)) | |
1255 | return 0; | |
1256 | ||
04c4fcd2 YW |
1257 | obj = get_ras_manager(adev, blk); |
1258 | if (!obj) | |
1259 | return -EINVAL; | |
1260 | ||
1261 | return amdgpu_aca_add_handle(adev, &obj->aca_handle, ras_block_str(blk), aca_info, data); | |
1262 | } | |
1263 | ||
1264 | int amdgpu_ras_unbind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk) | |
1265 | { | |
1266 | struct ras_manager *obj; | |
1267 | ||
1268 | obj = get_ras_manager(adev, blk); | |
1269 | if (!obj) | |
1270 | return -EINVAL; | |
1271 | ||
1272 | amdgpu_aca_remove_handle(&obj->aca_handle); | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | static int amdgpu_aca_log_ras_error_data(struct amdgpu_device *adev, enum amdgpu_ras_block blk, | |
31fd330b YW |
1278 | enum aca_error_type type, struct ras_err_data *err_data, |
1279 | struct ras_query_context *qctx) | |
04c4fcd2 YW |
1280 | { |
1281 | struct ras_manager *obj; | |
1282 | ||
1283 | obj = get_ras_manager(adev, blk); | |
1284 | if (!obj) | |
1285 | return -EINVAL; | |
1286 | ||
31fd330b | 1287 | return amdgpu_aca_get_error_data(adev, &obj->aca_handle, type, err_data, qctx); |
04c4fcd2 YW |
1288 | } |
1289 | ||
37973b69 YW |
1290 | ssize_t amdgpu_ras_aca_sysfs_read(struct device *dev, struct device_attribute *attr, |
1291 | struct aca_handle *handle, char *buf, void *data) | |
1292 | { | |
1293 | struct ras_manager *obj = container_of(handle, struct ras_manager, aca_handle); | |
1294 | struct ras_query_if info = { | |
1295 | .head = obj->head, | |
1296 | }; | |
1297 | ||
1298 | if (amdgpu_ras_query_error_status(obj->adev, &info)) | |
1299 | return -EINVAL; | |
1300 | ||
865d3397 YW |
1301 | return sysfs_emit(buf, "%s: %lu\n%s: %lu\n%s: %lu\n", "ue", info.ue_count, |
1302 | "ce", info.ce_count, "de", info.ue_count); | |
37973b69 YW |
1303 | } |
1304 | ||
8cc0f566 HZ |
1305 | static int amdgpu_ras_query_error_status_helper(struct amdgpu_device *adev, |
1306 | struct ras_query_if *info, | |
1307 | struct ras_err_data *err_data, | |
9dc57c2a | 1308 | struct ras_query_context *qctx, |
8cc0f566 | 1309 | unsigned int error_query_mode) |
c030f2e4 | 1310 | { |
8cc0f566 | 1311 | enum amdgpu_ras_block blk = info ? info->head.block : AMDGPU_RAS_BLOCK_COUNT; |
b6efdb02 | 1312 | struct amdgpu_ras_block_object *block_obj = NULL; |
04c4fcd2 | 1313 | int ret; |
8cc0f566 | 1314 | |
b8d55a90 SS |
1315 | if (blk == AMDGPU_RAS_BLOCK_COUNT) |
1316 | return -EINVAL; | |
1317 | ||
8cc0f566 HZ |
1318 | if (error_query_mode == AMDGPU_RAS_INVALID_ERROR_QUERY) |
1319 | return -EINVAL; | |
1320 | ||
1321 | if (error_query_mode == AMDGPU_RAS_DIRECT_ERROR_QUERY) { | |
1322 | if (info->head.block == AMDGPU_RAS_BLOCK__UMC) { | |
1323 | amdgpu_ras_get_ecc_info(adev, err_data); | |
1324 | } else { | |
1325 | block_obj = amdgpu_ras_get_ras_block(adev, info->head.block, 0); | |
1326 | if (!block_obj || !block_obj->hw_ops) { | |
1327 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", | |
1328 | get_ras_block_str(&info->head)); | |
1329 | return -EINVAL; | |
1330 | } | |
1331 | ||
1332 | if (block_obj->hw_ops->query_ras_error_count) | |
07ee43fa | 1333 | block_obj->hw_ops->query_ras_error_count(adev, err_data); |
8cc0f566 HZ |
1334 | |
1335 | if ((info->head.block == AMDGPU_RAS_BLOCK__SDMA) || | |
1336 | (info->head.block == AMDGPU_RAS_BLOCK__GFX) || | |
1337 | (info->head.block == AMDGPU_RAS_BLOCK__MMHUB)) { | |
1338 | if (block_obj->hw_ops->query_ras_error_status) | |
1339 | block_obj->hw_ops->query_ras_error_status(adev); | |
1340 | } | |
1341 | } | |
1342 | } else { | |
04c4fcd2 | 1343 | if (amdgpu_aca_is_enabled(adev)) { |
31fd330b | 1344 | ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_UE, err_data, qctx); |
04c4fcd2 YW |
1345 | if (ret) |
1346 | return ret; | |
1347 | ||
31fd330b | 1348 | ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_CE, err_data, qctx); |
04c4fcd2 YW |
1349 | if (ret) |
1350 | return ret; | |
865d3397 | 1351 | |
31fd330b | 1352 | ret = amdgpu_aca_log_ras_error_data(adev, blk, ACA_ERROR_TYPE_DEFERRED, err_data, qctx); |
865d3397 YW |
1353 | if (ret) |
1354 | return ret; | |
04c4fcd2 YW |
1355 | } else { |
1356 | /* FIXME: add code to check return value later */ | |
9dc57c2a YW |
1357 | amdgpu_mca_smu_log_ras_error(adev, blk, AMDGPU_MCA_ERROR_TYPE_UE, err_data, qctx); |
1358 | amdgpu_mca_smu_log_ras_error(adev, blk, AMDGPU_MCA_ERROR_TYPE_CE, err_data, qctx); | |
04c4fcd2 | 1359 | } |
8cc0f566 HZ |
1360 | } |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
1365 | /* query/inject/cure begin */ | |
1366 | int amdgpu_ras_query_error_status(struct amdgpu_device *adev, struct ras_query_if *info) | |
1367 | { | |
c030f2e4 | 1368 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); |
5b1270be | 1369 | struct ras_err_data err_data; |
9dc57c2a | 1370 | struct ras_query_context qctx; |
8cc0f566 | 1371 | unsigned int error_query_mode; |
5b1270be | 1372 | int ret; |
c030f2e4 | 1373 | |
1374 | if (!obj) | |
1375 | return -EINVAL; | |
c030f2e4 | 1376 | |
5b1270be YW |
1377 | ret = amdgpu_ras_error_data_init(&err_data); |
1378 | if (ret) | |
1379 | return ret; | |
1380 | ||
8cc0f566 HZ |
1381 | if (!amdgpu_ras_get_error_query_mode(adev, &error_query_mode)) |
1382 | return -EINVAL; | |
7389a5b8 | 1383 | |
9dc57c2a YW |
1384 | memset(&qctx, 0, sizeof(qctx)); |
1385 | qctx.event_id = amdgpu_ras_acquire_event_id(adev, amdgpu_ras_intr_triggered() ? | |
1386 | RAS_EVENT_TYPE_ISR : RAS_EVENT_TYPE_INVALID); | |
8cc0f566 HZ |
1387 | ret = amdgpu_ras_query_error_status_helper(adev, info, |
1388 | &err_data, | |
9dc57c2a | 1389 | &qctx, |
8cc0f566 HZ |
1390 | error_query_mode); |
1391 | if (ret) | |
1392 | goto out_fini_err_data; | |
05a58345 | 1393 | |
ec3e0a91 | 1394 | amdgpu_rasmgr_error_data_statistic_update(obj, &err_data); |
05a58345 | 1395 | |
c030f2e4 | 1396 | info->ue_count = obj->err_data.ue_count; |
1397 | info->ce_count = obj->err_data.ce_count; | |
46e2231c | 1398 | info->de_count = obj->err_data.de_count; |
c030f2e4 | 1399 | |
9dc57c2a | 1400 | amdgpu_ras_error_generate_report(adev, info, &err_data, &qctx); |
05a58345 | 1401 | |
5b1270be YW |
1402 | out_fini_err_data: |
1403 | amdgpu_ras_error_data_fini(&err_data); | |
1404 | ||
1405 | return ret; | |
c030f2e4 | 1406 | } |
1407 | ||
472c5fb2 | 1408 | int amdgpu_ras_reset_error_count(struct amdgpu_device *adev, |
761d86d3 DL |
1409 | enum amdgpu_ras_block block) |
1410 | { | |
b6efdb02 | 1411 | struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, block, 0); |
73582be1 TZ |
1412 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
1413 | const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; | |
04c4fcd2 | 1414 | const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; |
18eae367 TZ |
1415 | struct amdgpu_hive_info *hive; |
1416 | int hive_ras_recovery = 0; | |
8b0fb0e9 | 1417 | |
a83f2bf1 | 1418 | if (!block_obj || !block_obj->hw_ops) { |
afa37315 | 1419 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", |
472c5fb2 TZ |
1420 | ras_block_str(block)); |
1421 | return -EOPNOTSUPP; | |
761d86d3 DL |
1422 | } |
1423 | ||
d1d4c0b7 | 1424 | if (!amdgpu_ras_is_supported(adev, block) || |
04c4fcd2 | 1425 | !amdgpu_ras_get_aca_debug_mode(adev)) |
d1d4c0b7 TZ |
1426 | return -EOPNOTSUPP; |
1427 | ||
18eae367 TZ |
1428 | hive = amdgpu_get_xgmi_hive(adev); |
1429 | if (hive) { | |
1430 | hive_ras_recovery = atomic_read(&hive->ras_recovery); | |
1431 | amdgpu_put_xgmi_hive(hive); | |
1432 | } | |
1433 | ||
73582be1 | 1434 | /* skip ras error reset in gpu reset */ |
18eae367 TZ |
1435 | if ((amdgpu_in_reset(adev) || atomic_read(&ras->in_recovery) || |
1436 | hive_ras_recovery) && | |
04c4fcd2 YW |
1437 | ((smu_funcs && smu_funcs->set_debug_mode) || |
1438 | (mca_funcs && mca_funcs->mca_set_debug_mode))) | |
73582be1 TZ |
1439 | return -EOPNOTSUPP; |
1440 | ||
7389a5b8 | 1441 | if (block_obj->hw_ops->reset_ras_error_count) |
1442 | block_obj->hw_ops->reset_ras_error_count(adev); | |
5c23e9e0 | 1443 | |
472c5fb2 TZ |
1444 | return 0; |
1445 | } | |
1446 | ||
1447 | int amdgpu_ras_reset_error_status(struct amdgpu_device *adev, | |
1448 | enum amdgpu_ras_block block) | |
1449 | { | |
1450 | struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, block, 0); | |
1451 | ||
1452 | if (amdgpu_ras_reset_error_count(adev, block) == -EOPNOTSUPP) | |
1453 | return 0; | |
1454 | ||
7389a5b8 | 1455 | if ((block == AMDGPU_RAS_BLOCK__GFX) || |
1456 | (block == AMDGPU_RAS_BLOCK__MMHUB)) { | |
8b0fb0e9 | 1457 | if (block_obj->hw_ops->reset_ras_error_status) |
1458 | block_obj->hw_ops->reset_ras_error_status(adev); | |
761d86d3 | 1459 | } |
5c23e9e0 | 1460 | |
761d86d3 | 1461 | return 0; |
5c23e9e0 JC |
1462 | } |
1463 | ||
c030f2e4 | 1464 | /* wrapper of psp_ras_trigger_error */ |
1465 | int amdgpu_ras_error_inject(struct amdgpu_device *adev, | |
1466 | struct ras_inject_if *info) | |
1467 | { | |
1468 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); | |
1469 | struct ta_ras_trigger_error_input block_info = { | |
828cfa29 | 1470 | .block_id = amdgpu_ras_block_to_ta(info->head.block), |
1471 | .inject_error_type = amdgpu_ras_error_to_ta(info->head.type), | |
c030f2e4 | 1472 | .sub_block_index = info->head.sub_block_index, |
1473 | .address = info->address, | |
1474 | .value = info->value, | |
1475 | }; | |
ab3b9de6 YL |
1476 | int ret = -EINVAL; |
1477 | struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, | |
1478 | info->head.block, | |
1479 | info->head.sub_block_index); | |
c030f2e4 | 1480 | |
248c9635 TZ |
1481 | /* inject on guest isn't allowed, return success directly */ |
1482 | if (amdgpu_sriov_vf(adev)) | |
1483 | return 0; | |
1484 | ||
c030f2e4 | 1485 | if (!obj) |
1486 | return -EINVAL; | |
1487 | ||
22d4ba53 | 1488 | if (!block_obj || !block_obj->hw_ops) { |
afa37315 LT |
1489 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", |
1490 | get_ras_block_str(&info->head)); | |
22d4ba53 | 1491 | return -EINVAL; |
1492 | } | |
1493 | ||
a6c44d25 | 1494 | /* Calculate XGMI relative offset */ |
a80fe1a6 TZ |
1495 | if (adev->gmc.xgmi.num_physical_nodes > 1 && |
1496 | info->head.block != AMDGPU_RAS_BLOCK__GFX) { | |
19744f5f HZ |
1497 | block_info.address = |
1498 | amdgpu_xgmi_get_relative_phy_addr(adev, | |
1499 | block_info.address); | |
a6c44d25 JC |
1500 | } |
1501 | ||
27c5f295 TZ |
1502 | if (block_obj->hw_ops->ras_error_inject) { |
1503 | if (info->head.block == AMDGPU_RAS_BLOCK__GFX) | |
2c22ed0b | 1504 | ret = block_obj->hw_ops->ras_error_inject(adev, info, info->instance_mask); |
27c5f295 | 1505 | else /* Special ras_error_inject is defined (e.g: xgmi) */ |
2c22ed0b TZ |
1506 | ret = block_obj->hw_ops->ras_error_inject(adev, &block_info, |
1507 | info->instance_mask); | |
27c5f295 TZ |
1508 | } else { |
1509 | /* default path */ | |
1510 | ret = psp_ras_trigger_error(&adev->psp, &block_info, info->instance_mask); | |
a5dd40ca HZ |
1511 | } |
1512 | ||
011907fd DL |
1513 | if (ret) |
1514 | dev_err(adev->dev, "ras inject %s failed %d\n", | |
640ae42e | 1515 | get_ras_block_str(&info->head), ret); |
c030f2e4 | 1516 | |
1517 | return ret; | |
1518 | } | |
1519 | ||
4d9f771e | 1520 | /** |
4a1c9a44 HZ |
1521 | * amdgpu_ras_query_error_count_helper -- Get error counter for specific IP |
1522 | * @adev: pointer to AMD GPU device | |
1523 | * @ce_count: pointer to an integer to be set to the count of correctible errors. | |
1524 | * @ue_count: pointer to an integer to be set to the count of uncorrectible errors. | |
1525 | * @query_info: pointer to ras_query_if | |
1526 | * | |
1527 | * Return 0 for query success or do nothing, otherwise return an error | |
1528 | * on failures | |
1529 | */ | |
1530 | static int amdgpu_ras_query_error_count_helper(struct amdgpu_device *adev, | |
1531 | unsigned long *ce_count, | |
1532 | unsigned long *ue_count, | |
1533 | struct ras_query_if *query_info) | |
1534 | { | |
1535 | int ret; | |
1536 | ||
1537 | if (!query_info) | |
1538 | /* do nothing if query_info is not specified */ | |
1539 | return 0; | |
1540 | ||
1541 | ret = amdgpu_ras_query_error_status(adev, query_info); | |
1542 | if (ret) | |
1543 | return ret; | |
1544 | ||
1545 | *ce_count += query_info->ce_count; | |
1546 | *ue_count += query_info->ue_count; | |
1547 | ||
1548 | /* some hardware/IP supports read to clear | |
1549 | * no need to explictly reset the err status after the query call */ | |
4e8303cf LL |
1550 | if (amdgpu_ip_version(adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 2) && |
1551 | amdgpu_ip_version(adev, MP0_HWIP, 0) != IP_VERSION(11, 0, 4)) { | |
4a1c9a44 HZ |
1552 | if (amdgpu_ras_reset_error_status(adev, query_info->head.block)) |
1553 | dev_warn(adev->dev, | |
1554 | "Failed to reset error counter and error status\n"); | |
1555 | } | |
1556 | ||
1557 | return 0; | |
1558 | } | |
1559 | ||
1560 | /** | |
1561 | * amdgpu_ras_query_error_count -- Get error counts of all IPs or specific IP | |
bbe04dec IB |
1562 | * @adev: pointer to AMD GPU device |
1563 | * @ce_count: pointer to an integer to be set to the count of correctible errors. | |
1564 | * @ue_count: pointer to an integer to be set to the count of uncorrectible | |
4d9f771e | 1565 | * errors. |
4a1c9a44 HZ |
1566 | * @query_info: pointer to ras_query_if if the query request is only for |
1567 | * specific ip block; if info is NULL, then the qurey request is for | |
1568 | * all the ip blocks that support query ras error counters/status | |
4d9f771e LT |
1569 | * |
1570 | * If set, @ce_count or @ue_count, count and return the corresponding | |
1571 | * error counts in those integer pointers. Return 0 if the device | |
1572 | * supports RAS. Return -EOPNOTSUPP if the device doesn't support RAS. | |
1573 | */ | |
1574 | int amdgpu_ras_query_error_count(struct amdgpu_device *adev, | |
1575 | unsigned long *ce_count, | |
4a1c9a44 HZ |
1576 | unsigned long *ue_count, |
1577 | struct ras_query_if *query_info) | |
c030f2e4 | 1578 | { |
1579 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
1580 | struct ras_manager *obj; | |
a46751fb | 1581 | unsigned long ce, ue; |
4a1c9a44 | 1582 | int ret; |
c030f2e4 | 1583 | |
8ab0d6f0 | 1584 | if (!adev->ras_enabled || !con) |
4d9f771e LT |
1585 | return -EOPNOTSUPP; |
1586 | ||
1587 | /* Don't count since no reporting. | |
1588 | */ | |
1589 | if (!ce_count && !ue_count) | |
1590 | return 0; | |
c030f2e4 | 1591 | |
a46751fb LT |
1592 | ce = 0; |
1593 | ue = 0; | |
4a1c9a44 HZ |
1594 | if (!query_info) { |
1595 | /* query all the ip blocks that support ras query interface */ | |
1596 | list_for_each_entry(obj, &con->head, node) { | |
1597 | struct ras_query_if info = { | |
1598 | .head = obj->head, | |
1599 | }; | |
c030f2e4 | 1600 | |
4a1c9a44 | 1601 | ret = amdgpu_ras_query_error_count_helper(adev, &ce, &ue, &info); |
2a460963 | 1602 | } |
4a1c9a44 HZ |
1603 | } else { |
1604 | /* query specific ip block */ | |
1605 | ret = amdgpu_ras_query_error_count_helper(adev, &ce, &ue, query_info); | |
c030f2e4 | 1606 | } |
1607 | ||
4a1c9a44 HZ |
1608 | if (ret) |
1609 | return ret; | |
1610 | ||
a46751fb LT |
1611 | if (ce_count) |
1612 | *ce_count = ce; | |
1613 | ||
1614 | if (ue_count) | |
1615 | *ue_count = ue; | |
4d9f771e LT |
1616 | |
1617 | return 0; | |
c030f2e4 | 1618 | } |
1619 | /* query/inject/cure end */ | |
1620 | ||
1621 | ||
1622 | /* sysfs begin */ | |
1623 | ||
466b1793 | 1624 | static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, |
1625 | struct ras_badpage **bps, unsigned int *count); | |
1626 | ||
1627 | static char *amdgpu_ras_badpage_flags_str(unsigned int flags) | |
1628 | { | |
1629 | switch (flags) { | |
52dd95f2 | 1630 | case AMDGPU_RAS_RETIRE_PAGE_RESERVED: |
466b1793 | 1631 | return "R"; |
52dd95f2 | 1632 | case AMDGPU_RAS_RETIRE_PAGE_PENDING: |
466b1793 | 1633 | return "P"; |
52dd95f2 | 1634 | case AMDGPU_RAS_RETIRE_PAGE_FAULT: |
466b1793 | 1635 | default: |
1636 | return "F"; | |
aec576f9 | 1637 | } |
