File: | dev/pci/drm/amd/amdgpu/amdgpu_ras.c |
Warning: | line 763, column 3 1st function call argument is an uninitialized value |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
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> | |||
27 | #include <linux/uaccess.h> | |||
28 | #include <linux/reboot.h> | |||
29 | #include <linux/syscalls.h> | |||
30 | #include <linux/pm_runtime.h> | |||
31 | ||||
32 | #include "amdgpu.h" | |||
33 | #include "amdgpu_ras.h" | |||
34 | #include "amdgpu_atomfirmware.h" | |||
35 | #include "amdgpu_xgmi.h" | |||
36 | #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" | |||
37 | #include "atom.h" | |||
38 | #include "amdgpu_reset.h" | |||
39 | ||||
40 | #ifdef CONFIG_X86_MCE_AMD | |||
41 | #include <asm/mce.h> | |||
42 | ||||
43 | static bool_Bool notifier_registered; | |||
44 | #endif | |||
45 | static const char *RAS_FS_NAME = "ras"; | |||
46 | ||||
47 | const char *ras_error_string[] = { | |||
48 | "none", | |||
49 | "parity", | |||
50 | "single_correctable", | |||
51 | "multi_uncorrectable", | |||
52 | "poison", | |||
53 | }; | |||
54 | ||||
55 | const char *ras_block_string[] = { | |||
56 | "umc", | |||
57 | "sdma", | |||
58 | "gfx", | |||
59 | "mmhub", | |||
60 | "athub", | |||
61 | "pcie_bif", | |||
62 | "hdp", | |||
63 | "xgmi_wafl", | |||
64 | "df", | |||
65 | "smn", | |||
66 | "sem", | |||
67 | "mp0", | |||
68 | "mp1", | |||
69 | "fuse", | |||
70 | "mca", | |||
71 | "vcn", | |||
72 | "jpeg", | |||
73 | }; | |||
74 | ||||
75 | const char *ras_mca_block_string[] = { | |||
76 | "mca_mp0", | |||
77 | "mca_mp1", | |||
78 | "mca_mpio", | |||
79 | "mca_iohc", | |||
80 | }; | |||
81 | ||||
82 | struct amdgpu_ras_block_list { | |||
83 | /* ras block link */ | |||
84 | struct list_head node; | |||
85 | ||||
86 | struct amdgpu_ras_block_object *ras_obj; | |||
87 | }; | |||
88 | ||||
89 | const char *get_ras_block_str(struct ras_common_if *ras_block) | |||
90 | { | |||
91 | if (!ras_block) | |||
92 | return "NULL"; | |||
93 | ||||
94 | if (ras_block->block >= AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST) | |||
95 | return "OUT OF RANGE"; | |||
96 | ||||
97 | if (ras_block->block == AMDGPU_RAS_BLOCK__MCA) | |||
98 | return ras_mca_block_string[ras_block->sub_block_index]; | |||
99 | ||||
100 | return ras_block_string[ras_block->block]; | |||
101 | } | |||
102 | ||||
103 | #define ras_block_str(_BLOCK_)(((_BLOCK_) < (sizeof((ras_block_string)) / sizeof((ras_block_string )[0]))) ? ras_block_string[_BLOCK_] : "Out Of Range") \ | |||
104 | (((_BLOCK_) < ARRAY_SIZE(ras_block_string)(sizeof((ras_block_string)) / sizeof((ras_block_string)[0]))) ? ras_block_string[_BLOCK_] : "Out Of Range") | |||
105 | ||||
106 | #define ras_err_str(i)(ras_error_string[ffs(i)]) (ras_error_string[ffs(i)]) | |||
107 | ||||
108 | #define RAS_DEFAULT_FLAGS((0x1 << 0)) (AMDGPU_RAS_FLAG_INIT_BY_VBIOS(0x1 << 0)) | |||
109 | ||||
110 | /* inject address is 52 bits */ | |||
111 | #define RAS_UMC_INJECT_ADDR_LIMIT(0x1ULL << 52) (0x1ULL << 52) | |||
112 | ||||
113 | /* typical ECC bad page rate is 1 bad page per 100MB VRAM */ | |||
114 | #define RAS_BAD_PAGE_COVER(100 * 1024 * 1024ULL) (100 * 1024 * 1024ULL) | |||
115 | ||||
116 | enum amdgpu_ras_retire_page_reservation { | |||
117 | AMDGPU_RAS_RETIRE_PAGE_RESERVED, | |||
118 | AMDGPU_RAS_RETIRE_PAGE_PENDING, | |||
119 | AMDGPU_RAS_RETIRE_PAGE_FAULT, | |||
120 | }; | |||
121 | ||||
122 | atomic_t amdgpu_ras_in_intr = ATOMIC_INIT(0)(0); | |||
123 | ||||
124 | static bool_Bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, | |||
125 | uint64_t addr); | |||
126 | static bool_Bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, | |||
127 | uint64_t addr); | |||
128 | #ifdef CONFIG_X86_MCE_AMD | |||
129 | static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev); | |||
130 | struct mce_notifier_adev_list { | |||
131 | struct amdgpu_device *devs[MAX_GPU_INSTANCE16]; | |||
132 | int num_gpu; | |||
133 | }; | |||
134 | static struct mce_notifier_adev_list mce_adev_list; | |||
135 | #endif | |||
136 | ||||
137 | void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool_Bool ready) | |||
138 | { | |||
139 | if (adev && amdgpu_ras_get_context(adev)) | |||
140 | amdgpu_ras_get_context(adev)->error_query_ready = ready; | |||
141 | } | |||
142 | ||||
143 | static bool_Bool amdgpu_ras_get_error_query_ready(struct amdgpu_device *adev) | |||
144 | { | |||
145 | if (adev && amdgpu_ras_get_context(adev)) | |||
146 | return amdgpu_ras_get_context(adev)->error_query_ready; | |||
147 | ||||
148 | return false0; | |||
149 | } | |||
150 | ||||
151 | static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t address) | |||
152 | { | |||
153 | struct ras_err_data err_data = {0, 0, 0, NULL((void *)0)}; | |||
154 | struct eeprom_table_record err_rec; | |||
155 | ||||
156 | if ((address >= adev->gmc.mc_vram_size) || | |||
157 | (address >= RAS_UMC_INJECT_ADDR_LIMIT(0x1ULL << 52))) { | |||
158 | dev_warn(adev->dev,printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address 0x%llx is invalid.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ) | |||
159 | "RAS WARN: input address 0x%llx is invalid.\n",printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address 0x%llx is invalid.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ) | |||
160 | address)printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address 0x%llx is invalid.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ); | |||
161 | return -EINVAL22; | |||
162 | } | |||
163 | ||||
164 | if (amdgpu_ras_check_bad_page(adev, address)) { | |||
165 | dev_warn(adev->dev,printf("drm:pid%d:%s *WARNING* " "RAS WARN: 0x%llx has already been marked as bad page!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ) | |||
166 | "RAS WARN: 0x%llx has already been marked as bad page!\n",printf("drm:pid%d:%s *WARNING* " "RAS WARN: 0x%llx has already been marked as bad page!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ) | |||
167 | address)printf("drm:pid%d:%s *WARNING* " "RAS WARN: 0x%llx has already been marked as bad page!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , address ); | |||
168 | return 0; | |||
169 | } | |||
170 | ||||
171 | memset(&err_rec, 0x0, sizeof(struct eeprom_table_record))__builtin_memset((&err_rec), (0x0), (sizeof(struct eeprom_table_record ))); | |||
172 | err_data.err_addr = &err_rec; | |||
173 | amdgpu_umc_fill_error_record(&err_data, address, address, 0, 0); | |||
174 | ||||
175 | if (amdgpu_bad_page_threshold != 0) { | |||
176 | amdgpu_ras_add_bad_pages(adev, err_data.err_addr, | |||
177 | err_data.err_addr_cnt); | |||
178 | amdgpu_ras_save_bad_pages(adev); | |||
179 | } | |||
180 | ||||
181 | dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL CORRUPT RAS EEPROM\n")printf("drm:pid%d:%s *WARNING* " "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL CORRUPT RAS EEPROM\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
182 | dev_warn(adev->dev, "Clear EEPROM:\n")printf("drm:pid%d:%s *WARNING* " "Clear EEPROM:\n", ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__); | |||
183 | dev_warn(adev->dev, " echo 1 > /sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n")printf("drm:pid%d:%s *WARNING* " " echo 1 > /sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
184 | ||||
185 | return 0; | |||
186 | } | |||
187 | ||||
188 | #ifdef __linux__ | |||
189 | ||||
190 | static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf, | |||
191 | size_t size, loff_t *pos) | |||
192 | { | |||
193 | struct ras_manager *obj = (struct ras_manager *)file_inode(f)->i_private; | |||
194 | struct ras_query_if info = { | |||
195 | .head = obj->head, | |||
196 | }; | |||
197 | ssize_t s; | |||
198 | char val[128]; | |||
199 | ||||
200 | if (amdgpu_ras_query_error_status(obj->adev, &info)) | |||
201 | return -EINVAL22; | |||
202 | ||||
203 | /* Hardware counter will be reset automatically after the query on Vega20 and Arcturus */ | |||
204 | if (obj->adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2)(((11) << 16) | ((0) << 8) | (2)) && | |||
205 | obj->adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)(((11) << 16) | ((0) << 8) | (4))) { | |||
206 | if (amdgpu_ras_reset_error_status(obj->adev, info.head.block)) | |||
207 | dev_warn(obj->adev->dev, "Failed to reset error counter and error status")printf("drm:pid%d:%s *WARNING* " "Failed to reset error counter and error status" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
208 | } | |||
209 | ||||
210 | s = snprintf(val, sizeof(val), "%s: %lu\n%s: %lu\n", | |||
211 | "ue", info.ue_count, | |||
212 | "ce", info.ce_count); | |||
213 | if (*pos >= s) | |||
214 | return 0; | |||
215 | ||||
216 | s -= *pos; | |||
217 | s = min_t(u64, s, size)({ u64 __min_a = (s); u64 __min_b = (size); __min_a < __min_b ? __min_a : __min_b; }); | |||
218 | ||||
219 | ||||
220 | if (copy_to_user(buf, &val[*pos], s)) | |||
221 | return -EINVAL22; | |||
222 | ||||
223 | *pos += s; | |||
224 | ||||
225 | return s; | |||
226 | } | |||
227 | ||||
228 | static const struct file_operations amdgpu_ras_debugfs_ops = { | |||
229 | .owner = THIS_MODULE((void *)0), | |||
230 | .read = amdgpu_ras_debugfs_read, | |||
231 | .write = NULL((void *)0), | |||
232 | .llseek = default_llseek | |||
233 | }; | |||
234 | ||||
235 | static int amdgpu_ras_find_block_id_by_name(const char *name, int *block_id) | |||
236 | { | |||
237 | int i; | |||
238 | ||||
239 | for (i = 0; i < ARRAY_SIZE(ras_block_string)(sizeof((ras_block_string)) / sizeof((ras_block_string)[0])); i++) { | |||
240 | *block_id = i; | |||
241 | if (strcmp(name, ras_block_string[i]) == 0) | |||
242 | return 0; | |||
243 | } | |||
244 | return -EINVAL22; | |||
245 | } | |||
246 | ||||
247 | static int amdgpu_ras_debugfs_ctrl_parse_data(struct file *f, | |||
248 | const char __user *buf, size_t size, | |||
249 | loff_t *pos, struct ras_debug_if *data) | |||
250 | { | |||
251 | ssize_t s = min_t(u64, 64, size)({ u64 __min_a = (64); u64 __min_b = (size); __min_a < __min_b ? __min_a : __min_b; }); | |||
252 | char str[65]; | |||
253 | char block_name[33]; | |||
254 | char err[9] = "ue"; | |||
255 | int op = -1; | |||
256 | int block_id; | |||
257 | uint32_t sub_block; | |||
258 | u64 address, value; | |||
259 | ||||
260 | if (*pos) | |||
261 | return -EINVAL22; | |||
262 | *pos = size; | |||
263 | ||||
264 | memset(str, 0, sizeof(str))__builtin_memset((str), (0), (sizeof(str))); | |||
265 | memset(data, 0, sizeof(*data))__builtin_memset((data), (0), (sizeof(*data))); | |||
266 | ||||
267 | if (copy_from_user(str, buf, s)) | |||
268 | return -EINVAL22; | |||
269 | ||||
270 | if (sscanf(str, "disable %32s", block_name) == 1) | |||
271 | op = 0; | |||
272 | else if (sscanf(str, "enable %32s %8s", block_name, err) == 2) | |||
273 | op = 1; | |||
274 | else if (sscanf(str, "inject %32s %8s", block_name, err) == 2) | |||
275 | op = 2; | |||
276 | else if (strstr(str, "retire_page") != NULL((void *)0)) | |||
277 | op = 3; | |||
278 | else if (str[0] && str[1] && str[2] && str[3]) | |||
279 | /* ascii string, but commands are not matched. */ | |||
280 | return -EINVAL22; | |||
281 | ||||
282 | if (op != -1) { | |||
283 | if (op == 3) { | |||
284 | if (sscanf(str, "%*s 0x%llx", &address) != 1 && | |||
285 | sscanf(str, "%*s %llu", &address) != 1) | |||
286 | return -EINVAL22; | |||
287 | ||||
288 | data->op = op; | |||
289 | data->inject.address = address; | |||
290 | ||||
291 | return 0; | |||
292 | } | |||
293 | ||||
294 | if (amdgpu_ras_find_block_id_by_name(block_name, &block_id)) | |||
295 | return -EINVAL22; | |||
296 | ||||
297 | data->head.block = block_id; | |||
298 | /* only ue and ce errors are supported */ | |||
299 | if (!memcmp("ue", err, 2)__builtin_memcmp(("ue"), (err), (2))) | |||
300 | data->head.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; | |||
301 | else if (!memcmp("ce", err, 2)__builtin_memcmp(("ce"), (err), (2))) | |||
302 | data->head.type = AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE; | |||
303 | else | |||
304 | return -EINVAL22; | |||
305 | ||||
306 | data->op = op; | |||
307 | ||||
308 | if (op == 2) { | |||
309 | if (sscanf(str, "%*s %*s %*s 0x%x 0x%llx 0x%llx", | |||
310 | &sub_block, &address, &value) != 3 && | |||
311 | sscanf(str, "%*s %*s %*s %u %llu %llu", | |||
312 | &sub_block, &address, &value) != 3) | |||
313 | return -EINVAL22; | |||
314 | data->head.sub_block_index = sub_block; | |||
315 | data->inject.address = address; | |||
316 | data->inject.value = value; | |||
317 | } | |||
318 | } else { | |||
319 | if (size < sizeof(*data)) | |||
320 | return -EINVAL22; | |||
321 | ||||
322 | if (copy_from_user(data, buf, sizeof(*data))) | |||
323 | return -EINVAL22; | |||
324 | } | |||
325 | ||||
326 | return 0; | |||
327 | } | |||
328 | ||||
329 | /** | |||
330 | * DOC: AMDGPU RAS debugfs control interface | |||
331 | * | |||
332 | * The control interface accepts struct ras_debug_if which has two members. | |||
333 | * | |||
334 | * First member: ras_debug_if::head or ras_debug_if::inject. | |||
335 | * | |||
336 | * head is used to indicate which IP block will be under control. | |||
337 | * | |||
338 | * head has four members, they are block, type, sub_block_index, name. | |||
339 | * block: which IP will be under control. | |||
340 | * type: what kind of error will be enabled/disabled/injected. | |||
341 | * sub_block_index: some IPs have subcomponets. say, GFX, sDMA. | |||
342 | * name: the name of IP. | |||
343 | * | |||
344 | * inject has two more members than head, they are address, value. | |||
345 | * As their names indicate, inject operation will write the | |||
346 | * value to the address. | |||
347 | * | |||
348 | * The second member: struct ras_debug_if::op. | |||
349 | * It has three kinds of operations. | |||
350 | * | |||
351 | * - 0: disable RAS on the block. Take ::head as its data. | |||
352 | * - 1: enable RAS on the block. Take ::head as its data. | |||
353 | * - 2: inject errors on the block. Take ::inject as its data. | |||
354 | * | |||
355 | * How to use the interface? | |||
356 | * | |||
357 | * In a program | |||
358 | * | |||
359 | * Copy the struct ras_debug_if in your code and initialize it. | |||
360 | * Write the struct to the control interface. | |||
361 | * | |||
362 | * From shell | |||
363 | * | |||
364 | * .. code-block:: bash | |||
365 | * | |||
366 | * echo "disable <block>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl | |||
367 | * echo "enable <block> <error>" > /sys/kernel/debug/dri/<N>/ras/ras_ctrl | |||
368 | * echo "inject <block> <error> <sub-block> <address> <value> > /sys/kernel/debug/dri/<N>/ras/ras_ctrl | |||
369 | * | |||
370 | * Where N, is the card which you want to affect. | |||
371 | * | |||
372 | * "disable" requires only the block. | |||
373 | * "enable" requires the block and error type. | |||
374 | * "inject" requires the block, error type, address, and value. | |||
375 | * | |||
376 | * The block is one of: umc, sdma, gfx, etc. | |||
377 | * see ras_block_string[] for details | |||
378 | * | |||
379 | * The error type is one of: ue, ce, where, | |||
380 | * ue is multi-uncorrectable | |||
381 | * ce is single-correctable | |||
382 | * | |||
383 | * The sub-block is a the sub-block index, pass 0 if there is no sub-block. | |||
384 | * The address and value are hexadecimal numbers, leading 0x is optional. | |||
385 | * | |||
386 | * For instance, | |||
387 | * | |||
388 | * .. code-block:: bash | |||
389 | * | |||
390 | * echo inject umc ue 0x0 0x0 0x0 > /sys/kernel/debug/dri/0/ras/ras_ctrl | |||
391 | * echo inject umc ce 0 0 0 > /sys/kernel/debug/dri/0/ras/ras_ctrl | |||
392 | * echo disable umc > /sys/kernel/debug/dri/0/ras/ras_ctrl | |||
393 | * | |||
394 | * How to check the result of the operation? | |||
395 | * | |||
396 | * To check disable/enable, see "ras" features at, | |||
397 | * /sys/class/drm/card[0/1/2...]/device/ras/features | |||
398 | * | |||
399 | * To check inject, see the corresponding error count at, | |||
400 | * /sys/class/drm/card[0/1/2...]/device/ras/[gfx|sdma|umc|...]_err_count | |||
401 | * | |||
402 | * .. note:: | |||
403 | * Operations are only allowed on blocks which are supported. | |||
404 | * Check the "ras" mask at /sys/module/amdgpu/parameters/ras_mask | |||
405 | * to see which blocks support RAS on a particular asic. | |||
406 | * | |||
407 | */ | |||
408 | static ssize_t amdgpu_ras_debugfs_ctrl_write(struct file *f, | |||
409 | const char __user *buf, | |||
410 | size_t size, loff_t *pos) | |||
411 | { | |||
412 | struct amdgpu_device *adev = (struct amdgpu_device *)file_inode(f)->i_private; | |||
413 | struct ras_debug_if data; | |||
414 | int ret = 0; | |||
415 | ||||
416 | if (!amdgpu_ras_get_error_query_ready(adev)) { | |||
417 | dev_warn(adev->dev, "RAS WARN: error injection "printf("drm:pid%d:%s *WARNING* " "RAS WARN: error injection " "currently inaccessible\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__) | |||
