Loading...
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#ifndef _AMDGPU_RAS_H
25#define _AMDGPU_RAS_H
26
27#include <linux/debugfs.h>
28#include <linux/list.h>
29#include "ta_ras_if.h"
30#include "amdgpu_ras_eeprom.h"
31
32struct amdgpu_iv_entry;
33
34#define AMDGPU_RAS_FLAG_INIT_BY_VBIOS (0x1 << 0)
35
36enum amdgpu_ras_block {
37 AMDGPU_RAS_BLOCK__UMC = 0,
38 AMDGPU_RAS_BLOCK__SDMA,
39 AMDGPU_RAS_BLOCK__GFX,
40 AMDGPU_RAS_BLOCK__MMHUB,
41 AMDGPU_RAS_BLOCK__ATHUB,
42 AMDGPU_RAS_BLOCK__PCIE_BIF,
43 AMDGPU_RAS_BLOCK__HDP,
44 AMDGPU_RAS_BLOCK__XGMI_WAFL,
45 AMDGPU_RAS_BLOCK__DF,
46 AMDGPU_RAS_BLOCK__SMN,
47 AMDGPU_RAS_BLOCK__SEM,
48 AMDGPU_RAS_BLOCK__MP0,
49 AMDGPU_RAS_BLOCK__MP1,
50 AMDGPU_RAS_BLOCK__FUSE,
51 AMDGPU_RAS_BLOCK__MCA,
52 AMDGPU_RAS_BLOCK__VCN,
53 AMDGPU_RAS_BLOCK__JPEG,
54
55 AMDGPU_RAS_BLOCK__LAST
56};
57
58enum amdgpu_ras_mca_block {
59 AMDGPU_RAS_MCA_BLOCK__MP0 = 0,
60 AMDGPU_RAS_MCA_BLOCK__MP1,
61 AMDGPU_RAS_MCA_BLOCK__MPIO,
62 AMDGPU_RAS_MCA_BLOCK__IOHC,
63
64 AMDGPU_RAS_MCA_BLOCK__LAST
65};
66
67#define AMDGPU_RAS_BLOCK_COUNT AMDGPU_RAS_BLOCK__LAST
68#define AMDGPU_RAS_MCA_BLOCK_COUNT AMDGPU_RAS_MCA_BLOCK__LAST
69#define AMDGPU_RAS_BLOCK_MASK ((1ULL << AMDGPU_RAS_BLOCK_COUNT) - 1)
70
71enum amdgpu_ras_gfx_subblock {
72 /* CPC */
73 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
74 AMDGPU_RAS_BLOCK__GFX_CPC_SCRATCH =
75 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START,
76 AMDGPU_RAS_BLOCK__GFX_CPC_UCODE,
77 AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME1,
78 AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
79 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME1,
80 AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME2,
81 AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
82 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
83 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_END =
84 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
85 /* CPF */
86 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
87 AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME2 =
88 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
89 AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME1,
90 AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
91 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
92 /* CPG */
93 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
94 AMDGPU_RAS_BLOCK__GFX_CPG_DMA_ROQ =
95 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
96 AMDGPU_RAS_BLOCK__GFX_CPG_DMA_TAG,
97 AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
98 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
99 /* GDS */
100 AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
101 AMDGPU_RAS_BLOCK__GFX_GDS_MEM = AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
102 AMDGPU_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
103 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
104 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
105 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
106 AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_END =
107 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
108 /* SPI */
109 AMDGPU_RAS_BLOCK__GFX_SPI_SR_MEM,
110 /* SQ */
111 AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
112 AMDGPU_RAS_BLOCK__GFX_SQ_SGPR = AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
113 AMDGPU_RAS_BLOCK__GFX_SQ_LDS_D,
114 AMDGPU_RAS_BLOCK__GFX_SQ_LDS_I,
115 AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
116 AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_END = AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
117 /* SQC (3 ranges) */
118 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
119 /* SQC range 0 */
120 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START =
121 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
122 AMDGPU_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
123 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START,
124 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
125 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
126 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
127 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
128 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
129 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
130 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_END =
131 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
132 /* SQC range 1 */
133 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
134 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
135 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
136 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
137 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
138 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
139 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
140 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
141 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
142 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
143 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
144 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_END =
145 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
146 /* SQC range 2 */
147 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
148 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
149 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
150 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
151 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
152 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
153 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
154 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
155 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
156 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
157 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
158 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END =
159 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
160 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_END =