466b1793 | 1638 | } |
1639 | ||
f77c7109 AD |
1640 | /** |
1641 | * DOC: AMDGPU RAS sysfs gpu_vram_bad_pages Interface | |
466b1793 | 1642 | * |
1643 | * It allows user to read the bad pages of vram on the gpu through | |
1644 | * /sys/class/drm/card[0/1/2...]/device/ras/gpu_vram_bad_pages | |
1645 | * | |
1646 | * It outputs multiple lines, and each line stands for one gpu page. | |
1647 | * | |
1648 | * The format of one line is below, | |
1649 | * gpu pfn : gpu page size : flags | |
1650 | * | |
1651 | * gpu pfn and gpu page size are printed in hex format. | |
1652 | * flags can be one of below character, | |
f77c7109 | 1653 | * |
466b1793 | 1654 | * R: reserved, this gpu page is reserved and not able to use. |
f77c7109 | 1655 | * |
466b1793 | 1656 | * P: pending for reserve, this gpu page is marked as bad, will be reserved |
f77c7109 AD |
1657 | * in next window of page_reserve. |
1658 | * | |
466b1793 | 1659 | * F: unable to reserve. this gpu page can't be reserved due to some reasons. |
1660 | * | |
f77c7109 AD |
1661 | * Examples: |
1662 | * | |
1663 | * .. code-block:: bash | |
1664 | * | |
1665 | * 0x00000001 : 0x00001000 : R | |
1666 | * 0x00000002 : 0x00001000 : P | |
1667 | * | |
466b1793 | 1668 | */ |
1669 | ||
1670 | static ssize_t amdgpu_ras_sysfs_badpages_read(struct file *f, | |
1671 | struct kobject *kobj, struct bin_attribute *attr, | |
1672 | char *buf, loff_t ppos, size_t count) | |
1673 | { | |
1674 | struct amdgpu_ras *con = | |
1675 | container_of(attr, struct amdgpu_ras, badpages_attr); | |
1676 | struct amdgpu_device *adev = con->adev; | |
1677 | const unsigned int element_size = | |
1678 | sizeof("0xabcdabcd : 0x12345678 : R\n") - 1; | |
d6ee400e SA |
1679 | unsigned int start = div64_ul(ppos + element_size - 1, element_size); |
1680 | unsigned int end = div64_ul(ppos + count - 1, element_size); | |
466b1793 | 1681 | ssize_t s = 0; |
1682 | struct ras_badpage *bps = NULL; | |
1683 | unsigned int bps_count = 0; | |
1684 | ||
1685 | memset(buf, 0, count); | |
1686 | ||
1687 | if (amdgpu_ras_badpages_read(adev, &bps, &bps_count)) | |
1688 | return 0; | |
1689 | ||
1690 | for (; start < end && start < bps_count; start++) | |
1691 | s += scnprintf(&buf[s], element_size + 1, | |
1692 | "0x%08x : 0x%08x : %1s\n", | |
1693 | bps[start].bp, | |
1694 | bps[start].size, | |
1695 | amdgpu_ras_badpage_flags_str(bps[start].flags)); | |
1696 | ||
1697 | kfree(bps); | |
1698 | ||
1699 | return s; | |
1700 | } | |
1701 | ||
c030f2e4 | 1702 | static ssize_t amdgpu_ras_sysfs_features_read(struct device *dev, |
1703 | struct device_attribute *attr, char *buf) | |
1704 | { | |
1705 | struct amdgpu_ras *con = | |
1706 | container_of(attr, struct amdgpu_ras, features_attr); | |
c030f2e4 | 1707 | |
2cffcb66 | 1708 | return sysfs_emit(buf, "feature mask: 0x%x\n", con->features); |
c030f2e4 | 1709 | } |
1710 | ||
625e5f38 AK |
1711 | static ssize_t amdgpu_ras_sysfs_version_show(struct device *dev, |
1712 | struct device_attribute *attr, char *buf) | |
1713 | { | |
1714 | struct amdgpu_ras *con = | |
1715 | container_of(attr, struct amdgpu_ras, version_attr); | |
1716 | return sysfs_emit(buf, "table version: 0x%x\n", con->eeprom_control.tbl_hdr.version); | |
1717 | } | |
1718 | ||
1719 | static ssize_t amdgpu_ras_sysfs_schema_show(struct device *dev, | |
1720 | struct device_attribute *attr, char *buf) | |
1721 | { | |
1722 | struct amdgpu_ras *con = | |
1723 | container_of(attr, struct amdgpu_ras, schema_attr); | |
1724 | return sysfs_emit(buf, "schema: 0x%x\n", con->schema); | |
1725 | } | |
1726 | ||
f848159b GC |
1727 | static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device *adev) |
1728 | { | |
1729 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
1730 | ||
4638e0c2 VP |
1731 | if (adev->dev->kobj.sd) |
1732 | sysfs_remove_file_from_group(&adev->dev->kobj, | |
f848159b GC |
1733 | &con->badpages_attr.attr, |
1734 | RAS_FS_NAME); | |
1735 | } | |
1736 | ||
625e5f38 | 1737 | static int amdgpu_ras_sysfs_remove_dev_attr_node(struct amdgpu_device *adev) |
c030f2e4 | 1738 | { |
1739 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
1740 | struct attribute *attrs[] = { | |
1741 | &con->features_attr.attr, | |
625e5f38 AK |
1742 | &con->version_attr.attr, |
1743 | &con->schema_attr.attr, | |
c030f2e4 | 1744 | NULL |
1745 | }; | |
1746 | struct attribute_group group = { | |
eb0c3cd4 | 1747 | .name = RAS_FS_NAME, |
c030f2e4 | 1748 | .attrs = attrs, |
1749 | }; | |
1750 | ||
4638e0c2 VP |
1751 | if (adev->dev->kobj.sd) |
1752 | sysfs_remove_group(&adev->dev->kobj, &group); | |
c030f2e4 | 1753 | |
1754 | return 0; | |
1755 | } | |
1756 | ||
1757 | int amdgpu_ras_sysfs_create(struct amdgpu_device *adev, | |
9252d33d | 1758 | struct ras_common_if *head) |
c030f2e4 | 1759 | { |
9252d33d | 1760 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); |
c030f2e4 | 1761 | |
1762 | if (!obj || obj->attr_inuse) | |
1763 | return -EINVAL; | |
1764 | ||
1765 | get_obj(obj); | |
1766 | ||
9252d33d | 1767 | snprintf(obj->fs_data.sysfs_name, sizeof(obj->fs_data.sysfs_name), |
1768 | "%s_err_count", head->name); | |
c030f2e4 | 1769 | |
1770 | obj->sysfs_attr = (struct device_attribute){ | |
1771 | .attr = { | |
1772 | .name = obj->fs_data.sysfs_name, | |
1773 | .mode = S_IRUGO, | |
1774 | }, | |
1775 | .show = amdgpu_ras_sysfs_read, | |
1776 | }; | |
163def43 | 1777 | sysfs_attr_init(&obj->sysfs_attr.attr); |
c030f2e4 | 1778 | |
1779 | if (sysfs_add_file_to_group(&adev->dev->kobj, | |
1780 | &obj->sysfs_attr.attr, | |
eb0c3cd4 | 1781 | RAS_FS_NAME)) { |
c030f2e4 | 1782 | put_obj(obj); |
1783 | return -EINVAL; | |
1784 | } | |
1785 | ||
1786 | obj->attr_inuse = 1; | |
1787 | ||
1788 | return 0; | |
1789 | } | |
1790 | ||
1791 | int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev, | |
1792 | struct ras_common_if *head) | |
1793 | { | |
1794 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |
1795 | ||
1796 | if (!obj || !obj->attr_inuse) | |
1797 | return -EINVAL; | |
1798 | ||
4638e0c2 VP |
1799 | if (adev->dev->kobj.sd) |
1800 | sysfs_remove_file_from_group(&adev->dev->kobj, | |
c030f2e4 | 1801 | &obj->sysfs_attr.attr, |
eb0c3cd4 | 1802 | RAS_FS_NAME); |
c030f2e4 | 1803 | obj->attr_inuse = 0; |
1804 | put_obj(obj); | |
1805 | ||
1806 | return 0; | |
1807 | } | |
1808 | ||
1809 | static int amdgpu_ras_sysfs_remove_all(struct amdgpu_device *adev) | |
1810 | { | |
1811 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
1812 | struct ras_manager *obj, *tmp; | |
1813 | ||
1814 | list_for_each_entry_safe(obj, tmp, &con->head, node) { | |
1815 | amdgpu_ras_sysfs_remove(adev, &obj->head); | |
1816 | } | |
1817 | ||
f848159b GC |
1818 | if (amdgpu_bad_page_threshold != 0) |
1819 | amdgpu_ras_sysfs_remove_bad_page_node(adev); | |
1820 | ||
625e5f38 | 1821 | amdgpu_ras_sysfs_remove_dev_attr_node(adev); |
c030f2e4 | 1822 | |
1823 | return 0; | |
1824 | } | |
1825 | /* sysfs end */ | |
1826 | ||
ef177d11 AD |
1827 | /** |
1828 | * DOC: AMDGPU RAS Reboot Behavior for Unrecoverable Errors | |
1829 | * | |
1830 | * Normally when there is an uncorrectable error, the driver will reset | |
1831 | * the GPU to recover. However, in the event of an unrecoverable error, | |
1832 | * the driver provides an interface to reboot the system automatically | |
1833 | * in that event. | |
1834 | * | |
1835 | * The following file in debugfs provides that interface: | |
1836 | * /sys/kernel/debug/dri/[0/1/2...]/ras/auto_reboot | |
1837 | * | |
1838 | * Usage: | |
1839 | * | |
1840 | * .. code-block:: bash | |
1841 | * | |
1842 | * echo true > .../ras/auto_reboot | |
1843 | * | |
1844 | */ | |
c030f2e4 | 1845 | /* debugfs begin */ |
ea1b8c9b | 1846 | static struct dentry *amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev) |
36ea1bd2 | 1847 | { |
1848 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
740f42a2 | 1849 | struct amdgpu_ras_eeprom_control *eeprom = &con->eeprom_control; |
ef0d7d20 LT |
1850 | struct drm_minor *minor = adev_to_drm(adev)->primary; |
1851 | struct dentry *dir; | |
36ea1bd2 | 1852 | |
88293c03 ND |
1853 | dir = debugfs_create_dir(RAS_FS_NAME, minor->debugfs_root); |
1854 | debugfs_create_file("ras_ctrl", S_IWUGO | S_IRUGO, dir, adev, | |
1855 | &amdgpu_ras_debugfs_ctrl_ops); | |
1856 | debugfs_create_file("ras_eeprom_reset", S_IWUGO | S_IRUGO, dir, adev, | |
1857 | &amdgpu_ras_debugfs_eeprom_ops); | |
7fb64071 LT |
1858 | debugfs_create_u32("bad_page_cnt_threshold", 0444, dir, |
1859 | &con->bad_page_cnt_threshold); | |
740f42a2 | 1860 | debugfs_create_u32("ras_num_recs", 0444, dir, &eeprom->ras_num_recs); |
ef0d7d20 LT |
1861 | debugfs_create_x32("ras_hw_enabled", 0444, dir, &adev->ras_hw_enabled); |
1862 | debugfs_create_x32("ras_enabled", 0444, dir, &adev->ras_enabled); | |
c65b0805 LT |
1863 | debugfs_create_file("ras_eeprom_size", S_IRUGO, dir, adev, |
1864 | &amdgpu_ras_debugfs_eeprom_size_ops); | |
1865 | con->de_ras_eeprom_table = debugfs_create_file("ras_eeprom_table", | |
1866 | S_IRUGO, dir, adev, | |
1867 | &amdgpu_ras_debugfs_eeprom_table_ops); | |
1868 | amdgpu_ras_debugfs_set_ret_size(&con->eeprom_control); | |
c688a06b GC |
1869 | |
1870 | /* | |
1871 | * After one uncorrectable error happens, usually GPU recovery will | |
1872 | * be scheduled. But due to the known problem in GPU recovery failing | |
1873 | * to bring GPU back, below interface provides one direct way to | |
1874 | * user to reboot system automatically in such case within | |
1875 | * ERREVENT_ATHUB_INTERRUPT generated. Normal GPU recovery routine | |
1876 | * will never be called. | |
1877 | */ | |
88293c03 | 1878 | debugfs_create_bool("auto_reboot", S_IWUGO | S_IRUGO, dir, &con->reboot); |
66459e1d GC |
1879 | |
1880 | /* | |
1881 | * User could set this not to clean up hardware's error count register | |
1882 | * of RAS IPs during ras recovery. | |
1883 | */ | |
88293c03 ND |
1884 | debugfs_create_bool("disable_ras_err_cnt_harvest", 0644, dir, |
1885 | &con->disable_ras_err_cnt_harvest); | |
1886 | return dir; | |
36ea1bd2 | 1887 | } |
1888 | ||
cedf7884 | 1889 | static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev, |
88293c03 ND |
1890 | struct ras_fs_if *head, |
1891 | struct dentry *dir) | |
c030f2e4 | 1892 | { |
c030f2e4 | 1893 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head->head); |
c030f2e4 | 1894 | |
88293c03 | 1895 | if (!obj || !dir) |
450f30ea | 1896 | return; |
c030f2e4 | 1897 | |
1898 | get_obj(obj); | |
1899 | ||
1900 | memcpy(obj->fs_data.debugfs_name, | |
1901 | head->debugfs_name, | |
1902 | sizeof(obj->fs_data.debugfs_name)); | |
1903 | ||
88293c03 ND |
1904 | debugfs_create_file(obj->fs_data.debugfs_name, S_IWUGO | S_IRUGO, dir, |
1905 | obj, &amdgpu_ras_debugfs_ops); | |
c030f2e4 | 1906 | } |
1907 | ||
f9317014 TZ |
1908 | void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) |
1909 | { | |
1910 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
88293c03 | 1911 | struct dentry *dir; |
c1509f3f | 1912 | struct ras_manager *obj; |
f9317014 TZ |
1913 | struct ras_fs_if fs_info; |
1914 | ||
1915 | /* | |
1916 | * it won't be called in resume path, no need to check | |
1917 | * suspend and gpu reset status | |
1918 | */ | |
cedf7884 | 1919 | if (!IS_ENABLED(CONFIG_DEBUG_FS) || !con) |
f9317014 TZ |
1920 | return; |
1921 | ||
88293c03 | 1922 | dir = amdgpu_ras_debugfs_create_ctrl_node(adev); |
f9317014 | 1923 | |
c1509f3f | 1924 | list_for_each_entry(obj, &con->head, node) { |
f9317014 TZ |
1925 | if (amdgpu_ras_is_supported(adev, obj->head.block) && |
1926 | (obj->attr_inuse == 1)) { | |
1927 | sprintf(fs_info.debugfs_name, "%s_err_inject", | |
640ae42e | 1928 | get_ras_block_str(&obj->head)); |
f9317014 | 1929 | fs_info.head = obj->head; |
88293c03 | 1930 | amdgpu_ras_debugfs_create(adev, &fs_info, dir); |
f9317014 TZ |
1931 | } |
1932 | } | |
4051844c | 1933 | |
04c4fcd2 YW |
1934 | if (amdgpu_aca_is_enabled(adev)) |
1935 | amdgpu_aca_smu_debugfs_init(adev, dir); | |
1936 | else | |
1937 | amdgpu_mca_smu_debugfs_init(adev, dir); | |
f9317014 TZ |
1938 | } |
1939 | ||
c030f2e4 | 1940 | /* debugfs end */ |
1941 | ||
1942 | /* ras fs */ | |
c3d4d45d GC |
1943 | static BIN_ATTR(gpu_vram_bad_pages, S_IRUGO, |
1944 | amdgpu_ras_sysfs_badpages_read, NULL, 0); | |
1945 | static DEVICE_ATTR(features, S_IRUGO, | |
1946 | amdgpu_ras_sysfs_features_read, NULL); | |
625e5f38 AK |
1947 | static DEVICE_ATTR(version, 0444, |
1948 | amdgpu_ras_sysfs_version_show, NULL); | |
1949 | static DEVICE_ATTR(schema, 0444, | |
1950 | amdgpu_ras_sysfs_schema_show, NULL); | |
c030f2e4 | 1951 | static int amdgpu_ras_fs_init(struct amdgpu_device *adev) |
1952 | { | |
c3d4d45d GC |
1953 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
1954 | struct attribute_group group = { | |
1955 | .name = RAS_FS_NAME, | |
1956 | }; | |
1957 | struct attribute *attrs[] = { | |
1958 | &con->features_attr.attr, | |
625e5f38 AK |
1959 | &con->version_attr.attr, |
1960 | &con->schema_attr.attr, | |
c3d4d45d GC |
1961 | NULL |
1962 | }; | |
1963 | struct bin_attribute *bin_attrs[] = { | |
1964 | NULL, | |
1965 | NULL, | |
1966 | }; | |
a069a9eb | 1967 | int r; |
c030f2e4 | 1968 | |
625e5f38 AK |
1969 | group.attrs = attrs; |
1970 | ||
c3d4d45d GC |
1971 | /* add features entry */ |
1972 | con->features_attr = dev_attr_features; | |
c3d4d45d GC |
1973 | sysfs_attr_init(attrs[0]); |
1974 | ||
625e5f38 AK |
1975 | /* add version entry */ |
1976 | con->version_attr = dev_attr_version; | |
1977 | sysfs_attr_init(attrs[1]); | |
1978 | ||
1979 | /* add schema entry */ | |
1980 | con->schema_attr = dev_attr_schema; | |
1981 | sysfs_attr_init(attrs[2]); | |
1982 | ||
c3d4d45d GC |
1983 | if (amdgpu_bad_page_threshold != 0) { |
1984 | /* add bad_page_features entry */ | |
1985 | bin_attr_gpu_vram_bad_pages.private = NULL; | |
1986 | con->badpages_attr = bin_attr_gpu_vram_bad_pages; | |
1987 | bin_attrs[0] = &con->badpages_attr; | |
1988 | group.bin_attrs = bin_attrs; | |
1989 | sysfs_bin_attr_init(bin_attrs[0]); | |
1990 | } | |
1991 | ||
a069a9eb AD |
1992 | r = sysfs_create_group(&adev->dev->kobj, &group); |
1993 | if (r) | |
1994 | dev_err(adev->dev, "Failed to create RAS sysfs group!"); | |
f848159b | 1995 | |
c030f2e4 | 1996 | return 0; |
1997 | } | |
1998 | ||
1999 | static int amdgpu_ras_fs_fini(struct amdgpu_device *adev) | |
2000 | { | |
88293c03 ND |
2001 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
2002 | struct ras_manager *con_obj, *ip_obj, *tmp; | |
2003 | ||
2004 | if (IS_ENABLED(CONFIG_DEBUG_FS)) { | |
2005 | list_for_each_entry_safe(con_obj, tmp, &con->head, node) { | |
2006 | ip_obj = amdgpu_ras_find_obj(adev, &con_obj->head); | |
2007 | if (ip_obj) | |
2008 | put_obj(ip_obj); | |
2009 | } | |
2010 | } | |
2011 | ||
c030f2e4 | 2012 | amdgpu_ras_sysfs_remove_all(adev); |
2013 | return 0; | |
2014 | } | |
2015 | /* ras fs end */ | |
2016 | ||
2017 | /* ih begin */ | |
b3c76814 TZ |
2018 | |
2019 | /* For the hardware that cannot enable bif ring for both ras_controller_irq | |
2020 | * and ras_err_evnet_athub_irq ih cookies, the driver has to poll status | |
2021 | * register to check whether the interrupt is triggered or not, and properly | |
2022 | * ack the interrupt if it is there | |
2023 | */ | |
2024 | void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) | |
2025 | { | |
950d6425 | 2026 | /* Fatal error events are handled on host side */ |
8eba7205 | 2027 | if (amdgpu_sriov_vf(adev)) |
b3c76814 TZ |
2028 | return; |
2029 | ||
2030 | if (adev->nbio.ras && | |
2031 | adev->nbio.ras->handle_ras_controller_intr_no_bifring) | |
2032 | adev->nbio.ras->handle_ras_controller_intr_no_bifring(adev); | |
2033 | ||
2034 | if (adev->nbio.ras && | |
2035 | adev->nbio.ras->handle_ras_err_event_athub_intr_no_bifring) | |
2036 | adev->nbio.ras->handle_ras_err_event_athub_intr_no_bifring(adev); | |
2037 | } | |
2038 | ||
66f87949 TZ |