418 | "currently inaccessible\n")printf("drm:pid%d:%s *WARNING* " "RAS WARN: error injection " "currently inaccessible\n", ({struct cpu_info *__ci; asm volatile ("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid , __func__); | |||
419 | return size; | |||
420 | } | |||
421 | ||||
422 | ret = amdgpu_ras_debugfs_ctrl_parse_data(f, buf, size, pos, &data); | |||
423 | if (ret) | |||
424 | return ret; | |||
425 | ||||
426 | if (data.op == 3) { | |||
427 | ret = amdgpu_reserve_page_direct(adev, data.inject.address); | |||
428 | if (!ret) | |||
429 | return size; | |||
430 | else | |||
431 | return ret; | |||
432 | } | |||
433 | ||||
434 | if (!amdgpu_ras_is_supported(adev, data.head.block)) | |||
435 | return -EINVAL22; | |||
436 | ||||
437 | switch (data.op) { | |||
438 | case 0: | |||
439 | ret = amdgpu_ras_feature_enable(adev, &data.head, 0); | |||
440 | break; | |||
441 | case 1: | |||
442 | ret = amdgpu_ras_feature_enable(adev, &data.head, 1); | |||
443 | break; | |||
444 | case 2: | |||
445 | if ((data.inject.address >= adev->gmc.mc_vram_size) || | |||
446 | (data.inject.address >= RAS_UMC_INJECT_ADDR_LIMIT(0x1ULL << 52))) { | |||
447 | dev_warn(adev->dev, "RAS WARN: input address "printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address " "0x%llx is invalid." , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , data.inject .address) | |||
448 | "0x%llx is invalid.",printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address " "0x%llx is invalid." , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , data.inject .address) | |||
449 | data.inject.address)printf("drm:pid%d:%s *WARNING* " "RAS WARN: input address " "0x%llx is invalid." , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , data.inject .address); | |||
450 | ret = -EINVAL22; | |||
451 | break; | |||
452 | } | |||
453 | ||||
454 | /* umc ce/ue error injection for a bad page is not allowed */ | |||
455 | if ((data.head.block == AMDGPU_RAS_BLOCK__UMC) && | |||
456 | amdgpu_ras_check_bad_page(adev, data.inject.address)) { | |||
457 | dev_warn(adev->dev, "RAS WARN: inject: 0x%llx has "printf("drm:pid%d:%s *WARNING* " "RAS WARN: inject: 0x%llx has " "already been marked as bad!\n", ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , data.inject.address) | |||
458 | "already been marked as bad!\n",printf("drm:pid%d:%s *WARNING* " "RAS WARN: inject: 0x%llx has " "already been marked as bad!\n", ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , data.inject.address) | |||
459 | data.inject.address)printf("drm:pid%d:%s *WARNING* " "RAS WARN: inject: 0x%llx has " "already been marked as bad!\n", ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof (struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p-> ps_pid, __func__ , data.inject.address); | |||
460 | break; | |||
461 | } | |||
462 | ||||
463 | /* data.inject.address is offset instead of absolute gpu address */ | |||
464 | ret = amdgpu_ras_error_inject(adev, &data.inject); | |||
465 | break; | |||
466 | default: | |||
467 | ret = -EINVAL22; | |||
468 | break; | |||
469 | } | |||
470 | ||||
471 | if (ret) | |||
472 | return ret; | |||
473 | ||||
474 | return size; | |||
475 | } | |||
476 | ||||
477 | /** | |||
478 | * DOC: AMDGPU RAS debugfs EEPROM table reset interface | |||
479 | * | |||
480 | * Some boards contain an EEPROM which is used to persistently store a list of | |||
481 | * bad pages which experiences ECC errors in vram. This interface provides | |||
482 | * a way to reset the EEPROM, e.g., after testing error injection. | |||
483 | * | |||
484 | * Usage: | |||
485 | * | |||
486 | * .. code-block:: bash | |||
487 | * | |||
488 | * echo 1 > ../ras/ras_eeprom_reset | |||
489 | * | |||
490 | * will reset EEPROM table to 0 entries. | |||
491 | * | |||
492 | */ | |||
493 | static ssize_t amdgpu_ras_debugfs_eeprom_write(struct file *f, | |||
494 | const char __user *buf, | |||
495 | size_t size, loff_t *pos) | |||
496 | { | |||
497 | struct amdgpu_device *adev = | |||
498 | (struct amdgpu_device *)file_inode(f)->i_private; | |||
499 | int ret; | |||
500 | ||||
501 | ret = amdgpu_ras_eeprom_reset_table( | |||
502 | &(amdgpu_ras_get_context(adev)->eeprom_control)); | |||
503 | ||||
504 | if (!ret) { | |||
505 | /* Something was written to EEPROM. | |||
506 | */ | |||
507 | amdgpu_ras_get_context(adev)->flags = RAS_DEFAULT_FLAGS((0x1 << 0)); | |||
508 | return size; | |||
509 | } else { | |||
510 | return ret; | |||
511 | } | |||
512 | } | |||
513 | ||||
514 | static const struct file_operations amdgpu_ras_debugfs_ctrl_ops = { | |||
515 | .owner = THIS_MODULE((void *)0), | |||
516 | .read = NULL((void *)0), | |||
517 | .write = amdgpu_ras_debugfs_ctrl_write, | |||
518 | .llseek = default_llseek | |||
519 | }; | |||
520 | ||||
521 | static const struct file_operations amdgpu_ras_debugfs_eeprom_ops = { | |||
522 | .owner = THIS_MODULE((void *)0), | |||
523 | .read = NULL((void *)0), | |||
524 | .write = amdgpu_ras_debugfs_eeprom_write, | |||
525 | .llseek = default_llseek | |||
526 | }; | |||
527 | ||||
528 | /** | |||
529 | * DOC: AMDGPU RAS sysfs Error Count Interface | |||
530 | * | |||
531 | * It allows the user to read the error count for each IP block on the gpu through | |||
532 | * /sys/class/drm/card[0/1/2...]/device/ras/[gfx/sdma/...]_err_count | |||
533 | * | |||
534 | * It outputs the multiple lines which report the uncorrected (ue) and corrected | |||
535 | * (ce) error counts. | |||
536 | * | |||
537 | * The format of one line is below, | |||
538 | * | |||
539 | * [ce|ue]: count | |||
540 | * | |||
541 | * Example: | |||
542 | * | |||
543 | * .. code-block:: bash | |||
544 | * | |||
545 | * ue: 0 | |||
546 | * ce: 1 | |||
547 | * | |||
548 | */ | |||
549 | static ssize_t amdgpu_ras_sysfs_read(struct device *dev, | |||
550 | struct device_attribute *attr, char *buf) | |||
551 | { | |||
552 | struct ras_manager *obj = container_of(attr, struct ras_manager, sysfs_attr)({ const __typeof( ((struct ras_manager *)0)->sysfs_attr ) *__mptr = (attr); (struct ras_manager *)( (char *)__mptr - __builtin_offsetof (struct ras_manager, sysfs_attr) );}); | |||
553 | struct ras_query_if info = { | |||
554 | .head = obj->head, | |||
555 | }; | |||
556 | ||||
557 | if (!amdgpu_ras_get_error_query_ready(obj->adev)) | |||
558 | return sysfs_emit(buf, "Query currently inaccessible\n"); | |||
559 | ||||
560 | if (amdgpu_ras_query_error_status(obj->adev, &info)) | |||
561 | return -EINVAL22; | |||
562 | ||||
563 | if (obj->adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2)(((11) << 16) | ((0) << 8) | (2)) && | |||
564 | obj->adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)(((11) << 16) | ((0) << 8) | (4))) { | |||
565 | if (amdgpu_ras_reset_error_status(obj->adev, info.head.block)) | |||
566 | dev_warn(obj->adev->dev, "Failed to reset error counter and error status")printf("drm:pid%d:%s *WARNING* " "Failed to reset error counter and error status" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
567 | } | |||
568 | ||||
569 | return sysfs_emit(buf, "%s: %lu\n%s: %lu\n", "ue", info.ue_count, | |||
570 | "ce", info.ce_count); | |||
571 | } | |||
572 | ||||
573 | #endif /* __linux__ */ | |||
574 | ||||
575 | /* obj begin */ | |||
576 | ||||
577 | #define get_obj(obj)do { (obj)->use++; } while (0) do { (obj)->use++; } while (0) | |||
578 | #define alive_obj(obj)((obj)->use) ((obj)->use) | |||
579 | ||||
580 | static inline void put_obj(struct ras_manager *obj) | |||
581 | { | |||
582 | if (obj && (--obj->use == 0)) | |||
583 | list_del(&obj->node); | |||
584 | if (obj && (obj->use < 0)) | |||
585 | DRM_ERROR("RAS ERROR: Unbalance obj(%s) use\n", get_ras_block_str(&obj->head))__drm_err("RAS ERROR: Unbalance obj(%s) use\n", get_ras_block_str (&obj->head)); | |||
586 | } | |||
587 | ||||
588 | /* make one obj and return it. */ | |||
589 | static struct ras_manager *amdgpu_ras_create_obj(struct amdgpu_device *adev, | |||
590 | struct ras_common_if *head) | |||
591 | { | |||
592 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
593 | struct ras_manager *obj; | |||
594 | ||||
595 | if (!adev->ras_enabled || !con) | |||
596 | return NULL((void *)0); | |||
597 | ||||
598 | if (head->block >= AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST) | |||
599 | return NULL((void *)0); | |||
600 | ||||
601 | if (head->block == AMDGPU_RAS_BLOCK__MCA) { | |||
602 | if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST) | |||
603 | return NULL((void *)0); | |||
604 | ||||
605 | obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index]; | |||
606 | } else | |||
607 | obj = &con->objs[head->block]; | |||
608 | ||||
609 | /* already exist. return obj? */ | |||
610 | if (alive_obj(obj)((obj)->use)) | |||
611 | return NULL((void *)0); | |||
612 | ||||
613 | obj->head = *head; | |||
614 | obj->adev = adev; | |||
615 | list_add(&obj->node, &con->head); | |||
616 | get_obj(obj)do { (obj)->use++; } while (0); | |||
617 | ||||
618 | return obj; | |||
619 | } | |||
620 | ||||
621 | /* return an obj equal to head, or the first when head is NULL */ | |||
622 | struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev, | |||
623 | struct ras_common_if *head) | |||
624 | { | |||
625 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
626 | struct ras_manager *obj; | |||
627 | int i; | |||
628 | ||||
629 | if (!adev->ras_enabled || !con) | |||
630 | return NULL((void *)0); | |||
631 | ||||
632 | if (head) { | |||
633 | if (head->block >= AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST) | |||
634 | return NULL((void *)0); | |||
635 | ||||
636 | if (head->block == AMDGPU_RAS_BLOCK__MCA) { | |||
637 | if (head->sub_block_index >= AMDGPU_RAS_MCA_BLOCK__LAST) | |||
638 | return NULL((void *)0); | |||
639 | ||||
640 | obj = &con->objs[AMDGPU_RAS_BLOCK__LAST + head->sub_block_index]; | |||
641 | } else | |||
642 | obj = &con->objs[head->block]; | |||
643 | ||||
644 | if (alive_obj(obj)((obj)->use)) | |||
645 | return obj; | |||
646 | } else { | |||
647 | for (i = 0; i < AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST + AMDGPU_RAS_MCA_BLOCK_COUNTAMDGPU_RAS_MCA_BLOCK__LAST; i++) { | |||
648 | obj = &con->objs[i]; | |||
649 | if (alive_obj(obj)((obj)->use)) | |||
650 | return obj; | |||
651 | } | |||
652 | } | |||
653 | ||||
654 | return NULL((void *)0); | |||
655 | } | |||
656 | /* obj end */ | |||
657 | ||||
658 | /* feature ctl begin */ | |||
659 | static int amdgpu_ras_is_feature_allowed(struct amdgpu_device *adev, | |||
660 | struct ras_common_if *head) | |||
661 | { | |||
662 | return adev->ras_hw_enabled & BIT(head->block)(1UL << (head->block)); | |||
663 | } | |||
664 | ||||
665 | static int amdgpu_ras_is_feature_enabled(struct amdgpu_device *adev, | |||
666 | struct ras_common_if *head) | |||
667 | { | |||
668 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
669 | ||||
670 | return con->features & BIT(head->block)(1UL << (head->block)); | |||
671 | } | |||
672 | ||||
673 | /* | |||
674 | * if obj is not created, then create one. | |||
675 | * set feature enable flag. | |||
676 | */ | |||
677 | static int __amdgpu_ras_feature_enable(struct amdgpu_device *adev, | |||
678 | struct ras_common_if *head, int enable) | |||
679 | { | |||
680 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
681 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |||
682 | ||||
683 | /* If hardware does not support ras, then do not create obj. | |||
684 | * But if hardware support ras, we can create the obj. | |||
685 | * Ras framework checks con->hw_supported to see if it need do | |||
686 | * corresponding initialization. | |||
687 | * IP checks con->support to see if it need disable ras. | |||
688 | */ | |||
689 | if (!amdgpu_ras_is_feature_allowed(adev, head)) | |||
690 | return 0; | |||
691 | ||||
692 | if (enable) { | |||
693 | if (!obj) { | |||
694 | obj = amdgpu_ras_create_obj(adev, head); | |||
695 | if (!obj) | |||
696 | return -EINVAL22; | |||
697 | } else { | |||
698 | /* In case we create obj somewhere else */ | |||
699 | get_obj(obj)do { (obj)->use++; } while (0); | |||
700 | } | |||
701 | con->features |= BIT(head->block)(1UL << (head->block)); | |||
702 | } else { | |||
703 | if (obj && amdgpu_ras_is_feature_enabled(adev, head)) { | |||
704 | con->features &= ~BIT(head->block)(1UL << (head->block)); | |||
705 | put_obj(obj); | |||
706 | } | |||
707 | } | |||
708 | ||||
709 | return 0; | |||
710 | } | |||
711 | ||||
712 | /* wrapper of psp_ras_enable_features */ | |||
713 | int amdgpu_ras_feature_enable(struct amdgpu_device *adev, | |||
714 | struct ras_common_if *head, bool_Bool enable) | |||
715 | { | |||
716 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
717 | union ta_ras_cmd_input *info; | |||
718 | int ret; | |||
719 | ||||
720 | if (!con
| |||
721 | return -EINVAL22; | |||
722 | ||||
723 | if (head->block == AMDGPU_RAS_BLOCK__GFX) { | |||
724 | info = kzalloc(sizeof(union ta_ras_cmd_input), GFP_KERNEL(0x0001 | 0x0004)); | |||
725 | if (!info) | |||
726 | return -ENOMEM12; | |||
727 | ||||
728 | if (!enable) { | |||
729 | info->disable_features = (struct ta_ras_disable_features_input) { | |||
730 | .block_id = amdgpu_ras_block_to_ta(head->block), | |||
731 | .error_type = amdgpu_ras_error_to_ta(head->type), | |||
732 | }; | |||
733 | } else { | |||
734 | info->enable_features = (struct ta_ras_enable_features_input) { | |||
735 | .block_id = amdgpu_ras_block_to_ta(head->block), | |||
736 | .error_type = amdgpu_ras_error_to_ta(head->type), | |||
737 | }; | |||
738 | } | |||
739 | } | |||
740 | ||||
741 | /* Do not enable if it is not allowed. */ | |||
742 | WARN_ON(enable && !amdgpu_ras_is_feature_allowed(adev, head))({ int __ret = !!(enable && !amdgpu_ras_is_feature_allowed (adev, head)); if (__ret) printf("WARNING %s failed at %s:%d\n" , "enable && !amdgpu_ras_is_feature_allowed(adev, head)" , "/usr/src/sys/dev/pci/drm/amd/amdgpu/amdgpu_ras.c", 742); __builtin_expect (!!(__ret), 0); }); | |||
743 | ||||
744 | /* Only enable ras feature operation handle on host side */ | |||
745 | if (head->block
| |||
746 | !amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2)) && | |||
747 | !amdgpu_ras_intr_triggered()) { | |||
748 | ret = psp_ras_enable_features(&adev->psp, info, enable); | |||
749 | if (ret) { | |||
750 | dev_err(adev->dev, "ras %s %s failed poison:%d ret:%d\n",printf("drm:pid%d:%s *ERROR* " "ras %s %s failed poison:%d ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , enable ? "enable":"disable", get_ras_block_str(head), amdgpu_ras_is_poison_mode_supported (adev), ret) | |||
751 | enable ? "enable":"disable",printf("drm:pid%d:%s *ERROR* " "ras %s %s failed poison:%d ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , enable ? "enable":"disable", get_ras_block_str(head), amdgpu_ras_is_poison_mode_supported (adev), ret) | |||
752 | get_ras_block_str(head),printf("drm:pid%d:%s *ERROR* " "ras %s %s failed poison:%d ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , enable ? "enable":"disable", get_ras_block_str(head), amdgpu_ras_is_poison_mode_supported (adev), ret) | |||
753 | amdgpu_ras_is_poison_mode_supported(adev), ret)printf("drm:pid%d:%s *ERROR* " "ras %s %s failed poison:%d ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , enable ? "enable":"disable", get_ras_block_str(head), amdgpu_ras_is_poison_mode_supported (adev), ret); | |||
754 | goto out; | |||
755 | } | |||
756 | } | |||
757 | ||||
758 | /* setup the obj */ | |||
759 | __amdgpu_ras_feature_enable(adev, head, enable); | |||
760 | ret = 0; | |||
761 | out: | |||
762 | if (head->block == AMDGPU_RAS_BLOCK__GFX) | |||
763 | kfree(info); | |||
| ||||
764 | return ret; | |||
765 | } | |||
766 | ||||
767 | /* Only used in device probe stage and called only once. */ | |||
768 | int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev, | |||
769 | struct ras_common_if *head, bool_Bool enable) | |||
770 | { | |||
771 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
772 | int ret; | |||
773 | ||||
774 | if (!con) | |||
775 | return -EINVAL22; | |||
776 | ||||
777 | if (con->flags & AMDGPU_RAS_FLAG_INIT_BY_VBIOS(0x1 << 0)) { | |||
778 | if (enable) { | |||
779 | /* There is no harm to issue a ras TA cmd regardless of | |||
780 | * the currecnt ras state. | |||
781 | * If current state == target state, it will do nothing | |||
782 | * But sometimes it requests driver to reset and repost | |||
783 | * with error code -EAGAIN. | |||
784 | */ | |||
785 | ret = amdgpu_ras_feature_enable(adev, head, 1); | |||
786 | /* With old ras TA, we might fail to enable ras. | |||
787 | * Log it and just setup the object. | |||
788 | * TODO need remove this WA in the future. | |||
789 | */ | |||
790 | if (ret == -EINVAL22) { | |||
791 | ret = __amdgpu_ras_feature_enable(adev, head, 1); | |||
792 | if (!ret) | |||
793 | dev_info(adev->dev,do { } while(0) | |||