161 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END,
162 /* TA */
163 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
164 AMDGPU_RAS_BLOCK__GFX_TA_FS_DFIFO =
165 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
166 AMDGPU_RAS_BLOCK__GFX_TA_FS_AFIFO,
167 AMDGPU_RAS_BLOCK__GFX_TA_FL_LFIFO,
168 AMDGPU_RAS_BLOCK__GFX_TA_FX_LFIFO,
169 AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
170 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
171 /* TCA */
172 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
173 AMDGPU_RAS_BLOCK__GFX_TCA_HOLE_FIFO =
174 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
175 AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
176 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_END =
177 AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
178 /* TCC (5 sub-ranges) */
179 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
180 /* TCC range 0 */
181 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START =
182 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
183 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA =
184 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START,
185 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
186 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
187 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
188 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
189 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
190 AMDGPU_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
191 AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
192 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_END =
193 AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
194 /* TCC range 1 */
195 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
196 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_DEC =
197 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
198 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
199 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_END =
200 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
201 /* TCC range 2 */
202 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
203 AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_DATA =
204 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
205 AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
206 AMDGPU_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
207 AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
208 AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
209 AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO,
210 AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
211 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
212 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_END =
213 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
214 /* TCC range 3 */
215 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
216 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO =
217 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
218 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
219 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_END =
220 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
221 /* TCC range 4 */
222 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
223 AMDGPU_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
224 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
225 AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
226 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END =
227 AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
228 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_END =
229 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END,
230 /* TCI */
231 AMDGPU_RAS_BLOCK__GFX_TCI_WRITE_RAM,
232 /* TCP */
233 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
234 AMDGPU_RAS_BLOCK__GFX_TCP_CACHE_RAM =
235 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
236 AMDGPU_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
237 AMDGPU_RAS_BLOCK__GFX_TCP_CMD_FIFO,
238 AMDGPU_RAS_BLOCK__GFX_TCP_VM_FIFO,
239 AMDGPU_RAS_BLOCK__GFX_TCP_DB_RAM,
240 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
241 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
242 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_END =
243 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
244 /* TD */
245 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
246 AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_LO =
247 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
248 AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
249 AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
250 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
251 /* EA (3 sub-ranges) */
252 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
253 /* EA range 0 */
254 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START =
255 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
256 AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM =
257 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START,
258 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
259 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
260 AMDGPU_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
261 AMDGPU_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
262 AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
263 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
264 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
265 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_END =
266 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
267 /* EA range 1 */
268 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
269 AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM =
270 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
271 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
272 AMDGPU_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
273 AMDGPU_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
274 AMDGPU_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