2039 | static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *obj, |
2040 | struct amdgpu_iv_entry *entry) | |
2041 | { | |
b63ac5d3 | 2042 | bool poison_stat = false; |
66f87949 | 2043 | struct amdgpu_device *adev = obj->adev; |
66f87949 TZ |
2044 | struct amdgpu_ras_block_object *block_obj = |
2045 | amdgpu_ras_get_ras_block(adev, obj->head.block, 0); | |
2046 | ||
ac7b25d9 | 2047 | if (!block_obj) |
b63ac5d3 | 2048 | return; |
66f87949 | 2049 | |
b63ac5d3 TZ |
2050 | /* both query_poison_status and handle_poison_consumption are optional, |
2051 | * but at least one of them should be implemented if we need poison | |
2052 | * consumption handler | |
2053 | */ | |
ac7b25d9 | 2054 | if (block_obj->hw_ops && block_obj->hw_ops->query_poison_status) { |
b63ac5d3 TZ |
2055 | poison_stat = block_obj->hw_ops->query_poison_status(adev); |
2056 | if (!poison_stat) { | |
2057 | /* Not poison consumption interrupt, no need to handle it */ | |
2058 | dev_info(adev->dev, "No RAS poison status in %s poison IH.\n", | |
2059 | block_obj->ras_comm.name); | |
2060 | ||
2061 | return; | |
66f87949 TZ |
2062 | } |
2063 | } | |
2064 | ||
2fc46e0b | 2065 | amdgpu_umc_poison_handler(adev, obj->head.block, 0); |
b63ac5d3 | 2066 | |
ac7b25d9 | 2067 | if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption) |
b63ac5d3 TZ |
2068 | poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); |
2069 | ||
2070 | /* gpu reset is fallback for failed and default cases */ | |
2071 | if (poison_stat) { | |
2072 | dev_info(adev->dev, "GPU reset for %s RAS poison consumption is issued!\n", | |
2073 | block_obj->ras_comm.name); | |
66f87949 | 2074 | amdgpu_ras_reset_gpu(adev); |
ac7b25d9 YC |
2075 | } else { |
2076 | amdgpu_gfx_poison_consumption_handler(adev, entry); | |
b63ac5d3 | 2077 | } |
66f87949 TZ |
2078 | } |
2079 | ||
50a7d025 TZ |
2080 | static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj, |
2081 | struct amdgpu_iv_entry *entry) | |
2082 | { | |
2083 | dev_info(obj->adev->dev, | |
90bd0147 | 2084 | "Poison is created\n"); |
a734adfb YC |
2085 | |
2086 | if (amdgpu_ip_version(obj->adev, UMC_HWIP, 0) >= IP_VERSION(12, 0, 0)) { | |
2087 | struct amdgpu_ras *con = amdgpu_ras_get_context(obj->adev); | |
2088 | ||
2089 | amdgpu_ras_put_poison_req(obj->adev, | |
2090 | AMDGPU_RAS_BLOCK__UMC, 0, NULL, NULL, false); | |
2091 | ||
2092 | atomic_inc(&con->page_retirement_req_cnt); | |
2093 | ||
2094 | wake_up(&con->page_retirement_wq); | |
2095 | } | |
50a7d025 TZ |
2096 | } |
2097 | ||
2098 | static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, | |
2099 | struct amdgpu_iv_entry *entry) | |
2100 | { | |
2101 | struct ras_ih_data *data = &obj->ih_data; | |
5b1270be | 2102 | struct ras_err_data err_data; |
50a7d025 TZ |
2103 | int ret; |
2104 | ||
2105 | if (!data->cb) | |
2106 | return; | |
2107 | ||
5b1270be YW |
2108 | ret = amdgpu_ras_error_data_init(&err_data); |
2109 | if (ret) | |
2110 | return; | |
2111 | ||
50a7d025 TZ |
2112 | /* Let IP handle its data, maybe we need get the output |
2113 | * from the callback to update the error type/count, etc | |
2114 | */ | |
2115 | ret = data->cb(obj->adev, &err_data, entry); | |
2116 | /* ue will trigger an interrupt, and in that case | |
2117 | * we need do a reset to recovery the whole system. | |
2118 | * But leave IP do that recovery, here we just dispatch | |
2119 | * the error. | |
2120 | */ | |
2121 | if (ret == AMDGPU_RAS_SUCCESS) { | |
2122 | /* these counts could be left as 0 if | |
2123 | * some blocks do not count error number | |
2124 | */ | |
2125 | obj->err_data.ue_count += err_data.ue_count; | |
2126 | obj->err_data.ce_count += err_data.ce_count; | |
46e2231c | 2127 | obj->err_data.de_count += err_data.de_count; |
50a7d025 | 2128 | } |
5b1270be YW |
2129 | |
2130 | amdgpu_ras_error_data_fini(&err_data); | |
50a7d025 TZ |
2131 | } |
2132 | ||
c030f2e4 | 2133 | static void amdgpu_ras_interrupt_handler(struct ras_manager *obj) |
2134 | { | |
2135 | struct ras_ih_data *data = &obj->ih_data; | |
2136 | struct amdgpu_iv_entry entry; | |
c030f2e4 | 2137 | |
2138 | while (data->rptr != data->wptr) { | |
2139 | rmb(); | |
2140 | memcpy(&entry, &data->ring[data->rptr], | |
2141 | data->element_size); | |
2142 | ||
2143 | wmb(); | |
2144 | data->rptr = (data->aligned_element_size + | |
2145 | data->rptr) % data->ring_size; | |
2146 | ||
50a7d025 TZ |
2147 | if (amdgpu_ras_is_poison_mode_supported(obj->adev)) { |
2148 | if (obj->head.block == AMDGPU_RAS_BLOCK__UMC) | |
2149 | amdgpu_ras_interrupt_poison_creation_handler(obj, &entry); | |
66f87949 TZ |
2150 | else |
2151 | amdgpu_ras_interrupt_poison_consumption_handler(obj, &entry); | |
50a7d025 TZ |
2152 | } else { |
2153 | if (obj->head.block == AMDGPU_RAS_BLOCK__UMC) | |
2154 | amdgpu_ras_interrupt_umc_handler(obj, &entry); | |
2155 | else | |
2156 | dev_warn(obj->adev->dev, | |
2157 | "No RAS interrupt handler for non-UMC block with poison disabled.\n"); | |
c030f2e4 | 2158 | } |
2159 | } | |
2160 | } | |
2161 | ||
2162 | static void amdgpu_ras_interrupt_process_handler(struct work_struct *work) | |
2163 | { | |
2164 | struct ras_ih_data *data = | |
2165 | container_of(work, struct ras_ih_data, ih_work); | |
2166 | struct ras_manager *obj = | |
2167 | container_of(data, struct ras_manager, ih_data); | |
2168 | ||
2169 | amdgpu_ras_interrupt_handler(obj); | |
2170 | } | |
2171 | ||
2172 | int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev, | |
2173 | struct ras_dispatch_if *info) | |
2174 | { | |
2175 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); | |
2176 | struct ras_ih_data *data = &obj->ih_data; | |
2177 | ||
2178 | if (!obj) | |
2179 | return -EINVAL; | |
2180 | ||
2181 | if (data->inuse == 0) | |
2182 | return 0; | |
2183 | ||
2184 | /* Might be overflow... */ | |
2185 | memcpy(&data->ring[data->wptr], info->entry, | |
2186 | data->element_size); | |
2187 | ||
2188 | wmb(); | |
2189 | data->wptr = (data->aligned_element_size + | |
2190 | data->wptr) % data->ring_size; | |
2191 | ||
2192 | schedule_work(&data->ih_work); | |
2193 | ||
2194 | return 0; | |
2195 | } | |
2196 | ||
2197 | int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev, | |
9252d33d | 2198 | struct ras_common_if *head) |
c030f2e4 | 2199 | { |
9252d33d | 2200 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); |
c030f2e4 | 2201 | struct ras_ih_data *data; |
2202 | ||
2203 | if (!obj) | |
2204 | return -EINVAL; | |
2205 | ||
2206 | data = &obj->ih_data; | |
2207 | if (data->inuse == 0) | |
2208 | return 0; | |
2209 | ||
2210 | cancel_work_sync(&data->ih_work); | |
2211 | ||
2212 | kfree(data->ring); | |
2213 | memset(data, 0, sizeof(*data)); | |
2214 | put_obj(obj); | |
2215 | ||
2216 | return 0; | |
2217 | } | |
2218 | ||
2219 | int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev, | |
9252d33d | 2220 | struct ras_common_if *head) |
c030f2e4 | 2221 | { |
9252d33d | 2222 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); |
c030f2e4 | 2223 | struct ras_ih_data *data; |
9252d33d | 2224 | struct amdgpu_ras_block_object *ras_obj; |
c030f2e4 | 2225 | |
2226 | if (!obj) { | |
2227 | /* in case we registe the IH before enable ras feature */ | |
9252d33d | 2228 | obj = amdgpu_ras_create_obj(adev, head); |
c030f2e4 | 2229 | if (!obj) |
2230 | return -EINVAL; | |
2231 | } else | |
2232 | get_obj(obj); | |
2233 | ||
9252d33d | 2234 | ras_obj = container_of(head, struct amdgpu_ras_block_object, ras_comm); |
2235 | ||
c030f2e4 | 2236 | data = &obj->ih_data; |
2237 | /* add the callback.etc */ | |
2238 | *data = (struct ras_ih_data) { | |
2239 | .inuse = 0, | |
9252d33d | 2240 | .cb = ras_obj->ras_cb, |
c030f2e4 | 2241 | .element_size = sizeof(struct amdgpu_iv_entry), |
2242 | .rptr = 0, | |
2243 | .wptr = 0, | |
2244 | }; | |
2245 | ||
2246 | INIT_WORK(&data->ih_work, amdgpu_ras_interrupt_process_handler); | |
2247 | ||
2248 | data->aligned_element_size = ALIGN(data->element_size, 8); | |
2249 | /* the ring can store 64 iv entries. */ | |
2250 | data->ring_size = 64 * data->aligned_element_size; | |
2251 | data->ring = kmalloc(data->ring_size, GFP_KERNEL); | |
2252 | if (!data->ring) { | |
2253 | put_obj(obj); | |
2254 | return -ENOMEM; | |
2255 | } | |
2256 | ||
2257 | /* IH is ready */ | |
2258 | data->inuse = 1; | |
2259 | ||
2260 | return 0; | |
2261 | } | |
2262 | ||
2263 | static int amdgpu_ras_interrupt_remove_all(struct amdgpu_device *adev) | |
2264 | { | |
2265 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2266 | struct ras_manager *obj, *tmp; | |
2267 | ||
2268 | list_for_each_entry_safe(obj, tmp, &con->head, node) { | |
9252d33d | 2269 | amdgpu_ras_interrupt_remove_handler(adev, &obj->head); |
c030f2e4 | 2270 | } |
2271 | ||
2272 | return 0; | |
2273 | } | |
2274 | /* ih end */ | |
2275 | ||
313c8fd3 GC |
2276 | /* traversal all IPs except NBIO to query error counter */ |
2277 | static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) | |
2278 | { | |
2279 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2280 | struct ras_manager *obj; | |
2281 | ||
8ab0d6f0 | 2282 | if (!adev->ras_enabled || !con) |
313c8fd3 GC |
2283 | return; |
2284 | ||
2285 | list_for_each_entry(obj, &con->head, node) { | |
2286 | struct ras_query_if info = { | |
2287 | .head = obj->head, | |
2288 | }; | |
2289 | ||
2290 | /* | |
2291 | * PCIE_BIF IP has one different isr by ras controller | |
2292 | * interrupt, the specific ras counter query will be | |
2293 | * done in that isr. So skip such block from common | |
2294 | * sync flood interrupt isr calling. | |
2295 | */ | |
2296 | if (info.head.block == AMDGPU_RAS_BLOCK__PCIE_BIF) | |
2297 | continue; | |
2298 | ||
cf63b702 SY |
2299 | /* |
2300 | * this is a workaround for aldebaran, skip send msg to | |
2301 | * smu to get ecc_info table due to smu handle get ecc | |
2302 | * info table failed temporarily. | |
2303 | * should be removed until smu fix handle ecc_info table. | |
2304 | */ | |
2305 | if ((info.head.block == AMDGPU_RAS_BLOCK__UMC) && | |
4e8303cf LL |
2306 | (amdgpu_ip_version(adev, MP1_HWIP, 0) == |
2307 | IP_VERSION(13, 0, 2))) | |
cf63b702 SY |
2308 | continue; |
2309 | ||
761d86d3 | 2310 | amdgpu_ras_query_error_status(adev, &info); |
2a460963 | 2311 | |
4e8303cf LL |
2312 | if (amdgpu_ip_version(adev, MP0_HWIP, 0) != |
2313 | IP_VERSION(11, 0, 2) && | |
2314 | amdgpu_ip_version(adev, MP0_HWIP, 0) != | |
2315 | IP_VERSION(11, 0, 4) && | |
2316 | amdgpu_ip_version(adev, MP0_HWIP, 0) != | |
2317 | IP_VERSION(13, 0, 0)) { | |
2a460963 CL |
2318 | if (amdgpu_ras_reset_error_status(adev, info.head.block)) |
2319 | dev_warn(adev->dev, "Failed to reset error counter and error status"); | |
2320 | } | |
313c8fd3 GC |
2321 | } |
2322 | } | |
2323 | ||
3f975d0f | 2324 | /* Parse RdRspStatus and WrRspStatus */ |
cd92df93 LJ |
2325 | static void amdgpu_ras_error_status_query(struct amdgpu_device *adev, |
2326 | struct ras_query_if *info) | |
3f975d0f | 2327 | { |
8eb53bb2 | 2328 | struct amdgpu_ras_block_object *block_obj; |
3f975d0f SY |
2329 | /* |
2330 | * Only two block need to query read/write | |
2331 | * RspStatus at current state | |
2332 | */ | |
5e67bba3 | 2333 | if ((info->head.block != AMDGPU_RAS_BLOCK__GFX) && |
2334 | (info->head.block != AMDGPU_RAS_BLOCK__MMHUB)) | |
b6efdb02 | 2335 | return; |
2336 | ||
2337 | block_obj = amdgpu_ras_get_ras_block(adev, | |
2338 | info->head.block, | |
2339 | info->head.sub_block_index); | |
5e67bba3 | 2340 | |
5e67bba3 | 2341 | if (!block_obj || !block_obj->hw_ops) { |
afa37315 LT |
2342 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", |
2343 | get_ras_block_str(&info->head)); | |
b6efdb02 | 2344 | return; |
3f975d0f | 2345 | } |
5e67bba3 | 2346 | |
2347 | if (block_obj->hw_ops->query_ras_error_status) | |
ab3b9de6 | 2348 | block_obj->hw_ops->query_ras_error_status(adev); |
5e67bba3 | 2349 | |
3f975d0f SY |
2350 | } |
2351 | ||
2352 | static void amdgpu_ras_query_err_status(struct amdgpu_device *adev) | |
2353 | { | |
2354 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2355 | struct ras_manager *obj; | |
2356 | ||
8ab0d6f0 | 2357 | if (!adev->ras_enabled || !con) |
3f975d0f SY |
2358 | return; |
2359 | ||
2360 | list_for_each_entry(obj, &con->head, node) { | |
2361 | struct ras_query_if info = { | |
2362 | .head = obj->head, | |
2363 | }; | |
2364 | ||
2365 | amdgpu_ras_error_status_query(adev, &info); | |
2366 | } | |
2367 | } | |
2368 | ||
c030f2e4 | 2369 | /* recovery begin */ |
466b1793 | 2370 | |
2371 | /* return 0 on success. | |
2372 | * caller need free bps. | |
2373 | */ | |
2374 | static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, | |
2375 | struct ras_badpage **bps, unsigned int *count) | |
2376 | { | |
2377 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2378 | struct ras_err_handler_data *data; | |
2379 | int i = 0; | |
732f2a30 | 2380 | int ret = 0, status; |
466b1793 | 2381 | |
2382 | if (!con || !con->eh_data || !bps || !count) | |
2383 | return -EINVAL; | |
2384 | ||
2385 | mutex_lock(&con->recovery_lock); | |
2386 | data = con->eh_data; | |
2387 | if (!data || data->count == 0) { | |
2388 | *bps = NULL; | |
46cf2fec | 2389 | ret = -EINVAL; |
466b1793 | 2390 | goto out; |
2391 | } | |
2392 | ||
2393 | *bps = kmalloc(sizeof(struct ras_badpage) * data->count, GFP_KERNEL); | |
2394 | if (!*bps) { | |
2395 | ret = -ENOMEM; | |
2396 | goto out; | |
2397 | } | |
2398 | ||
2399 | for (; i < data->count; i++) { | |
2400 | (*bps)[i] = (struct ras_badpage){ | |
9dc23a63 | 2401 | .bp = data->bps[i].retired_page, |
466b1793 | 2402 | .size = AMDGPU_GPU_PAGE_SIZE, |
52dd95f2 | 2403 | .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, |
466b1793 | 2404 | }; |
ec6aae97 | 2405 | status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, |
bcc09348 | 2406 | data->bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT); |
732f2a30 | 2407 | if (status == -EBUSY) |
52dd95f2 | 2408 | (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; |
732f2a30 | 2409 | else if (status == -ENOENT) |
52dd95f2 | 2410 | (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_FAULT; |
466b1793 | 2411 | } |
2412 | ||
2413 | *count = data->count; | |
2414 | out: | |
2415 | mutex_unlock(&con->recovery_lock); | |
2416 | return ret; | |
2417 | } | |
2418 | ||
b41f742d LL |
2419 | static void amdgpu_ras_set_fed_all(struct amdgpu_device *adev, |
2420 | struct amdgpu_hive_info *hive, bool status) | |
2421 | { | |
2422 | struct amdgpu_device *tmp_adev; | |
2423 | ||
2424 | if (hive) { | |
2425 | list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) | |
2426 | amdgpu_ras_set_fed(tmp_adev, status); | |
2427 | } else { | |
2428 | amdgpu_ras_set_fed(adev, status); | |
2429 | } | |
2430 | } | |
2431 | ||
c030f2e4 | 2432 | static void amdgpu_ras_do_recovery(struct work_struct *work) |
2433 | { | |
2434 | struct amdgpu_ras *ras = | |
2435 | container_of(work, struct amdgpu_ras, recovery_work); | |
b3dbd6d3 JC |
2436 | struct amdgpu_device *remote_adev = NULL; |
2437 | struct amdgpu_device *adev = ras->adev; | |
2438 | struct list_head device_list, *device_list_handle = NULL; | |
53dd920c | 2439 | struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); |
b3dbd6d3 | 2440 | |
b41f742d | 2441 | if (hive) { |
53dd920c | 2442 | atomic_set(&hive->ras_recovery, 1); |
b41f742d LL |
2443 | |
2444 | /* If any device which is part of the hive received RAS fatal | |
2445 | * error interrupt, set fatal error status on all. This | |
2446 | * condition will need a recovery, and flag will be cleared | |
2447 | * as part of recovery. | |
2448 | */ | |
2449 | list_for_each_entry(remote_adev, &hive->device_list, | |
2450 | gmc.xgmi.head) | |
2451 | if (amdgpu_ras_get_fed_status(remote_adev)) { | |
2452 | amdgpu_ras_set_fed_all(adev, hive, true); | |
2453 | break; | |
2454 | } | |
2455 | } | |
f75e94d8 | 2456 | if (!ras->disable_ras_err_cnt_harvest) { |
d95e8e97 | 2457 | |
f75e94d8 GC |