794 | "RAS INFO: %s setup object\n",do { } while(0) | |||
795 | get_ras_block_str(head))do { } while(0); | |||
796 | } | |||
797 | } else { | |||
798 | /* setup the object then issue a ras TA disable cmd.*/ | |||
799 | ret = __amdgpu_ras_feature_enable(adev, head, 1); | |||
800 | if (ret) | |||
801 | return ret; | |||
802 | ||||
803 | /* gfx block ras dsiable cmd must send to ras-ta */ | |||
804 | if (head->block == AMDGPU_RAS_BLOCK__GFX) | |||
805 | con->features |= BIT(head->block)(1UL << (head->block)); | |||
806 | ||||
807 | ret = amdgpu_ras_feature_enable(adev, head, 0); | |||
808 | ||||
809 | /* clean gfx block ras features flag */ | |||
810 | if (adev->ras_enabled && head->block == AMDGPU_RAS_BLOCK__GFX) | |||
811 | con->features &= ~BIT(head->block)(1UL << (head->block)); | |||
812 | } | |||
813 | } else | |||
814 | ret = amdgpu_ras_feature_enable(adev, head, enable); | |||
815 | ||||
816 | return ret; | |||
817 | } | |||
818 | ||||
819 | static int amdgpu_ras_disable_all_features(struct amdgpu_device *adev, | |||
820 | bool_Bool bypass) | |||
821 | { | |||
822 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
823 | struct ras_manager *obj, *tmp; | |||
824 | ||||
825 | list_for_each_entry_safe(obj, tmp, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}), tmp = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj->node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) );}); &obj-> node != (&con->head); obj = tmp, tmp = ({ const __typeof ( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next ); (__typeof(*tmp) *)( (char *)__mptr - __builtin_offsetof(__typeof (*tmp), node) );})) { | |||
826 | /* bypass psp. | |||
827 | * aka just release the obj and corresponding flags | |||
828 | */ | |||
829 | if (bypass
| |||
830 | if (__amdgpu_ras_feature_enable(adev, &obj->head, 0)) | |||
831 | break; | |||
832 | } else { | |||
833 | if (amdgpu_ras_feature_enable(adev, &obj->head, 0)) | |||
834 | break; | |||
835 | } | |||
836 | } | |||
837 | ||||
838 | return con->features; | |||
839 | } | |||
840 | ||||
841 | static int amdgpu_ras_enable_all_features(struct amdgpu_device *adev, | |||
842 | bool_Bool bypass) | |||
843 | { | |||
844 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
845 | int i; | |||
846 | const enum amdgpu_ras_error_type default_ras_type = AMDGPU_RAS_ERROR__NONE; | |||
847 | ||||
848 | for (i = 0; i < AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST; i++) { | |||
849 | struct ras_common_if head = { | |||
850 | .block = i, | |||
851 | .type = default_ras_type, | |||
852 | .sub_block_index = 0, | |||
853 | }; | |||
854 | ||||
855 | if (i == AMDGPU_RAS_BLOCK__MCA) | |||
856 | continue; | |||
857 | ||||
858 | if (bypass) { | |||
859 | /* | |||
860 | * bypass psp. vbios enable ras for us. | |||
861 | * so just create the obj | |||
862 | */ | |||
863 | if (__amdgpu_ras_feature_enable(adev, &head, 1)) | |||
864 | break; | |||
865 | } else { | |||
866 | if (amdgpu_ras_feature_enable(adev, &head, 1)) | |||
867 | break; | |||
868 | } | |||
869 | } | |||
870 | ||||
871 | for (i = 0; i < AMDGPU_RAS_MCA_BLOCK_COUNTAMDGPU_RAS_MCA_BLOCK__LAST; i++) { | |||
872 | struct ras_common_if head = { | |||
873 | .block = AMDGPU_RAS_BLOCK__MCA, | |||
874 | .type = default_ras_type, | |||
875 | .sub_block_index = i, | |||
876 | }; | |||
877 | ||||
878 | if (bypass) { | |||
879 | /* | |||
880 | * bypass psp. vbios enable ras for us. | |||
881 | * so just create the obj | |||
882 | */ | |||
883 | if (__amdgpu_ras_feature_enable(adev, &head, 1)) | |||
884 | break; | |||
885 | } else { | |||
886 | if (amdgpu_ras_feature_enable(adev, &head, 1)) | |||
887 | break; | |||
888 | } | |||
889 | } | |||
890 | ||||
891 | return con->features; | |||
892 | } | |||
893 | /* feature ctl end */ | |||
894 | ||||
895 | static int amdgpu_ras_block_match_default(struct amdgpu_ras_block_object *block_obj, | |||
896 | enum amdgpu_ras_block block) | |||
897 | { | |||
898 | if (!block_obj) | |||
899 | return -EINVAL22; | |||
900 | ||||
901 | if (block_obj->ras_comm.block == block) | |||
902 | return 0; | |||
903 | ||||
904 | return -EINVAL22; | |||
905 | } | |||
906 | ||||
907 | static struct amdgpu_ras_block_object *amdgpu_ras_get_ras_block(struct amdgpu_device *adev, | |||
908 | enum amdgpu_ras_block block, uint32_t sub_block_index) | |||
909 | { | |||
910 | struct amdgpu_ras_block_list *node, *tmp; | |||
911 | struct amdgpu_ras_block_object *obj; | |||
912 | ||||
913 | if (block >= AMDGPU_RAS_BLOCK__LAST) | |||
914 | return NULL((void *)0); | |||
915 | ||||
916 | if (!amdgpu_ras_is_supported(adev, block)) | |||
917 | return NULL((void *)0); | |||
918 | ||||
919 | list_for_each_entry_safe(node, tmp, &adev->ras_list, node)for (node = ({ const __typeof( ((__typeof(*node) *)0)->node ) *__mptr = ((&adev->ras_list)->next); (__typeof(* node) *)( (char *)__mptr - __builtin_offsetof(__typeof(*node) , node) );}), tmp = ({ const __typeof( ((__typeof(*node) *)0) ->node ) *__mptr = (node->node.next); (__typeof(*node) * )( (char *)__mptr - __builtin_offsetof(__typeof(*node), node) );}); &node->node != (&adev->ras_list); node = tmp, tmp = ({ const __typeof( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next); (__typeof(*tmp) *)( (char * )__mptr - __builtin_offsetof(__typeof(*tmp), node) );})) { | |||
920 | if (!node->ras_obj) { | |||
921 | dev_warn(adev->dev, "Warning: abnormal ras list node.\n")printf("drm:pid%d:%s *WARNING* " "Warning: abnormal ras list node.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
922 | continue; | |||
923 | } | |||
924 | ||||
925 | obj = node->ras_obj; | |||
926 | if (obj->ras_block_match) { | |||
927 | if (obj->ras_block_match(obj, block, sub_block_index) == 0) | |||
928 | return obj; | |||
929 | } else { | |||
930 | if (amdgpu_ras_block_match_default(obj, block) == 0) | |||
931 | return obj; | |||
932 | } | |||
933 | } | |||
934 | ||||
935 | return NULL((void *)0); | |||
936 | } | |||
937 | ||||
938 | static void amdgpu_ras_get_ecc_info(struct amdgpu_device *adev, struct ras_err_data *err_data) | |||
939 | { | |||
940 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |||
941 | int ret = 0; | |||
942 | ||||
943 | /* | |||
944 | * choosing right query method according to | |||
945 | * whether smu support query error information | |||
946 | */ | |||
947 | ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(ras->umc_ecc)); | |||
948 | if (ret == -EOPNOTSUPP45) { | |||
949 | if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && | |||
950 | adev->umc.ras->ras_block.hw_ops->query_ras_error_count) | |||
951 | adev->umc.ras->ras_block.hw_ops->query_ras_error_count(adev, err_data); | |||
952 | ||||
953 | /* umc query_ras_error_address is also responsible for clearing | |||
954 | * error status | |||
955 | */ | |||
956 | if (adev->umc.ras && adev->umc.ras->ras_block.hw_ops && | |||
957 | adev->umc.ras->ras_block.hw_ops->query_ras_error_address) | |||
958 | adev->umc.ras->ras_block.hw_ops->query_ras_error_address(adev, err_data); | |||
959 | } else if (!ret) { | |||
960 | if (adev->umc.ras && | |||
961 | adev->umc.ras->ecc_info_query_ras_error_count) | |||
962 | adev->umc.ras->ecc_info_query_ras_error_count(adev, err_data); | |||
963 | ||||
964 | if (adev->umc.ras && | |||
965 | adev->umc.ras->ecc_info_query_ras_error_address) | |||
966 | adev->umc.ras->ecc_info_query_ras_error_address(adev, err_data); | |||
967 | } | |||
968 | } | |||
969 | ||||
970 | /* query/inject/cure begin */ | |||
971 | int amdgpu_ras_query_error_status(struct amdgpu_device *adev, | |||
972 | struct ras_query_if *info) | |||
973 | { | |||
974 | struct amdgpu_ras_block_object *block_obj = NULL((void *)0); | |||
975 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); | |||
976 | struct ras_err_data err_data = {0, 0, 0, NULL((void *)0)}; | |||
977 | ||||
978 | if (!obj) | |||
979 | return -EINVAL22; | |||
980 | ||||
981 | if (info->head.block == AMDGPU_RAS_BLOCK__UMC) { | |||
982 | amdgpu_ras_get_ecc_info(adev, &err_data); | |||
983 | } else { | |||
984 | block_obj = amdgpu_ras_get_ras_block(adev, info->head.block, 0); | |||
985 | if (!block_obj || !block_obj->hw_ops) { | |||
986 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",do { } while(0) | |||
987 | get_ras_block_str(&info->head))do { } while(0); | |||
988 | return -EINVAL22; | |||
989 | } | |||
990 | ||||
991 | if (block_obj->hw_ops->query_ras_error_count) | |||
992 | block_obj->hw_ops->query_ras_error_count(adev, &err_data); | |||
993 | ||||
994 | if ((info->head.block == AMDGPU_RAS_BLOCK__SDMA) || | |||
995 | (info->head.block == AMDGPU_RAS_BLOCK__GFX) || | |||
996 | (info->head.block == AMDGPU_RAS_BLOCK__MMHUB)) { | |||
997 | if (block_obj->hw_ops->query_ras_error_status) | |||
998 | block_obj->hw_ops->query_ras_error_status(adev); | |||
999 | } | |||
1000 | } | |||
1001 | ||||
1002 | obj->err_data.ue_count += err_data.ue_count; | |||
1003 | obj->err_data.ce_count += err_data.ce_count; | |||
1004 | ||||
1005 | info->ue_count = obj->err_data.ue_count; | |||
1006 | info->ce_count = obj->err_data.ce_count; | |||
1007 | ||||
1008 | if (err_data.ce_count) { | |||
1009 | if (adev->smuio.funcs && | |||
1010 | adev->smuio.funcs->get_socket_id && | |||
1011 | adev->smuio.funcs->get_die_id) { | |||
1012 | dev_info(adev->dev, "socket: %d, die: %d "do { } while(0) | |||
1013 | "%ld correctable hardware errors "do { } while(0) | |||
1014 | "detected in %s block, no user "do { } while(0) | |||
1015 | "action is needed.\n",do { } while(0) | |||
1016 | adev->smuio.funcs->get_socket_id(adev),do { } while(0) | |||
1017 | adev->smuio.funcs->get_die_id(adev),do { } while(0) | |||
1018 | obj->err_data.ce_count,do { } while(0) | |||
1019 | get_ras_block_str(&info->head))do { } while(0); | |||
1020 | } else { | |||
1021 | dev_info(adev->dev, "%ld correctable hardware errors "do { } while(0) | |||
1022 | "detected in %s block, no user "do { } while(0) | |||
1023 | "action is needed.\n",do { } while(0) | |||
1024 | obj->err_data.ce_count,do { } while(0) | |||
1025 | get_ras_block_str(&info->head))do { } while(0); | |||
1026 | } | |||
1027 | } | |||
1028 | if (err_data.ue_count) { | |||
1029 | if (adev->smuio.funcs && | |||
1030 | adev->smuio.funcs->get_socket_id && | |||
1031 | adev->smuio.funcs->get_die_id) { | |||
1032 | dev_info(adev->dev, "socket: %d, die: %d "do { } while(0) | |||
1033 | "%ld uncorrectable hardware errors "do { } while(0) | |||
1034 | "detected in %s block\n",do { } while(0) | |||
1035 | adev->smuio.funcs->get_socket_id(adev),do { } while(0) | |||
1036 | adev->smuio.funcs->get_die_id(adev),do { } while(0) | |||
1037 | obj->err_data.ue_count,do { } while(0) | |||
1038 | get_ras_block_str(&info->head))do { } while(0); | |||
1039 | } else { | |||
1040 | dev_info(adev->dev, "%ld uncorrectable hardware errors "do { } while(0) | |||
1041 | "detected in %s block\n",do { } while(0) | |||
1042 | obj->err_data.ue_count,do { } while(0) | |||
1043 | get_ras_block_str(&info->head))do { } while(0); | |||
1044 | } | |||
1045 | } | |||
1046 | ||||
1047 | return 0; | |||
1048 | } | |||
1049 | ||||
1050 | int amdgpu_ras_reset_error_status(struct amdgpu_device *adev, | |||
1051 | enum amdgpu_ras_block block) | |||
1052 | { | |||
1053 | struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, block, 0); | |||
1054 | ||||
1055 | if (!amdgpu_ras_is_supported(adev, block)) | |||
1056 | return -EINVAL22; | |||
1057 | ||||
1058 | if (!block_obj || !block_obj->hw_ops) { | |||
1059 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",do { } while(0) | |||
1060 | ras_block_str(block))do { } while(0); | |||
1061 | return -EINVAL22; | |||
1062 | } | |||
1063 | ||||
1064 | if (block_obj->hw_ops->reset_ras_error_count) | |||
1065 | block_obj->hw_ops->reset_ras_error_count(adev); | |||
1066 | ||||
1067 | if ((block == AMDGPU_RAS_BLOCK__GFX) || | |||
1068 | (block == AMDGPU_RAS_BLOCK__MMHUB)) { | |||
1069 | if (block_obj->hw_ops->reset_ras_error_status) | |||
1070 | block_obj->hw_ops->reset_ras_error_status(adev); | |||
1071 | } | |||
1072 | ||||
1073 | return 0; | |||
1074 | } | |||
1075 | ||||
1076 | /* wrapper of psp_ras_trigger_error */ | |||
1077 | int amdgpu_ras_error_inject(struct amdgpu_device *adev, | |||
1078 | struct ras_inject_if *info) | |||
1079 | { | |||
1080 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); | |||
1081 | struct ta_ras_trigger_error_input block_info = { | |||
1082 | .block_id = amdgpu_ras_block_to_ta(info->head.block), | |||
1083 | .inject_error_type = amdgpu_ras_error_to_ta(info->head.type), | |||
1084 | .sub_block_index = info->head.sub_block_index, | |||
1085 | .address = info->address, | |||
1086 | .value = info->value, | |||
1087 | }; | |||
1088 | int ret = -EINVAL22; | |||
1089 | struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, | |||
1090 | info->head.block, | |||
1091 | info->head.sub_block_index); | |||
1092 | ||||
1093 | if (!obj) | |||
1094 | return -EINVAL22; | |||
1095 | ||||
1096 | if (!block_obj || !block_obj->hw_ops) { | |||
1097 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",do { } while(0) | |||
1098 | get_ras_block_str(&info->head))do { } while(0); | |||
1099 | return -EINVAL22; | |||
1100 | } | |||
1101 | ||||
1102 | /* Calculate XGMI relative offset */ | |||
1103 | if (adev->gmc.xgmi.num_physical_nodes > 1) { | |||
1104 | block_info.address = | |||
1105 | amdgpu_xgmi_get_relative_phy_addr(adev, | |||
1106 | block_info.address); | |||
1107 | } | |||
1108 | ||||
1109 | if (info->head.block == AMDGPU_RAS_BLOCK__GFX) { | |||
1110 | if (block_obj->hw_ops->ras_error_inject) | |||
1111 | ret = block_obj->hw_ops->ras_error_inject(adev, info); | |||
1112 | } else { | |||
1113 | /* If defined special ras_error_inject(e.g: xgmi), implement special ras_error_inject */ | |||
1114 | if (block_obj->hw_ops->ras_error_inject) | |||
1115 | ret = block_obj->hw_ops->ras_error_inject(adev, &block_info); | |||
1116 | else /*If not defined .ras_error_inject, use default ras_error_inject*/ | |||
1117 | ret = psp_ras_trigger_error(&adev->psp, &block_info); | |||
1118 | } | |||
1119 | ||||
1120 | if (ret) | |||
1121 | dev_err(adev->dev, "ras inject %s failed %d\n",printf("drm:pid%d:%s *ERROR* " "ras inject %s failed %d\n", ( {struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , get_ras_block_str (&info->head), ret) | |||
1122 | get_ras_block_str(&info->head), ret)printf("drm:pid%d:%s *ERROR* " "ras inject %s failed %d\n", ( {struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , get_ras_block_str (&info->head), ret); | |||
1123 | ||||
1124 | return ret; | |||
1125 | } | |||
1126 | ||||
1127 | /** | |||
1128 | * amdgpu_ras_query_error_count -- Get error counts of all IPs | |||
1129 | * @adev: pointer to AMD GPU device | |||
1130 | * @ce_count: pointer to an integer to be set to the count of correctible errors. | |||
1131 | * @ue_count: pointer to an integer to be set to the count of uncorrectible | |||
1132 | * errors. | |||
1133 | * | |||
1134 | * If set, @ce_count or @ue_count, count and return the corresponding | |||
1135 | * error counts in those integer pointers. Return 0 if the device | |||
1136 | * supports RAS. Return -EOPNOTSUPP if the device doesn't support RAS. | |||
1137 | */ | |||
1138 | int amdgpu_ras_query_error_count(struct amdgpu_device *adev, | |||
1139 | unsigned long *ce_count, | |||
1140 | unsigned long *ue_count) | |||
1141 | { | |||
1142 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1143 | struct ras_manager *obj; | |||
1144 | unsigned long ce, ue; | |||
1145 | ||||
1146 | if (!adev->ras_enabled || !con) | |||
1147 | return -EOPNOTSUPP45; | |||
1148 | ||||
1149 | /* Don't count since no reporting. | |||
1150 | */ | |||
1151 | if (!ce_count && !ue_count) | |||
1152 | return 0; | |||
1153 | ||||
1154 | ce = 0; | |||
1155 | ue = 0; | |||
1156 | list_for_each_entry(obj, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}); &obj->node != (&con->head); obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj-> node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof (__typeof(*obj), node) );})) { | |||
1157 | struct ras_query_if info = { | |||
1158 | .head = obj->head, | |||
1159 | }; | |||
1160 | int res; | |||
1161 | ||||
1162 | res = amdgpu_ras_query_error_status(adev, &info); | |||
1163 | if (res) | |||
1164 | return res; | |||
1165 | ||||
1166 | if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2)(((11) << 16) | ((0) << 8) | (2)) && | |||
1167 | adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)(((11) << 16) | ((0) << 8) | (4))) { | |||
1168 | if (amdgpu_ras_reset_error_status(adev, info.head.block)) | |||
1169 | dev_warn(adev->dev, "Failed to reset error counter and error status")printf("drm:pid%d:%s *WARNING* " "Failed to reset error counter and error status" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
1170 | } | |||
1171 | ||||
1172 | ce += info.ce_count; | |||
1173 | ue += info.ue_count; | |||
1174 | } | |||