275 AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
276 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
277 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_END =
278 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
279 /* EA range 2 */
280 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
281 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D0MEM =
282 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
283 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D1MEM,
284 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D2MEM,
285 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
286 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END =
287 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
288 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_END =
289 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END,
290 /* UTC VM L2 bank */
291 AMDGPU_RAS_BLOCK__UTC_VML2_BANK_CACHE,
292 /* UTC VM walker */
293 AMDGPU_RAS_BLOCK__UTC_VML2_WALKER,
294 /* UTC ATC L2 2MB cache */
295 AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
296 /* UTC ATC L2 4KB cache */
297 AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
298 AMDGPU_RAS_BLOCK__GFX_MAX
299};
300
301enum amdgpu_ras_error_type {
302 AMDGPU_RAS_ERROR__NONE = 0,
303 AMDGPU_RAS_ERROR__PARITY = 1,
304 AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE = 2,
305 AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE = 4,
306 AMDGPU_RAS_ERROR__POISON = 8,
307};
308
309enum amdgpu_ras_ret {
310 AMDGPU_RAS_SUCCESS = 0,
311 AMDGPU_RAS_FAIL,
312 AMDGPU_RAS_UE,
313 AMDGPU_RAS_CE,
314 AMDGPU_RAS_PT,
315};
316
317struct ras_common_if {
318 enum amdgpu_ras_block block;
319 enum amdgpu_ras_error_type type;
320 uint32_t sub_block_index;
321 char name[32];
322};
323
324#define MAX_UMC_CHANNEL_NUM 32
325
326struct ecc_info_per_ch {
327 uint16_t ce_count_lo_chip;
328 uint16_t ce_count_hi_chip;
329 uint64_t mca_umc_status;
330 uint64_t mca_umc_addr;
331 uint64_t mca_ceumc_addr;
332};
333
334struct umc_ecc_info {
335 struct ecc_info_per_ch ecc[MAX_UMC_CHANNEL_NUM];
336
337 /* Determine smu ecctable whether support
338 * record correctable error address
339 */
340 int record_ce_addr_supported;
341};
342
343struct amdgpu_ras {
344 /* ras infrastructure */
345 /* for ras itself. */
346 uint32_t features;
347 struct list_head head;
348 /* sysfs */
349 struct device_attribute features_attr;
350 struct bin_attribute badpages_attr;
351 struct dentry *de_ras_eeprom_table;
352 /* block array */
353 struct ras_manager *objs;
354
355 /* gpu recovery */
356 struct work_struct recovery_work;
357 atomic_t in_recovery;
358 struct amdgpu_device *adev;
359 /* error handler data */
360 struct ras_err_handler_data *eh_data;
361 struct mutex recovery_lock;
362
363 uint32_t flags;
364 bool reboot;
365 struct amdgpu_ras_eeprom_control eeprom_control;
366
367 bool error_query_ready;
368
369 /* bad page count threshold */
370 uint32_t bad_page_cnt_threshold;
371
372 /* disable ras error count harvest in recovery */
373 bool disable_ras_err_cnt_harvest;
374
375 /* is poison mode supported */
376 bool poison_supported;
377
378 /* RAS count errors delayed work */
379 struct delayed_work ras_counte_delay_work;
380 atomic_t ras_ue_count;
381 atomic_t ras_ce_count;
382
383 /* record umc error info queried from smu */
384 struct umc_ecc_info umc_ecc;
385
386 /* Indicates smu whether need update bad channel info */
387 bool update_channel_flag;
388};
389
390struct ras_fs_data {
391 char sysfs_name[32];
392 char debugfs_name[32];
393};
394
395struct ras_err_data {
396 unsigned long ue_count;
397 unsigned long ce_count;
398 unsigned long err_addr_cnt;
399 struct eeprom_table_record *err_addr;
400};
401
402struct ras_err_handler_data {
403 /* point to bad page records array */
404 struct eeprom_table_record *bps;
405 /* the count of entries */
406 int count;
407 /* the space can place new entries */
408 int space_left;
409};
410
411typedef int (*ras_ih_cb)(struct amdgpu_device *adev,
412 void *err_data,
413 struct amdgpu_iv_entry *entry);
414
415struct ras_ih_data {
416 /* interrupt bottom half */
417 struct work_struct ih_work;
418 int inuse;
419 /* IP callback */
420 ras_ih_cb cb;
421 /* full of entries */
422 unsigned char *ring;
423 unsigned int ring_size;
424 unsigned int element_size;
425 unsigned int aligned_element_size;
426 unsigned int rptr;
427 unsigned int wptr;
428};
429
430struct ras_manager {
431 struct ras_common_if head;
432 /* reference count */
433 int use;
434 /* ras block link */
435 struct list_head node;
436 /* the device */
437 struct amdgpu_device *adev;
438 /* sysfs */
439 struct device_attribute sysfs_attr;
440 int attr_inuse;
441
442 /* fs node name */
443 struct ras_fs_data fs_data;
444
445 /* IH data */
446 struct ras_ih_data ih_data;
447
448 struct ras_err_data err_data;
449};
450
451struct ras_badpage {
452 unsigned int bp;
453 unsigned int size;
454 unsigned int flags;
455};
456
457/* interfaces for IP */
458struct ras_fs_if {
459 struct ras_common_if head;
460 const char* sysfs_name;
461 char debugfs_name[32];
462};
463
464struct ras_query_if {
465 struct ras_common_if head;
466 unsigned long ue_count;
467 unsigned long ce_count;
468};
469
470struct ras_inject_if {
471 struct ras_common_if head;
472 uint64_t address;
473 uint64_t value;
474};
475
476struct ras_cure_if {
477 struct ras_common_if head;
478 uint64_t address;
479};
480
481struct ras_ih_if {
482 struct ras_common_if head;
483 ras_ih_cb cb;
484};
485
486struct ras_dispatch_if {
487 struct ras_common_if head;
488 struct amdgpu_iv_entry *entry;
489};
490
491struct ras_debug_if {
492 union {
493 struct ras_common_if head;
494 struct ras_inject_if inject;
495 };
496 int op;
497};
498
499struct amdgpu_ras_block_object {
500 struct ras_common_if ras_comm;
501
502 int (*ras_block_match)(struct amdgpu_ras_block_object *block_obj,
503 enum amdgpu_ras_block block, uint32_t sub_block_index);
504 int (*ras_late_init)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
505 void (*ras_fini)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