2458 | /* Build list of devices to query RAS related errors */ |
2459 | if (hive && adev->gmc.xgmi.num_physical_nodes > 1) { | |
2460 | device_list_handle = &hive->device_list; | |
2461 | } else { | |
2462 | INIT_LIST_HEAD(&device_list); | |
2463 | list_add_tail(&adev->gmc.xgmi.head, &device_list); | |
2464 | device_list_handle = &device_list; | |
2465 | } | |
c030f2e4 | 2466 | |
f75e94d8 | 2467 | list_for_each_entry(remote_adev, |
3f975d0f SY |
2468 | device_list_handle, gmc.xgmi.head) { |
2469 | amdgpu_ras_query_err_status(remote_adev); | |
f75e94d8 | 2470 | amdgpu_ras_log_on_err_counter(remote_adev); |
3f975d0f | 2471 | } |
d95e8e97 | 2472 | |
b3dbd6d3 | 2473 | } |
313c8fd3 | 2474 | |
f1549c09 LG |
2475 | if (amdgpu_device_should_recover_gpu(ras->adev)) { |
2476 | struct amdgpu_reset_context reset_context; | |
2477 | memset(&reset_context, 0, sizeof(reset_context)); | |
2478 | ||
2479 | reset_context.method = AMD_RESET_METHOD_NONE; | |
2480 | reset_context.reset_req_dev = adev; | |
1a11a65d YC |
2481 | |
2482 | /* Perform full reset in fatal error mode */ | |
2483 | if (!amdgpu_ras_is_poison_mode_supported(ras->adev)) | |
2484 | set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); | |
6c47a79b | 2485 | else { |
1a11a65d | 2486 | clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); |
f1549c09 | 2487 | |
6c47a79b YC |
2488 | if (ras->gpu_reset_flags & AMDGPU_RAS_GPU_RESET_MODE2_RESET) { |
2489 | ras->gpu_reset_flags &= ~AMDGPU_RAS_GPU_RESET_MODE2_RESET; | |
2490 | reset_context.method = AMD_RESET_METHOD_MODE2; | |
2491 | } | |
2c7cd280 YC |
2492 | |
2493 | /* Fatal error occurs in poison mode, mode1 reset is used to | |
2494 | * recover gpu. | |
2495 | */ | |
2496 | if (ras->gpu_reset_flags & AMDGPU_RAS_GPU_RESET_MODE1_RESET) { | |
2497 | ras->gpu_reset_flags &= ~AMDGPU_RAS_GPU_RESET_MODE1_RESET; | |
2498 | set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); | |
1b98a5f8 YC |
2499 | |
2500 | psp_fatal_error_recovery_quirk(&adev->psp); | |
2c7cd280 | 2501 | } |
6c47a79b YC |
2502 | } |
2503 | ||
f1549c09 LG |
2504 | amdgpu_device_gpu_recover(ras->adev, NULL, &reset_context); |
2505 | } | |
c030f2e4 | 2506 | atomic_set(&ras->in_recovery, 0); |
53dd920c AK |
2507 | if (hive) { |
2508 | atomic_set(&hive->ras_recovery, 0); | |
2509 | amdgpu_put_xgmi_hive(hive); | |
2510 | } | |
c030f2e4 | 2511 | } |
2512 | ||
c030f2e4 | 2513 | /* alloc/realloc bps array */ |
2514 | static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev, | |
2515 | struct ras_err_handler_data *data, int pages) | |
2516 | { | |
2517 | unsigned int old_space = data->count + data->space_left; | |
2518 | unsigned int new_space = old_space + pages; | |
9dc23a63 TZ |
2519 | unsigned int align_space = ALIGN(new_space, 512); |
2520 | void *bps = kmalloc(align_space * sizeof(*data->bps), GFP_KERNEL); | |
9dc23a63 | 2521 | |
676deb38 | 2522 | if (!bps) { |
c030f2e4 | 2523 | return -ENOMEM; |
9dc23a63 | 2524 | } |
c030f2e4 | 2525 | |
2526 | if (data->bps) { | |
9dc23a63 | 2527 | memcpy(bps, data->bps, |
c030f2e4 | 2528 | data->count * sizeof(*data->bps)); |
2529 | kfree(data->bps); | |
2530 | } | |
2531 | ||
9dc23a63 | 2532 | data->bps = bps; |
c030f2e4 | 2533 | data->space_left += align_space - old_space; |
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | /* it deal with vram only. */ | |
2538 | int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, | |
9dc23a63 | 2539 | struct eeprom_table_record *bps, int pages) |
c030f2e4 | 2540 | { |
2541 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
73aa8e1a | 2542 | struct ras_err_handler_data *data; |
c030f2e4 | 2543 | int ret = 0; |
676deb38 | 2544 | uint32_t i; |
c030f2e4 | 2545 | |
73aa8e1a | 2546 | if (!con || !con->eh_data || !bps || pages <= 0) |
c030f2e4 | 2547 | return 0; |
2548 | ||
2549 | mutex_lock(&con->recovery_lock); | |
73aa8e1a | 2550 | data = con->eh_data; |
c030f2e4 | 2551 | if (!data) |
2552 | goto out; | |
2553 | ||
676deb38 DL |
2554 | for (i = 0; i < pages; i++) { |
2555 | if (amdgpu_ras_check_bad_page_unlock(con, | |
2556 | bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT)) | |
2557 | continue; | |
2558 | ||
2559 | if (!data->space_left && | |
2560 | amdgpu_ras_realloc_eh_data_space(adev, data, 256)) { | |
c030f2e4 | 2561 | ret = -ENOMEM; |
2562 | goto out; | |
2563 | } | |
2564 | ||
48fa9071 | 2565 | amdgpu_ras_reserve_page(adev, bps[i].retired_page); |
9dc23a63 | 2566 | |
676deb38 DL |
2567 | memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps)); |
2568 | data->count++; | |
2569 | data->space_left--; | |
2570 | } | |
c030f2e4 | 2571 | out: |
2572 | mutex_unlock(&con->recovery_lock); | |
2573 | ||
2574 | return ret; | |
2575 | } | |
2576 | ||
78ad00c9 TZ |
2577 | /* |
2578 | * write error record array to eeprom, the function should be | |
2579 | * protected by recovery_lock | |
4d33e0f1 | 2580 | * new_cnt: new added UE count, excluding reserved bad pages, can be NULL |
78ad00c9 | 2581 | */ |
4d33e0f1 TZ |
2582 | int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev, |
2583 | unsigned long *new_cnt) | |
78ad00c9 TZ |
2584 | { |
2585 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2586 | struct ras_err_handler_data *data; | |
8a3e801f | 2587 | struct amdgpu_ras_eeprom_control *control; |
78ad00c9 TZ |
2588 | int save_count; |
2589 | ||
4d33e0f1 TZ |
2590 | if (!con || !con->eh_data) { |
2591 | if (new_cnt) | |
2592 | *new_cnt = 0; | |
2593 | ||
78ad00c9 | 2594 | return 0; |
4d33e0f1 | 2595 | } |
78ad00c9 | 2596 | |
d9a69fe5 | 2597 | mutex_lock(&con->recovery_lock); |
8a3e801f | 2598 | control = &con->eeprom_control; |
78ad00c9 | 2599 | data = con->eh_data; |
0686627b | 2600 | save_count = data->count - control->ras_num_recs; |
d9a69fe5 | 2601 | mutex_unlock(&con->recovery_lock); |
4d33e0f1 TZ |
2602 | |
2603 | if (new_cnt) | |
2604 | *new_cnt = save_count / adev->umc.retire_unit; | |
2605 | ||
78ad00c9 | 2606 | /* only new entries are saved */ |
b1628425 | 2607 | if (save_count > 0) { |
63d4c081 LT |
2608 | if (amdgpu_ras_eeprom_append(control, |
2609 | &data->bps[control->ras_num_recs], | |
2610 | save_count)) { | |
6952e99c | 2611 | dev_err(adev->dev, "Failed to save EEPROM table data!"); |
78ad00c9 TZ |
2612 | return -EIO; |
2613 | } | |
2614 | ||
b1628425 GC |
2615 | dev_info(adev->dev, "Saved %d pages to EEPROM table.\n", save_count); |
2616 | } | |
2617 | ||
78ad00c9 TZ |
2618 | return 0; |
2619 | } | |
2620 | ||
2621 | /* | |
2622 | * read error record array in eeprom and reserve enough space for | |
2623 | * storing new bad pages | |
2624 | */ | |
2625 | static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev) | |
2626 | { | |
2627 | struct amdgpu_ras_eeprom_control *control = | |
6457205c | 2628 | &adev->psp.ras_context.ras->eeprom_control; |
e4e6a589 LT |
2629 | struct eeprom_table_record *bps; |
2630 | int ret; | |
78ad00c9 TZ |
2631 | |
2632 | /* no bad page record, skip eeprom access */ | |
0686627b | 2633 | if (control->ras_num_recs == 0 || amdgpu_bad_page_threshold == 0) |
e4e6a589 | 2634 | return 0; |
78ad00c9 | 2635 | |
0686627b | 2636 | bps = kcalloc(control->ras_num_recs, sizeof(*bps), GFP_KERNEL); |
78ad00c9 TZ |
2637 | if (!bps) |
2638 | return -ENOMEM; | |
2639 | ||
0686627b | 2640 | ret = amdgpu_ras_eeprom_read(control, bps, control->ras_num_recs); |
e4e6a589 | 2641 | if (ret) |
6952e99c | 2642 | dev_err(adev->dev, "Failed to load EEPROM table records!"); |
e4e6a589 | 2643 | else |
0686627b | 2644 | ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs); |
78ad00c9 | 2645 | |
78ad00c9 TZ |
2646 | kfree(bps); |
2647 | return ret; | |
2648 | } | |
2649 | ||
676deb38 DL |
2650 | static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, |
2651 | uint64_t addr) | |
2652 | { | |
2653 | struct ras_err_handler_data *data = con->eh_data; | |
2654 | int i; | |
2655 | ||
2656 | addr >>= AMDGPU_GPU_PAGE_SHIFT; | |
2657 | for (i = 0; i < data->count; i++) | |
2658 | if (addr == data->bps[i].retired_page) | |
2659 | return true; | |
2660 | ||
2661 | return false; | |
2662 | } | |
2663 | ||
6e4be987 TZ |
2664 | /* |
2665 | * check if an address belongs to bad page | |
2666 | * | |
2667 | * Note: this check is only for umc block | |
2668 | */ | |
2669 | static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, | |
2670 | uint64_t addr) | |
2671 | { | |
2672 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
6e4be987 TZ |
2673 | bool ret = false; |
2674 | ||
2675 | if (!con || !con->eh_data) | |
2676 | return ret; | |
2677 | ||
2678 | mutex_lock(&con->recovery_lock); | |
676deb38 | 2679 | ret = amdgpu_ras_check_bad_page_unlock(con, addr); |
6e4be987 TZ |
2680 | mutex_unlock(&con->recovery_lock); |
2681 | return ret; | |
2682 | } | |
2683 | ||
e5c04edf | 2684 | static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, |
e4e6a589 | 2685 | uint32_t max_count) |
c84d4670 | 2686 | { |
e5c04edf | 2687 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
c84d4670 GC |
2688 | |
2689 | /* | |
2690 | * Justification of value bad_page_cnt_threshold in ras structure | |
2691 | * | |
f3cbe70e TZ |
2692 | * Generally, 0 <= amdgpu_bad_page_threshold <= max record length |
2693 | * in eeprom or amdgpu_bad_page_threshold == -2, introduce two | |
2694 | * scenarios accordingly. | |
c84d4670 GC |
2695 | * |
2696 | * Bad page retirement enablement: | |
f3cbe70e | 2697 | * - If amdgpu_bad_page_threshold = -2, |
c84d4670 GC |
2698 | * bad_page_cnt_threshold = typical value by formula. |
2699 | * | |
2700 | * - When the value from user is 0 < amdgpu_bad_page_threshold < | |
2701 | * max record length in eeprom, use it directly. | |
2702 | * | |
2703 | * Bad page retirement disablement: | |
2704 | * - If amdgpu_bad_page_threshold = 0, bad page retirement | |
2705 | * functionality is disabled, and bad_page_cnt_threshold will | |
2706 | * take no effect. | |
2707 | */ | |
2708 | ||
e4e6a589 LT |
2709 | if (amdgpu_bad_page_threshold < 0) { |
2710 | u64 val = adev->gmc.mc_vram_size; | |
c84d4670 | 2711 | |
e4e6a589 | 2712 | do_div(val, RAS_BAD_PAGE_COVER); |
e5c04edf | 2713 | con->bad_page_cnt_threshold = min(lower_32_bits(val), |
e4e6a589 | 2714 | max_count); |
e5c04edf | 2715 | } else { |
e4e6a589 LT |
2716 | con->bad_page_cnt_threshold = min_t(int, max_count, |
2717 | amdgpu_bad_page_threshold); | |
c84d4670 GC |
2718 | } |
2719 | } | |
2720 | ||
98b5bc87 YC |
2721 | int amdgpu_ras_put_poison_req(struct amdgpu_device *adev, |
2722 | enum amdgpu_ras_block block, uint16_t pasid, | |
2723 | pasid_notify pasid_fn, void *data, uint32_t reset) | |
2724 | { | |
2725 | int ret = 0; | |
2726 | struct ras_poison_msg poison_msg; | |
2727 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2728 | ||
2729 | memset(&poison_msg, 0, sizeof(poison_msg)); | |
2730 | poison_msg.block = block; | |
2731 | poison_msg.pasid = pasid; | |
2732 | poison_msg.reset = reset; | |
2733 | poison_msg.pasid_fn = pasid_fn; | |
2734 | poison_msg.data = data; | |
2735 | ||
2736 | ret = kfifo_put(&con->poison_fifo, poison_msg); | |
2737 | if (!ret) { | |
2738 | dev_err(adev->dev, "Poison message fifo is full!\n"); | |
2739 | return -ENOSPC; | |
2740 | } | |
2741 | ||
2742 | return 0; | |
2743 | } | |
2744 | ||
98b5bc87 YC |
2745 | static int amdgpu_ras_get_poison_req(struct amdgpu_device *adev, |
2746 | struct ras_poison_msg *poison_msg) | |
2747 | { | |
2748 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2749 | ||
2750 | return kfifo_get(&con->poison_fifo, poison_msg); | |
2751 | } | |
98b5bc87 | 2752 | |
f493dd64 YC |
2753 | static void amdgpu_ras_ecc_log_init(struct ras_ecc_log_info *ecc_log) |
2754 | { | |
2755 | mutex_init(&ecc_log->lock); | |
2756 | ||
2757 | /* Set any value as siphash key */ | |
2758 | memset(&ecc_log->ecc_key, 0xad, sizeof(ecc_log->ecc_key)); | |
2759 | ||
2760 | INIT_RADIX_TREE(&ecc_log->de_page_tree, GFP_KERNEL); | |
2761 | ecc_log->de_updated = false; | |
2762 | } | |
2763 | ||
2764 | static void amdgpu_ras_ecc_log_fini(struct ras_ecc_log_info *ecc_log) | |
2765 | { | |
2766 | struct radix_tree_iter iter; | |
2767 | void __rcu **slot; | |
2768 | struct ras_ecc_err *ecc_err; | |
2769 | ||
2770 | mutex_lock(&ecc_log->lock); | |
2771 | radix_tree_for_each_slot(slot, &ecc_log->de_page_tree, &iter, 0) { | |
2772 | ecc_err = radix_tree_deref_slot(slot); | |
2773 | kfree(ecc_err->err_pages.pfn); | |
2774 | kfree(ecc_err); | |
2775 | radix_tree_iter_delete(&ecc_log->de_page_tree, &iter, slot); | |
2776 | } | |
2777 | mutex_unlock(&ecc_log->lock); | |
2778 | ||
2779 | mutex_destroy(&ecc_log->lock); | |
2780 | ecc_log->de_updated = false; | |
2781 | } | |
a734adfb | 2782 | |
2cf8e50e YC |
2783 | static void amdgpu_ras_do_page_retirement(struct work_struct *work) |
2784 | { | |
2785 | struct amdgpu_ras *con = container_of(work, struct amdgpu_ras, | |
2786 | page_retirement_dwork.work); | |
2787 | struct amdgpu_device *adev = con->adev; | |
2788 | struct ras_err_data err_data; | |
2789 | ||
2790 | if (amdgpu_in_reset(adev) || atomic_read(&con->in_recovery)) | |
2791 | return; | |
2792 | ||
2793 | amdgpu_ras_error_data_init(&err_data); | |
2794 | ||
2795 | amdgpu_umc_handle_bad_pages(adev, &err_data); | |
2796 | ||
2797 | amdgpu_ras_error_data_fini(&err_data); | |
2798 | ||
2799 | mutex_lock(&con->umc_ecc_log.lock); | |
2800 | if (radix_tree_tagged(&con->umc_ecc_log.de_page_tree, | |
2801 | UMC_ECC_NEW_DETECTED_TAG)) | |
2802 | schedule_delayed_work(&con->page_retirement_dwork, | |
2803 | msecs_to_jiffies(AMDGPU_RAS_RETIRE_PAGE_INTERVAL)); | |
2804 | mutex_unlock(&con->umc_ecc_log.lock); | |
2805 | } | |
2806 | ||
a734adfb YC |
2807 | static int amdgpu_ras_query_ecc_status(struct amdgpu_device *adev, |
2808 | enum amdgpu_ras_block ras_block, uint32_t timeout_ms) | |
2809 | { | |
2810 | int ret = 0; | |
2811 | struct ras_ecc_log_info *ecc_log; | |
2812 | struct ras_query_if info; | |
2813 | uint32_t timeout = timeout_ms; | |
2814 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |
2815 | ||
2816 | memset(&info, 0, sizeof(info)); | |
2817 | info.head.block = ras_block; | |
2818 | ||
2819 | ecc_log = &ras->umc_ecc_log; | |
2820 | ecc_log->de_updated = false; | |
2821 | do { | |
2822 | ret = amdgpu_ras_query_error_status(adev, &info); | |
2823 | if (ret) { | |
2824 | dev_err(adev->dev, "Failed to query ras error! ret:%d\n", ret); | |
2825 | return ret; | |
2826 | } | |
2827 | ||
2828 | if (timeout && !ecc_log->de_updated) { | |
2829 | msleep(1); | |
2830 | timeout--; | |
2831 | } | |
2832 | } while (timeout && !ecc_log->de_updated); | |
2833 | ||
2834 | if (timeout_ms && !timeout) { | |
2835 | dev_warn(adev->dev, "Can't find deferred error\n"); | |
2836 | return -ETIMEDOUT; | |
2837 | } | |
2838 | ||
2839 | return 0; | |
2840 | } | |
2841 | ||
2842 | static void amdgpu_ras_poison_creation_handler(struct amdgpu_device *adev, | |
2843 | uint32_t timeout) | |
2844 | { | |
2cf8e50e YC |
2845 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
2846 | int ret; | |
2847 | ||
2848 | ret = amdgpu_ras_query_ecc_status(adev, AMDGPU_RAS_BLOCK__UMC, timeout); | |
2849 | if (!ret) | |
2850 | schedule_delayed_work(&con->page_retirement_dwork, 0); | |
a734adfb YC |
2851 | } |
2852 | ||
370fbff4 YC |
2853 | static int amdgpu_ras_poison_consumption_handler(struct amdgpu_device *adev, |
2854 | struct ras_poison_msg *poison_msg) | |
2855 | { | |
2856 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
2857 | uint32_t reset = poison_msg->reset; | |
2858 | uint16_t pasid = poison_msg->pasid; | |
2859 | ||
2860 | kgd2kfd_set_sram_ecc_flag(adev->kfd.dev); | |
2861 | ||
2862 | if (poison_msg->pasid_fn) | |
2863 | poison_msg->pasid_fn(adev, pasid, poison_msg->data); | |
2864 | ||
2865 | if (reset) { | |
2866 | flush_delayed_work(&con->page_retirement_dwork); | |
2867 | ||
2868 | con->gpu_reset_flags |= reset; | |
2869 | amdgpu_ras_reset_gpu(adev); | |
2870 | } | |
2871 | ||
2872 | return 0; | |
2873 | } | |
2874 | ||
3fdcd0a3 YC |