1175 | ||||
1176 | if (ce_count) | |||
1177 | *ce_count = ce; | |||
1178 | ||||
1179 | if (ue_count) | |||
1180 | *ue_count = ue; | |||
1181 | ||||
1182 | return 0; | |||
1183 | } | |||
1184 | /* query/inject/cure end */ | |||
1185 | ||||
1186 | #ifdef __linux__ | |||
1187 | ||||
1188 | /* sysfs begin */ | |||
1189 | ||||
1190 | static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, | |||
1191 | struct ras_badpage **bps, unsigned int *count); | |||
1192 | ||||
1193 | static char *amdgpu_ras_badpage_flags_str(unsigned int flags) | |||
1194 | { | |||
1195 | switch (flags) { | |||
1196 | case AMDGPU_RAS_RETIRE_PAGE_RESERVED: | |||
1197 | return "R"; | |||
1198 | case AMDGPU_RAS_RETIRE_PAGE_PENDING: | |||
1199 | return "P"; | |||
1200 | case AMDGPU_RAS_RETIRE_PAGE_FAULT: | |||
1201 | default: | |||
1202 | return "F"; | |||
1203 | } | |||
1204 | } | |||
1205 | ||||
1206 | /** | |||
1207 | * DOC: AMDGPU RAS sysfs gpu_vram_bad_pages Interface | |||
1208 | * | |||
1209 | * It allows user to read the bad pages of vram on the gpu through | |||
1210 | * /sys/class/drm/card[0/1/2...]/device/ras/gpu_vram_bad_pages | |||
1211 | * | |||
1212 | * It outputs multiple lines, and each line stands for one gpu page. | |||
1213 | * | |||
1214 | * The format of one line is below, | |||
1215 | * gpu pfn : gpu page size : flags | |||
1216 | * | |||
1217 | * gpu pfn and gpu page size are printed in hex format. | |||
1218 | * flags can be one of below character, | |||
1219 | * | |||
1220 | * R: reserved, this gpu page is reserved and not able to use. | |||
1221 | * | |||
1222 | * P: pending for reserve, this gpu page is marked as bad, will be reserved | |||
1223 | * in next window of page_reserve. | |||
1224 | * | |||
1225 | * F: unable to reserve. this gpu page can't be reserved due to some reasons. | |||
1226 | * | |||
1227 | * Examples: | |||
1228 | * | |||
1229 | * .. code-block:: bash | |||
1230 | * | |||
1231 | * 0x00000001 : 0x00001000 : R | |||
1232 | * 0x00000002 : 0x00001000 : P | |||
1233 | * | |||
1234 | */ | |||
1235 | ||||
1236 | static ssize_t amdgpu_ras_sysfs_badpages_read(struct file *f, | |||
1237 | struct kobject *kobj, struct bin_attribute *attr, | |||
1238 | char *buf, loff_t ppos, size_t count) | |||
1239 | { | |||
1240 | struct amdgpu_ras *con = | |||
1241 | container_of(attr, struct amdgpu_ras, badpages_attr)({ const __typeof( ((struct amdgpu_ras *)0)->badpages_attr ) *__mptr = (attr); (struct amdgpu_ras *)( (char *)__mptr - __builtin_offsetof (struct amdgpu_ras, badpages_attr) );}); | |||
1242 | struct amdgpu_device *adev = con->adev; | |||
1243 | const unsigned int element_size = | |||
1244 | sizeof("0xabcdabcd : 0x12345678 : R\n") - 1; | |||
1245 | unsigned int start = div64_ul(ppos + element_size - 1, element_size); | |||
1246 | unsigned int end = div64_ul(ppos + count - 1, element_size); | |||
1247 | ssize_t s = 0; | |||
1248 | struct ras_badpage *bps = NULL((void *)0); | |||
1249 | unsigned int bps_count = 0; | |||
1250 | ||||
1251 | memset(buf, 0, count)__builtin_memset((buf), (0), (count)); | |||
1252 | ||||
1253 | if (amdgpu_ras_badpages_read(adev, &bps, &bps_count)) | |||
1254 | return 0; | |||
1255 | ||||
1256 | for (; start < end && start < bps_count; start++) | |||
1257 | s += scnprintf(&buf[s], element_size + 1,snprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n" , bps[start].bp, bps[start].size, amdgpu_ras_badpage_flags_str (bps[start].flags)) | |||
1258 | "0x%08x : 0x%08x : %1s\n",snprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n" , bps[start].bp, bps[start].size, amdgpu_ras_badpage_flags_str (bps[start].flags)) | |||
1259 | bps[start].bp,snprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n" , bps[start].bp, bps[start].size, amdgpu_ras_badpage_flags_str (bps[start].flags)) | |||
1260 | bps[start].size,snprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n" , bps[start].bp, bps[start].size, amdgpu_ras_badpage_flags_str (bps[start].flags)) | |||
1261 | amdgpu_ras_badpage_flags_str(bps[start].flags))snprintf(&buf[s], element_size + 1, "0x%08x : 0x%08x : %1s\n" , bps[start].bp, bps[start].size, amdgpu_ras_badpage_flags_str (bps[start].flags)); | |||
1262 | ||||
1263 | kfree(bps); | |||
1264 | ||||
1265 | return s; | |||
1266 | } | |||
1267 | ||||
1268 | static ssize_t amdgpu_ras_sysfs_features_read(struct device *dev, | |||
1269 | struct device_attribute *attr, char *buf) | |||
1270 | { | |||
1271 | struct amdgpu_ras *con = | |||
1272 | container_of(attr, struct amdgpu_ras, features_attr)({ const __typeof( ((struct amdgpu_ras *)0)->features_attr ) *__mptr = (attr); (struct amdgpu_ras *)( (char *)__mptr - __builtin_offsetof (struct amdgpu_ras, features_attr) );}); | |||
1273 | ||||
1274 | return scnprintf(buf, PAGE_SIZE, "feature mask: 0x%x\n", con->features)snprintf(buf, (1 << 12), "feature mask: 0x%x\n", con-> features); | |||
1275 | } | |||
1276 | ||||
1277 | static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device *adev) | |||
1278 | { | |||
1279 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1280 | ||||
1281 | if (adev->dev->kobj.sd) | |||
1282 | sysfs_remove_file_from_group(&adev->dev->kobj, | |||
1283 | &con->badpages_attr.attr, | |||
1284 | RAS_FS_NAME); | |||
1285 | } | |||
1286 | ||||
1287 | static int amdgpu_ras_sysfs_remove_feature_node(struct amdgpu_device *adev) | |||
1288 | { | |||
1289 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1290 | struct attribute *attrs[] = { | |||
1291 | &con->features_attr.attr, | |||
1292 | NULL((void *)0) | |||
1293 | }; | |||
1294 | struct attribute_group group = { | |||
1295 | .name = RAS_FS_NAME, | |||
1296 | .attrs = attrs, | |||
1297 | }; | |||
1298 | ||||
1299 | if (adev->dev->kobj.sd) | |||
1300 | sysfs_remove_group(&adev->dev->kobj, &group); | |||
1301 | ||||
1302 | return 0; | |||
1303 | } | |||
1304 | ||||
1305 | #endif /* __linux__ */ | |||
1306 | ||||
1307 | int amdgpu_ras_sysfs_create(struct amdgpu_device *adev, | |||
1308 | struct ras_common_if *head) | |||
1309 | { | |||
1310 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |||
1311 | ||||
1312 | if (!obj || obj->attr_inuse) | |||
1313 | return -EINVAL22; | |||
1314 | ||||
1315 | STUB()do { printf("%s: stub\n", __func__); } while(0); | |||
1316 | return -ENOSYS78; | |||
1317 | #ifdef notyet | |||
1318 | get_obj(obj)do { (obj)->use++; } while (0); | |||
1319 | ||||
1320 | snprintf(obj->fs_data.sysfs_name, sizeof(obj->fs_data.sysfs_name), | |||
1321 | "%s_err_count", head->name); | |||
1322 | ||||
1323 | obj->sysfs_attr = (struct device_attribute){ | |||
1324 | .attr = { | |||
1325 | .name = obj->fs_data.sysfs_name, | |||
1326 | .mode = S_IRUGO, | |||
1327 | }, | |||
1328 | .show = amdgpu_ras_sysfs_read, | |||
1329 | }; | |||
1330 | sysfs_attr_init(&obj->sysfs_attr.attr); | |||
1331 | ||||
1332 | if (sysfs_add_file_to_group(&adev->dev->kobj, | |||
1333 | &obj->sysfs_attr.attr, | |||
1334 | RAS_FS_NAME)) { | |||
1335 | put_obj(obj); | |||
1336 | return -EINVAL22; | |||
1337 | } | |||
1338 | ||||
1339 | obj->attr_inuse = 1; | |||
1340 | ||||
1341 | return 0; | |||
1342 | #endif | |||
1343 | } | |||
1344 | ||||
1345 | int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev, | |||
1346 | struct ras_common_if *head) | |||
1347 | { | |||
1348 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |||
1349 | ||||
1350 | if (!obj || !obj->attr_inuse) | |||
1351 | return -EINVAL22; | |||
1352 | ||||
1353 | #ifdef __linux__ | |||
1354 | if (adev->dev->kobj.sd) | |||
1355 | sysfs_remove_file_from_group(&adev->dev->kobj, | |||
1356 | &obj->sysfs_attr.attr, | |||
1357 | RAS_FS_NAME); | |||
1358 | #endif | |||
1359 | obj->attr_inuse = 0; | |||
1360 | put_obj(obj); | |||
1361 | ||||
1362 | return 0; | |||
1363 | } | |||
1364 | ||||
1365 | #ifdef __linux__ | |||
1366 | ||||
1367 | static int amdgpu_ras_sysfs_remove_all(struct amdgpu_device *adev) | |||
1368 | { | |||
1369 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1370 | struct ras_manager *obj, *tmp; | |||
1371 | ||||
1372 | list_for_each_entry_safe(obj, tmp, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}), tmp = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj->node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) );}); &obj-> node != (&con->head); obj = tmp, tmp = ({ const __typeof ( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next ); (__typeof(*tmp) *)( (char *)__mptr - __builtin_offsetof(__typeof (*tmp), node) );})) { | |||
1373 | amdgpu_ras_sysfs_remove(adev, &obj->head); | |||
1374 | } | |||
1375 | ||||
1376 | if (amdgpu_bad_page_threshold != 0) | |||
1377 | amdgpu_ras_sysfs_remove_bad_page_node(adev); | |||
1378 | ||||
1379 | amdgpu_ras_sysfs_remove_feature_node(adev); | |||
1380 | ||||
1381 | return 0; | |||
1382 | } | |||
1383 | /* sysfs end */ | |||
1384 | ||||
1385 | /** | |||
1386 | * DOC: AMDGPU RAS Reboot Behavior for Unrecoverable Errors | |||
1387 | * | |||
1388 | * Normally when there is an uncorrectable error, the driver will reset | |||
1389 | * the GPU to recover. However, in the event of an unrecoverable error, | |||
1390 | * the driver provides an interface to reboot the system automatically | |||
1391 | * in that event. | |||
1392 | * | |||
1393 | * The following file in debugfs provides that interface: | |||
1394 | * /sys/kernel/debug/dri/[0/1/2...]/ras/auto_reboot | |||
1395 | * | |||
1396 | * Usage: | |||
1397 | * | |||
1398 | * .. code-block:: bash | |||
1399 | * | |||
1400 | * echo true > .../ras/auto_reboot | |||
1401 | * | |||
1402 | */ | |||
1403 | /* debugfs begin */ | |||
1404 | static struct dentry *amdgpu_ras_debugfs_create_ctrl_node(struct amdgpu_device *adev) | |||
1405 | { | |||
1406 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1407 | struct drm_minor *minor = adev_to_drm(adev)->primary; | |||
1408 | struct dentry *dir; | |||
1409 | ||||
1410 | dir = debugfs_create_dir(RAS_FS_NAME, minor->debugfs_root)ERR_PTR(-78); | |||
1411 | debugfs_create_file("ras_ctrl", S_IWUGO | S_IRUGO, dir, adev,ERR_PTR(-78) | |||
1412 | &amdgpu_ras_debugfs_ctrl_ops)ERR_PTR(-78); | |||
1413 | debugfs_create_file("ras_eeprom_reset", S_IWUGO | S_IRUGO, dir, adev,ERR_PTR(-78) | |||
1414 | &amdgpu_ras_debugfs_eeprom_ops)ERR_PTR(-78); | |||
1415 | debugfs_create_u32("bad_page_cnt_threshold", 0444, dir, | |||
1416 | &con->bad_page_cnt_threshold); | |||
1417 | debugfs_create_x32("ras_hw_enabled", 0444, dir, &adev->ras_hw_enabled); | |||
1418 | debugfs_create_x32("ras_enabled", 0444, dir, &adev->ras_enabled); | |||
1419 | debugfs_create_file("ras_eeprom_size", S_IRUGO, dir, adev,ERR_PTR(-78) | |||
1420 | &amdgpu_ras_debugfs_eeprom_size_ops)ERR_PTR(-78); | |||
1421 | con->de_ras_eeprom_table = debugfs_create_file("ras_eeprom_table",ERR_PTR(-78) | |||
1422 | S_IRUGO, dir, adev,ERR_PTR(-78) | |||
1423 | &amdgpu_ras_debugfs_eeprom_table_ops)ERR_PTR(-78); | |||
1424 | amdgpu_ras_debugfs_set_ret_size(&con->eeprom_control); | |||
1425 | ||||
1426 | /* | |||
1427 | * After one uncorrectable error happens, usually GPU recovery will | |||
1428 | * be scheduled. But due to the known problem in GPU recovery failing | |||
1429 | * to bring GPU back, below interface provides one direct way to | |||
1430 | * user to reboot system automatically in such case within | |||
1431 | * ERREVENT_ATHUB_INTERRUPT generated. Normal GPU recovery routine | |||
1432 | * will never be called. | |||
1433 | */ | |||
1434 | debugfs_create_bool("auto_reboot", S_IWUGO | S_IRUGO, dir, &con->reboot); | |||
1435 | ||||
1436 | /* | |||
1437 | * User could set this not to clean up hardware's error count register | |||
1438 | * of RAS IPs during ras recovery. | |||
1439 | */ | |||
1440 | debugfs_create_bool("disable_ras_err_cnt_harvest", 0644, dir, | |||
1441 | &con->disable_ras_err_cnt_harvest); | |||
1442 | return dir; | |||
1443 | } | |||
1444 | ||||
1445 | static void amdgpu_ras_debugfs_create(struct amdgpu_device *adev, | |||
1446 | struct ras_fs_if *head, | |||
1447 | struct dentry *dir) | |||
1448 | { | |||
1449 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &head->head); | |||
1450 | ||||
1451 | if (!obj || !dir) | |||
1452 | return; | |||
1453 | ||||
1454 | get_obj(obj)do { (obj)->use++; } while (0); | |||
1455 | ||||
1456 | memcpy(obj->fs_data.debugfs_name,__builtin_memcpy((obj->fs_data.debugfs_name), (head->debugfs_name ), (sizeof(obj->fs_data.debugfs_name))) | |||
1457 | head->debugfs_name,__builtin_memcpy((obj->fs_data.debugfs_name), (head->debugfs_name ), (sizeof(obj->fs_data.debugfs_name))) | |||
1458 | sizeof(obj->fs_data.debugfs_name))__builtin_memcpy((obj->fs_data.debugfs_name), (head->debugfs_name ), (sizeof(obj->fs_data.debugfs_name))); | |||
1459 | ||||
1460 | debugfs_create_file(obj->fs_data.debugfs_name, S_IWUGO | S_IRUGO, dir,ERR_PTR(-78) | |||
1461 | obj, &amdgpu_ras_debugfs_ops)ERR_PTR(-78); | |||
1462 | } | |||
1463 | ||||
1464 | void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) | |||
1465 | { | |||
1466 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1467 | struct dentry *dir; | |||
1468 | struct ras_manager *obj; | |||
1469 | struct ras_fs_if fs_info; | |||
1470 | ||||
1471 | /* | |||
1472 | * it won't be called in resume path, no need to check | |||
1473 | * suspend and gpu reset status | |||
1474 | */ | |||
1475 | if (!IS_ENABLED(CONFIG_DEBUG_FS)0 || !con) | |||
1476 | return; | |||
1477 | ||||
1478 | dir = amdgpu_ras_debugfs_create_ctrl_node(adev); | |||
1479 | ||||
1480 | list_for_each_entry(obj, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}); &obj->node != (&con->head); obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj-> node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof (__typeof(*obj), node) );})) { | |||
1481 | if (amdgpu_ras_is_supported(adev, obj->head.block) && | |||
1482 | (obj->attr_inuse == 1)) { | |||
1483 | sprintf(fs_info.debugfs_name, "%s_err_inject", | |||
1484 | get_ras_block_str(&obj->head)); | |||
1485 | fs_info.head = obj->head; | |||
1486 | amdgpu_ras_debugfs_create(adev, &fs_info, dir); | |||
1487 | } | |||
1488 | } | |||
1489 | } | |||
1490 | ||||
1491 | /* debugfs end */ | |||
1492 | ||||
1493 | /* ras fs */ | |||
1494 | static BIN_ATTR(gpu_vram_bad_pages, S_IRUGO, | |||
1495 | amdgpu_ras_sysfs_badpages_read, NULL((void *)0), 0); | |||
1496 | #endif /* __linux__ */ | |||
1497 | static DEVICE_ATTR(features, S_IRUGO,struct device_attribute dev_attr_features | |||
1498 | amdgpu_ras_sysfs_features_read, NULL)struct device_attribute dev_attr_features; | |||
1499 | static int amdgpu_ras_fs_init(struct amdgpu_device *adev) | |||
1500 | { | |||
1501 | #ifdef __linux__ | |||
1502 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1503 | struct attribute_group group = { | |||
1504 | .name = RAS_FS_NAME, | |||
1505 | }; | |||
1506 | struct attribute *attrs[] = { | |||
1507 | &con->features_attr.attr, | |||
1508 | NULL((void *)0) | |||
1509 | }; | |||
1510 | struct bin_attribute *bin_attrs[] = { | |||
1511 | NULL((void *)0), | |||
1512 | NULL((void *)0), | |||
1513 | }; | |||
1514 | int r; | |||
1515 | ||||
1516 | /* add features entry */ | |||
1517 | con->features_attr = dev_attr_features; | |||
1518 | group.attrs = attrs; | |||
1519 | sysfs_attr_init(attrs[0]); | |||
1520 | ||||
1521 | if (amdgpu_bad_page_threshold != 0) { | |||
1522 | /* add bad_page_features entry */ | |||
1523 | bin_attr_gpu_vram_bad_pages.private = NULL((void *)0); | |||
1524 | con->badpages_attr = bin_attr_gpu_vram_bad_pages; | |||
1525 | bin_attrs[0] = &con->badpages_attr; | |||
1526 | group.bin_attrs = bin_attrs; | |||
1527 | sysfs_bin_attr_init(bin_attrs[0]); | |||
1528 | } | |||
1529 | ||||
1530 | r = sysfs_create_group(&adev->dev->kobj, &group)0; | |||
1531 | if (r) | |||
1532 | dev_err(adev->dev, "Failed to create RAS sysfs group!")printf("drm:pid%d:%s *ERROR* " "Failed to create RAS sysfs group!" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
1533 | #endif | |||
1534 | ||||
1535 | return 0; | |||
1536 | } | |||
1537 | ||||
1538 | static int amdgpu_ras_fs_fini(struct amdgpu_device *adev) | |||
1539 | { | |||
1540 | #ifdef __linux__ | |||
1541 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1542 | struct ras_manager *con_obj, *ip_obj, *tmp; | |||
1543 | ||||
1544 | if (IS_ENABLED(CONFIG_DEBUG_FS)0) { | |||
1545 | list_for_each_entry_safe(con_obj, tmp, &con->head, node)for (con_obj = ({ const __typeof( ((__typeof(*con_obj) *)0)-> node ) *__mptr = ((&con->head)->next); (__typeof(*con_obj ) *)( (char *)__mptr - __builtin_offsetof(__typeof(*con_obj), node) );}), tmp = ({ const __typeof( ((__typeof(*con_obj) *) 0)->node ) *__mptr = (con_obj->node.next); (__typeof(*con_obj ) *)( (char *)__mptr - __builtin_offsetof(__typeof(*con_obj), node) );}); &con_obj->node != (&con->head); con_obj = tmp, tmp = ({ const __typeof( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next); (__typeof(*tmp) *)( (char * )__mptr - __builtin_offsetof(__typeof(*tmp), node) );})) { | |||