506 ras_ih_cb ras_cb;
507 const struct amdgpu_ras_block_hw_ops *hw_ops;
508};
509
510struct amdgpu_ras_block_hw_ops {
511 int (*ras_error_inject)(struct amdgpu_device *adev, void *inject_if);
512 void (*query_ras_error_count)(struct amdgpu_device *adev, void *ras_error_status);
513 void (*query_ras_error_status)(struct amdgpu_device *adev);
514 void (*query_ras_error_address)(struct amdgpu_device *adev, void *ras_error_status);
515 void (*reset_ras_error_count)(struct amdgpu_device *adev);
516 void (*reset_ras_error_status)(struct amdgpu_device *adev);
517 bool (*query_poison_status)(struct amdgpu_device *adev);
518 bool (*handle_poison_consumption)(struct amdgpu_device *adev);
519};
520
521/* work flow
522 * vbios
523 * 1: ras feature enable (enabled by default)
524 * psp
525 * 2: ras framework init (in ip_init)
526 * IP
527 * 3: IH add
528 * 4: debugfs/sysfs create
529 * 5: query/inject
530 * 6: debugfs/sysfs remove
531 * 7: IH remove
532 * 8: feature disable
533 */
534
535
536int amdgpu_ras_recovery_init(struct amdgpu_device *adev);
537
538void amdgpu_ras_resume(struct amdgpu_device *adev);
539void amdgpu_ras_suspend(struct amdgpu_device *adev);
540
541int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
542 unsigned long *ce_count,
543 unsigned long *ue_count);
544
545/* error handling functions */
546int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
547 struct eeprom_table_record *bps, int pages);
548
549int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev);
550
551static inline enum ta_ras_block
552amdgpu_ras_block_to_ta(enum amdgpu_ras_block block) {
553 switch (block) {
554 case AMDGPU_RAS_BLOCK__UMC:
555 return TA_RAS_BLOCK__UMC;
556 case AMDGPU_RAS_BLOCK__SDMA:
557 return TA_RAS_BLOCK__SDMA;
558 case AMDGPU_RAS_BLOCK__GFX:
559 return TA_RAS_BLOCK__GFX;
560 case AMDGPU_RAS_BLOCK__MMHUB:
561 return TA_RAS_BLOCK__MMHUB;
562 case AMDGPU_RAS_BLOCK__ATHUB:
563 return TA_RAS_BLOCK__ATHUB;
564 case AMDGPU_RAS_BLOCK__PCIE_BIF:
565 return TA_RAS_BLOCK__PCIE_BIF;
566 case AMDGPU_RAS_BLOCK__HDP:
567 return TA_RAS_BLOCK__HDP;
568 case AMDGPU_RAS_BLOCK__XGMI_WAFL:
569 return TA_RAS_BLOCK__XGMI_WAFL;
570 case AMDGPU_RAS_BLOCK__DF:
571 return TA_RAS_BLOCK__DF;
572 case AMDGPU_RAS_BLOCK__SMN:
573 return TA_RAS_BLOCK__SMN;
574 case AMDGPU_RAS_BLOCK__SEM:
575 return TA_RAS_BLOCK__SEM;
576 case AMDGPU_RAS_BLOCK__MP0:
577 return TA_RAS_BLOCK__MP0;
578 case AMDGPU_RAS_BLOCK__MP1:
579 return TA_RAS_BLOCK__MP1;
580 case AMDGPU_RAS_BLOCK__FUSE:
581 return TA_RAS_BLOCK__FUSE;
582 case AMDGPU_RAS_BLOCK__MCA:
583 return TA_RAS_BLOCK__MCA;
584 default:
585 WARN_ONCE(1, "RAS ERROR: unexpected block id %d\n", block);
586 return TA_RAS_BLOCK__UMC;
587 }
588}
589
590static inline enum ta_ras_error_type
591amdgpu_ras_error_to_ta(enum amdgpu_ras_error_type error) {
592 switch (error) {
593 case AMDGPU_RAS_ERROR__NONE:
594 return TA_RAS_ERROR__NONE;
595 case AMDGPU_RAS_ERROR__PARITY:
596 return TA_RAS_ERROR__PARITY;
597 case AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE:
598 return TA_RAS_ERROR__SINGLE_CORRECTABLE;
599 case AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE:
600 return TA_RAS_ERROR__MULTI_UNCORRECTABLE;
601 case AMDGPU_RAS_ERROR__POISON:
602 return TA_RAS_ERROR__POISON;
603 default:
604 WARN_ONCE(1, "RAS ERROR: unexpected error type %d\n", error);
605 return TA_RAS_ERROR__NONE;
606 }
607}
608
609/* called in ip_init and ip_fini */
610int amdgpu_ras_init(struct amdgpu_device *adev);
611int amdgpu_ras_late_init(struct amdgpu_device *adev);
612int amdgpu_ras_fini(struct amdgpu_device *adev);
613int amdgpu_ras_pre_fini(struct amdgpu_device *adev);
614
615int amdgpu_ras_block_late_init(struct amdgpu_device *adev,
616 struct ras_common_if *ras_block);
617
618void amdgpu_ras_block_late_fini(struct amdgpu_device *adev,
619 struct ras_common_if *ras_block);
620
621int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
622 struct ras_common_if *head, bool enable);
623
624int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev,
625 struct ras_common_if *head, bool enable);
626
627int amdgpu_ras_sysfs_create(struct amdgpu_device *adev,
628 struct ras_common_if *head);
629
630int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev,
631 struct ras_common_if *head);
632
633void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev);
634
635int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
636 struct ras_query_if *info);
637
638int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
639 enum amdgpu_ras_block block);
640
641int amdgpu_ras_error_inject(struct amdgpu_device *adev,
642 struct ras_inject_if *info);
643
644int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev,
645 struct ras_common_if *head);
646
647int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev,
648 struct ras_common_if *head);
649
650int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev,
651 struct ras_dispatch_if *info);
652
653struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
654 struct ras_common_if *head);
655
656extern atomic_t amdgpu_ras_in_intr;
657
658static inline bool amdgpu_ras_intr_triggered(void)
659{
660 return !!atomic_read(&amdgpu_ras_in_intr);
661}
662
663static inline void amdgpu_ras_intr_cleared(void)
664{
665 atomic_set(&amdgpu_ras_in_intr, 0);
666}
667
668void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev);
669
670void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready);
671
672bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev);
673
674void amdgpu_release_ras_context(struct amdgpu_device *adev);
675
676int amdgpu_persistent_edc_harvesting_supported(struct amdgpu_device *adev);
677
678const char *get_ras_block_str(struct ras_common_if *ras_block);
679
680bool amdgpu_ras_is_poison_mode_supported(struct amdgpu_device *adev);
681
682int amdgpu_ras_is_supported(struct amdgpu_device *adev, unsigned int block);
683
684int amdgpu_ras_reset_gpu(struct amdgpu_device *adev);
685
686struct amdgpu_ras* amdgpu_ras_get_context(struct amdgpu_device *adev);
687
688int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_con);