2875 | static int amdgpu_ras_page_retirement_thread(void *param) |
2876 | { | |
2877 | struct amdgpu_device *adev = (struct amdgpu_device *)param; | |
2878 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
a734adfb YC |
2879 | struct ras_poison_msg poison_msg; |
2880 | enum amdgpu_ras_block ras_block; | |
370fbff4 | 2881 | bool poison_creation_is_handled = false; |
3fdcd0a3 YC |
2882 | |
2883 | while (!kthread_should_stop()) { | |
2884 | ||
2885 | wait_event_interruptible(con->page_retirement_wq, | |
c84a7e21 | 2886 | kthread_should_stop() || |
3fdcd0a3 YC |
2887 | atomic_read(&con->page_retirement_req_cnt)); |
2888 | ||
c84a7e21 MJ |
2889 | if (kthread_should_stop()) |
2890 | break; | |
2891 | ||
3fdcd0a3 | 2892 | atomic_dec(&con->page_retirement_req_cnt); |
6c23f3d1 | 2893 | |
a734adfb YC |
2894 | if (!amdgpu_ras_get_poison_req(adev, &poison_msg)) |
2895 | continue; | |
2896 | ||
2897 | ras_block = poison_msg.block; | |
2898 | ||
2899 | dev_info(adev->dev, "Start processing ras block %s(%d)\n", | |
2900 | ras_block_str(ras_block), ras_block); | |
2901 | ||
370fbff4 | 2902 | if (ras_block == AMDGPU_RAS_BLOCK__UMC) { |
a734adfb YC |
2903 | amdgpu_ras_poison_creation_handler(adev, |
2904 | MAX_UMC_POISON_POLLING_TIME_ASYNC); | |
370fbff4 YC |
2905 | poison_creation_is_handled = true; |
2906 | } else { | |
2907 | /* poison_creation_is_handled: | |
2908 | * false: no poison creation interrupt, but it has poison | |
2909 | * consumption interrupt. | |
2910 | * true: It has poison creation interrupt at the beginning, | |
2911 | * but it has no poison creation interrupt later. | |
2912 | */ | |
2913 | amdgpu_ras_poison_creation_handler(adev, | |
2914 | poison_creation_is_handled ? | |
2915 | 0 : MAX_UMC_POISON_POLLING_TIME_ASYNC); | |
2916 | ||
2917 | amdgpu_ras_poison_consumption_handler(adev, &poison_msg); | |
2918 | poison_creation_is_handled = false; | |
2919 | } | |
3fdcd0a3 YC |
2920 | } |
2921 | ||
2922 | return 0; | |
2923 | } | |
2924 | ||
1a6fc071 | 2925 | int amdgpu_ras_recovery_init(struct amdgpu_device *adev) |
c030f2e4 | 2926 | { |
2927 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
4d1337d2 | 2928 | struct ras_err_handler_data **data; |
e4e6a589 | 2929 | u32 max_eeprom_records_count = 0; |
b82e65a9 | 2930 | bool exc_err_limit = false; |
78ad00c9 | 2931 | int ret; |
c030f2e4 | 2932 | |
e0e146d5 | 2933 | if (!con || amdgpu_sriov_vf(adev)) |
1d9d2ca8 LT |
2934 | return 0; |
2935 | ||
2936 | /* Allow access to RAS EEPROM via debugfs, when the ASIC | |
2937 | * supports RAS and debugfs is enabled, but when | |
2938 | * adev->ras_enabled is unset, i.e. when "ras_enable" | |
2939 | * module parameter is set to 0. | |
2940 | */ | |
2941 | con->adev = adev; | |
2942 | ||
2943 | if (!adev->ras_enabled) | |
4d1337d2 AG |
2944 | return 0; |
2945 | ||
1d9d2ca8 | 2946 | data = &con->eh_data; |
091411be | 2947 | *data = kzalloc(sizeof(**data), GFP_KERNEL); |
1a6fc071 TZ |
2948 | if (!*data) { |
2949 | ret = -ENOMEM; | |
2950 | goto out; | |
2951 | } | |
c030f2e4 | 2952 | |
2953 | mutex_init(&con->recovery_lock); | |
2954 | INIT_WORK(&con->recovery_work, amdgpu_ras_do_recovery); | |
2955 | atomic_set(&con->in_recovery, 0); | |
69691c82 | 2956 | con->eeprom_control.bad_channel_bitmap = 0; |
c030f2e4 | 2957 | |
7f599fed | 2958 | max_eeprom_records_count = amdgpu_ras_eeprom_max_record_count(&con->eeprom_control); |
e4e6a589 | 2959 | amdgpu_ras_validate_threshold(adev, max_eeprom_records_count); |
c84d4670 | 2960 | |
e5086659 | 2961 | /* Todo: During test the SMU might fail to read the eeprom through I2C |
2962 | * when the GPU is pending on XGMI reset during probe time | |
2963 | * (Mostly after second bus reset), skip it now | |
2964 | */ | |
2965 | if (adev->gmc.xgmi.pending_reset) | |
2966 | return 0; | |
b82e65a9 GC |
2967 | ret = amdgpu_ras_eeprom_init(&con->eeprom_control, &exc_err_limit); |
2968 | /* | |
2969 | * This calling fails when exc_err_limit is true or | |
2970 | * ret != 0. | |
2971 | */ | |
2972 | if (exc_err_limit || ret) | |
1a6fc071 | 2973 | goto free; |
78ad00c9 | 2974 | |
0686627b | 2975 | if (con->eeprom_control.ras_num_recs) { |
78ad00c9 TZ |
2976 | ret = amdgpu_ras_load_bad_pages(adev); |
2977 | if (ret) | |
1a6fc071 | 2978 | goto free; |
513befa6 | 2979 | |
bc143d8b | 2980 | amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs); |
69691c82 SY |
2981 | |
2982 | if (con->update_channel_flag == true) { | |
2983 | amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap); | |
2984 | con->update_channel_flag = false; | |
2985 | } | |
78ad00c9 | 2986 | } |
c030f2e4 | 2987 | |
af730e08 | 2988 | mutex_init(&con->page_rsv_lock); |
98b5bc87 | 2989 | INIT_KFIFO(con->poison_fifo); |
3fdcd0a3 YC |
2990 | mutex_init(&con->page_retirement_lock); |
2991 | init_waitqueue_head(&con->page_retirement_wq); | |
2992 | atomic_set(&con->page_retirement_req_cnt, 0); | |
2993 | con->page_retirement_thread = | |
2994 | kthread_run(amdgpu_ras_page_retirement_thread, adev, "umc_page_retirement"); | |
2995 | if (IS_ERR(con->page_retirement_thread)) { | |
2996 | con->page_retirement_thread = NULL; | |
2997 | dev_warn(adev->dev, "Failed to create umc_page_retirement thread!!!\n"); | |
2998 | } | |
2999 | ||
2cf8e50e | 3000 | INIT_DELAYED_WORK(&con->page_retirement_dwork, amdgpu_ras_do_page_retirement); |
f493dd64 | 3001 | amdgpu_ras_ecc_log_init(&con->umc_ecc_log); |
12b2cab7 MJ |
3002 | #ifdef CONFIG_X86_MCE_AMD |
3003 | if ((adev->asic_type == CHIP_ALDEBARAN) && | |
3004 | (adev->gmc.xgmi.connected_to_cpu)) | |
91a1a52d | 3005 | amdgpu_register_bad_pages_mca_notifier(adev); |
12b2cab7 | 3006 | #endif |
c030f2e4 | 3007 | return 0; |
1a6fc071 | 3008 | |
1a6fc071 | 3009 | free: |
1a6fc071 | 3010 | kfree((*data)->bps); |
1a6fc071 | 3011 | kfree(*data); |
1995b3a3 | 3012 | con->eh_data = NULL; |
1a6fc071 | 3013 | out: |
cf696091 | 3014 | dev_warn(adev->dev, "Failed to initialize ras recovery! (%d)\n", ret); |
1a6fc071 | 3015 | |
b82e65a9 GC |
3016 | /* |
3017 | * Except error threshold exceeding case, other failure cases in this | |
3018 | * function would not fail amdgpu driver init. | |
3019 | */ | |
3020 | if (!exc_err_limit) | |
3021 | ret = 0; | |
3022 | else | |
3023 | ret = -EINVAL; | |
3024 | ||
1a6fc071 | 3025 | return ret; |
c030f2e4 | 3026 | } |
3027 | ||
3028 | static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) | |
3029 | { | |
3030 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3031 | struct ras_err_handler_data *data = con->eh_data; | |
3032 | ||
1a6fc071 TZ |
3033 | /* recovery_init failed to init it, fini is useless */ |
3034 | if (!data) | |
3035 | return 0; | |
3036 | ||
3fdcd0a3 YC |
3037 | if (con->page_retirement_thread) |
3038 | kthread_stop(con->page_retirement_thread); | |
3039 | ||
3040 | atomic_set(&con->page_retirement_req_cnt, 0); | |
3041 | ||
af730e08 YC |
3042 | mutex_destroy(&con->page_rsv_lock); |
3043 | ||
c030f2e4 | 3044 | cancel_work_sync(&con->recovery_work); |
c030f2e4 | 3045 | |
2cf8e50e YC |
3046 | cancel_delayed_work_sync(&con->page_retirement_dwork); |
3047 | ||
f493dd64 YC |
3048 | amdgpu_ras_ecc_log_fini(&con->umc_ecc_log); |
3049 | ||
c030f2e4 | 3050 | mutex_lock(&con->recovery_lock); |
3051 | con->eh_data = NULL; | |
3052 | kfree(data->bps); | |
3053 | kfree(data); | |
3054 | mutex_unlock(&con->recovery_lock); | |
3055 | ||
3056 | return 0; | |
3057 | } | |
3058 | /* recovery end */ | |
3059 | ||
084e2640 | 3060 | static bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) |
5436ab94 | 3061 | { |
82835055 | 3062 | if (amdgpu_sriov_vf(adev)) { |
4e8303cf | 3063 | switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { |
82835055 | 3064 | case IP_VERSION(13, 0, 2): |
80578f16 | 3065 | case IP_VERSION(13, 0, 6): |
82835055 YC |
3066 | return true; |
3067 | default: | |
3068 | return false; | |
3069 | } | |
3070 | } | |
3071 | ||
073285ef | 3072 | if (adev->asic_type == CHIP_IP_DISCOVERY) { |
4e8303cf | 3073 | switch (amdgpu_ip_version(adev, MP0_HWIP, 0)) { |
073285ef | 3074 | case IP_VERSION(13, 0, 0): |
cb906ce3 | 3075 | case IP_VERSION(13, 0, 6): |
073285ef YC |
3076 | case IP_VERSION(13, 0, 10): |
3077 | return true; | |
3078 | default: | |
3079 | return false; | |
3080 | } | |
3081 | } | |
3082 | ||
084e2640 LT |
3083 | return adev->asic_type == CHIP_VEGA10 || |
3084 | adev->asic_type == CHIP_VEGA20 || | |
3085 | adev->asic_type == CHIP_ARCTURUS || | |
75f06251 | 3086 | adev->asic_type == CHIP_ALDEBARAN || |
084e2640 | 3087 | adev->asic_type == CHIP_SIENNA_CICHLID; |
5436ab94 SY |
3088 | } |
3089 | ||
f50160cf SY |
3090 | /* |
3091 | * this is workaround for vega20 workstation sku, | |
3092 | * force enable gfx ras, ignore vbios gfx ras flag | |
3093 | * due to GC EDC can not write | |
3094 | */ | |
e509965e | 3095 | static void amdgpu_ras_get_quirks(struct amdgpu_device *adev) |
f50160cf SY |
3096 | { |
3097 | struct atom_context *ctx = adev->mode_info.atom_context; | |
3098 | ||
3099 | if (!ctx) | |
3100 | return; | |
3101 | ||
adf64e21 ML |
3102 | if (strnstr(ctx->vbios_pn, "D16406", |
3103 | sizeof(ctx->vbios_pn)) || | |
3104 | strnstr(ctx->vbios_pn, "D36002", | |
3105 | sizeof(ctx->vbios_pn))) | |
8ab0d6f0 | 3106 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__GFX); |
f50160cf SY |
3107 | } |
3108 | ||
4e2965bd HZ |
3109 | /* Query ras capablity via atomfirmware interface */ |
3110 | static void amdgpu_ras_query_ras_capablity_from_vbios(struct amdgpu_device *adev) | |
3111 | { | |
3112 | /* mem_ecc cap */ | |
3113 | if (amdgpu_atomfirmware_mem_ecc_supported(adev)) { | |
3114 | dev_info(adev->dev, "MEM ECC is active.\n"); | |
3115 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__UMC | | |
3116 | 1 << AMDGPU_RAS_BLOCK__DF); | |
3117 | } else { | |
3118 | dev_info(adev->dev, "MEM ECC is not presented.\n"); | |
3119 | } | |
3120 | ||
3121 | /* sram_ecc cap */ | |
3122 | if (amdgpu_atomfirmware_sram_ecc_supported(adev)) { | |
3123 | dev_info(adev->dev, "SRAM ECC is active.\n"); | |
3124 | if (!amdgpu_sriov_vf(adev)) | |
3125 | adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC | | |
3126 | 1 << AMDGPU_RAS_BLOCK__DF); | |
3127 | else | |
3128 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__PCIE_BIF | | |
3129 | 1 << AMDGPU_RAS_BLOCK__SDMA | | |
3130 | 1 << AMDGPU_RAS_BLOCK__GFX); | |
3131 | ||
3132 | /* | |
3133 | * VCN/JPEG RAS can be supported on both bare metal and | |
3134 | * SRIOV environment | |
3135 | */ | |
3136 | if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(2, 6, 0) || | |
3137 | amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 0) || | |
3138 | amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 3)) | |
3139 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN | | |
3140 | 1 << AMDGPU_RAS_BLOCK__JPEG); | |
3141 | else | |
3142 | adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__VCN | | |
3143 | 1 << AMDGPU_RAS_BLOCK__JPEG); | |
3144 | ||
3145 | /* | |
3146 | * XGMI RAS is not supported if xgmi num physical nodes | |
3147 | * is zero | |
3148 | */ | |
3149 | if (!adev->gmc.xgmi.num_physical_nodes) | |
3150 | adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__XGMI_WAFL); | |
3151 | } else { | |
3152 | dev_info(adev->dev, "SRAM ECC is not presented.\n"); | |
3153 | } | |
3154 | } | |
3155 | ||
3156 | /* Query poison mode from umc/df IP callbacks */ | |
3157 | static void amdgpu_ras_query_poison_mode(struct amdgpu_device *adev) | |
3158 | { | |
3159 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3160 | bool df_poison, umc_poison; | |
3161 | ||
3162 | /* poison setting is useless on SRIOV guest */ | |
3163 | if (amdgpu_sriov_vf(adev) || !con) | |
3164 | return; | |
3165 | ||
3166 | /* Init poison supported flag, the default value is false */ | |
3167 | if (adev->gmc.xgmi.connected_to_cpu || | |
3168 | adev->gmc.is_app_apu) { | |
3169 | /* enabled by default when GPU is connected to CPU */ | |
3170 | con->poison_supported = true; | |
3171 | } else if (adev->df.funcs && | |
3172 | adev->df.funcs->query_ras_poison_mode && | |
3173 | adev->umc.ras && | |
3174 | adev->umc.ras->query_ras_poison_mode) { | |
3175 | df_poison = | |
3176 | adev->df.funcs->query_ras_poison_mode(adev); | |
3177 | umc_poison = | |
3178 | adev->umc.ras->query_ras_poison_mode(adev); | |
3179 | ||
3180 | /* Only poison is set in both DF and UMC, we can support it */ | |
3181 | if (df_poison && umc_poison) | |
3182 | con->poison_supported = true; | |
3183 | else if (df_poison != umc_poison) | |
3184 | dev_warn(adev->dev, | |
3185 | "Poison setting is inconsistent in DF/UMC(%d:%d)!\n", | |
3186 | df_poison, umc_poison); | |
3187 | } | |
3188 | } | |
3189 | ||
5caf466a | 3190 | /* |
3191 | * check hardware's ras ability which will be saved in hw_supported. | |
3192 | * if hardware does not support ras, we can skip some ras initializtion and | |
3193 | * forbid some ras operations from IP. | |
3194 | * if software itself, say boot parameter, limit the ras ability. We still | |
3195 | * need allow IP do some limited operations, like disable. In such case, | |
3196 | * we have to initialize ras as normal. but need check if operation is | |
3197 | * allowed or not in each function. | |
3198 | */ | |
e509965e | 3199 | static void amdgpu_ras_check_supported(struct amdgpu_device *adev) |
c030f2e4 | 3200 | { |
8ab0d6f0 | 3201 | adev->ras_hw_enabled = adev->ras_enabled = 0; |
c030f2e4 | 3202 | |
38298ce6 | 3203 | if (!amdgpu_ras_asic_supported(adev)) |
5caf466a | 3204 | return; |
b404ae82 | 3205 | |
4e2965bd HZ |
3206 | /* query ras capability from psp */ |
3207 | if (amdgpu_psp_get_ras_capability(&adev->psp)) | |
3208 | goto init_ras_enabled_flag; | |
58bc2a9c | 3209 | |
4e2965bd HZ |
3210 | /* query ras capablity from bios */ |
3211 | if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) { | |
3212 | amdgpu_ras_query_ras_capablity_from_vbios(adev); | |
75f06251 HZ |
3213 | } else { |
3214 | /* driver only manages a few IP blocks RAS feature | |
3215 | * when GPU is connected cpu through XGMI */ | |
8ab0d6f0 | 3216 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__GFX | |
e509965e LT |
3217 | 1 << AMDGPU_RAS_BLOCK__SDMA | |
3218 | 1 << AMDGPU_RAS_BLOCK__MMHUB); | |
75f06251 | 3219 | } |
88474cca | 3220 | |
4e2965bd | 3221 | /* apply asic specific settings (vega20 only for now) */ |
e509965e | 3222 | amdgpu_ras_get_quirks(adev); |
f50160cf | 3223 | |
4e2965bd HZ |
3224 | /* query poison mode from umc/df ip callback */ |
3225 | amdgpu_ras_query_poison_mode(adev); | |
3226 | ||
3227 | init_ras_enabled_flag: | |
88474cca | 3228 | /* hw_supported needs to be aligned with RAS block mask. */ |
8ab0d6f0 | 3229 | adev->ras_hw_enabled &= AMDGPU_RAS_BLOCK_MASK; |
b404ae82 | 3230 | |
66d64e4e SY |
3231 | adev->ras_enabled = amdgpu_ras_enable == 0 ? 0 : |
3232 | adev->ras_hw_enabled & amdgpu_ras_mask; | |
04c4fcd2 YW |
3233 | |
3234 | /* aca is disabled by default */ | |
3235 | adev->aca.is_enabled = false; | |
c030f2e4 | 3236 | } |
3237 | ||
05adfd80 LT |
3238 | static void amdgpu_ras_counte_dw(struct work_struct *work) |
3239 | { | |
3240 | struct amdgpu_ras *con = container_of(work, struct amdgpu_ras, | |
3241 | ras_counte_delay_work.work); | |
3242 | struct amdgpu_device *adev = con->adev; | |
a3fbb0d8 | 3243 | struct drm_device *dev = adev_to_drm(adev); |
05adfd80 LT |
3244 | unsigned long ce_count, ue_count; |
3245 | int res; | |
3246 | ||
3247 | res = pm_runtime_get_sync(dev->dev); | |
3248 | if (res < 0) | |
3249 | goto Out; | |
3250 | ||
3251 | /* Cache new values. | |
3252 | */ | |
4a1c9a44 | 3253 | if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count, NULL) == 0) { |
4d9f771e LT |
3254 | atomic_set(&con->ras_ce_count, ce_count); |
3255 | atomic_set(&con->ras_ue_count, ue_count); | |
3256 | } | |
05adfd80 LT |
3257 | |
3258 | pm_runtime_mark_last_busy(dev->dev); | |
3259 | Out: | |
3260 | pm_runtime_put_autosuspend(dev->dev); | |
3261 | } | |
3262 | ||
625e5f38 AK |