1546 | ip_obj = amdgpu_ras_find_obj(adev, &con_obj->head); | |||
1547 | if (ip_obj) | |||
1548 | put_obj(ip_obj); | |||
1549 | } | |||
1550 | } | |||
1551 | ||||
1552 | amdgpu_ras_sysfs_remove_all(adev); | |||
1553 | #endif | |||
1554 | return 0; | |||
1555 | } | |||
1556 | /* ras fs end */ | |||
1557 | ||||
1558 | /* ih begin */ | |||
1559 | ||||
1560 | /* For the hardware that cannot enable bif ring for both ras_controller_irq | |||
1561 | * and ras_err_evnet_athub_irq ih cookies, the driver has to poll status | |||
1562 | * register to check whether the interrupt is triggered or not, and properly | |||
1563 | * ack the interrupt if it is there | |||
1564 | */ | |||
1565 | void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev) | |||
1566 | { | |||
1567 | /* Fatal error events are handled on host side */ | |||
1568 | if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2)) || | |||
1569 | !amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__PCIE_BIF)) | |||
1570 | return; | |||
1571 | ||||
1572 | if (adev->nbio.ras && | |||
1573 | adev->nbio.ras->handle_ras_controller_intr_no_bifring) | |||
1574 | adev->nbio.ras->handle_ras_controller_intr_no_bifring(adev); | |||
1575 | ||||
1576 | if (adev->nbio.ras && | |||
1577 | adev->nbio.ras->handle_ras_err_event_athub_intr_no_bifring) | |||
1578 | adev->nbio.ras->handle_ras_err_event_athub_intr_no_bifring(adev); | |||
1579 | } | |||
1580 | ||||
1581 | static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager *obj, | |||
1582 | struct amdgpu_iv_entry *entry) | |||
1583 | { | |||
1584 | bool_Bool poison_stat = false0; | |||
1585 | struct amdgpu_device *adev = obj->adev; | |||
1586 | struct ras_err_data err_data = {0, 0, 0, NULL((void *)0)}; | |||
1587 | struct amdgpu_ras_block_object *block_obj = | |||
1588 | amdgpu_ras_get_ras_block(adev, obj->head.block, 0); | |||
1589 | ||||
1590 | if (!block_obj || !block_obj->hw_ops) | |||
1591 | return; | |||
1592 | ||||
1593 | /* both query_poison_status and handle_poison_consumption are optional, | |||
1594 | * but at least one of them should be implemented if we need poison | |||
1595 | * consumption handler | |||
1596 | */ | |||
1597 | if (block_obj->hw_ops->query_poison_status) { | |||
1598 | poison_stat = block_obj->hw_ops->query_poison_status(adev); | |||
1599 | if (!poison_stat) { | |||
1600 | /* Not poison consumption interrupt, no need to handle it */ | |||
1601 | dev_info(adev->dev, "No RAS poison status in %s poison IH.\n",do { } while(0) | |||
1602 | block_obj->ras_comm.name)do { } while(0); | |||
1603 | ||||
1604 | return; | |||
1605 | } | |||
1606 | } | |||
1607 | ||||
1608 | if (!adev->gmc.xgmi.connected_to_cpu) | |||
1609 | amdgpu_umc_poison_handler(adev, &err_data, false0); | |||
1610 | ||||
1611 | if (block_obj->hw_ops->handle_poison_consumption) | |||
1612 | poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); | |||
1613 | ||||
1614 | /* gpu reset is fallback for failed and default cases */ | |||
1615 | if (poison_stat) { | |||
1616 | dev_info(adev->dev, "GPU reset for %s RAS poison consumption is issued!\n",do { } while(0) | |||
1617 | block_obj->ras_comm.name)do { } while(0); | |||
1618 | amdgpu_ras_reset_gpu(adev); | |||
1619 | } | |||
1620 | } | |||
1621 | ||||
1622 | static void amdgpu_ras_interrupt_poison_creation_handler(struct ras_manager *obj, | |||
1623 | struct amdgpu_iv_entry *entry) | |||
1624 | { | |||
1625 | dev_info(obj->adev->dev,do { } while(0) | |||
1626 | "Poison is created, no user action is needed.\n")do { } while(0); | |||
1627 | } | |||
1628 | ||||
1629 | static void amdgpu_ras_interrupt_umc_handler(struct ras_manager *obj, | |||
1630 | struct amdgpu_iv_entry *entry) | |||
1631 | { | |||
1632 | struct ras_ih_data *data = &obj->ih_data; | |||
1633 | struct ras_err_data err_data = {0, 0, 0, NULL((void *)0)}; | |||
1634 | int ret; | |||
1635 | ||||
1636 | if (!data->cb) | |||
1637 | return; | |||
1638 | ||||
1639 | /* Let IP handle its data, maybe we need get the output | |||
1640 | * from the callback to update the error type/count, etc | |||
1641 | */ | |||
1642 | ret = data->cb(obj->adev, &err_data, entry); | |||
1643 | /* ue will trigger an interrupt, and in that case | |||
1644 | * we need do a reset to recovery the whole system. | |||
1645 | * But leave IP do that recovery, here we just dispatch | |||
1646 | * the error. | |||
1647 | */ | |||
1648 | if (ret == AMDGPU_RAS_SUCCESS) { | |||
1649 | /* these counts could be left as 0 if | |||
1650 | * some blocks do not count error number | |||
1651 | */ | |||
1652 | obj->err_data.ue_count += err_data.ue_count; | |||
1653 | obj->err_data.ce_count += err_data.ce_count; | |||
1654 | } | |||
1655 | } | |||
1656 | ||||
1657 | static void amdgpu_ras_interrupt_handler(struct ras_manager *obj) | |||
1658 | { | |||
1659 | struct ras_ih_data *data = &obj->ih_data; | |||
1660 | struct amdgpu_iv_entry entry; | |||
1661 | ||||
1662 | while (data->rptr != data->wptr) { | |||
1663 | rmb()do { __asm volatile("lfence" ::: "memory"); } while (0); | |||
1664 | memcpy(&entry, &data->ring[data->rptr],__builtin_memcpy((&entry), (&data->ring[data->rptr ]), (data->element_size)) | |||
1665 | data->element_size)__builtin_memcpy((&entry), (&data->ring[data->rptr ]), (data->element_size)); | |||
1666 | ||||
1667 | wmb()do { __asm volatile("sfence" ::: "memory"); } while (0); | |||
1668 | data->rptr = (data->aligned_element_size + | |||
1669 | data->rptr) % data->ring_size; | |||
1670 | ||||
1671 | if (amdgpu_ras_is_poison_mode_supported(obj->adev)) { | |||
1672 | if (obj->head.block == AMDGPU_RAS_BLOCK__UMC) | |||
1673 | amdgpu_ras_interrupt_poison_creation_handler(obj, &entry); | |||
1674 | else | |||
1675 | amdgpu_ras_interrupt_poison_consumption_handler(obj, &entry); | |||
1676 | } else { | |||
1677 | if (obj->head.block == AMDGPU_RAS_BLOCK__UMC) | |||
1678 | amdgpu_ras_interrupt_umc_handler(obj, &entry); | |||
1679 | else | |||
1680 | dev_warn(obj->adev->dev,printf("drm:pid%d:%s *WARNING* " "No RAS interrupt handler for non-UMC block with poison disabled.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__) | |||
1681 | "No RAS interrupt handler for non-UMC block with poison disabled.\n")printf("drm:pid%d:%s *WARNING* " "No RAS interrupt handler for non-UMC block with poison disabled.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
1682 | } | |||
1683 | } | |||
1684 | } | |||
1685 | ||||
1686 | static void amdgpu_ras_interrupt_process_handler(struct work_struct *work) | |||
1687 | { | |||
1688 | struct ras_ih_data *data = | |||
1689 | container_of(work, struct ras_ih_data, ih_work)({ const __typeof( ((struct ras_ih_data *)0)->ih_work ) *__mptr = (work); (struct ras_ih_data *)( (char *)__mptr - __builtin_offsetof (struct ras_ih_data, ih_work) );}); | |||
1690 | struct ras_manager *obj = | |||
1691 | container_of(data, struct ras_manager, ih_data)({ const __typeof( ((struct ras_manager *)0)->ih_data ) *__mptr = (data); (struct ras_manager *)( (char *)__mptr - __builtin_offsetof (struct ras_manager, ih_data) );}); | |||
1692 | ||||
1693 | amdgpu_ras_interrupt_handler(obj); | |||
1694 | } | |||
1695 | ||||
1696 | int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev, | |||
1697 | struct ras_dispatch_if *info) | |||
1698 | { | |||
1699 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, &info->head); | |||
1700 | struct ras_ih_data *data = &obj->ih_data; | |||
1701 | ||||
1702 | if (!obj) | |||
1703 | return -EINVAL22; | |||
1704 | ||||
1705 | if (data->inuse == 0) | |||
1706 | return 0; | |||
1707 | ||||
1708 | /* Might be overflow... */ | |||
1709 | memcpy(&data->ring[data->wptr], info->entry,__builtin_memcpy((&data->ring[data->wptr]), (info-> entry), (data->element_size)) | |||
1710 | data->element_size)__builtin_memcpy((&data->ring[data->wptr]), (info-> entry), (data->element_size)); | |||
1711 | ||||
1712 | wmb()do { __asm volatile("sfence" ::: "memory"); } while (0); | |||
1713 | data->wptr = (data->aligned_element_size + | |||
1714 | data->wptr) % data->ring_size; | |||
1715 | ||||
1716 | schedule_work(&data->ih_work); | |||
1717 | ||||
1718 | return 0; | |||
1719 | } | |||
1720 | ||||
1721 | int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev, | |||
1722 | struct ras_common_if *head) | |||
1723 | { | |||
1724 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |||
1725 | struct ras_ih_data *data; | |||
1726 | ||||
1727 | if (!obj) | |||
1728 | return -EINVAL22; | |||
1729 | ||||
1730 | data = &obj->ih_data; | |||
1731 | if (data->inuse == 0) | |||
1732 | return 0; | |||
1733 | ||||
1734 | cancel_work_sync(&data->ih_work); | |||
1735 | ||||
1736 | kfree(data->ring); | |||
1737 | memset(data, 0, sizeof(*data))__builtin_memset((data), (0), (sizeof(*data))); | |||
1738 | put_obj(obj); | |||
1739 | ||||
1740 | return 0; | |||
1741 | } | |||
1742 | ||||
1743 | int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev, | |||
1744 | struct ras_common_if *head) | |||
1745 | { | |||
1746 | struct ras_manager *obj = amdgpu_ras_find_obj(adev, head); | |||
1747 | struct ras_ih_data *data; | |||
1748 | struct amdgpu_ras_block_object *ras_obj; | |||
1749 | ||||
1750 | if (!obj) { | |||
1751 | /* in case we registe the IH before enable ras feature */ | |||
1752 | obj = amdgpu_ras_create_obj(adev, head); | |||
1753 | if (!obj) | |||
1754 | return -EINVAL22; | |||
1755 | } else | |||
1756 | get_obj(obj)do { (obj)->use++; } while (0); | |||
1757 | ||||
1758 | ras_obj = container_of(head, struct amdgpu_ras_block_object, ras_comm)({ const __typeof( ((struct amdgpu_ras_block_object *)0)-> ras_comm ) *__mptr = (head); (struct amdgpu_ras_block_object * )( (char *)__mptr - __builtin_offsetof(struct amdgpu_ras_block_object , ras_comm) );}); | |||
1759 | ||||
1760 | data = &obj->ih_data; | |||
1761 | /* add the callback.etc */ | |||
1762 | *data = (struct ras_ih_data) { | |||
1763 | .inuse = 0, | |||
1764 | .cb = ras_obj->ras_cb, | |||
1765 | .element_size = sizeof(struct amdgpu_iv_entry), | |||
1766 | .rptr = 0, | |||
1767 | .wptr = 0, | |||
1768 | }; | |||
1769 | ||||
1770 | INIT_WORK(&data->ih_work, amdgpu_ras_interrupt_process_handler); | |||
1771 | ||||
1772 | data->aligned_element_size = roundup2(data->element_size, 8)(((data->element_size) + ((8) - 1)) & (~((__typeof(data ->element_size))(8) - 1))); | |||
1773 | /* the ring can store 64 iv entries. */ | |||
1774 | data->ring_size = 64 * data->aligned_element_size; | |||
1775 | data->ring = kmalloc(data->ring_size, GFP_KERNEL(0x0001 | 0x0004)); | |||
1776 | if (!data->ring) { | |||
1777 | put_obj(obj); | |||
1778 | return -ENOMEM12; | |||
1779 | } | |||
1780 | ||||
1781 | /* IH is ready */ | |||
1782 | data->inuse = 1; | |||
1783 | ||||
1784 | return 0; | |||
1785 | } | |||
1786 | ||||
1787 | static int amdgpu_ras_interrupt_remove_all(struct amdgpu_device *adev) | |||
1788 | { | |||
1789 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1790 | struct ras_manager *obj, *tmp; | |||
1791 | ||||
1792 | list_for_each_entry_safe(obj, tmp, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}), tmp = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj->node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) );}); &obj-> node != (&con->head); obj = tmp, tmp = ({ const __typeof ( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next ); (__typeof(*tmp) *)( (char *)__mptr - __builtin_offsetof(__typeof (*tmp), node) );})) { | |||
1793 | amdgpu_ras_interrupt_remove_handler(adev, &obj->head); | |||
1794 | } | |||
1795 | ||||
1796 | return 0; | |||
1797 | } | |||
1798 | /* ih end */ | |||
1799 | ||||
1800 | /* traversal all IPs except NBIO to query error counter */ | |||
1801 | static void amdgpu_ras_log_on_err_counter(struct amdgpu_device *adev) | |||
1802 | { | |||
1803 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1804 | struct ras_manager *obj; | |||
1805 | ||||
1806 | if (!adev->ras_enabled || !con) | |||
1807 | return; | |||
1808 | ||||
1809 | list_for_each_entry(obj, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}); &obj->node != (&con->head); obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj-> node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof (__typeof(*obj), node) );})) { | |||
1810 | struct ras_query_if info = { | |||
1811 | .head = obj->head, | |||
1812 | }; | |||
1813 | ||||
1814 | /* | |||
1815 | * PCIE_BIF IP has one different isr by ras controller | |||
1816 | * interrupt, the specific ras counter query will be | |||
1817 | * done in that isr. So skip such block from common | |||
1818 | * sync flood interrupt isr calling. | |||
1819 | */ | |||
1820 | if (info.head.block == AMDGPU_RAS_BLOCK__PCIE_BIF) | |||
1821 | continue; | |||
1822 | ||||
1823 | /* | |||
1824 | * this is a workaround for aldebaran, skip send msg to | |||
1825 | * smu to get ecc_info table due to smu handle get ecc | |||
1826 | * info table failed temporarily. | |||
1827 | * should be removed until smu fix handle ecc_info table. | |||
1828 | */ | |||
1829 | if ((info.head.block == AMDGPU_RAS_BLOCK__UMC) && | |||
1830 | (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(13, 0, 2)(((13) << 16) | ((0) << 8) | (2)))) | |||
1831 | continue; | |||
1832 | ||||
1833 | amdgpu_ras_query_error_status(adev, &info); | |||
1834 | ||||
1835 | if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 2)(((11) << 16) | ((0) << 8) | (2)) && | |||
1836 | adev->ip_versions[MP0_HWIP][0] != IP_VERSION(11, 0, 4)(((11) << 16) | ((0) << 8) | (4)) && | |||
1837 | adev->ip_versions[MP0_HWIP][0] != IP_VERSION(13, 0, 0)(((13) << 16) | ((0) << 8) | (0))) { | |||
1838 | if (amdgpu_ras_reset_error_status(adev, info.head.block)) | |||
1839 | dev_warn(adev->dev, "Failed to reset error counter and error status")printf("drm:pid%d:%s *WARNING* " "Failed to reset error counter and error status" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
1840 | } | |||
1841 | } | |||
1842 | } | |||
1843 | ||||
1844 | /* Parse RdRspStatus and WrRspStatus */ | |||
1845 | static void amdgpu_ras_error_status_query(struct amdgpu_device *adev, | |||
1846 | struct ras_query_if *info) | |||
1847 | { | |||
1848 | struct amdgpu_ras_block_object *block_obj; | |||
1849 | /* | |||
1850 | * Only two block need to query read/write | |||
1851 | * RspStatus at current state | |||
1852 | */ | |||
1853 | if ((info->head.block != AMDGPU_RAS_BLOCK__GFX) && | |||
1854 | (info->head.block != AMDGPU_RAS_BLOCK__MMHUB)) | |||
1855 | return; | |||
1856 | ||||
1857 | block_obj = amdgpu_ras_get_ras_block(adev, | |||
1858 | info->head.block, | |||
1859 | info->head.sub_block_index); | |||
1860 | ||||
1861 | if (!block_obj || !block_obj->hw_ops) { | |||
1862 | dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",do { } while(0) | |||
1863 | get_ras_block_str(&info->head))do { } while(0); | |||
1864 | return; | |||
1865 | } | |||
1866 | ||||
1867 | if (block_obj->hw_ops->query_ras_error_status) | |||
1868 | block_obj->hw_ops->query_ras_error_status(adev); | |||
1869 | ||||
1870 | } | |||
1871 | ||||
1872 | static void amdgpu_ras_query_err_status(struct amdgpu_device *adev) | |||
1873 | { | |||
1874 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1875 | struct ras_manager *obj; | |||
1876 | ||||
1877 | if (!adev->ras_enabled || !con) | |||
1878 | return; | |||
1879 | ||||
1880 | list_for_each_entry(obj, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}); &obj->node != (&con->head); obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj-> node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof (__typeof(*obj), node) );})) { | |||
1881 | struct ras_query_if info = { | |||
1882 | .head = obj->head, | |||
1883 | }; | |||
1884 | ||||
1885 | amdgpu_ras_error_status_query(adev, &info); | |||
1886 | } | |||
1887 | } | |||
1888 | ||||
1889 | /* recovery begin */ | |||
1890 | ||||
1891 | /* return 0 on success. | |||
1892 | * caller need free bps. | |||
1893 | */ | |||
1894 | static int amdgpu_ras_badpages_read(struct amdgpu_device *adev, | |||
1895 | struct ras_badpage **bps, unsigned int *count) | |||
1896 | { | |||
1897 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
1898 | struct ras_err_handler_data *data; | |||
1899 | int i = 0; | |||
1900 | int ret = 0, status; | |||
1901 | ||||
1902 | if (!con || !con->eh_data || !bps || !count) | |||
1903 | return -EINVAL22; | |||
1904 | ||||
1905 | mutex_lock(&con->recovery_lock)rw_enter_write(&con->recovery_lock); | |||
1906 | data = con->eh_data; | |||
1907 | if (!data || data->count == 0) { | |||
1908 | *bps = NULL((void *)0); | |||
1909 | ret = -EINVAL22; | |||
1910 | goto out; | |||
1911 | } | |||
1912 | ||||
1913 | *bps = kmalloc(sizeof(struct ras_badpage) * data->count, GFP_KERNEL(0x0001 | 0x0004)); | |||
1914 | if (!*bps) { | |||
1915 | ret = -ENOMEM12; | |||
1916 | goto out; | |||
1917 | } | |||
1918 | ||||
1919 | for (; i < data->count; i++) { | |||
1920 | (*bps)[i] = (struct ras_badpage){ | |||
1921 | .bp = data->bps[i].retired_page, | |||