689
690int amdgpu_ras_register_ras_block(struct amdgpu_device *adev,
691 struct amdgpu_ras_block_object *ras_block_obj);
692void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev);
693#endif
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#ifndef _AMDGPU_RAS_H
25#define _AMDGPU_RAS_H
26
27#include <linux/debugfs.h>
28#include <linux/list.h>
29#include <linux/kfifo.h>
30#include <linux/radix-tree.h>
31#include "ta_ras_if.h"
32#include "amdgpu_ras_eeprom.h"
33#include "amdgpu_smuio.h"
34#include "amdgpu_aca.h"
35
36struct amdgpu_iv_entry;
37
38#define AMDGPU_RAS_GPU_ERR_MEM_TRAINING(x) AMDGPU_GET_REG_FIELD(x, 0, 0)
39#define AMDGPU_RAS_GPU_ERR_FW_LOAD(x) AMDGPU_GET_REG_FIELD(x, 1, 1)
40#define AMDGPU_RAS_GPU_ERR_WAFL_LINK_TRAINING(x) AMDGPU_GET_REG_FIELD(x, 2, 2)
41#define AMDGPU_RAS_GPU_ERR_XGMI_LINK_TRAINING(x) AMDGPU_GET_REG_FIELD(x, 3, 3)
42#define AMDGPU_RAS_GPU_ERR_USR_CP_LINK_TRAINING(x) AMDGPU_GET_REG_FIELD(x, 4, 4)
43#define AMDGPU_RAS_GPU_ERR_USR_DP_LINK_TRAINING(x) AMDGPU_GET_REG_FIELD(x, 5, 5)
44#define AMDGPU_RAS_GPU_ERR_HBM_MEM_TEST(x) AMDGPU_GET_REG_FIELD(x, 6, 6)
45#define AMDGPU_RAS_GPU_ERR_HBM_BIST_TEST(x) AMDGPU_GET_REG_FIELD(x, 7, 7)
46#define AMDGPU_RAS_GPU_ERR_SOCKET_ID(x) AMDGPU_GET_REG_FIELD(x, 10, 8)
47#define AMDGPU_RAS_GPU_ERR_AID_ID(x) AMDGPU_GET_REG_FIELD(x, 12, 11)
48#define AMDGPU_RAS_GPU_ERR_HBM_ID(x) AMDGPU_GET_REG_FIELD(x, 14, 13)
49#define AMDGPU_RAS_GPU_ERR_DATA_ABORT(x) AMDGPU_GET_REG_FIELD(x, 29, 29)
50#define AMDGPU_RAS_GPU_ERR_UNKNOWN(x) AMDGPU_GET_REG_FIELD(x, 30, 30)
51
52#define AMDGPU_RAS_BOOT_STATUS_POLLING_LIMIT 100
53#define AMDGPU_RAS_BOOT_STEADY_STATUS 0xBA
54#define AMDGPU_RAS_BOOT_STATUS_MASK 0xFF
55
56#define AMDGPU_RAS_FLAG_INIT_BY_VBIOS (0x1 << 0)
57/* position of instance value in sub_block_index of
58 * ta_ras_trigger_error_input, the sub block uses lower 12 bits
59 */
60#define AMDGPU_RAS_INST_MASK 0xfffff000
61#define AMDGPU_RAS_INST_SHIFT 0xc
62
63#define AMDGPU_RAS_FEATURES_SOCKETID_SHIFT 29
64#define AMDGPU_RAS_FEATURES_SOCKETID_MASK 0xe0000000
65
66/* Reserve 8 physical dram row for possible retirement.
67 * In worst cases, it will lose 8 * 2MB memory in vram domain */
68#define AMDGPU_RAS_RESERVED_VRAM_SIZE (16ULL << 20)
69/* The high three bits indicates socketid */
70#define AMDGPU_RAS_GET_FEATURES(val) ((val) & ~AMDGPU_RAS_FEATURES_SOCKETID_MASK)
71
72#define RAS_EVENT_INVALID_ID (BIT_ULL(63))
73#define RAS_EVENT_ID_IS_VALID(x) (!((x) & BIT_ULL(63)))
74
75#define RAS_EVENT_LOG(adev, id, fmt, ...) \
76 amdgpu_ras_event_log_print((adev), (id), (fmt), ##__VA_ARGS__)
77
78#define amdgpu_ras_mark_ras_event(adev, type) \
79 (amdgpu_ras_mark_ras_event_caller((adev), (type), __builtin_return_address(0)))
80
81enum amdgpu_ras_block {
82 AMDGPU_RAS_BLOCK__UMC = 0,
83 AMDGPU_RAS_BLOCK__SDMA,
84 AMDGPU_RAS_BLOCK__GFX,
85 AMDGPU_RAS_BLOCK__MMHUB,
86 AMDGPU_RAS_BLOCK__ATHUB,
87 AMDGPU_RAS_BLOCK__PCIE_BIF,
88 AMDGPU_RAS_BLOCK__HDP,
89 AMDGPU_RAS_BLOCK__XGMI_WAFL,
90 AMDGPU_RAS_BLOCK__DF,
91 AMDGPU_RAS_BLOCK__SMN,
92 AMDGPU_RAS_BLOCK__SEM,
93 AMDGPU_RAS_BLOCK__MP0,
94 AMDGPU_RAS_BLOCK__MP1,
95 AMDGPU_RAS_BLOCK__FUSE,
96 AMDGPU_RAS_BLOCK__MCA,
97 AMDGPU_RAS_BLOCK__VCN,
98 AMDGPU_RAS_BLOCK__JPEG,
99 AMDGPU_RAS_BLOCK__IH,
100 AMDGPU_RAS_BLOCK__MPIO,
101
102 AMDGPU_RAS_BLOCK__LAST
103};
104
105enum amdgpu_ras_mca_block {
106 AMDGPU_RAS_MCA_BLOCK__MP0 = 0,
107 AMDGPU_RAS_MCA_BLOCK__MP1,
108 AMDGPU_RAS_MCA_BLOCK__MPIO,
109 AMDGPU_RAS_MCA_BLOCK__IOHC,
110
111 AMDGPU_RAS_MCA_BLOCK__LAST
112};
113
114#define AMDGPU_RAS_BLOCK_COUNT AMDGPU_RAS_BLOCK__LAST
115#define AMDGPU_RAS_MCA_BLOCK_COUNT AMDGPU_RAS_MCA_BLOCK__LAST
116#define AMDGPU_RAS_BLOCK_MASK ((1ULL << AMDGPU_RAS_BLOCK_COUNT) - 1)
117
118enum amdgpu_ras_gfx_subblock {
119 /* CPC */
120 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
121 AMDGPU_RAS_BLOCK__GFX_CPC_SCRATCH =
122 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START,
123 AMDGPU_RAS_BLOCK__GFX_CPC_UCODE,
124 AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME1,
125 AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
126 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME1,
127 AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME2,
128 AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
129 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
130 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_END =
131 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
132 /* CPF */
133 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
134 AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME2 =
135 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
136 AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME1,
137 AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
138 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
139 /* CPG */
140 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
141 AMDGPU_RAS_BLOCK__GFX_CPG_DMA_ROQ =
142 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
143 AMDGPU_RAS_BLOCK__GFX_CPG_DMA_TAG,
144 AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
145 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
146 /* GDS */
147 AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
148 AMDGPU_RAS_BLOCK__GFX_GDS_MEM = AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
149 AMDGPU_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
150 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
151 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
152 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
153 AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_END =
154 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
155 /* SPI */
156 AMDGPU_RAS_BLOCK__GFX_SPI_SR_MEM,
157 /* SQ */
158 AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
159 AMDGPU_RAS_BLOCK__GFX_SQ_SGPR = AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
160 AMDGPU_RAS_BLOCK__GFX_SQ_LDS_D,
161 AMDGPU_RAS_BLOCK__GFX_SQ_LDS_I,
162 AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