3263 | static int amdgpu_get_ras_schema(struct amdgpu_device *adev) |
3264 | { | |
3265 | return amdgpu_ras_is_poison_mode_supported(adev) ? AMDGPU_RAS_ERROR__POISON : 0 | | |
3266 | AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE | | |
3267 | AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE | | |
3268 | AMDGPU_RAS_ERROR__PARITY; | |
3269 | } | |
3270 | ||
9dc57c2a YW |
3271 | static void ras_event_mgr_init(struct ras_event_manager *mgr) |
3272 | { | |
3273 | int i; | |
3274 | ||
3275 | for (i = 0; i < ARRAY_SIZE(mgr->seqnos); i++) | |
3276 | atomic64_set(&mgr->seqnos[i], 0); | |
3277 | } | |
3278 | ||
3279 | static void amdgpu_ras_event_mgr_init(struct amdgpu_device *adev) | |
3280 | { | |
3281 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |
3282 | struct amdgpu_hive_info *hive; | |
3283 | ||
3284 | if (!ras) | |
3285 | return; | |
3286 | ||
3287 | hive = amdgpu_get_xgmi_hive(adev); | |
3288 | ras->event_mgr = hive ? &hive->event_mgr : &ras->__event_mgr; | |
3289 | ||
3290 | /* init event manager with node 0 on xgmi system */ | |
3291 | if (!amdgpu_in_reset(adev)) { | |
3292 | if (!hive || adev->gmc.xgmi.node_id == 0) | |
3293 | ras_event_mgr_init(ras->event_mgr); | |
3294 | } | |
3295 | ||
3296 | if (hive) | |
3297 | amdgpu_put_xgmi_hive(hive); | |
3298 | } | |
3299 | ||
c030f2e4 | 3300 | int amdgpu_ras_init(struct amdgpu_device *adev) |
3301 | { | |
3302 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
4e644fff | 3303 | int r; |
c030f2e4 | 3304 | |
b404ae82 | 3305 | if (con) |
c030f2e4 | 3306 | return 0; |
3307 | ||
091411be | 3308 | con = kzalloc(sizeof(*con) + |
640ae42e JC |
3309 | sizeof(struct ras_manager) * AMDGPU_RAS_BLOCK_COUNT + |
3310 | sizeof(struct ras_manager) * AMDGPU_RAS_MCA_BLOCK_COUNT, | |
091411be | 3311 | GFP_KERNEL); |
c030f2e4 | 3312 | if (!con) |
3313 | return -ENOMEM; | |
3314 | ||
05adfd80 LT |
3315 | con->adev = adev; |
3316 | INIT_DELAYED_WORK(&con->ras_counte_delay_work, amdgpu_ras_counte_dw); | |
3317 | atomic_set(&con->ras_ce_count, 0); | |
3318 | atomic_set(&con->ras_ue_count, 0); | |
3319 | ||
c030f2e4 | 3320 | con->objs = (struct ras_manager *)(con + 1); |
3321 | ||
3322 | amdgpu_ras_set_context(adev, con); | |
3323 | ||
e509965e LT |
3324 | amdgpu_ras_check_supported(adev); |
3325 | ||
7ddd9770 | 3326 | if (!adev->ras_enabled || adev->asic_type == CHIP_VEGA10) { |
970fd197 SY |
3327 | /* set gfx block ras context feature for VEGA20 Gaming |
3328 | * send ras disable cmd to ras ta during ras late init. | |
3329 | */ | |
8ab0d6f0 | 3330 | if (!adev->ras_enabled && adev->asic_type == CHIP_VEGA20) { |
970fd197 SY |
3331 | con->features |= BIT(AMDGPU_RAS_BLOCK__GFX); |
3332 | ||
3333 | return 0; | |
3334 | } | |
3335 | ||
5e91160a | 3336 | r = 0; |
5436ab94 | 3337 | goto release_con; |
fb2a3607 HZ |
3338 | } |
3339 | ||
69691c82 | 3340 | con->update_channel_flag = false; |
c030f2e4 | 3341 | con->features = 0; |
625e5f38 | 3342 | con->schema = 0; |
c030f2e4 | 3343 | INIT_LIST_HEAD(&con->head); |
108c6a63 | 3344 | /* Might need get this flag from vbios. */ |
3345 | con->flags = RAS_DEFAULT_FLAGS; | |
c030f2e4 | 3346 | |
6e36f231 HZ |
3347 | /* initialize nbio ras function ahead of any other |
3348 | * ras functions so hardware fatal error interrupt | |
3349 | * can be enabled as early as possible */ | |
4e8303cf | 3350 | switch (amdgpu_ip_version(adev, NBIO_HWIP, 0)) { |
fdc94d3a HZ |
3351 | case IP_VERSION(7, 4, 0): |
3352 | case IP_VERSION(7, 4, 1): | |
3353 | case IP_VERSION(7, 4, 4): | |
3354 | if (!adev->gmc.xgmi.connected_to_cpu) | |
2e54fe5d | 3355 | adev->nbio.ras = &nbio_v7_4_ras; |
6e36f231 | 3356 | break; |
9af357bc HZ |
3357 | case IP_VERSION(4, 3, 0): |
3358 | if (adev->ras_hw_enabled & (1 << AMDGPU_RAS_BLOCK__DF)) | |
3359 | /* unlike other generation of nbio ras, | |
3360 | * nbio v4_3 only support fatal error interrupt | |
3361 | * to inform software that DF is freezed due to | |
3362 | * system fatal error event. driver should not | |
3363 | * enable nbio ras in such case. Instead, | |
3364 | * check DF RAS */ | |
3365 | adev->nbio.ras = &nbio_v4_3_ras; | |
3366 | break; | |
7692e1ee TZ |
3367 | case IP_VERSION(7, 9, 0): |
3368 | if (!adev->gmc.is_app_apu) | |
3369 | adev->nbio.ras = &nbio_v7_9_ras; | |
3370 | break; | |
6e36f231 HZ |
3371 | default: |
3372 | /* nbio ras is not available */ | |
3373 | break; | |
3374 | } | |
3375 | ||
fdc94d3a HZ |
3376 | /* nbio ras block needs to be enabled ahead of other ras blocks |
3377 | * to handle fatal error */ | |
3378 | r = amdgpu_nbio_ras_sw_init(adev); | |
3379 | if (r) | |
3380 | return r; | |
3381 | ||
2e54fe5d | 3382 | if (adev->nbio.ras && |
3383 | adev->nbio.ras->init_ras_controller_interrupt) { | |
3384 | r = adev->nbio.ras->init_ras_controller_interrupt(adev); | |
4e644fff | 3385 | if (r) |
5436ab94 | 3386 | goto release_con; |
4e644fff HZ |
3387 | } |
3388 | ||
2e54fe5d | 3389 | if (adev->nbio.ras && |
3390 | adev->nbio.ras->init_ras_err_event_athub_interrupt) { | |
3391 | r = adev->nbio.ras->init_ras_err_event_athub_interrupt(adev); | |
4e644fff | 3392 | if (r) |
5436ab94 | 3393 | goto release_con; |
4e644fff HZ |
3394 | } |
3395 | ||
73cb81dc HZ |
3396 | /* Packed socket_id to ras feature mask bits[31:29] */ |
3397 | if (adev->smuio.funcs && | |
3398 | adev->smuio.funcs->get_socket_id) | |
ee9c3031 SY |
3399 | con->features |= ((adev->smuio.funcs->get_socket_id(adev)) << |
3400 | AMDGPU_RAS_FEATURES_SOCKETID_SHIFT); | |
73cb81dc | 3401 | |
625e5f38 AK |
3402 | /* Get RAS schema for particular SOC */ |
3403 | con->schema = amdgpu_get_ras_schema(adev); | |
3404 | ||
5e91160a GC |
3405 | if (amdgpu_ras_fs_init(adev)) { |
3406 | r = -EINVAL; | |
5436ab94 | 3407 | goto release_con; |
5e91160a | 3408 | } |
c030f2e4 | 3409 | |
6952e99c | 3410 | dev_info(adev->dev, "RAS INFO: ras initialized successfully, " |
e509965e | 3411 | "hardware ability[%x] ras_mask[%x]\n", |
8ab0d6f0 | 3412 | adev->ras_hw_enabled, adev->ras_enabled); |
e509965e | 3413 | |
c030f2e4 | 3414 | return 0; |
5436ab94 | 3415 | release_con: |
c030f2e4 | 3416 | amdgpu_ras_set_context(adev, NULL); |
3417 | kfree(con); | |
3418 | ||
5e91160a | 3419 | return r; |
c030f2e4 | 3420 | } |
3421 | ||
8f6368a9 | 3422 | int amdgpu_persistent_edc_harvesting_supported(struct amdgpu_device *adev) |
134d16d5 | 3423 | { |
8107e499 HZ |
3424 | if (adev->gmc.xgmi.connected_to_cpu || |
3425 | adev->gmc.is_app_apu) | |
134d16d5 JC |
3426 | return 1; |
3427 | return 0; | |
3428 | } | |
3429 | ||
3430 | static int amdgpu_persistent_edc_harvesting(struct amdgpu_device *adev, | |
3431 | struct ras_common_if *ras_block) | |
3432 | { | |
3433 | struct ras_query_if info = { | |
3434 | .head = *ras_block, | |
3435 | }; | |
3436 | ||
3437 | if (!amdgpu_persistent_edc_harvesting_supported(adev)) | |
3438 | return 0; | |
3439 | ||
3440 | if (amdgpu_ras_query_error_status(adev, &info) != 0) | |
3441 | DRM_WARN("RAS init harvest failure"); | |
3442 | ||
3443 | if (amdgpu_ras_reset_error_status(adev, ras_block->block) != 0) | |
3444 | DRM_WARN("RAS init harvest reset failure"); | |
3445 | ||
3446 | return 0; | |
3447 | } | |
3448 | ||
e4348849 TZ |
3449 | bool amdgpu_ras_is_poison_mode_supported(struct amdgpu_device *adev) |
3450 | { | |
3451 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3452 | ||
3453 | if (!con) | |
3454 | return false; | |
3455 | ||
3456 | return con->poison_supported; | |
3457 | } | |
3458 | ||
b293e891 | 3459 | /* helper function to handle common stuff in ip late init phase */ |
563285c8 | 3460 | int amdgpu_ras_block_late_init(struct amdgpu_device *adev, |
3461 | struct ras_common_if *ras_block) | |
b293e891 | 3462 | { |
29c9b6cd | 3463 | struct amdgpu_ras_block_object *ras_obj = NULL; |
05adfd80 | 3464 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
4a1c9a44 | 3465 | struct ras_query_if *query_info; |
05adfd80 | 3466 | unsigned long ue_count, ce_count; |
b293e891 HZ |
3467 | int r; |
3468 | ||
3469 | /* disable RAS feature per IP block if it is not supported */ | |
3470 | if (!amdgpu_ras_is_supported(adev, ras_block->block)) { | |
3471 | amdgpu_ras_feature_enable_on_boot(adev, ras_block, 0); | |
3472 | return 0; | |
3473 | } | |
3474 | ||
3475 | r = amdgpu_ras_feature_enable_on_boot(adev, ras_block, 1); | |
3476 | if (r) { | |
9080a18f | 3477 | if (adev->in_suspend || amdgpu_in_reset(adev)) { |
b293e891 HZ |
3478 | /* in resume phase, if fail to enable ras, |
3479 | * clean up all ras fs nodes, and disable ras */ | |
3480 | goto cleanup; | |
3481 | } else | |
3482 | return r; | |
3483 | } | |
3484 | ||
134d16d5 JC |
3485 | /* check for errors on warm reset edc persisant supported ASIC */ |
3486 | amdgpu_persistent_edc_harvesting(adev, ras_block); | |
3487 | ||
b293e891 | 3488 | /* in resume phase, no need to create ras fs node */ |
53b3f8f4 | 3489 | if (adev->in_suspend || amdgpu_in_reset(adev)) |
b293e891 HZ |
3490 | return 0; |
3491 | ||
563285c8 | 3492 | ras_obj = container_of(ras_block, struct amdgpu_ras_block_object, ras_comm); |
36780606 TZ |
3493 | if (ras_obj->ras_cb || (ras_obj->hw_ops && |
3494 | (ras_obj->hw_ops->query_poison_status || | |
3495 | ras_obj->hw_ops->handle_poison_consumption))) { | |
9252d33d | 3496 | r = amdgpu_ras_interrupt_add_handler(adev, ras_block); |
b293e891 | 3497 | if (r) |
779596ce | 3498 | goto cleanup; |
b293e891 HZ |
3499 | } |
3500 | ||
f957138c HZ |
3501 | if (ras_obj->hw_ops && |
3502 | (ras_obj->hw_ops->query_ras_error_count || | |
3503 | ras_obj->hw_ops->query_ras_error_status)) { | |
3504 | r = amdgpu_ras_sysfs_create(adev, ras_block); | |
3505 | if (r) | |
3506 | goto interrupt; | |
b293e891 | 3507 | |
f957138c HZ |
3508 | /* Those are the cached values at init. |
3509 | */ | |
3510 | query_info = kzalloc(sizeof(*query_info), GFP_KERNEL); | |
3511 | if (!query_info) | |
3512 | return -ENOMEM; | |
3513 | memcpy(&query_info->head, ras_block, sizeof(struct ras_common_if)); | |
4a1c9a44 | 3514 | |
f957138c HZ |
3515 | if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count, query_info) == 0) { |
3516 | atomic_set(&con->ras_ce_count, ce_count); | |
3517 | atomic_set(&con->ras_ue_count, ue_count); | |
3518 | } | |
3519 | ||
3520 | kfree(query_info); | |
4d9f771e | 3521 | } |
05adfd80 | 3522 | |
b293e891 | 3523 | return 0; |
779596ce TR |
3524 | |
3525 | interrupt: | |
563285c8 | 3526 | if (ras_obj->ras_cb) |
9252d33d | 3527 | amdgpu_ras_interrupt_remove_handler(adev, ras_block); |
779596ce | 3528 | cleanup: |
b293e891 HZ |
3529 | amdgpu_ras_feature_enable(adev, ras_block, 0); |
3530 | return r; | |
3531 | } | |
3532 | ||
d41ff22a | 3533 | static int amdgpu_ras_block_late_init_default(struct amdgpu_device *adev, |
418abce2 | 3534 | struct ras_common_if *ras_block) |
3535 | { | |
3536 | return amdgpu_ras_block_late_init(adev, ras_block); | |
3537 | } | |
3538 | ||
b293e891 | 3539 | /* helper function to remove ras fs node and interrupt handler */ |
bdb3489c | 3540 | void amdgpu_ras_block_late_fini(struct amdgpu_device *adev, |
3541 | struct ras_common_if *ras_block) | |
3542 | { | |
563285c8 | 3543 | struct amdgpu_ras_block_object *ras_obj; |
bdb3489c | 3544 | if (!ras_block) |
3545 | return; | |
3546 | ||
563285c8 | 3547 | amdgpu_ras_sysfs_remove(adev, ras_block); |
bdb3489c | 3548 | |
563285c8 | 3549 | ras_obj = container_of(ras_block, struct amdgpu_ras_block_object, ras_comm); |
3550 | if (ras_obj->ras_cb) | |
3551 | amdgpu_ras_interrupt_remove_handler(adev, ras_block); | |
bdb3489c | 3552 | } |
3553 | ||
80e0c2cb | 3554 | static void amdgpu_ras_block_late_fini_default(struct amdgpu_device *adev, |
3555 | struct ras_common_if *ras_block) | |
3556 | { | |
3557 | return amdgpu_ras_block_late_fini(adev, ras_block); | |
3558 | } | |
3559 | ||
a564808e | 3560 | /* do some init work after IP late init as dependence. |
511fdbc3 | 3561 | * and it runs in resume/gpu reset/booting up cases. |
a564808e | 3562 | */ |
511fdbc3 | 3563 | void amdgpu_ras_resume(struct amdgpu_device *adev) |
108c6a63 | 3564 | { |
3565 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3566 | struct ras_manager *obj, *tmp; | |
3567 | ||
8ab0d6f0 | 3568 | if (!adev->ras_enabled || !con) { |
970fd197 SY |
3569 | /* clean ras context for VEGA20 Gaming after send ras disable cmd */ |
3570 | amdgpu_release_ras_context(adev); | |
3571 | ||
108c6a63 | 3572 | return; |
970fd197 | 3573 | } |
108c6a63 | 3574 | |
108c6a63 | 3575 | if (con->flags & AMDGPU_RAS_FLAG_INIT_BY_VBIOS) { |
191051a1 | 3576 | /* Set up all other IPs which are not implemented. There is a |
3577 | * tricky thing that IP's actual ras error type should be | |
3578 | * MULTI_UNCORRECTABLE, but as driver does not handle it, so | |
3579 | * ERROR_NONE make sense anyway. | |
3580 | */ | |
3581 | amdgpu_ras_enable_all_features(adev, 1); | |
3582 | ||
3583 | /* We enable ras on all hw_supported block, but as boot | |
3584 | * parameter might disable some of them and one or more IP has | |
3585 | * not implemented yet. So we disable them on behalf. | |
3586 | */ | |
108c6a63 | 3587 | list_for_each_entry_safe(obj, tmp, &con->head, node) { |
3588 | if (!amdgpu_ras_is_supported(adev, obj->head.block)) { | |
3589 | amdgpu_ras_feature_enable(adev, &obj->head, 0); | |
3590 | /* there should be no any reference. */ | |
3591 | WARN_ON(alive_obj(obj)); | |
3592 | } | |
191051a1 | 3593 | } |
108c6a63 | 3594 | } |
3595 | } | |
3596 | ||
511fdbc3 | 3597 | void amdgpu_ras_suspend(struct amdgpu_device *adev) |
3598 | { | |
3599 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3600 | ||
8ab0d6f0 | 3601 | if (!adev->ras_enabled || !con) |
511fdbc3 | 3602 | return; |
3603 | ||
3604 | amdgpu_ras_disable_all_features(adev, 0); | |
3605 | /* Make sure all ras objects are disabled. */ | |
ee9c3031 | 3606 | if (AMDGPU_RAS_GET_FEATURES(con->features)) |
511fdbc3 | 3607 | amdgpu_ras_disable_all_features(adev, 1); |
3608 | } | |
3609 | ||
867e24ca | 3610 | int amdgpu_ras_late_init(struct amdgpu_device *adev) |
3611 | { | |
3612 | struct amdgpu_ras_block_list *node, *tmp; | |
3613 | struct amdgpu_ras_block_object *obj; | |
3614 | int r; | |
3615 | ||
950d6425 SY |
3616 | /* Guest side doesn't need init ras feature */ |
3617 | if (amdgpu_sriov_vf(adev)) | |
3618 | return 0; | |
3619 | ||
9dc57c2a YW |
3620 | amdgpu_ras_event_mgr_init(adev); |
3621 | ||
c0c48f0d YW |
3622 | if (amdgpu_aca_is_enabled(adev)) { |
3623 | if (amdgpu_in_reset(adev)) | |
3624 | r = amdgpu_aca_reset(adev); | |
3625 | else | |
3626 | r = amdgpu_aca_init(adev); | |
3627 | if (r) | |
3628 | return r; | |
3629 | ||
04c4fcd2 | 3630 | amdgpu_ras_set_aca_debug_mode(adev, false); |
c0c48f0d | 3631 | } else { |
04c4fcd2 | 3632 | amdgpu_ras_set_mca_debug_mode(adev, false); |
c0c48f0d | 3633 | } |
201761b5 | 3634 | |
867e24ca | 3635 | list_for_each_entry_safe(node, tmp, &adev->ras_list, node) { |
2866a454 YW |
3636 | obj = node->ras_obj; |
3637 | if (!obj) { | |
867e24ca | 3638 | dev_warn(adev->dev, "Warning: abnormal ras list node.\n"); |
3639 | continue; | |
3640 | } | |
418abce2 | 3641 | |
2866a454 YW |
3642 | if (!amdgpu_ras_is_supported(adev, obj->ras_comm.block)) |
3643 | continue; | |
3644 | ||
867e24ca | 3645 | if (obj->ras_late_init) { |
3646 | r = obj->ras_late_init(adev, &obj->ras_comm); | |
3647 | if (r) { | |
3648 | dev_err(adev->dev, "%s failed to execute ras_late_init! ret:%d\n", | |
3649 | obj->ras_comm.name, r); | |
3650 | return r; | |
3651 | } | |
418abce2 | 3652 | } else |
3653 | amdgpu_ras_block_late_init_default(adev, &obj->ras_comm); | |
867e24ca | 3654 | } |
3655 | ||
3656 | return 0; | |
3657 | } | |
3658 | ||
c030f2e4 | 3659 | /* do some fini work before IP fini as dependence */ |
3660 | int amdgpu_ras_pre_fini(struct amdgpu_device *adev) | |
3661 | { | |
3662 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3663 | ||
8ab0d6f0 | 3664 | if (!adev->ras_enabled || !con) |
c030f2e4 | 3665 | return 0; |
3666 | ||
72c8c97b | 3667 | |