1922 | .size = AMDGPU_GPU_PAGE_SIZE4096, | |||
1923 | .flags = AMDGPU_RAS_RETIRE_PAGE_RESERVED, | |||
1924 | }; | |||
1925 | status = amdgpu_vram_mgr_query_page_status(&adev->mman.vram_mgr, | |||
1926 | data->bps[i].retired_page); | |||
1927 | if (status == -EBUSY16) | |||
1928 | (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_PENDING; | |||
1929 | else if (status == -ENOENT2) | |||
1930 | (*bps)[i].flags = AMDGPU_RAS_RETIRE_PAGE_FAULT; | |||
1931 | } | |||
1932 | ||||
1933 | *count = data->count; | |||
1934 | out: | |||
1935 | mutex_unlock(&con->recovery_lock)rw_exit_write(&con->recovery_lock); | |||
1936 | return ret; | |||
1937 | } | |||
1938 | ||||
1939 | static void amdgpu_ras_do_recovery(struct work_struct *work) | |||
1940 | { | |||
1941 | struct amdgpu_ras *ras = | |||
1942 | container_of(work, struct amdgpu_ras, recovery_work)({ const __typeof( ((struct amdgpu_ras *)0)->recovery_work ) *__mptr = (work); (struct amdgpu_ras *)( (char *)__mptr - __builtin_offsetof (struct amdgpu_ras, recovery_work) );}); | |||
1943 | struct amdgpu_device *remote_adev = NULL((void *)0); | |||
1944 | struct amdgpu_device *adev = ras->adev; | |||
1945 | struct list_head device_list, *device_list_handle = NULL((void *)0); | |||
1946 | ||||
1947 | if (!ras->disable_ras_err_cnt_harvest) { | |||
1948 | struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); | |||
1949 | ||||
1950 | /* Build list of devices to query RAS related errors */ | |||
1951 | if (hive && adev->gmc.xgmi.num_physical_nodes > 1) { | |||
1952 | device_list_handle = &hive->device_list; | |||
1953 | } else { | |||
1954 | INIT_LIST_HEAD(&device_list); | |||
1955 | list_add_tail(&adev->gmc.xgmi.head, &device_list); | |||
1956 | device_list_handle = &device_list; | |||
1957 | } | |||
1958 | ||||
1959 | list_for_each_entry(remote_adev,for (remote_adev = ({ const __typeof( ((__typeof(*remote_adev ) *)0)->gmc.xgmi.head ) *__mptr = ((device_list_handle)-> next); (__typeof(*remote_adev) *)( (char *)__mptr - __builtin_offsetof (__typeof(*remote_adev), gmc.xgmi.head) );}); &remote_adev ->gmc.xgmi.head != (device_list_handle); remote_adev = ({ const __typeof( ((__typeof(*remote_adev) *)0)->gmc.xgmi.head ) * __mptr = (remote_adev->gmc.xgmi.head.next); (__typeof(*remote_adev ) *)( (char *)__mptr - __builtin_offsetof(__typeof(*remote_adev ), gmc.xgmi.head) );})) | |||
1960 | device_list_handle, gmc.xgmi.head)for (remote_adev = ({ const __typeof( ((__typeof(*remote_adev ) *)0)->gmc.xgmi.head ) *__mptr = ((device_list_handle)-> next); (__typeof(*remote_adev) *)( (char *)__mptr - __builtin_offsetof (__typeof(*remote_adev), gmc.xgmi.head) );}); &remote_adev ->gmc.xgmi.head != (device_list_handle); remote_adev = ({ const __typeof( ((__typeof(*remote_adev) *)0)->gmc.xgmi.head ) * __mptr = (remote_adev->gmc.xgmi.head.next); (__typeof(*remote_adev ) *)( (char *)__mptr - __builtin_offsetof(__typeof(*remote_adev ), gmc.xgmi.head) );})) { | |||
1961 | amdgpu_ras_query_err_status(remote_adev); | |||
1962 | amdgpu_ras_log_on_err_counter(remote_adev); | |||
1963 | } | |||
1964 | ||||
1965 | amdgpu_put_xgmi_hive(hive); | |||
1966 | } | |||
1967 | ||||
1968 | if (amdgpu_device_should_recover_gpu(ras->adev)) { | |||
1969 | struct amdgpu_reset_context reset_context; | |||
1970 | memset(&reset_context, 0, sizeof(reset_context))__builtin_memset((&reset_context), (0), (sizeof(reset_context ))); | |||
1971 | ||||
1972 | reset_context.method = AMD_RESET_METHOD_NONE; | |||
1973 | reset_context.reset_req_dev = adev; | |||
1974 | clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); | |||
1975 | ||||
1976 | amdgpu_device_gpu_recover(ras->adev, NULL((void *)0), &reset_context); | |||
1977 | } | |||
1978 | atomic_set(&ras->in_recovery, 0)({ typeof(*(&ras->in_recovery)) __tmp = ((0)); *(volatile typeof(*(&ras->in_recovery)) *)&(*(&ras->in_recovery )) = __tmp; __tmp; }); | |||
1979 | } | |||
1980 | ||||
1981 | /* alloc/realloc bps array */ | |||
1982 | static int amdgpu_ras_realloc_eh_data_space(struct amdgpu_device *adev, | |||
1983 | struct ras_err_handler_data *data, int pages) | |||
1984 | { | |||
1985 | unsigned int old_space = data->count + data->space_left; | |||
1986 | unsigned int new_space = old_space + pages; | |||
1987 | unsigned int align_space = roundup2(new_space, 512)(((new_space) + ((512) - 1)) & (~((__typeof(new_space))(512 ) - 1))); | |||
1988 | void *bps = kmalloc(align_space * sizeof(*data->bps), GFP_KERNEL(0x0001 | 0x0004)); | |||
1989 | ||||
1990 | if (!bps) { | |||
1991 | return -ENOMEM12; | |||
1992 | } | |||
1993 | ||||
1994 | if (data->bps) { | |||
1995 | memcpy(bps, data->bps,__builtin_memcpy((bps), (data->bps), (data->count * sizeof (*data->bps))) | |||
1996 | data->count * sizeof(*data->bps))__builtin_memcpy((bps), (data->bps), (data->count * sizeof (*data->bps))); | |||
1997 | kfree(data->bps); | |||
1998 | } | |||
1999 | ||||
2000 | data->bps = bps; | |||
2001 | data->space_left += align_space - old_space; | |||
2002 | return 0; | |||
2003 | } | |||
2004 | ||||
2005 | /* it deal with vram only. */ | |||
2006 | int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev, | |||
2007 | struct eeprom_table_record *bps, int pages) | |||
2008 | { | |||
2009 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2010 | struct ras_err_handler_data *data; | |||
2011 | int ret = 0; | |||
2012 | uint32_t i; | |||
2013 | ||||
2014 | if (!con || !con->eh_data || !bps || pages <= 0) | |||
2015 | return 0; | |||
2016 | ||||
2017 | mutex_lock(&con->recovery_lock)rw_enter_write(&con->recovery_lock); | |||
2018 | data = con->eh_data; | |||
2019 | if (!data) | |||
2020 | goto out; | |||
2021 | ||||
2022 | for (i = 0; i < pages; i++) { | |||
2023 | if (amdgpu_ras_check_bad_page_unlock(con, | |||
2024 | bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT12)) | |||
2025 | continue; | |||
2026 | ||||
2027 | if (!data->space_left && | |||
2028 | amdgpu_ras_realloc_eh_data_space(adev, data, 256)) { | |||
2029 | ret = -ENOMEM12; | |||
2030 | goto out; | |||
2031 | } | |||
2032 | ||||
2033 | amdgpu_vram_mgr_reserve_range(&adev->mman.vram_mgr, | |||
2034 | bps[i].retired_page << AMDGPU_GPU_PAGE_SHIFT12, | |||
2035 | AMDGPU_GPU_PAGE_SIZE4096); | |||
2036 | ||||
2037 | memcpy(&data->bps[data->count], &bps[i], sizeof(*data->bps))__builtin_memcpy((&data->bps[data->count]), (&bps [i]), (sizeof(*data->bps))); | |||
2038 | data->count++; | |||
2039 | data->space_left--; | |||
2040 | } | |||
2041 | out: | |||
2042 | mutex_unlock(&con->recovery_lock)rw_exit_write(&con->recovery_lock); | |||
2043 | ||||
2044 | return ret; | |||
2045 | } | |||
2046 | ||||
2047 | /* | |||
2048 | * write error record array to eeprom, the function should be | |||
2049 | * protected by recovery_lock | |||
2050 | */ | |||
2051 | int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev) | |||
2052 | { | |||
2053 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2054 | struct ras_err_handler_data *data; | |||
2055 | struct amdgpu_ras_eeprom_control *control; | |||
2056 | int save_count; | |||
2057 | ||||
2058 | if (!con || !con->eh_data) | |||
2059 | return 0; | |||
2060 | ||||
2061 | mutex_lock(&con->recovery_lock)rw_enter_write(&con->recovery_lock); | |||
2062 | control = &con->eeprom_control; | |||
2063 | data = con->eh_data; | |||
2064 | save_count = data->count - control->ras_num_recs; | |||
2065 | mutex_unlock(&con->recovery_lock)rw_exit_write(&con->recovery_lock); | |||
2066 | /* only new entries are saved */ | |||
2067 | if (save_count > 0) { | |||
2068 | if (amdgpu_ras_eeprom_append(control, | |||
2069 | &data->bps[control->ras_num_recs], | |||
2070 | save_count)) { | |||
2071 | dev_err(adev->dev, "Failed to save EEPROM table data!")printf("drm:pid%d:%s *ERROR* " "Failed to save EEPROM table data!" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
2072 | return -EIO5; | |||
2073 | } | |||
2074 | ||||
2075 | dev_info(adev->dev, "Saved %d pages to EEPROM table.\n", save_count)do { } while(0); | |||
2076 | } | |||
2077 | ||||
2078 | return 0; | |||
2079 | } | |||
2080 | ||||
2081 | /* | |||
2082 | * read error record array in eeprom and reserve enough space for | |||
2083 | * storing new bad pages | |||
2084 | */ | |||
2085 | static int amdgpu_ras_load_bad_pages(struct amdgpu_device *adev) | |||
2086 | { | |||
2087 | struct amdgpu_ras_eeprom_control *control = | |||
2088 | &adev->psp.ras_context.ras->eeprom_control; | |||
2089 | struct eeprom_table_record *bps; | |||
2090 | int ret; | |||
2091 | ||||
2092 | /* no bad page record, skip eeprom access */ | |||
2093 | if (control->ras_num_recs == 0 || amdgpu_bad_page_threshold == 0) | |||
2094 | return 0; | |||
2095 | ||||
2096 | bps = kcalloc(control->ras_num_recs, sizeof(*bps), GFP_KERNEL(0x0001 | 0x0004)); | |||
2097 | if (!bps) | |||
2098 | return -ENOMEM12; | |||
2099 | ||||
2100 | ret = amdgpu_ras_eeprom_read(control, bps, control->ras_num_recs); | |||
2101 | if (ret) | |||
2102 | dev_err(adev->dev, "Failed to load EEPROM table records!")printf("drm:pid%d:%s *ERROR* " "Failed to load EEPROM table records!" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
2103 | else | |||
2104 | ret = amdgpu_ras_add_bad_pages(adev, bps, control->ras_num_recs); | |||
2105 | ||||
2106 | kfree(bps); | |||
2107 | return ret; | |||
2108 | } | |||
2109 | ||||
2110 | static bool_Bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, | |||
2111 | uint64_t addr) | |||
2112 | { | |||
2113 | struct ras_err_handler_data *data = con->eh_data; | |||
2114 | int i; | |||
2115 | ||||
2116 | addr >>= AMDGPU_GPU_PAGE_SHIFT12; | |||
2117 | for (i = 0; i < data->count; i++) | |||
2118 | if (addr == data->bps[i].retired_page) | |||
2119 | return true1; | |||
2120 | ||||
2121 | return false0; | |||
2122 | } | |||
2123 | ||||
2124 | /* | |||
2125 | * check if an address belongs to bad page | |||
2126 | * | |||
2127 | * Note: this check is only for umc block | |||
2128 | */ | |||
2129 | static bool_Bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev, | |||
2130 | uint64_t addr) | |||
2131 | { | |||
2132 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2133 | bool_Bool ret = false0; | |||
2134 | ||||
2135 | if (!con || !con->eh_data) | |||
2136 | return ret; | |||
2137 | ||||
2138 | mutex_lock(&con->recovery_lock)rw_enter_write(&con->recovery_lock); | |||
2139 | ret = amdgpu_ras_check_bad_page_unlock(con, addr); | |||
2140 | mutex_unlock(&con->recovery_lock)rw_exit_write(&con->recovery_lock); | |||
2141 | return ret; | |||
2142 | } | |||
2143 | ||||
2144 | static void amdgpu_ras_validate_threshold(struct amdgpu_device *adev, | |||
2145 | uint32_t max_count) | |||
2146 | { | |||
2147 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2148 | ||||
2149 | /* | |||
2150 | * Justification of value bad_page_cnt_threshold in ras structure | |||
2151 | * | |||
2152 | * Generally, -1 <= amdgpu_bad_page_threshold <= max record length | |||
2153 | * in eeprom, and introduce two scenarios accordingly. | |||
2154 | * | |||
2155 | * Bad page retirement enablement: | |||
2156 | * - If amdgpu_bad_page_threshold = -1, | |||
2157 | * bad_page_cnt_threshold = typical value by formula. | |||
2158 | * | |||
2159 | * - When the value from user is 0 < amdgpu_bad_page_threshold < | |||
2160 | * max record length in eeprom, use it directly. | |||
2161 | * | |||
2162 | * Bad page retirement disablement: | |||
2163 | * - If amdgpu_bad_page_threshold = 0, bad page retirement | |||
2164 | * functionality is disabled, and bad_page_cnt_threshold will | |||
2165 | * take no effect. | |||
2166 | */ | |||
2167 | ||||
2168 | if (amdgpu_bad_page_threshold < 0) { | |||
2169 | u64 val = adev->gmc.mc_vram_size; | |||
2170 | ||||
2171 | do_div(val, RAS_BAD_PAGE_COVER)({ uint32_t __base = ((100 * 1024 * 1024ULL)); uint32_t __rem = ((uint64_t)(val)) % __base; (val) = ((uint64_t)(val)) / __base ; __rem; }); | |||
2172 | con->bad_page_cnt_threshold = min(lower_32_bits(val),(((((u32)(val)))<(max_count))?(((u32)(val))):(max_count)) | |||
2173 | max_count)(((((u32)(val)))<(max_count))?(((u32)(val))):(max_count)); | |||
2174 | } else { | |||
2175 | con->bad_page_cnt_threshold = min_t(int, max_count,({ int __min_a = (max_count); int __min_b = (amdgpu_bad_page_threshold ); __min_a < __min_b ? __min_a : __min_b; }) | |||
2176 | amdgpu_bad_page_threshold)({ int __min_a = (max_count); int __min_b = (amdgpu_bad_page_threshold ); __min_a < __min_b ? __min_a : __min_b; }); | |||
2177 | } | |||
2178 | } | |||
2179 | ||||
2180 | int amdgpu_ras_recovery_init(struct amdgpu_device *adev) | |||
2181 | { | |||
2182 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2183 | struct ras_err_handler_data **data; | |||
2184 | u32 max_eeprom_records_count = 0; | |||
2185 | bool_Bool exc_err_limit = false0; | |||
2186 | int ret; | |||
2187 | ||||
2188 | if (!con || amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) | |||
2189 | return 0; | |||
2190 | ||||
2191 | /* Allow access to RAS EEPROM via debugfs, when the ASIC | |||
2192 | * supports RAS and debugfs is enabled, but when | |||
2193 | * adev->ras_enabled is unset, i.e. when "ras_enable" | |||
2194 | * module parameter is set to 0. | |||
2195 | */ | |||
2196 | con->adev = adev; | |||
2197 | ||||
2198 | if (!adev->ras_enabled) | |||
2199 | return 0; | |||
2200 | ||||
2201 | data = &con->eh_data; | |||
2202 | *data = kmalloc(sizeof(**data), GFP_KERNEL(0x0001 | 0x0004) | __GFP_ZERO0x0008); | |||
2203 | if (!*data) { | |||
2204 | ret = -ENOMEM12; | |||
2205 | goto out; | |||
2206 | } | |||
2207 | ||||
2208 | rw_init(&con->recovery_lock, "rasrec")_rw_init_flags(&con->recovery_lock, "rasrec", 0, ((void *)0)); | |||
2209 | INIT_WORK(&con->recovery_work, amdgpu_ras_do_recovery); | |||
2210 | atomic_set(&con->in_recovery, 0)({ typeof(*(&con->in_recovery)) __tmp = ((0)); *(volatile typeof(*(&con->in_recovery)) *)&(*(&con->in_recovery )) = __tmp; __tmp; }); | |||
2211 | con->eeprom_control.bad_channel_bitmap = 0; | |||
2212 | ||||
2213 | max_eeprom_records_count = amdgpu_ras_eeprom_max_record_count(); | |||
2214 | amdgpu_ras_validate_threshold(adev, max_eeprom_records_count); | |||
2215 | ||||
2216 | /* Todo: During test the SMU might fail to read the eeprom through I2C | |||
2217 | * when the GPU is pending on XGMI reset during probe time | |||
2218 | * (Mostly after second bus reset), skip it now | |||
2219 | */ | |||
2220 | if (adev->gmc.xgmi.pending_reset) | |||
2221 | return 0; | |||
2222 | ret = amdgpu_ras_eeprom_init(&con->eeprom_control, &exc_err_limit); | |||
2223 | /* | |||
2224 | * This calling fails when exc_err_limit is true or | |||
2225 | * ret != 0. | |||
2226 | */ | |||
2227 | if (exc_err_limit || ret) | |||
2228 | goto free; | |||
2229 | ||||
2230 | if (con->eeprom_control.ras_num_recs) { | |||
2231 | ret = amdgpu_ras_load_bad_pages(adev); | |||
2232 | if (ret) | |||
2233 | goto free; | |||
2234 | ||||
2235 | amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs); | |||
2236 | ||||
2237 | if (con->update_channel_flag == true1) { | |||
2238 | amdgpu_dpm_send_hbm_bad_channel_flag(adev, con->eeprom_control.bad_channel_bitmap); | |||
2239 | con->update_channel_flag = false0; | |||
2240 | } | |||
2241 | } | |||
2242 | ||||
2243 | #ifdef CONFIG_X86_MCE_AMD | |||
2244 | if ((adev->asic_type == CHIP_ALDEBARAN) && | |||
2245 | (adev->gmc.xgmi.connected_to_cpu)) | |||
2246 | amdgpu_register_bad_pages_mca_notifier(adev); | |||
2247 | #endif | |||
2248 | return 0; | |||
2249 | ||||
2250 | free: | |||
2251 | kfree((*data)->bps); | |||
2252 | kfree(*data); | |||
2253 | con->eh_data = NULL((void *)0); | |||
2254 | out: | |||
2255 | dev_warn(adev->dev, "Failed to initialize ras recovery! (%d)\n", ret)printf("drm:pid%d:%s *WARNING* " "Failed to initialize ras recovery! (%d)\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , ret); | |||
2256 | ||||
2257 | /* | |||
2258 | * Except error threshold exceeding case, other failure cases in this | |||
2259 | * function would not fail amdgpu driver init. | |||
2260 | */ | |||
2261 | if (!exc_err_limit) | |||
2262 | ret = 0; | |||
2263 | else | |||
2264 | ret = -EINVAL22; | |||
2265 | ||||
2266 | return ret; | |||
2267 | } | |||
2268 | ||||
2269 | static int amdgpu_ras_recovery_fini(struct amdgpu_device *adev) | |||
2270 | { | |||
2271 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2272 | struct ras_err_handler_data *data = con->eh_data; | |||
2273 | ||||
2274 | /* recovery_init failed to init it, fini is useless */ | |||
2275 | if (!data) | |||
2276 | return 0; | |||
2277 | ||||
2278 | cancel_work_sync(&con->recovery_work); | |||
2279 | ||||
2280 | mutex_lock(&con->recovery_lock)rw_enter_write(&con->recovery_lock); | |||
2281 | con->eh_data = NULL((void *)0); | |||
2282 | kfree(data->bps); | |||
2283 | kfree(data); | |||
2284 | mutex_unlock(&con->recovery_lock)rw_exit_write(&con->recovery_lock); | |||
2285 | ||||
2286 | return 0; | |||
2287 | } | |||