163 AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_END = AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
164 /* SQC (3 ranges) */
165 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
166 /* SQC range 0 */
167 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START =
168 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
169 AMDGPU_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
170 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START,
171 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
172 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
173 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
174 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
175 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
176 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
177 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_END =
178 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
179 /* SQC range 1 */
180 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
181 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
182 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
183 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
184 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
185 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
186 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
187 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
188 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
189 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
190 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
191 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_END =
192 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
193 /* SQC range 2 */
194 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
195 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
196 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
197 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
198 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
199 AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
200 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
201 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
202 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
203 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
204 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
205 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END =
206 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
207 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_END =
208 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END,
209 /* TA */
210 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
211 AMDGPU_RAS_BLOCK__GFX_TA_FS_DFIFO =
212 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
213 AMDGPU_RAS_BLOCK__GFX_TA_FS_AFIFO,
214 AMDGPU_RAS_BLOCK__GFX_TA_FL_LFIFO,
215 AMDGPU_RAS_BLOCK__GFX_TA_FX_LFIFO,
216 AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
217 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
218 /* TCA */
219 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
220 AMDGPU_RAS_BLOCK__GFX_TCA_HOLE_FIFO =
221 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
222 AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
223 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_END =
224 AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
225 /* TCC (5 sub-ranges) */
226 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
227 /* TCC range 0 */
228 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START =
229 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
230 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA =
231 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START,
232 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
233 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
234 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
235 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
236 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
237 AMDGPU_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
238 AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
239 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_END =
240 AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
241 /* TCC range 1 */
242 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
243 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_DEC =
244 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
245 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
246 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_END =
247 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
248 /* TCC range 2 */
249 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
250 AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_DATA =
251 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
252 AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
253 AMDGPU_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
254 AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
255 AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
256 AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO,
257 AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
258 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
259 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_END =
260 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
261 /* TCC range 3 */
262 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
263 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO =
264 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
265 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
266 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_END =
267 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
268 /* TCC range 4 */
269 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
270 AMDGPU_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
271 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
272 AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
273 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END =