c030f2e4 | 3668 | /* Need disable ras on all IPs here before ip [hw/sw]fini */ |
ee9c3031 | 3669 | if (AMDGPU_RAS_GET_FEATURES(con->features)) |
642c0401 | 3670 | amdgpu_ras_disable_all_features(adev, 0); |
c030f2e4 | 3671 | amdgpu_ras_recovery_fini(adev); |
3672 | return 0; | |
3673 | } | |
3674 | ||
3675 | int amdgpu_ras_fini(struct amdgpu_device *adev) | |
3676 | { | |
d5e8ff5f | 3677 | struct amdgpu_ras_block_list *ras_node, *tmp; |
1f211a82 | 3678 | struct amdgpu_ras_block_object *obj = NULL; |
c030f2e4 | 3679 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); |
3680 | ||
8ab0d6f0 | 3681 | if (!adev->ras_enabled || !con) |
c030f2e4 | 3682 | return 0; |
3683 | ||
1f211a82 | 3684 | list_for_each_entry_safe(ras_node, tmp, &adev->ras_list, node) { |
3685 | if (ras_node->ras_obj) { | |
3686 | obj = ras_node->ras_obj; | |
3687 | if (amdgpu_ras_is_supported(adev, obj->ras_comm.block) && | |
3688 | obj->ras_fini) | |
3689 | obj->ras_fini(adev, &obj->ras_comm); | |
80e0c2cb | 3690 | else |
3691 | amdgpu_ras_block_late_fini_default(adev, &obj->ras_comm); | |
1f211a82 | 3692 | } |
3693 | ||
3694 | /* Clear ras blocks from ras_list and free ras block list node */ | |
3695 | list_del(&ras_node->node); | |
3696 | kfree(ras_node); | |
3697 | } | |
3698 | ||
c030f2e4 | 3699 | amdgpu_ras_fs_fini(adev); |
3700 | amdgpu_ras_interrupt_remove_all(adev); | |
3701 | ||
c0c48f0d YW |
3702 | if (amdgpu_aca_is_enabled(adev)) |
3703 | amdgpu_aca_fini(adev); | |
3704 | ||
ee9c3031 | 3705 | WARN(AMDGPU_RAS_GET_FEATURES(con->features), "Feature mask is not cleared"); |
c030f2e4 | 3706 | |
ee9c3031 | 3707 | if (AMDGPU_RAS_GET_FEATURES(con->features)) |
edfdde90 | 3708 | amdgpu_ras_disable_all_features(adev, 0); |
c030f2e4 | 3709 | |
05adfd80 LT |
3710 | cancel_delayed_work_sync(&con->ras_counte_delay_work); |
3711 | ||
c030f2e4 | 3712 | amdgpu_ras_set_context(adev, NULL); |
3713 | kfree(con); | |
3714 | ||
3715 | return 0; | |
3716 | } | |
7c6e68c7 | 3717 | |
1b6ef74b LL |
3718 | bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev) |
3719 | { | |
3720 | struct amdgpu_ras *ras; | |
3721 | ||
3722 | ras = amdgpu_ras_get_context(adev); | |
3723 | if (!ras) | |
3724 | return false; | |
3725 | ||
3726 | return atomic_read(&ras->fed); | |
3727 | } | |
3728 | ||
3729 | void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status) | |
3730 | { | |
3731 | struct amdgpu_ras *ras; | |
3732 | ||
3733 | ras = amdgpu_ras_get_context(adev); | |
3734 | if (ras) | |
3735 | atomic_set(&ras->fed, !!status); | |
3736 | } | |
3737 | ||
9dc57c2a YW |
3738 | bool amdgpu_ras_event_id_is_valid(struct amdgpu_device *adev, u64 id) |
3739 | { | |
3740 | return !(id & BIT_ULL(63)); | |
3741 | } | |
3742 | ||
3743 | u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type type) | |
3744 | { | |
3745 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |
3746 | u64 id; | |
3747 | ||
3748 | switch (type) { | |
3749 | case RAS_EVENT_TYPE_ISR: | |
3750 | id = (u64)atomic64_read(&ras->event_mgr->seqnos[type]); | |
3751 | break; | |
3752 | case RAS_EVENT_TYPE_INVALID: | |
3753 | default: | |
3754 | id = BIT_ULL(63) | 0ULL; | |
3755 | break; | |
3756 | } | |
3757 | ||
3758 | return id; | |
3759 | } | |
3760 | ||
7c6e68c7 AG |
3761 | void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) |
3762 | { | |
3763 | if (atomic_cmpxchg(&amdgpu_ras_in_intr, 0, 1) == 0) { | |
2c7cd280 | 3764 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
9dc57c2a | 3765 | u64 event_id = (u64)atomic64_inc_return(&ras->event_mgr->seqnos[RAS_EVENT_TYPE_ISR]); |
2c7cd280 | 3766 | |
9dc57c2a YW |
3767 | RAS_EVENT_LOG(adev, event_id, "uncorrectable hardware error" |
3768 | "(ERREVENT_ATHUB_INTERRUPT) detected!\n"); | |
d5ea093e | 3769 | |
b41f742d | 3770 | amdgpu_ras_set_fed(adev, true); |
2c7cd280 | 3771 | ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET; |
61934624 | 3772 | amdgpu_ras_reset_gpu(adev); |
7c6e68c7 AG |
3773 | } |
3774 | } | |
bb5c7235 WS |
3775 | |
3776 | bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev) | |
3777 | { | |
3778 | if (adev->asic_type == CHIP_VEGA20 && | |
3779 | adev->pm.fw_version <= 0x283400) { | |
3780 | return !(amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) && | |
3781 | amdgpu_ras_intr_triggered(); | |
3782 | } | |
3783 | ||
3784 | return false; | |
3785 | } | |
970fd197 SY |
3786 | |
3787 | void amdgpu_release_ras_context(struct amdgpu_device *adev) | |
3788 | { | |
3789 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3790 | ||
3791 | if (!con) | |
3792 | return; | |
3793 | ||
8ab0d6f0 | 3794 | if (!adev->ras_enabled && con->features & BIT(AMDGPU_RAS_BLOCK__GFX)) { |
970fd197 SY |
3795 | con->features &= ~BIT(AMDGPU_RAS_BLOCK__GFX); |
3796 | amdgpu_ras_set_context(adev, NULL); | |
3797 | kfree(con); | |
3798 | } | |
3799 | } | |
12b2cab7 MJ |
3800 | |
3801 | #ifdef CONFIG_X86_MCE_AMD | |
3802 | static struct amdgpu_device *find_adev(uint32_t node_id) | |
3803 | { | |
12b2cab7 MJ |
3804 | int i; |
3805 | struct amdgpu_device *adev = NULL; | |
3806 | ||
91a1a52d MJ |
3807 | for (i = 0; i < mce_adev_list.num_gpu; i++) { |
3808 | adev = mce_adev_list.devs[i]; | |
12b2cab7 | 3809 | |
91a1a52d | 3810 | if (adev && adev->gmc.xgmi.connected_to_cpu && |
12b2cab7 MJ |
3811 | adev->gmc.xgmi.physical_node_id == node_id) |
3812 | break; | |
3813 | adev = NULL; | |
3814 | } | |
3815 | ||
12b2cab7 MJ |
3816 | return adev; |
3817 | } | |
3818 | ||
3819 | #define GET_MCA_IPID_GPUID(m) (((m) >> 44) & 0xF) | |
3820 | #define GET_UMC_INST(m) (((m) >> 21) & 0x7) | |
3821 | #define GET_CHAN_INDEX(m) ((((m) >> 12) & 0x3) | (((m) >> 18) & 0x4)) | |
3822 | #define GPU_ID_OFFSET 8 | |
3823 | ||
3824 | static int amdgpu_bad_page_notifier(struct notifier_block *nb, | |
3825 | unsigned long val, void *data) | |
3826 | { | |
3827 | struct mce *m = (struct mce *)data; | |
3828 | struct amdgpu_device *adev = NULL; | |
3829 | uint32_t gpu_id = 0; | |
cd4c99f1 | 3830 | uint32_t umc_inst = 0, ch_inst = 0; |
12b2cab7 MJ |
3831 | |
3832 | /* | |
3833 | * If the error was generated in UMC_V2, which belongs to GPU UMCs, | |
3834 | * and error occurred in DramECC (Extended error code = 0) then only | |
3835 | * process the error, else bail out. | |
3836 | */ | |
91f75eb4 | 3837 | if (!m || !((smca_get_bank_type(m->extcpu, m->bank) == SMCA_UMC_V2) && |
12b2cab7 MJ |
3838 | (XEC(m->status, 0x3f) == 0x0))) |
3839 | return NOTIFY_DONE; | |
3840 | ||
3841 | /* | |
3842 | * If it is correctable error, return. | |
3843 | */ | |
3844 | if (mce_is_correctable(m)) | |
3845 | return NOTIFY_OK; | |
3846 | ||
3847 | /* | |
3848 | * GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. | |
3849 | */ | |
3850 | gpu_id = GET_MCA_IPID_GPUID(m->ipid) - GPU_ID_OFFSET; | |
3851 | ||
3852 | adev = find_adev(gpu_id); | |
3853 | if (!adev) { | |
3854 | DRM_WARN("%s: Unable to find adev for gpu_id: %d\n", __func__, | |
3855 | gpu_id); | |
3856 | return NOTIFY_DONE; | |
3857 | } | |
3858 | ||
3859 | /* | |
3860 | * If it is uncorrectable error, then find out UMC instance and | |
3861 | * channel index. | |
3862 | */ | |
3863 | umc_inst = GET_UMC_INST(m->ipid); | |
3864 | ch_inst = GET_CHAN_INDEX(m->ipid); | |
3865 | ||
3866 | dev_info(adev->dev, "Uncorrectable error detected in UMC inst: %d, chan_idx: %d", | |
3867 | umc_inst, ch_inst); | |
3868 | ||
24b82292 TZ |
3869 | if (!amdgpu_umc_page_retirement_mca(adev, m->addr, ch_inst, umc_inst)) |
3870 | return NOTIFY_OK; | |
3871 | else | |
6c0ca748 | 3872 | return NOTIFY_DONE; |
12b2cab7 MJ |
3873 | } |
3874 | ||
3875 | static struct notifier_block amdgpu_bad_page_nb = { | |
3876 | .notifier_call = amdgpu_bad_page_notifier, | |
3877 | .priority = MCE_PRIO_UC, | |
3878 | }; | |
3879 | ||
91a1a52d | 3880 | static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev) |
12b2cab7 | 3881 | { |
91a1a52d MJ |
3882 | /* |
3883 | * Add the adev to the mce_adev_list. | |
3884 | * During mode2 reset, amdgpu device is temporarily | |
3885 | * removed from the mgpu_info list which can cause | |
3886 | * page retirement to fail. | |
3887 | * Use this list instead of mgpu_info to find the amdgpu | |
3888 | * device on which the UMC error was reported. | |
3889 | */ | |
3890 | mce_adev_list.devs[mce_adev_list.num_gpu++] = adev; | |
3891 | ||
12b2cab7 MJ |
3892 | /* |
3893 | * Register the x86 notifier only once | |
3894 | * with MCE subsystem. | |
3895 | */ | |
3896 | if (notifier_registered == false) { | |
3897 | mce_register_decode_chain(&amdgpu_bad_page_nb); | |
3898 | notifier_registered = true; | |
3899 | } | |
3900 | } | |
3901 | #endif | |
7cab2124 | 3902 | |
b6efdb02 | 3903 | struct amdgpu_ras *amdgpu_ras_get_context(struct amdgpu_device *adev) |
7cab2124 | 3904 | { |
3905 | if (!adev) | |
3906 | return NULL; | |
3907 | ||
3908 | return adev->psp.ras_context.ras; | |
3909 | } | |
3910 | ||
b6efdb02 | 3911 | int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_con) |
7cab2124 | 3912 | { |
3913 | if (!adev) | |
69f91d32 | 3914 | return -EINVAL; |
7cab2124 | 3915 | |
3916 | adev->psp.ras_context.ras = ras_con; | |
3917 | return 0; | |
3918 | } | |
3919 | ||
3920 | /* check if ras is supported on block, say, sdma, gfx */ | |
3921 | int amdgpu_ras_is_supported(struct amdgpu_device *adev, | |
3922 | unsigned int block) | |
3923 | { | |
8f453c51 | 3924 | int ret = 0; |
7cab2124 | 3925 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
3926 | ||
3927 | if (block >= AMDGPU_RAS_BLOCK_COUNT) | |
3928 | return 0; | |
8f453c51 YC |
3929 | |
3930 | ret = ras && (adev->ras_enabled & (1 << block)); | |
3931 | ||
3932 | /* For the special asic with mem ecc enabled but sram ecc | |
3933 | * not enabled, even if the ras block is not supported on | |
3934 | * .ras_enabled, if the asic supports poison mode and the | |
3935 | * ras block has ras configuration, it can be considered | |
3936 | * that the ras block supports ras function. | |
3937 | */ | |
3938 | if (!ret && | |
bc0f8080 CL |
3939 | (block == AMDGPU_RAS_BLOCK__GFX || |
3940 | block == AMDGPU_RAS_BLOCK__SDMA || | |
3941 | block == AMDGPU_RAS_BLOCK__VCN || | |
3942 | block == AMDGPU_RAS_BLOCK__JPEG) && | |
7ec11c2f | 3943 | (amdgpu_ras_mask & (1 << block)) && |
8f453c51 YC |
3944 | amdgpu_ras_is_poison_mode_supported(adev) && |
3945 | amdgpu_ras_get_ras_block(adev, block, 0)) | |
3946 | ret = 1; | |
3947 | ||
3948 | return ret; | |
7cab2124 | 3949 | } |
3950 | ||
3951 | int amdgpu_ras_reset_gpu(struct amdgpu_device *adev) | |
3952 | { | |
3953 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |
3954 | ||
3955 | if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0) | |
25a2b22e | 3956 | amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work); |
7cab2124 | 3957 | return 0; |
3958 | } | |
3959 | ||
201761b5 | 3960 | int amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable) |
8096df76 TZ |
3961 | { |
3962 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
201761b5 | 3963 | int ret = 0; |
8096df76 | 3964 | |
201761b5 LL |
3965 | if (con) { |
3966 | ret = amdgpu_mca_smu_set_debug_mode(adev, enable); | |
3967 | if (!ret) | |
04c4fcd2 | 3968 | con->is_aca_debug_mode = enable; |
201761b5 LL |
3969 | } |
3970 | ||
3971 | return ret; | |
8096df76 TZ |
3972 | } |
3973 | ||
33dcda51 YW |
3974 | int amdgpu_ras_set_aca_debug_mode(struct amdgpu_device *adev, bool enable) |
3975 | { | |
3976 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
3977 | int ret = 0; | |
3978 | ||
3979 | if (con) { | |
04c4fcd2 YW |
3980 | if (amdgpu_aca_is_enabled(adev)) |
3981 | ret = amdgpu_aca_smu_set_debug_mode(adev, enable); | |
3982 | else | |
3983 | ret = amdgpu_mca_smu_set_debug_mode(adev, enable); | |
33dcda51 | 3984 | if (!ret) |
04c4fcd2 | 3985 | con->is_aca_debug_mode = enable; |
33dcda51 YW |
3986 | } |
3987 | ||
3988 | return ret; | |
3989 | } | |
3990 | ||
04c4fcd2 | 3991 | bool amdgpu_ras_get_aca_debug_mode(struct amdgpu_device *adev) |
8096df76 TZ |
3992 | { |
3993 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
04c4fcd2 | 3994 | const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; |
8096df76 TZ |
3995 | const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; |
3996 | ||
3997 | if (!con) | |
3998 | return false; | |
3999 | ||
04c4fcd2 YW |
4000 | if ((amdgpu_aca_is_enabled(adev) && smu_funcs && smu_funcs->set_debug_mode) || |
4001 | (!amdgpu_aca_is_enabled(adev) && mca_funcs && mca_funcs->mca_set_debug_mode)) | |
4002 | return con->is_aca_debug_mode; | |
8096df76 TZ |
4003 | else |
4004 | return true; | |
4005 | } | |
7cab2124 | 4006 | |
8cc0f566 HZ |
4007 | bool amdgpu_ras_get_error_query_mode(struct amdgpu_device *adev, |
4008 | unsigned int *error_query_mode) | |
4009 | { | |
4010 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
4011 | const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs; | |
04c4fcd2 | 4012 | const struct aca_smu_funcs *smu_funcs = adev->aca.smu_funcs; |
8cc0f566 HZ |
4013 | |
4014 | if (!con) { | |
4015 | *error_query_mode = AMDGPU_RAS_INVALID_ERROR_QUERY; | |
4016 | return false; | |
4017 | } | |
4018 | ||
04c4fcd2 | 4019 | if ((smu_funcs && smu_funcs->set_debug_mode) || (mca_funcs && mca_funcs->mca_set_debug_mode)) |
8cc0f566 | 4020 | *error_query_mode = |
04c4fcd2 | 4021 | (con->is_aca_debug_mode) ? AMDGPU_RAS_DIRECT_ERROR_QUERY : AMDGPU_RAS_FIRMWARE_ERROR_QUERY; |
8cc0f566 HZ |
4022 | else |
4023 | *error_query_mode = AMDGPU_RAS_DIRECT_ERROR_QUERY; | |
4024 | ||
4025 | return true; | |
4026 | } | |
4027 | ||
6492e1b0 | 4028 | /* Register each ip ras block into amdgpu ras */ |
4029 | int amdgpu_ras_register_ras_block(struct amdgpu_device *adev, | |
b6efdb02 | 4030 | struct amdgpu_ras_block_object *ras_block_obj) |
6492e1b0 | 4031 | { |
d5e8ff5f | 4032 | struct amdgpu_ras_block_list *ras_node; |
6492e1b0 | 4033 | if (!adev || !ras_block_obj) |
4034 | return -EINVAL; | |
4035 | ||
d5e8ff5f | 4036 | ras_node = kzalloc(sizeof(*ras_node), GFP_KERNEL); |
4037 | if (!ras_node) | |
4038 | return -ENOMEM; | |
4039 | ||
4040 | INIT_LIST_HEAD(&ras_node->node); | |
4041 | ras_node->ras_obj = ras_block_obj; | |
4042 | list_add_tail(&ras_node->node, &adev->ras_list); | |
6492e1b0 | 4043 | |
4044 | return 0; | |
4045 | } | |
322a7e00 HZ |
4046 | |
4047 | void amdgpu_ras_get_error_type_name(uint32_t err_type, char *err_type_name) | |
4048 | { | |
4049 | if (!err_type_name) | |
4050 | return; | |
4051 | ||
4052 | switch (err_type) { | |
4053 | case AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE: | |
4054 | sprintf(err_type_name, "correctable"); | |
4055 | break; | |
4056 | case AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE: | |
4057 | sprintf(err_type_name, "uncorrectable"); | |
4058 | break; | |
4059 | default: | |
4060 | sprintf(err_type_name, "unknown"); | |
4061 | break; | |
4062 | } | |
4063 | } | |
4064 | ||
4065 | bool amdgpu_ras_inst_get_memory_id_field(struct amdgpu_device *adev, | |
4066 | const struct amdgpu_ras_err_status_reg_entry *reg_entry, | |
4067 | uint32_t instance, | |
4068 | uint32_t *memory_id) | |
4069 | { | |
4070 | uint32_t err_status_lo_data, err_status_lo_offset; | |
4071 | ||
4072 | if (!reg_entry) | |
4073 | return false; | |
4074 | ||
4075 | err_status_lo_offset = | |
4076 | AMDGPU_RAS_REG_ENTRY_OFFSET(reg_entry->hwip, instance, | |
4077 | reg_entry->seg_lo, reg_entry->reg_lo); | |
4078 | err_status_lo_data = RREG32(err_status_lo_offset); | |
4079 | ||
4080 | if ((reg_entry->flags & AMDGPU_RAS_ERR_STATUS_VALID) && | |
4081 | !REG_GET_FIELD(err_status_lo_data, ERR_STATUS_LO, ERR_STATUS_VALID_FLAG)) | |
4082 | return false; | |
4083 | ||
4084 | *memory_id = REG_GET_FIELD(err_status_lo_data, ERR_STATUS_LO, MEMORY_ID); | |
4085 | ||
4086 | return true; | |
4087 | } | |
4088 | ||
4089 | bool amdgpu_ras_inst_get_err_cnt_field(struct amdgpu_device *adev, | |
4090 | const struct amdgpu_ras_err_status_reg_entry *reg_entry, | |
4091 | uint32_t instance, | |
4092 | unsigned long *err_cnt) | |