2288 | /* recovery end */ | |||
2289 | ||||
2290 | static bool_Bool amdgpu_ras_asic_supported(struct amdgpu_device *adev) | |||
2291 | { | |||
2292 | if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) { | |||
2293 | switch (adev->ip_versions[MP0_HWIP][0]) { | |||
2294 | case IP_VERSION(13, 0, 2)(((13) << 16) | ((0) << 8) | (2)): | |||
2295 | return true1; | |||
2296 | default: | |||
2297 | return false0; | |||
2298 | } | |||
2299 | } | |||
2300 | ||||
2301 | if (adev->asic_type == CHIP_IP_DISCOVERY) { | |||
2302 | switch (adev->ip_versions[MP0_HWIP][0]) { | |||
2303 | case IP_VERSION(13, 0, 0)(((13) << 16) | ((0) << 8) | (0)): | |||
2304 | case IP_VERSION(13, 0, 10)(((13) << 16) | ((0) << 8) | (10)): | |||
2305 | return true1; | |||
2306 | default: | |||
2307 | return false0; | |||
2308 | } | |||
2309 | } | |||
2310 | ||||
2311 | return adev->asic_type == CHIP_VEGA10 || | |||
2312 | adev->asic_type == CHIP_VEGA20 || | |||
2313 | adev->asic_type == CHIP_ARCTURUS || | |||
2314 | adev->asic_type == CHIP_ALDEBARAN || | |||
2315 | adev->asic_type == CHIP_SIENNA_CICHLID; | |||
2316 | } | |||
2317 | ||||
2318 | /* | |||
2319 | * this is workaround for vega20 workstation sku, | |||
2320 | * force enable gfx ras, ignore vbios gfx ras flag | |||
2321 | * due to GC EDC can not write | |||
2322 | */ | |||
2323 | static void amdgpu_ras_get_quirks(struct amdgpu_device *adev) | |||
2324 | { | |||
2325 | struct atom_context *ctx = adev->mode_info.atom_context; | |||
2326 | ||||
2327 | if (!ctx) | |||
2328 | return; | |||
2329 | ||||
2330 | if (strnstr(ctx->vbios_version, "D16406", | |||
2331 | sizeof(ctx->vbios_version)) || | |||
2332 | strnstr(ctx->vbios_version, "D36002", | |||
2333 | sizeof(ctx->vbios_version))) | |||
2334 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__GFX); | |||
2335 | } | |||
2336 | ||||
2337 | /* | |||
2338 | * check hardware's ras ability which will be saved in hw_supported. | |||
2339 | * if hardware does not support ras, we can skip some ras initializtion and | |||
2340 | * forbid some ras operations from IP. | |||
2341 | * if software itself, say boot parameter, limit the ras ability. We still | |||
2342 | * need allow IP do some limited operations, like disable. In such case, | |||
2343 | * we have to initialize ras as normal. but need check if operation is | |||
2344 | * allowed or not in each function. | |||
2345 | */ | |||
2346 | static void amdgpu_ras_check_supported(struct amdgpu_device *adev) | |||
2347 | { | |||
2348 | adev->ras_hw_enabled = adev->ras_enabled = 0; | |||
2349 | ||||
2350 | if (!adev->is_atom_fw || | |||
2351 | !amdgpu_ras_asic_supported(adev)) | |||
2352 | return; | |||
2353 | ||||
2354 | if (!adev->gmc.xgmi.connected_to_cpu) { | |||
2355 | if (amdgpu_atomfirmware_mem_ecc_supported(adev)) { | |||
2356 | dev_info(adev->dev, "MEM ECC is active.\n")do { } while(0); | |||
2357 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__UMC | | |||
2358 | 1 << AMDGPU_RAS_BLOCK__DF); | |||
2359 | } else { | |||
2360 | dev_info(adev->dev, "MEM ECC is not presented.\n")do { } while(0); | |||
2361 | } | |||
2362 | ||||
2363 | if (amdgpu_atomfirmware_sram_ecc_supported(adev)) { | |||
2364 | dev_info(adev->dev, "SRAM ECC is active.\n")do { } while(0); | |||
2365 | if (!amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) { | |||
2366 | adev->ras_hw_enabled |= ~(1 << AMDGPU_RAS_BLOCK__UMC | | |||
2367 | 1 << AMDGPU_RAS_BLOCK__DF); | |||
2368 | ||||
2369 | if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(2, 6, 0)(((2) << 16) | ((6) << 8) | (0))) | |||
2370 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__VCN | | |||
2371 | 1 << AMDGPU_RAS_BLOCK__JPEG); | |||
2372 | else | |||
2373 | adev->ras_hw_enabled &= ~(1 << AMDGPU_RAS_BLOCK__VCN | | |||
2374 | 1 << AMDGPU_RAS_BLOCK__JPEG); | |||
2375 | } else { | |||
2376 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__PCIE_BIF | | |||
2377 | 1 << AMDGPU_RAS_BLOCK__SDMA | | |||
2378 | 1 << AMDGPU_RAS_BLOCK__GFX); | |||
2379 | } | |||
2380 | } else { | |||
2381 | dev_info(adev->dev, "SRAM ECC is not presented.\n")do { } while(0); | |||
2382 | } | |||
2383 | } else { | |||
2384 | /* driver only manages a few IP blocks RAS feature | |||
2385 | * when GPU is connected cpu through XGMI */ | |||
2386 | adev->ras_hw_enabled |= (1 << AMDGPU_RAS_BLOCK__GFX | | |||
2387 | 1 << AMDGPU_RAS_BLOCK__SDMA | | |||
2388 | 1 << AMDGPU_RAS_BLOCK__MMHUB); | |||
2389 | } | |||
2390 | ||||
2391 | amdgpu_ras_get_quirks(adev); | |||
2392 | ||||
2393 | /* hw_supported needs to be aligned with RAS block mask. */ | |||
2394 | adev->ras_hw_enabled &= AMDGPU_RAS_BLOCK_MASK((1ULL << AMDGPU_RAS_BLOCK__LAST) - 1); | |||
2395 | ||||
2396 | adev->ras_enabled = amdgpu_ras_enable == 0 ? 0 : | |||
2397 | adev->ras_hw_enabled & amdgpu_ras_mask; | |||
2398 | } | |||
2399 | ||||
2400 | static void amdgpu_ras_counte_dw(struct work_struct *work) | |||
2401 | { | |||
2402 | struct amdgpu_ras *con = container_of(work, struct amdgpu_ras,({ const __typeof( ((struct amdgpu_ras *)0)->ras_counte_delay_work .work ) *__mptr = (work); (struct amdgpu_ras *)( (char *)__mptr - __builtin_offsetof(struct amdgpu_ras, ras_counte_delay_work .work) );}) | |||
2403 | ras_counte_delay_work.work)({ const __typeof( ((struct amdgpu_ras *)0)->ras_counte_delay_work .work ) *__mptr = (work); (struct amdgpu_ras *)( (char *)__mptr - __builtin_offsetof(struct amdgpu_ras, ras_counte_delay_work .work) );}); | |||
2404 | struct amdgpu_device *adev = con->adev; | |||
2405 | struct drm_device *dev = adev_to_drm(adev); | |||
2406 | unsigned long ce_count, ue_count; | |||
2407 | int res; | |||
2408 | ||||
2409 | res = pm_runtime_get_sync(dev->dev); | |||
2410 | if (res < 0) | |||
2411 | goto Out; | |||
2412 | ||||
2413 | /* Cache new values. | |||
2414 | */ | |||
2415 | if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count) == 0) { | |||
2416 | atomic_set(&con->ras_ce_count, ce_count)({ typeof(*(&con->ras_ce_count)) __tmp = ((ce_count)); *(volatile typeof(*(&con->ras_ce_count)) *)&(*(& con->ras_ce_count)) = __tmp; __tmp; }); | |||
2417 | atomic_set(&con->ras_ue_count, ue_count)({ typeof(*(&con->ras_ue_count)) __tmp = ((ue_count)); *(volatile typeof(*(&con->ras_ue_count)) *)&(*(& con->ras_ue_count)) = __tmp; __tmp; }); | |||
2418 | } | |||
2419 | ||||
2420 | pm_runtime_mark_last_busy(dev->dev); | |||
2421 | Out: | |||
2422 | pm_runtime_put_autosuspend(dev->dev); | |||
2423 | } | |||
2424 | ||||
2425 | int amdgpu_ras_init(struct amdgpu_device *adev) | |||
2426 | { | |||
2427 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2428 | int r; | |||
2429 | bool_Bool df_poison, umc_poison; | |||
2430 | ||||
2431 | if (con) | |||
2432 | return 0; | |||
2433 | ||||
2434 | con = kmalloc(sizeof(struct amdgpu_ras) + | |||
2435 | sizeof(struct ras_manager) * AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST + | |||
2436 | sizeof(struct ras_manager) * AMDGPU_RAS_MCA_BLOCK_COUNTAMDGPU_RAS_MCA_BLOCK__LAST, | |||
2437 | GFP_KERNEL(0x0001 | 0x0004)|__GFP_ZERO0x0008); | |||
2438 | if (!con) | |||
2439 | return -ENOMEM12; | |||
2440 | ||||
2441 | con->adev = adev; | |||
2442 | INIT_DELAYED_WORK(&con->ras_counte_delay_work, amdgpu_ras_counte_dw); | |||
2443 | atomic_set(&con->ras_ce_count, 0)({ typeof(*(&con->ras_ce_count)) __tmp = ((0)); *(volatile typeof(*(&con->ras_ce_count)) *)&(*(&con-> ras_ce_count)) = __tmp; __tmp; }); | |||
2444 | atomic_set(&con->ras_ue_count, 0)({ typeof(*(&con->ras_ue_count)) __tmp = ((0)); *(volatile typeof(*(&con->ras_ue_count)) *)&(*(&con-> ras_ue_count)) = __tmp; __tmp; }); | |||
2445 | ||||
2446 | con->objs = (struct ras_manager *)(con + 1); | |||
2447 | ||||
2448 | amdgpu_ras_set_context(adev, con); | |||
2449 | ||||
2450 | amdgpu_ras_check_supported(adev); | |||
2451 | ||||
2452 | if (!adev->ras_enabled || adev->asic_type == CHIP_VEGA10) { | |||
2453 | /* set gfx block ras context feature for VEGA20 Gaming | |||
2454 | * send ras disable cmd to ras ta during ras late init. | |||
2455 | */ | |||
2456 | if (!adev->ras_enabled && adev->asic_type == CHIP_VEGA20) { | |||
2457 | con->features |= BIT(AMDGPU_RAS_BLOCK__GFX)(1UL << (AMDGPU_RAS_BLOCK__GFX)); | |||
2458 | ||||
2459 | return 0; | |||
2460 | } | |||
2461 | ||||
2462 | r = 0; | |||
2463 | goto release_con; | |||
2464 | } | |||
2465 | ||||
2466 | con->update_channel_flag = false0; | |||
2467 | con->features = 0; | |||
2468 | INIT_LIST_HEAD(&con->head); | |||
2469 | /* Might need get this flag from vbios. */ | |||
2470 | con->flags = RAS_DEFAULT_FLAGS((0x1 << 0)); | |||
2471 | ||||
2472 | /* initialize nbio ras function ahead of any other | |||
2473 | * ras functions so hardware fatal error interrupt | |||
2474 | * can be enabled as early as possible */ | |||
2475 | switch (adev->asic_type) { | |||
2476 | case CHIP_VEGA20: | |||
2477 | case CHIP_ARCTURUS: | |||
2478 | case CHIP_ALDEBARAN: | |||
2479 | if (!adev->gmc.xgmi.connected_to_cpu) { | |||
2480 | adev->nbio.ras = &nbio_v7_4_ras; | |||
2481 | amdgpu_ras_register_ras_block(adev, &adev->nbio.ras->ras_block); | |||
2482 | adev->nbio.ras_if = &adev->nbio.ras->ras_block.ras_comm; | |||
2483 | } | |||
2484 | break; | |||
2485 | default: | |||
2486 | /* nbio ras is not available */ | |||
2487 | break; | |||
2488 | } | |||
2489 | ||||
2490 | if (adev->nbio.ras && | |||
2491 | adev->nbio.ras->init_ras_controller_interrupt) { | |||
2492 | r = adev->nbio.ras->init_ras_controller_interrupt(adev); | |||
2493 | if (r) | |||
2494 | goto release_con; | |||
2495 | } | |||
2496 | ||||
2497 | if (adev->nbio.ras && | |||
2498 | adev->nbio.ras->init_ras_err_event_athub_interrupt) { | |||
2499 | r = adev->nbio.ras->init_ras_err_event_athub_interrupt(adev); | |||
2500 | if (r) | |||
2501 | goto release_con; | |||
2502 | } | |||
2503 | ||||
2504 | /* Init poison supported flag, the default value is false */ | |||
2505 | if (adev->gmc.xgmi.connected_to_cpu) { | |||
2506 | /* enabled by default when GPU is connected to CPU */ | |||
2507 | con->poison_supported = true1; | |||
2508 | } | |||
2509 | else if (adev->df.funcs && | |||
2510 | adev->df.funcs->query_ras_poison_mode && | |||
2511 | adev->umc.ras && | |||
2512 | adev->umc.ras->query_ras_poison_mode) { | |||
2513 | df_poison = | |||
2514 | adev->df.funcs->query_ras_poison_mode(adev); | |||
2515 | umc_poison = | |||
2516 | adev->umc.ras->query_ras_poison_mode(adev); | |||
2517 | /* Only poison is set in both DF and UMC, we can support it */ | |||
2518 | if (df_poison && umc_poison) | |||
2519 | con->poison_supported = true1; | |||
2520 | else if (df_poison != umc_poison) | |||
2521 | dev_warn(adev->dev, "Poison setting is inconsistent in DF/UMC(%d:%d)!\n",printf("drm:pid%d:%s *WARNING* " "Poison setting is inconsistent in DF/UMC(%d:%d)!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , df_poison , umc_poison) | |||
2522 | df_poison, umc_poison)printf("drm:pid%d:%s *WARNING* " "Poison setting is inconsistent in DF/UMC(%d:%d)!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , df_poison , umc_poison); | |||
2523 | } | |||
2524 | ||||
2525 | if (amdgpu_ras_fs_init(adev)) { | |||
2526 | r = -EINVAL22; | |||
2527 | goto release_con; | |||
2528 | } | |||
2529 | ||||
2530 | dev_info(adev->dev, "RAS INFO: ras initialized successfully, "do { } while(0) | |||
2531 | "hardware ability[%x] ras_mask[%x]\n",do { } while(0) | |||
2532 | adev->ras_hw_enabled, adev->ras_enabled)do { } while(0); | |||
2533 | ||||
2534 | return 0; | |||
2535 | release_con: | |||
2536 | amdgpu_ras_set_context(adev, NULL((void *)0)); | |||
2537 | kfree(con); | |||
2538 | ||||
2539 | return r; | |||
2540 | } | |||
2541 | ||||
2542 | int amdgpu_persistent_edc_harvesting_supported(struct amdgpu_device *adev) | |||
2543 | { | |||
2544 | if (adev->gmc.xgmi.connected_to_cpu) | |||
2545 | return 1; | |||
2546 | return 0; | |||
2547 | } | |||
2548 | ||||
2549 | static int amdgpu_persistent_edc_harvesting(struct amdgpu_device *adev, | |||
2550 | struct ras_common_if *ras_block) | |||
2551 | { | |||
2552 | struct ras_query_if info = { | |||
2553 | .head = *ras_block, | |||
2554 | }; | |||
2555 | ||||
2556 | if (!amdgpu_persistent_edc_harvesting_supported(adev)) | |||
2557 | return 0; | |||
2558 | ||||
2559 | if (amdgpu_ras_query_error_status(adev, &info) != 0) | |||
2560 | DRM_WARN("RAS init harvest failure")printk("\0014" "[" "drm" "] " "RAS init harvest failure"); | |||
2561 | ||||
2562 | if (amdgpu_ras_reset_error_status(adev, ras_block->block) != 0) | |||
2563 | DRM_WARN("RAS init harvest reset failure")printk("\0014" "[" "drm" "] " "RAS init harvest reset failure" ); | |||
2564 | ||||
2565 | return 0; | |||
2566 | } | |||
2567 | ||||
2568 | bool_Bool amdgpu_ras_is_poison_mode_supported(struct amdgpu_device *adev) | |||
2569 | { | |||
2570 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2571 | ||||
2572 | if (!con) | |||
2573 | return false0; | |||
2574 | ||||
2575 | return con->poison_supported; | |||
2576 | } | |||
2577 | ||||
2578 | /* helper function to handle common stuff in ip late init phase */ | |||
2579 | int amdgpu_ras_block_late_init(struct amdgpu_device *adev, | |||
2580 | struct ras_common_if *ras_block) | |||
2581 | { | |||
2582 | struct amdgpu_ras_block_object *ras_obj = NULL((void *)0); | |||
2583 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2584 | unsigned long ue_count, ce_count; | |||
2585 | int r; | |||
2586 | ||||
2587 | /* disable RAS feature per IP block if it is not supported */ | |||
2588 | if (!amdgpu_ras_is_supported(adev, ras_block->block)) { | |||
2589 | amdgpu_ras_feature_enable_on_boot(adev, ras_block, 0); | |||
2590 | return 0; | |||
2591 | } | |||
2592 | ||||
2593 | r = amdgpu_ras_feature_enable_on_boot(adev, ras_block, 1); | |||
2594 | if (r) { | |||
2595 | if (adev->in_suspend || amdgpu_in_reset(adev)) { | |||
2596 | /* in resume phase, if fail to enable ras, | |||
2597 | * clean up all ras fs nodes, and disable ras */ | |||
2598 | goto cleanup; | |||
2599 | } else | |||
2600 | return r; | |||
2601 | } | |||
2602 | ||||
2603 | /* check for errors on warm reset edc persisant supported ASIC */ | |||
2604 | amdgpu_persistent_edc_harvesting(adev, ras_block); | |||
2605 | ||||
2606 | /* in resume phase, no need to create ras fs node */ | |||
2607 | if (adev->in_suspend || amdgpu_in_reset(adev)) | |||
2608 | return 0; | |||
2609 | ||||
2610 | ras_obj = container_of(ras_block, struct amdgpu_ras_block_object, ras_comm)({ const __typeof( ((struct amdgpu_ras_block_object *)0)-> ras_comm ) *__mptr = (ras_block); (struct amdgpu_ras_block_object *)( (char *)__mptr - __builtin_offsetof(struct amdgpu_ras_block_object , ras_comm) );}); | |||
2611 | if (ras_obj->ras_cb || (ras_obj->hw_ops && | |||
2612 | (ras_obj->hw_ops->query_poison_status || | |||
2613 | ras_obj->hw_ops->handle_poison_consumption))) { | |||
2614 | r = amdgpu_ras_interrupt_add_handler(adev, ras_block); | |||
2615 | if (r) | |||
2616 | goto cleanup; | |||
2617 | } | |||
2618 | ||||
2619 | r = amdgpu_ras_sysfs_create(adev, ras_block); | |||
2620 | if (r) | |||
2621 | goto interrupt; | |||
2622 | ||||
2623 | /* Those are the cached values at init. | |||
2624 | */ | |||
2625 | if (amdgpu_ras_query_error_count(adev, &ce_count, &ue_count) == 0) { | |||
2626 | atomic_set(&con->ras_ce_count, ce_count)({ typeof(*(&con->ras_ce_count)) __tmp = ((ce_count)); *(volatile typeof(*(&con->ras_ce_count)) *)&(*(& con->ras_ce_count)) = __tmp; __tmp; }); | |||
2627 | atomic_set(&con->ras_ue_count, ue_count)({ typeof(*(&con->ras_ue_count)) __tmp = ((ue_count)); *(volatile typeof(*(&con->ras_ue_count)) *)&(*(& con->ras_ue_count)) = __tmp; __tmp; }); | |||
2628 | } | |||
2629 | ||||
2630 | return 0; | |||
2631 | ||||
2632 | interrupt: | |||
2633 | if (ras_obj->ras_cb) | |||
2634 | amdgpu_ras_interrupt_remove_handler(adev, ras_block); | |||
2635 | cleanup: | |||
2636 | amdgpu_ras_feature_enable(adev, ras_block, 0); | |||
2637 | return r; | |||
2638 | } | |||
2639 | ||||
2640 | static int amdgpu_ras_block_late_init_default(struct amdgpu_device *adev, | |||
2641 | struct ras_common_if *ras_block) | |||
2642 | { | |||
2643 | return amdgpu_ras_block_late_init(adev, ras_block); | |||
2644 | } | |||
2645 | ||||
2646 | /* helper function to remove ras fs node and interrupt handler */ | |||
2647 | void amdgpu_ras_block_late_fini(struct amdgpu_device *adev, | |||
2648 | struct ras_common_if *ras_block) | |||
2649 | { | |||
2650 | struct amdgpu_ras_block_object *ras_obj; | |||
2651 | if (!ras_block) | |||
2652 | return; | |||
2653 | ||||
2654 | amdgpu_ras_sysfs_remove(adev, ras_block); | |||
2655 | ||||
2656 | ras_obj = container_of(ras_block, struct amdgpu_ras_block_object, ras_comm)({ const __typeof( ((struct amdgpu_ras_block_object *)0)-> ras_comm ) *__mptr = (ras_block); (struct amdgpu_ras_block_object *)( (char *)__mptr - __builtin_offsetof(struct amdgpu_ras_block_object , ras_comm) );}); | |||
2657 | if (ras_obj->ras_cb) | |||