274 AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
275 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_END =
276 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END,
277 /* TCI */
278 AMDGPU_RAS_BLOCK__GFX_TCI_WRITE_RAM,
279 /* TCP */
280 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
281 AMDGPU_RAS_BLOCK__GFX_TCP_CACHE_RAM =
282 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
283 AMDGPU_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
284 AMDGPU_RAS_BLOCK__GFX_TCP_CMD_FIFO,
285 AMDGPU_RAS_BLOCK__GFX_TCP_VM_FIFO,
286 AMDGPU_RAS_BLOCK__GFX_TCP_DB_RAM,
287 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
288 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
289 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_END =
290 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
291 /* TD */
292 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
293 AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_LO =
294 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
295 AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
296 AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
297 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
298 /* EA (3 sub-ranges) */
299 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
300 /* EA range 0 */
301 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START =
302 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
303 AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM =
304 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START,
305 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
306 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
307 AMDGPU_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
308 AMDGPU_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
309 AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
310 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
311 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
312 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_END =
313 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
314 /* EA range 1 */
315 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
316 AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM =
317 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
318 AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
319 AMDGPU_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
320 AMDGPU_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
321 AMDGPU_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
322 AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
323 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
324 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_END =
325 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
326 /* EA range 2 */
327 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
328 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D0MEM =
329 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
330 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D1MEM,
331 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D2MEM,
332 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
333 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END =
334 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
335 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_END =
336 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END,
337 /* UTC VM L2 bank */
338 AMDGPU_RAS_BLOCK__UTC_VML2_BANK_CACHE,
339 /* UTC VM walker */
340 AMDGPU_RAS_BLOCK__UTC_VML2_WALKER,
341 /* UTC ATC L2 2MB cache */
342 AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
343 /* UTC ATC L2 4KB cache */
344 AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
345 AMDGPU_RAS_BLOCK__GFX_MAX
346};
347
348enum amdgpu_ras_error_type {
349 AMDGPU_RAS_ERROR__NONE = 0,
350 AMDGPU_RAS_ERROR__PARITY = 1,
351 AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE = 2,
352 AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE = 4,
353 AMDGPU_RAS_ERROR__POISON = 8,
354};
355
356enum amdgpu_ras_ret {
357 AMDGPU_RAS_SUCCESS = 0,
358 AMDGPU_RAS_FAIL,
359 AMDGPU_RAS_UE,
360 AMDGPU_RAS_CE,
361 AMDGPU_RAS_PT,
362};
363
364enum amdgpu_ras_error_query_mode {
365 AMDGPU_RAS_INVALID_ERROR_QUERY = 0,
366 AMDGPU_RAS_DIRECT_ERROR_QUERY = 1,
367 AMDGPU_RAS_FIRMWARE_ERROR_QUERY = 2,
368 AMDGPU_RAS_VIRT_ERROR_COUNT_QUERY = 3,
369};
370
371/* ras error status reisger fields */
372#define ERR_STATUS_LO__ERR_STATUS_VALID_FLAG__SHIFT 0x0
373#define ERR_STATUS_LO__ERR_STATUS_VALID_FLAG_MASK 0x00000001L
374#define ERR_STATUS_LO__MEMORY_ID__SHIFT 0x18
375#define ERR_STATUS_LO__MEMORY_ID_MASK 0xFF000000L
376#define ERR_STATUS_HI__ERR_INFO_VALID_FLAG__SHIFT 0x2
377#define ERR_STATUS_HI__ERR_INFO_VALID_FLAG_MASK 0x00000004L
378#define ERR_STATUS__ERR_CNT__SHIFT 0x17
379#define ERR_STATUS__ERR_CNT_MASK 0x03800000L
380
381#define AMDGPU_RAS_REG_ENTRY(ip, inst, reg_lo, reg_hi) \
382 ip##_HWIP, inst, reg_lo##_BASE_IDX, reg_lo, reg_hi##_BASE_IDX, reg_hi
383
384#define AMDGPU_RAS_REG_ENTRY_OFFSET(hwip, ip_inst, segment, reg) \
385 (adev->reg_offset[hwip][ip_inst][segment] + (reg))
386
387#define AMDGPU_RAS_ERR_INFO_VALID (1 << 0)
388#define AMDGPU_RAS_ERR_STATUS_VALID (1 << 1)
389#define AMDGPU_RAS_ERR_ADDRESS_VALID (1 << 2)
390
391#define AMDGPU_RAS_GPU_RESET_MODE2_RESET (0x1 << 0)
392#define AMDGPU_RAS_GPU_RESET_MODE1_RESET (0x1 << 1)
393
394struct amdgpu_ras_err_status_reg_entry {
395 uint32_t hwip;
396 uint32_t ip_inst;
397 uint32_t seg_lo;
398 uint32_t reg_lo;
399 uint32_t seg_hi;
400 uint32_t reg_hi;
401 uint32_t reg_inst;
402 uint32_t flags;
403 const char *block_name;
404};
405
406struct amdgpu_ras_memory_id_entry {
407 uint32_t memory_id;
408 const char *name;
409};
410
411struct ras_common_if {
412 enum amdgpu_ras_block block;
413 enum amdgpu_ras_error_type type;
414 uint32_t sub_block_index;
415 char name[32];
416};
417
418#define MAX_UMC_CHANNEL_NUM 32
419
420struct ecc_info_per_ch {
421 uint16_t ce_count_lo_chip;
422 uint16_t ce_count_hi_chip;
423 uint64_t mca_umc_status;
424 uint64_t mca_umc_addr;
425 uint64_t mca_ceumc_addr;