4093 | { | |
4094 | uint32_t err_status_hi_data, err_status_hi_offset; | |
4095 | ||
4096 | if (!reg_entry) | |
4097 | return false; | |
4098 | ||
4099 | err_status_hi_offset = | |
4100 | AMDGPU_RAS_REG_ENTRY_OFFSET(reg_entry->hwip, instance, | |
4101 | reg_entry->seg_hi, reg_entry->reg_hi); | |
4102 | err_status_hi_data = RREG32(err_status_hi_offset); | |
4103 | ||
4104 | if ((reg_entry->flags & AMDGPU_RAS_ERR_INFO_VALID) && | |
4105 | !REG_GET_FIELD(err_status_hi_data, ERR_STATUS_HI, ERR_INFO_VALID_FLAG)) | |
9b337b7d HZ |
4106 | /* keep the check here in case we need to refer to the result later */ |
4107 | dev_dbg(adev->dev, "Invalid err_info field\n"); | |
322a7e00 HZ |
4108 | |
4109 | /* read err count */ | |
4110 | *err_cnt = REG_GET_FIELD(err_status_hi_data, ERR_STATUS, ERR_CNT); | |
4111 | ||
4112 | return true; | |
4113 | } | |
4114 | ||
4115 | void amdgpu_ras_inst_query_ras_error_count(struct amdgpu_device *adev, | |
4116 | const struct amdgpu_ras_err_status_reg_entry *reg_list, | |
4117 | uint32_t reg_list_size, | |
4118 | const struct amdgpu_ras_memory_id_entry *mem_list, | |
4119 | uint32_t mem_list_size, | |
4120 | uint32_t instance, | |
4121 | uint32_t err_type, | |
4122 | unsigned long *err_count) | |
4123 | { | |
4124 | uint32_t memory_id; | |
4125 | unsigned long err_cnt; | |
4126 | char err_type_name[16]; | |
4127 | uint32_t i, j; | |
4128 | ||
4129 | for (i = 0; i < reg_list_size; i++) { | |
9b337b7d HZ |
4130 | /* query memory_id from err_status_lo */ |
4131 | if (!amdgpu_ras_inst_get_memory_id_field(adev, ®_list[i], | |
4132 | instance, &memory_id)) | |
4133 | continue; | |
4134 | ||
322a7e00 HZ |
4135 | /* query err_cnt from err_status_hi */ |
4136 | if (!amdgpu_ras_inst_get_err_cnt_field(adev, ®_list[i], | |
4137 | instance, &err_cnt) || | |
4138 | !err_cnt) | |
4139 | continue; | |
4140 | ||
322a7e00 HZ |
4141 | *err_count += err_cnt; |
4142 | ||
4143 | /* log the errors */ | |
4144 | amdgpu_ras_get_error_type_name(err_type, err_type_name); | |
4145 | if (!mem_list) { | |
4146 | /* memory_list is not supported */ | |
4147 | dev_info(adev->dev, | |
4148 | "%ld %s hardware errors detected in %s, instance: %d, memory_id: %d\n", | |
4149 | err_cnt, err_type_name, | |
4150 | reg_list[i].block_name, | |
4151 | instance, memory_id); | |
4152 | } else { | |
4153 | for (j = 0; j < mem_list_size; j++) { | |
4154 | if (memory_id == mem_list[j].memory_id) { | |
4155 | dev_info(adev->dev, | |
4156 | "%ld %s hardware errors detected in %s, instance: %d, memory block: %s\n", | |
4157 | err_cnt, err_type_name, | |
4158 | reg_list[i].block_name, | |
4159 | instance, mem_list[j].name); | |
4160 | break; | |
4161 | } | |
4162 | } | |
4163 | } | |
4164 | } | |
4165 | } | |
e53a3250 HZ |
4166 | |
4167 | void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev, | |
4168 | const struct amdgpu_ras_err_status_reg_entry *reg_list, | |
4169 | uint32_t reg_list_size, | |
4170 | uint32_t instance) | |
4171 | { | |
4172 | uint32_t err_status_lo_offset, err_status_hi_offset; | |
4173 | uint32_t i; | |
4174 | ||
4175 | for (i = 0; i < reg_list_size; i++) { | |
4176 | err_status_lo_offset = | |
4177 | AMDGPU_RAS_REG_ENTRY_OFFSET(reg_list[i].hwip, instance, | |
4178 | reg_list[i].seg_lo, reg_list[i].reg_lo); | |
4179 | err_status_hi_offset = | |
4180 | AMDGPU_RAS_REG_ENTRY_OFFSET(reg_list[i].hwip, instance, | |
4181 | reg_list[i].seg_hi, reg_list[i].reg_hi); | |
4182 | WREG32(err_status_lo_offset, 0); | |
4183 | WREG32(err_status_hi_offset, 0); | |
4184 | } | |
4185 | } | |
5b1270be YW |
4186 | |
4187 | int amdgpu_ras_error_data_init(struct ras_err_data *err_data) | |
4188 | { | |
4189 | memset(err_data, 0, sizeof(*err_data)); | |
4190 | ||
4191 | INIT_LIST_HEAD(&err_data->err_node_list); | |
4192 | ||
4193 | return 0; | |
4194 | } | |
4195 | ||
4196 | static void amdgpu_ras_error_node_release(struct ras_err_node *err_node) | |
4197 | { | |
4198 | if (!err_node) | |
4199 | return; | |
4200 | ||
4201 | list_del(&err_node->node); | |
4202 | kvfree(err_node); | |
4203 | } | |
4204 | ||
4205 | void amdgpu_ras_error_data_fini(struct ras_err_data *err_data) | |
4206 | { | |
4207 | struct ras_err_node *err_node, *tmp; | |
4208 | ||
8a656611 | 4209 | list_for_each_entry_safe(err_node, tmp, &err_data->err_node_list, node) |
5b1270be | 4210 | amdgpu_ras_error_node_release(err_node); |
5b1270be YW |
4211 | } |
4212 | ||
4213 | static struct ras_err_node *amdgpu_ras_error_find_node_by_id(struct ras_err_data *err_data, | |
4214 | struct amdgpu_smuio_mcm_config_info *mcm_info) | |
4215 | { | |
4216 | struct ras_err_node *err_node; | |
4217 | struct amdgpu_smuio_mcm_config_info *ref_id; | |
4218 | ||
4219 | if (!err_data || !mcm_info) | |
4220 | return NULL; | |
4221 | ||
4222 | for_each_ras_error(err_node, err_data) { | |
4223 | ref_id = &err_node->err_info.mcm_info; | |
5b1270be | 4224 | |
53d4d779 YW |
4225 | if (mcm_info->socket_id == ref_id->socket_id && |
4226 | mcm_info->die_id == ref_id->die_id) | |
4227 | return err_node; | |
5b1270be YW |
4228 | } |
4229 | ||
4230 | return NULL; | |
4231 | } | |
4232 | ||
4233 | static struct ras_err_node *amdgpu_ras_error_node_new(void) | |
4234 | { | |
4235 | struct ras_err_node *err_node; | |
4236 | ||
4237 | err_node = kvzalloc(sizeof(*err_node), GFP_KERNEL); | |
4238 | if (!err_node) | |
4239 | return NULL; | |
4240 | ||
4241 | INIT_LIST_HEAD(&err_node->node); | |
4242 | ||
4243 | return err_node; | |
4244 | } | |
4245 | ||
dbf3850d YW |
4246 | static int ras_err_info_cmp(void *priv, const struct list_head *a, const struct list_head *b) |
4247 | { | |
4248 | struct ras_err_node *nodea = container_of(a, struct ras_err_node, node); | |
4249 | struct ras_err_node *nodeb = container_of(b, struct ras_err_node, node); | |
4250 | struct amdgpu_smuio_mcm_config_info *infoa = &nodea->err_info.mcm_info; | |
4251 | struct amdgpu_smuio_mcm_config_info *infob = &nodeb->err_info.mcm_info; | |
4252 | ||
4253 | if (unlikely(infoa->socket_id != infob->socket_id)) | |
4254 | return infoa->socket_id - infob->socket_id; | |
4255 | else | |
4256 | return infoa->die_id - infob->die_id; | |
4257 | ||
4258 | return 0; | |
4259 | } | |
4260 | ||
5b1270be | 4261 | static struct ras_err_info *amdgpu_ras_error_get_info(struct ras_err_data *err_data, |
0795b5d2 | 4262 | struct amdgpu_smuio_mcm_config_info *mcm_info) |
5b1270be YW |
4263 | { |
4264 | struct ras_err_node *err_node; | |
4265 | ||
4266 | err_node = amdgpu_ras_error_find_node_by_id(err_data, mcm_info); | |
4267 | if (err_node) | |
4268 | return &err_node->err_info; | |
4269 | ||
4270 | err_node = amdgpu_ras_error_node_new(); | |
4271 | if (!err_node) | |
4272 | return NULL; | |
4273 | ||
0795b5d2 | 4274 | INIT_LIST_HEAD(&err_node->err_info.err_addr_list); |
5b1270be | 4275 | |
0795b5d2 | 4276 | memcpy(&err_node->err_info.mcm_info, mcm_info, sizeof(*mcm_info)); |
9f91e983 | 4277 | |
5b1270be YW |
4278 | err_data->err_list_count++; |
4279 | list_add_tail(&err_node->node, &err_data->err_node_list); | |
dbf3850d | 4280 | list_sort(NULL, &err_data->err_node_list, ras_err_info_cmp); |
5b1270be YW |
4281 | |
4282 | return &err_node->err_info; | |
4283 | } | |
4284 | ||
0795b5d2 YC |
4285 | void amdgpu_ras_add_mca_err_addr(struct ras_err_info *err_info, struct ras_err_addr *err_addr) |
4286 | { | |
4287 | struct ras_err_addr *mca_err_addr; | |
4288 | ||
95b4063d YC |
4289 | /* This function will be retired. */ |
4290 | return; | |
0795b5d2 YC |
4291 | mca_err_addr = kzalloc(sizeof(*mca_err_addr), GFP_KERNEL); |
4292 | if (!mca_err_addr) | |
4293 | return; | |
4294 | ||
4295 | INIT_LIST_HEAD(&mca_err_addr->node); | |
4296 | ||
4297 | mca_err_addr->err_status = err_addr->err_status; | |
4298 | mca_err_addr->err_ipid = err_addr->err_ipid; | |
4299 | mca_err_addr->err_addr = err_addr->err_addr; | |
4300 | ||
4301 | list_add_tail(&mca_err_addr->node, &err_info->err_addr_list); | |
4302 | } | |
4303 | ||
4304 | void amdgpu_ras_del_mca_err_addr(struct ras_err_info *err_info, struct ras_err_addr *mca_err_addr) | |
4305 | { | |
4306 | list_del(&mca_err_addr->node); | |
4307 | kfree(mca_err_addr); | |
4308 | } | |
4309 | ||
5b1270be | 4310 | int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data, |
9f91e983 YC |
4311 | struct amdgpu_smuio_mcm_config_info *mcm_info, |
4312 | struct ras_err_addr *err_addr, u64 count) | |
5b1270be YW |
4313 | { |
4314 | struct ras_err_info *err_info; | |
4315 | ||
4316 | if (!err_data || !mcm_info) | |
4317 | return -EINVAL; | |
4318 | ||
4319 | if (!count) | |
4320 | return 0; | |
4321 | ||
0795b5d2 | 4322 | err_info = amdgpu_ras_error_get_info(err_data, mcm_info); |
5b1270be YW |
4323 | if (!err_info) |
4324 | return -EINVAL; | |
4325 | ||
0795b5d2 YC |
4326 | if (err_addr && err_addr->err_status) |
4327 | amdgpu_ras_add_mca_err_addr(err_info, err_addr); | |
4328 | ||
5b1270be YW |
4329 | err_info->ue_count += count; |
4330 | err_data->ue_count += count; | |
4331 | ||
4332 | return 0; | |
4333 | } | |
4334 | ||
4335 | int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data, | |
9f91e983 YC |
4336 | struct amdgpu_smuio_mcm_config_info *mcm_info, |
4337 | struct ras_err_addr *err_addr, u64 count) | |
5b1270be YW |
4338 | { |
4339 | struct ras_err_info *err_info; | |
4340 | ||
4341 | if (!err_data || !mcm_info) | |
4342 | return -EINVAL; | |
4343 | ||
4344 | if (!count) | |
4345 | return 0; | |
4346 | ||
0795b5d2 | 4347 | err_info = amdgpu_ras_error_get_info(err_data, mcm_info); |
5b1270be YW |
4348 | if (!err_info) |
4349 | return -EINVAL; | |
4350 | ||
4351 | err_info->ce_count += count; | |
4352 | err_data->ce_count += count; | |
4353 | ||
4354 | return 0; | |
4355 | } | |
cce4febb | 4356 | |
46e2231c CL |
4357 | int amdgpu_ras_error_statistic_de_count(struct ras_err_data *err_data, |
4358 | struct amdgpu_smuio_mcm_config_info *mcm_info, | |
4359 | struct ras_err_addr *err_addr, u64 count) | |
4360 | { | |
4361 | struct ras_err_info *err_info; | |
4362 | ||
4363 | if (!err_data || !mcm_info) | |
4364 | return -EINVAL; | |
4365 | ||
4366 | if (!count) | |
4367 | return 0; | |
4368 | ||
0795b5d2 | 4369 | err_info = amdgpu_ras_error_get_info(err_data, mcm_info); |
46e2231c CL |
4370 | if (!err_info) |
4371 | return -EINVAL; | |
4372 | ||
0795b5d2 YC |
4373 | if (err_addr && err_addr->err_status) |
4374 | amdgpu_ras_add_mca_err_addr(err_info, err_addr); | |
4375 | ||
46e2231c CL |
4376 | err_info->de_count += count; |
4377 | err_data->de_count += count; | |
4378 | ||
4379 | return 0; | |
4380 | } | |
4381 | ||
cce4febb HZ |
4382 | #define mmMP0_SMN_C2PMSG_92 0x1609C |
4383 | #define mmMP0_SMN_C2PMSG_126 0x160BE | |
4384 | static void amdgpu_ras_boot_time_error_reporting(struct amdgpu_device *adev, | |
4385 | u32 instance, u32 boot_error) | |
4386 | { | |
4387 | u32 socket_id, aid_id, hbm_id; | |
4388 | u32 reg_data; | |
4389 | u64 reg_addr; | |
4390 | ||
4391 | socket_id = AMDGPU_RAS_GPU_ERR_SOCKET_ID(boot_error); | |
4392 | aid_id = AMDGPU_RAS_GPU_ERR_AID_ID(boot_error); | |
4393 | hbm_id = AMDGPU_RAS_GPU_ERR_HBM_ID(boot_error); | |
4394 | ||
4395 | /* The pattern for smn addressing in other SOC could be different from | |
4396 | * the one for aqua_vanjaram. We should revisit the code if the pattern | |
4397 | * is changed. In such case, replace the aqua_vanjaram implementation | |
4398 | * with more common helper */ | |
4399 | reg_addr = (mmMP0_SMN_C2PMSG_92 << 2) + | |
4400 | aqua_vanjaram_encode_ext_smn_addressing(instance); | |
4401 | ||
4402 | reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); | |
4403 | dev_err(adev->dev, "socket: %d, aid: %d, firmware boot failed, fw status is 0x%x\n", | |
4404 | socket_id, aid_id, reg_data); | |
4405 | ||
4406 | if (AMDGPU_RAS_GPU_ERR_MEM_TRAINING(boot_error)) | |
4407 | dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, memory training failed\n", | |
4408 | socket_id, aid_id, hbm_id); | |
4409 | ||
4410 | if (AMDGPU_RAS_GPU_ERR_FW_LOAD(boot_error)) | |
4411 | dev_info(adev->dev, "socket: %d, aid: %d, firmware load failed at boot time\n", | |
4412 | socket_id, aid_id); | |
4413 | ||
4414 | if (AMDGPU_RAS_GPU_ERR_WAFL_LINK_TRAINING(boot_error)) | |
4415 | dev_info(adev->dev, "socket: %d, aid: %d, wafl link training failed\n", | |
4416 | socket_id, aid_id); | |
4417 | ||
4418 | if (AMDGPU_RAS_GPU_ERR_XGMI_LINK_TRAINING(boot_error)) | |
4419 | dev_info(adev->dev, "socket: %d, aid: %d, xgmi link training failed\n", | |
4420 | socket_id, aid_id); | |
4421 | ||
4422 | if (AMDGPU_RAS_GPU_ERR_USR_CP_LINK_TRAINING(boot_error)) | |
4423 | dev_info(adev->dev, "socket: %d, aid: %d, usr cp link training failed\n", | |
4424 | socket_id, aid_id); | |
4425 | ||
4426 | if (AMDGPU_RAS_GPU_ERR_USR_DP_LINK_TRAINING(boot_error)) | |
4427 | dev_info(adev->dev, "socket: %d, aid: %d, usr dp link training failed\n", | |
4428 | socket_id, aid_id); | |
4429 | ||
4430 | if (AMDGPU_RAS_GPU_ERR_HBM_MEM_TEST(boot_error)) | |
4431 | dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, hbm memory test failed\n", | |
4432 | socket_id, aid_id, hbm_id); | |
4433 | ||
4434 | if (AMDGPU_RAS_GPU_ERR_HBM_BIST_TEST(boot_error)) | |
4435 | dev_info(adev->dev, "socket: %d, aid: %d, hbm: %d, hbm bist test failed\n", | |
4436 | socket_id, aid_id, hbm_id); | |
4437 | } | |
4438 | ||
4439 | static int amdgpu_ras_wait_for_boot_complete(struct amdgpu_device *adev, | |
4440 | u32 instance, u32 *boot_error) | |
4441 | { | |
4442 | u32 reg_addr; | |
4443 | u32 reg_data; | |
4444 | int retry_loop; | |
4445 | ||
1731ba9b HZ |
4446 | reg_addr = (mmMP0_SMN_C2PMSG_92 << 2) + |
4447 | aqua_vanjaram_encode_ext_smn_addressing(instance); | |
4448 | ||
4449 | for (retry_loop = 0; retry_loop < AMDGPU_RAS_BOOT_STATUS_POLLING_LIMIT; retry_loop++) { | |
4450 | reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); | |
4451 | if ((reg_data & AMDGPU_RAS_BOOT_STATUS_MASK) == AMDGPU_RAS_BOOT_STEADY_STATUS) { | |
4452 | *boot_error = AMDGPU_RAS_BOOT_SUCEESS; | |
4453 | return 0; | |
4454 | } | |
4455 | msleep(1); | |
4456 | } | |
4457 | ||
cce4febb HZ |
4458 | /* The pattern for smn addressing in other SOC could be different from |
4459 | * the one for aqua_vanjaram. We should revisit the code if the pattern | |
4460 | * is changed. In such case, replace the aqua_vanjaram implementation | |
4461 | * with more common helper */ | |
4462 | reg_addr = (mmMP0_SMN_C2PMSG_126 << 2) + | |
4463 | aqua_vanjaram_encode_ext_smn_addressing(instance); | |
4464 | ||
1731ba9b | 4465 | for (retry_loop = 0; retry_loop < AMDGPU_RAS_BOOT_STATUS_POLLING_LIMIT; retry_loop++) { |
cce4febb HZ |
4466 | reg_data = amdgpu_device_indirect_rreg_ext(adev, reg_addr); |
4467 | if (AMDGPU_RAS_GPU_ERR_BOOT_STATUS(reg_data)) { | |
4468 | *boot_error = reg_data; | |
4469 | return 0; | |
4470 | } | |
4471 | msleep(1); | |
4472 | } | |
4473 | ||
4474 | *boot_error = reg_data; | |
4475 | return -ETIME; | |
4476 | } | |
4477 | ||
4478 | void amdgpu_ras_query_boot_status(struct amdgpu_device *adev, u32 num_instances) | |
4479 | { | |
4480 | u32 boot_error = 0; | |
4481 | u32 i; | |
4482 | ||
4483 | for (i = 0; i < num_instances; i++) { | |
4484 | if (amdgpu_ras_wait_for_boot_complete(adev, i, &boot_error)) | |
4485 | amdgpu_ras_boot_time_error_reporting(adev, i, boot_error); | |
4486 | } | |
4487 | } | |
af730e08 YC |
4488 | |
4489 | int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn) | |
4490 | { | |
4491 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
4492 | struct amdgpu_vram_mgr *mgr = &adev->mman.vram_mgr; | |
4493 | uint64_t start = pfn << AMDGPU_GPU_PAGE_SHIFT; | |
4494 | int ret = 0; | |
4495 | ||
4496 | mutex_lock(&con->page_rsv_lock); | |
4497 | ret = amdgpu_vram_mgr_query_page_status(mgr, start); | |
4498 | if (ret == -ENOENT) | |
4499 | ret = amdgpu_vram_mgr_reserve_range(mgr, start, AMDGPU_GPU_PAGE_SIZE); | |
4500 | mutex_unlock(&con->page_rsv_lock); | |
4501 | ||
4502 | return ret; | |
4503 | } |