2658 | amdgpu_ras_interrupt_remove_handler(adev, ras_block); | |||
2659 | } | |||
2660 | ||||
2661 | static void amdgpu_ras_block_late_fini_default(struct amdgpu_device *adev, | |||
2662 | struct ras_common_if *ras_block) | |||
2663 | { | |||
2664 | return amdgpu_ras_block_late_fini(adev, ras_block); | |||
2665 | } | |||
2666 | ||||
2667 | /* do some init work after IP late init as dependence. | |||
2668 | * and it runs in resume/gpu reset/booting up cases. | |||
2669 | */ | |||
2670 | void amdgpu_ras_resume(struct amdgpu_device *adev) | |||
2671 | { | |||
2672 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2673 | struct ras_manager *obj, *tmp; | |||
2674 | ||||
2675 | if (!adev->ras_enabled || !con) { | |||
2676 | /* clean ras context for VEGA20 Gaming after send ras disable cmd */ | |||
2677 | amdgpu_release_ras_context(adev); | |||
2678 | ||||
2679 | return; | |||
2680 | } | |||
2681 | ||||
2682 | if (con->flags & AMDGPU_RAS_FLAG_INIT_BY_VBIOS(0x1 << 0)) { | |||
2683 | /* Set up all other IPs which are not implemented. There is a | |||
2684 | * tricky thing that IP's actual ras error type should be | |||
2685 | * MULTI_UNCORRECTABLE, but as driver does not handle it, so | |||
2686 | * ERROR_NONE make sense anyway. | |||
2687 | */ | |||
2688 | amdgpu_ras_enable_all_features(adev, 1); | |||
2689 | ||||
2690 | /* We enable ras on all hw_supported block, but as boot | |||
2691 | * parameter might disable some of them and one or more IP has | |||
2692 | * not implemented yet. So we disable them on behalf. | |||
2693 | */ | |||
2694 | list_for_each_entry_safe(obj, tmp, &con->head, node)for (obj = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = ((&con->head)->next); (__typeof(*obj) *) ( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) ) ;}), tmp = ({ const __typeof( ((__typeof(*obj) *)0)->node ) *__mptr = (obj->node.next); (__typeof(*obj) *)( (char *)__mptr - __builtin_offsetof(__typeof(*obj), node) );}); &obj-> node != (&con->head); obj = tmp, tmp = ({ const __typeof ( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next ); (__typeof(*tmp) *)( (char *)__mptr - __builtin_offsetof(__typeof (*tmp), node) );})) { | |||
2695 | if (!amdgpu_ras_is_supported(adev, obj->head.block)) { | |||
2696 | amdgpu_ras_feature_enable(adev, &obj->head, 0); | |||
2697 | /* there should be no any reference. */ | |||
2698 | WARN_ON(alive_obj(obj))({ int __ret = !!(((obj)->use)); if (__ret) printf("WARNING %s failed at %s:%d\n" , "((obj)->use)", "/usr/src/sys/dev/pci/drm/amd/amdgpu/amdgpu_ras.c" , 2698); __builtin_expect(!!(__ret), 0); }); | |||
2699 | } | |||
2700 | } | |||
2701 | } | |||
2702 | } | |||
2703 | ||||
2704 | void amdgpu_ras_suspend(struct amdgpu_device *adev) | |||
2705 | { | |||
2706 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2707 | ||||
2708 | if (!adev->ras_enabled || !con) | |||
2709 | return; | |||
2710 | ||||
2711 | amdgpu_ras_disable_all_features(adev, 0); | |||
2712 | /* Make sure all ras objects are disabled. */ | |||
2713 | if (con->features) | |||
2714 | amdgpu_ras_disable_all_features(adev, 1); | |||
2715 | } | |||
2716 | ||||
2717 | int amdgpu_ras_late_init(struct amdgpu_device *adev) | |||
2718 | { | |||
2719 | struct amdgpu_ras_block_list *node, *tmp; | |||
2720 | struct amdgpu_ras_block_object *obj; | |||
2721 | int r; | |||
2722 | ||||
2723 | /* Guest side doesn't need init ras feature */ | |||
2724 | if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) | |||
2725 | return 0; | |||
2726 | ||||
2727 | list_for_each_entry_safe(node, tmp, &adev->ras_list, node)for (node = ({ const __typeof( ((__typeof(*node) *)0)->node ) *__mptr = ((&adev->ras_list)->next); (__typeof(* node) *)( (char *)__mptr - __builtin_offsetof(__typeof(*node) , node) );}), tmp = ({ const __typeof( ((__typeof(*node) *)0) ->node ) *__mptr = (node->node.next); (__typeof(*node) * )( (char *)__mptr - __builtin_offsetof(__typeof(*node), node) );}); &node->node != (&adev->ras_list); node = tmp, tmp = ({ const __typeof( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next); (__typeof(*tmp) *)( (char * )__mptr - __builtin_offsetof(__typeof(*tmp), node) );})) { | |||
2728 | if (!node->ras_obj) { | |||
2729 | dev_warn(adev->dev, "Warning: abnormal ras list node.\n")printf("drm:pid%d:%s *WARNING* " "Warning: abnormal ras list node.\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
2730 | continue; | |||
2731 | } | |||
2732 | ||||
2733 | obj = node->ras_obj; | |||
2734 | if (obj->ras_late_init) { | |||
2735 | r = obj->ras_late_init(adev, &obj->ras_comm); | |||
2736 | if (r) { | |||
2737 | dev_err(adev->dev, "%s failed to execute ras_late_init! ret:%d\n",printf("drm:pid%d:%s *ERROR* " "%s failed to execute ras_late_init! ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , obj-> ras_comm.name, r) | |||
2738 | obj->ras_comm.name, r)printf("drm:pid%d:%s *ERROR* " "%s failed to execute ras_late_init! ret:%d\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__ , obj-> ras_comm.name, r); | |||
2739 | return r; | |||
2740 | } | |||
2741 | } else | |||
2742 | amdgpu_ras_block_late_init_default(adev, &obj->ras_comm); | |||
2743 | } | |||
2744 | ||||
2745 | return 0; | |||
2746 | } | |||
2747 | ||||
2748 | /* do some fini work before IP fini as dependence */ | |||
2749 | int amdgpu_ras_pre_fini(struct amdgpu_device *adev) | |||
2750 | { | |||
2751 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
| ||||
2752 | ||||
2753 | if (!adev->ras_enabled || !con) | |||
2754 | return 0; | |||
2755 | ||||
2756 | ||||
2757 | /* Need disable ras on all IPs here before ip [hw/sw]fini */ | |||
2758 | if (con->features) | |||
2759 | amdgpu_ras_disable_all_features(adev, 0); | |||
2760 | amdgpu_ras_recovery_fini(adev); | |||
2761 | return 0; | |||
2762 | } | |||
2763 | ||||
2764 | int amdgpu_ras_fini(struct amdgpu_device *adev) | |||
2765 | { | |||
2766 | struct amdgpu_ras_block_list *ras_node, *tmp; | |||
2767 | struct amdgpu_ras_block_object *obj = NULL((void *)0); | |||
2768 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2769 | ||||
2770 | if (!adev->ras_enabled || !con) | |||
2771 | return 0; | |||
2772 | ||||
2773 | list_for_each_entry_safe(ras_node, tmp, &adev->ras_list, node)for (ras_node = ({ const __typeof( ((__typeof(*ras_node) *)0) ->node ) *__mptr = ((&adev->ras_list)->next); (__typeof (*ras_node) *)( (char *)__mptr - __builtin_offsetof(__typeof( *ras_node), node) );}), tmp = ({ const __typeof( ((__typeof(* ras_node) *)0)->node ) *__mptr = (ras_node->node.next); (__typeof(*ras_node) *)( (char *)__mptr - __builtin_offsetof (__typeof(*ras_node), node) );}); &ras_node->node != ( &adev->ras_list); ras_node = tmp, tmp = ({ const __typeof ( ((__typeof(*tmp) *)0)->node ) *__mptr = (tmp->node.next ); (__typeof(*tmp) *)( (char *)__mptr - __builtin_offsetof(__typeof (*tmp), node) );})) { | |||
2774 | if (ras_node->ras_obj) { | |||
2775 | obj = ras_node->ras_obj; | |||
2776 | if (amdgpu_ras_is_supported(adev, obj->ras_comm.block) && | |||
2777 | obj->ras_fini) | |||
2778 | obj->ras_fini(adev, &obj->ras_comm); | |||
2779 | else | |||
2780 | amdgpu_ras_block_late_fini_default(adev, &obj->ras_comm); | |||
2781 | } | |||
2782 | ||||
2783 | /* Clear ras blocks from ras_list and free ras block list node */ | |||
2784 | list_del(&ras_node->node); | |||
2785 | kfree(ras_node); | |||
2786 | } | |||
2787 | ||||
2788 | amdgpu_ras_fs_fini(adev); | |||
2789 | amdgpu_ras_interrupt_remove_all(adev); | |||
2790 | ||||
2791 | WARN(con->features, "Feature mask is not cleared")({ int __ret = !!(con->features); if (__ret) printf("Feature mask is not cleared" ); __builtin_expect(!!(__ret), 0); }); | |||
2792 | ||||
2793 | if (con->features) | |||
2794 | amdgpu_ras_disable_all_features(adev, 1); | |||
2795 | ||||
2796 | cancel_delayed_work_sync(&con->ras_counte_delay_work); | |||
2797 | ||||
2798 | amdgpu_ras_set_context(adev, NULL((void *)0)); | |||
2799 | kfree(con); | |||
2800 | ||||
2801 | return 0; | |||
2802 | } | |||
2803 | ||||
2804 | void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev) | |||
2805 | { | |||
2806 | amdgpu_ras_check_supported(adev); | |||
2807 | if (!adev->ras_hw_enabled) | |||
2808 | return; | |||
2809 | ||||
2810 | if (atomic_cmpxchg(&amdgpu_ras_in_intr, 0, 1)__sync_val_compare_and_swap(&amdgpu_ras_in_intr, 0, 1) == 0) { | |||
2811 | dev_info(adev->dev, "uncorrectable hardware error"do { } while(0) | |||
2812 | "(ERREVENT_ATHUB_INTERRUPT) detected!\n")do { } while(0); | |||
2813 | ||||
2814 | amdgpu_ras_reset_gpu(adev); | |||
2815 | } | |||
2816 | } | |||
2817 | ||||
2818 | bool_Bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev) | |||
2819 | { | |||
2820 | if (adev->asic_type == CHIP_VEGA20 && | |||
2821 | adev->pm.fw_version <= 0x283400) { | |||
2822 | return !(amdgpu_asic_reset_method(adev)(adev)->asic_funcs->reset_method((adev)) == AMD_RESET_METHOD_BACO) && | |||
2823 | amdgpu_ras_intr_triggered(); | |||
2824 | } | |||
2825 | ||||
2826 | return false0; | |||
2827 | } | |||
2828 | ||||
2829 | void amdgpu_release_ras_context(struct amdgpu_device *adev) | |||
2830 | { | |||
2831 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |||
2832 | ||||
2833 | if (!con) | |||
2834 | return; | |||
2835 | ||||
2836 | if (!adev->ras_enabled && con->features & BIT(AMDGPU_RAS_BLOCK__GFX)(1UL << (AMDGPU_RAS_BLOCK__GFX))) { | |||
2837 | con->features &= ~BIT(AMDGPU_RAS_BLOCK__GFX)(1UL << (AMDGPU_RAS_BLOCK__GFX)); | |||
2838 | amdgpu_ras_set_context(adev, NULL((void *)0)); | |||
2839 | kfree(con); | |||
2840 | } | |||
2841 | } | |||
2842 | ||||
2843 | #ifdef CONFIG_X86_MCE_AMD | |||
2844 | static struct amdgpu_device *find_adev(uint32_t node_id) | |||
2845 | { | |||
2846 | int i; | |||
2847 | struct amdgpu_device *adev = NULL((void *)0); | |||
2848 | ||||
2849 | for (i = 0; i < mce_adev_list.num_gpu; i++) { | |||
2850 | adev = mce_adev_list.devs[i]; | |||
2851 | ||||
2852 | if (adev && adev->gmc.xgmi.connected_to_cpu && | |||
2853 | adev->gmc.xgmi.physical_node_id == node_id) | |||
2854 | break; | |||
2855 | adev = NULL((void *)0); | |||
2856 | } | |||
2857 | ||||
2858 | return adev; | |||
2859 | } | |||
2860 | ||||
2861 | #define GET_MCA_IPID_GPUID(m) (((m) >> 44) & 0xF) | |||
2862 | #define GET_UMC_INST(m) (((m) >> 21) & 0x7) | |||
2863 | #define GET_CHAN_INDEX(m) ((((m) >> 12) & 0x3) | (((m) >> 18) & 0x4)) | |||
2864 | #define GPU_ID_OFFSET 8 | |||
2865 | ||||
2866 | static int amdgpu_bad_page_notifier(struct notifier_block *nb, | |||
2867 | unsigned long val, void *data) | |||
2868 | { | |||
2869 | struct mce *m = (struct mce *)data; | |||
2870 | struct amdgpu_device *adev = NULL((void *)0); | |||
2871 | uint32_t gpu_id = 0; | |||
2872 | uint32_t umc_inst = 0, ch_inst = 0; | |||
2873 | struct ras_err_data err_data = {0, 0, 0, NULL((void *)0)}; | |||
2874 | ||||
2875 | /* | |||
2876 | * If the error was generated in UMC_V2, which belongs to GPU UMCs, | |||
2877 | * and error occurred in DramECC (Extended error code = 0) then only | |||
2878 | * process the error, else bail out. | |||
2879 | */ | |||
2880 | if (!m || !((smca_get_bank_type(m->extcpu, m->bank) == SMCA_UMC_V2) && | |||
2881 | (XEC(m->status, 0x3f) == 0x0))) | |||
2882 | return NOTIFY_DONE0; | |||
2883 | ||||
2884 | /* | |||
2885 | * If it is correctable error, return. | |||
2886 | */ | |||
2887 | if (mce_is_correctable(m)) | |||
2888 | return NOTIFY_OK1; | |||
2889 | ||||
2890 | /* | |||
2891 | * GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. | |||
2892 | */ | |||
2893 | gpu_id = GET_MCA_IPID_GPUID(m->ipid) - GPU_ID_OFFSET; | |||
2894 | ||||
2895 | adev = find_adev(gpu_id); | |||
2896 | if (!adev) { | |||
2897 | DRM_WARN("%s: Unable to find adev for gpu_id: %d\n", __func__,printk("\0014" "[" "drm" "] " "%s: Unable to find adev for gpu_id: %d\n" , __func__, gpu_id) | |||
2898 | gpu_id)printk("\0014" "[" "drm" "] " "%s: Unable to find adev for gpu_id: %d\n" , __func__, gpu_id); | |||
2899 | return NOTIFY_DONE0; | |||
2900 | } | |||
2901 | ||||
2902 | /* | |||
2903 | * If it is uncorrectable error, then find out UMC instance and | |||
2904 | * channel index. | |||
2905 | */ | |||
2906 | umc_inst = GET_UMC_INST(m->ipid); | |||
2907 | ch_inst = GET_CHAN_INDEX(m->ipid); | |||
2908 | ||||
2909 | dev_info(adev->dev, "Uncorrectable error detected in UMC inst: %d, chan_idx: %d",do { } while(0) | |||
2910 | umc_inst, ch_inst)do { } while(0); | |||
2911 | ||||
2912 | err_data.err_addr = | |||
2913 | kcalloc(adev->umc.max_ras_err_cnt_per_query, | |||
2914 | sizeof(struct eeprom_table_record), GFP_KERNEL(0x0001 | 0x0004)); | |||
2915 | if (!err_data.err_addr) { | |||
2916 | dev_warn(adev->dev,printf("drm:pid%d:%s *WARNING* " "Failed to alloc memory for umc error record in mca notifier!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__) | |||
2917 | "Failed to alloc memory for umc error record in mca notifier!\n")printf("drm:pid%d:%s *WARNING* " "Failed to alloc memory for umc error record in mca notifier!\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); | |||
2918 | return NOTIFY_DONE0; | |||
2919 | } | |||
2920 | ||||
2921 | /* | |||
2922 | * Translate UMC channel address to Physical address | |||
2923 | */ | |||
2924 | if (adev->umc.ras && | |||
2925 | adev->umc.ras->convert_ras_error_address) | |||
2926 | adev->umc.ras->convert_ras_error_address(adev, | |||
2927 | &err_data, m->addr, ch_inst, umc_inst); | |||
2928 | ||||
2929 | if (amdgpu_bad_page_threshold != 0) { | |||
2930 | amdgpu_ras_add_bad_pages(adev, err_data.err_addr, | |||
2931 | err_data.err_addr_cnt); | |||
2932 | amdgpu_ras_save_bad_pages(adev); | |||
2933 | } | |||
2934 | ||||
2935 | kfree(err_data.err_addr); | |||
2936 | return NOTIFY_OK1; | |||
2937 | } | |||
2938 | ||||
2939 | static struct notifier_block amdgpu_bad_page_nb = { | |||
2940 | .notifier_call = amdgpu_bad_page_notifier, | |||
2941 | .priority = MCE_PRIO_UC, | |||
2942 | }; | |||
2943 | ||||
2944 | static void amdgpu_register_bad_pages_mca_notifier(struct amdgpu_device *adev) | |||
2945 | { | |||
2946 | /* | |||
2947 | * Add the adev to the mce_adev_list. | |||
2948 | * During mode2 reset, amdgpu device is temporarily | |||
2949 | * removed from the mgpu_info list which can cause | |||
2950 | * page retirement to fail. | |||
2951 | * Use this list instead of mgpu_info to find the amdgpu | |||
2952 | * device on which the UMC error was reported. | |||
2953 | */ | |||
2954 | mce_adev_list.devs[mce_adev_list.num_gpu++] = adev; | |||
2955 | ||||
2956 | /* | |||
2957 | * Register the x86 notifier only once | |||
2958 | * with MCE subsystem. | |||
2959 | */ | |||
2960 | if (notifier_registered == false0) { | |||
2961 | mce_register_decode_chain(&amdgpu_bad_page_nb); | |||
2962 | notifier_registered = true1; | |||
2963 | } | |||
2964 | } | |||
2965 | #endif | |||
2966 | ||||
2967 | struct amdgpu_ras *amdgpu_ras_get_context(struct amdgpu_device *adev) | |||
2968 | { | |||
2969 | if (!adev
| |||
2970 | return NULL((void *)0); | |||
2971 | ||||
2972 | return adev->psp.ras_context.ras; | |||
2973 | } | |||
2974 | ||||
2975 | int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_con) | |||
2976 | { | |||
2977 | if (!adev) | |||
2978 | return -EINVAL22; | |||
2979 | ||||
2980 | adev->psp.ras_context.ras = ras_con; | |||
2981 | return 0; | |||
2982 | } | |||
2983 | ||||
2984 | /* check if ras is supported on block, say, sdma, gfx */ | |||
2985 | int amdgpu_ras_is_supported(struct amdgpu_device *adev, | |||
2986 | unsigned int block) | |||
2987 | { | |||
2988 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |||
2989 | ||||
2990 | if (block >= AMDGPU_RAS_BLOCK_COUNTAMDGPU_RAS_BLOCK__LAST) | |||
2991 | return 0; | |||
2992 | return ras && (adev->ras_enabled & (1 << block)); | |||
2993 | } | |||
2994 | ||||
2995 | int amdgpu_ras_reset_gpu(struct amdgpu_device *adev) | |||
2996 | { | |||
2997 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); | |||
2998 | ||||
2999 | if (atomic_cmpxchg(&ras->in_recovery, 0, 1)__sync_val_compare_and_swap(&ras->in_recovery, 0, 1) == 0) | |||
3000 | amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work); | |||
3001 | return 0; | |||
3002 | } | |||
3003 | ||||
3004 | ||||
3005 | /* Register each ip ras block into amdgpu ras */ | |||
3006 | int amdgpu_ras_register_ras_block(struct amdgpu_device *adev, | |||
3007 | struct amdgpu_ras_block_object *ras_block_obj) | |||
3008 | { | |||
3009 | struct amdgpu_ras_block_list *ras_node; | |||
3010 | if (!adev || !ras_block_obj) | |||
3011 | return -EINVAL22; | |||
3012 | ||||
3013 | if (!amdgpu_ras_asic_supported(adev)) | |||
3014 | return 0; | |||
3015 | ||||
3016 | ras_node = kzalloc(sizeof(*ras_node), GFP_KERNEL(0x0001 | 0x0004)); | |||
3017 | if (!ras_node) | |||
3018 | return -ENOMEM12; | |||
3019 | ||||
3020 | INIT_LIST_HEAD(&ras_node->node); | |||
3021 | ras_node->ras_obj = ras_block_obj; | |||
3022 | list_add_tail(&ras_node->node, &adev->ras_list); | |||
3023 | ||||
3024 | return 0; | |||
3025 | } |