426};
427
428struct umc_ecc_info {
429 struct ecc_info_per_ch ecc[MAX_UMC_CHANNEL_NUM];
430
431 /* Determine smu ecctable whether support
432 * record correctable error address
433 */
434 int record_ce_addr_supported;
435};
436
437enum ras_event_type {
438 RAS_EVENT_TYPE_INVALID = 0,
439 RAS_EVENT_TYPE_FATAL,
440 RAS_EVENT_TYPE_POISON_CREATION,
441 RAS_EVENT_TYPE_POISON_CONSUMPTION,
442 RAS_EVENT_TYPE_COUNT,
443};
444
445struct ras_event_state {
446 u64 last_seqno;
447 atomic64_t count;
448};
449
450struct ras_event_manager {
451 atomic64_t seqno;
452 struct ras_event_state event_state[RAS_EVENT_TYPE_COUNT];
453};
454
455struct ras_event_id {
456 enum ras_event_type type;
457 u64 event_id;
458};
459
460struct ras_query_context {
461 struct ras_event_id evid;
462};
463
464typedef int (*pasid_notify)(struct amdgpu_device *adev,
465 uint16_t pasid, void *data);
466
467struct ras_poison_msg {
468 enum amdgpu_ras_block block;
469 uint16_t pasid;
470 uint32_t reset;
471 pasid_notify pasid_fn;
472 void *data;
473};
474
475struct ras_err_pages {
476 uint32_t count;
477 uint64_t *pfn;
478};
479
480struct ras_ecc_err {
481 uint64_t status;
482 uint64_t ipid;
483 uint64_t addr;
484 uint64_t pa_pfn;
485 struct ras_err_pages err_pages;
486};
487
488struct ras_ecc_log_info {
489 struct mutex lock;
490 struct radix_tree_root de_page_tree;
491 uint64_t de_queried_count;
492 uint64_t prev_de_queried_count;
493};
494
495struct amdgpu_ras {
496 /* ras infrastructure */
497 /* for ras itself. */
498 uint32_t features;
499 uint32_t schema;
500 struct list_head head;
501 /* sysfs */
502 struct device_attribute features_attr;
503 struct device_attribute version_attr;
504 struct device_attribute schema_attr;
505 struct device_attribute event_state_attr;
506 struct bin_attribute badpages_attr;
507 struct dentry *de_ras_eeprom_table;
508 /* block array */
509 struct ras_manager *objs;
510
511 /* gpu recovery */
512 struct work_struct recovery_work;
513 atomic_t in_recovery;
514 struct amdgpu_device *adev;
515 /* error handler data */
516 struct ras_err_handler_data *eh_data;
517 struct mutex recovery_lock;
518
519 uint32_t flags;
520 bool reboot;
521 struct amdgpu_ras_eeprom_control eeprom_control;
522
523 bool error_query_ready;
524
525 /* bad page count threshold */
526 uint32_t bad_page_cnt_threshold;
527
528 /* disable ras error count harvest in recovery */
529 bool disable_ras_err_cnt_harvest;
530
531 /* is poison mode supported */
532 bool poison_supported;
533
534 /* RAS count errors delayed work */
535 struct delayed_work ras_counte_delay_work;
536 atomic_t ras_ue_count;
537 atomic_t ras_ce_count;
538
539 /* record umc error info queried from smu */
540 struct umc_ecc_info umc_ecc;
541
542 /* Indicates smu whether need update bad channel info */
543 bool update_channel_flag;
544 /* Record status of smu mca debug mode */
545 bool is_aca_debug_mode;
546 bool is_rma;
547
548 /* Record special requirements of gpu reset caller */
549 uint32_t gpu_reset_flags;
550
551 struct task_struct *page_retirement_thread;
552 wait_queue_head_t page_retirement_wq;
553 struct mutex page_retirement_lock;
554 atomic_t page_retirement_req_cnt;
555 atomic_t poison_creation_count;
556 struct mutex page_rsv_lock;
557 DECLARE_KFIFO(poison_fifo, struct ras_poison_msg, 128);
558 struct ras_ecc_log_info umc_ecc_log;
559 struct delayed_work page_retirement_dwork;
560
561 /* Fatal error detected flag */
562 atomic_t fed;
563
564 /* RAS event manager */
565 struct ras_event_manager __event_mgr;
566 struct ras_event_manager *event_mgr;
567
568 uint64_t reserved_pages_in_bytes;
569};
570
571struct ras_fs_data {
572 char sysfs_name[48];
573 char debugfs_name[32];
574};
575
576struct ras_err_info {
577 struct amdgpu_smuio_mcm_config_info mcm_info;
578 u64 ce_count;
579 u64 ue_count;
580 u64 de_count;
581};
582
583struct ras_err_node {
584 struct list_head node;
585 struct ras_err_info err_info;
586};
587
588struct ras_err_data {
589 unsigned long ue_count;
590 unsigned long ce_count;
591 unsigned long de_count;
592 unsigned long err_addr_cnt;
593 struct eeprom_table_record *err_addr;
594 unsigned long err_addr_len;
595 u32 err_list_count;
596 struct list_head err_node_list;
597};
598
599#define for_each_ras_error(err_node, err_data) \
600 list_for_each_entry(err_node, &(err_data)->err_node_list, node)
601
602struct ras_err_handler_data {
603 /* point to bad page records array */
604 struct eeprom_table_record *bps;
605 /* the count of entries */
606 int count;
607 /* the space can place new entries */
608 int space_left;
609};
610
611typedef int (*ras_ih_cb)(struct amdgpu_device *adev,
612 void *err_data,
613 struct amdgpu_iv_entry *entry);
614
615struct ras_ih_data {
616 /* interrupt bottom half */
617 struct work_struct ih_work;
618 int inuse;
619 /* IP callback */
620 ras_ih_cb cb;
621 /* full of entries */
622 unsigned char *ring;
623 unsigned int ring_size;
624 unsigned int element_size;
625 unsigned int aligned_element_size;
626 unsigned int rptr;
627 unsigned int wptr;
628};
629
630struct ras_manager {
631 struct ras_common_if head;
632 /* reference count */
633 int use;
634 /* ras block link */
635 struct list_head node;
636 /* the device */
637 struct amdgpu_device *adev;
638 /* sysfs */
639 struct device_attribute sysfs_attr;
640 int attr_inuse;
641
642 /* fs node name */
643 struct ras_fs_data fs_data;
644
645 /* IH data */
646 struct ras_ih_data ih_data;
647
648 struct ras_err_data err_data;
649
650 struct aca_handle aca_handle;
651};
652
653struct ras_badpage {
654 unsigned int bp;
655 unsigned int size;
656 unsigned int flags;
657};
658
659/* interfaces for IP */
660struct ras_fs_if {
661 struct ras_common_if head;
662 const char* sysfs_name;
663 char debugfs_name[32];
664};
665
666struct ras_query_if {
667 struct ras_common_if head;
668 unsigned long ue_count;
669 unsigned long ce_count;
670 unsigned long de_count;
671};
672
673struct ras_inject_if {
674 struct ras_common_if head;
675 uint64_t address;
676 uint64_t value;
677 uint32_t instance_mask;
678};
679
680struct ras_cure_if {
681 struct ras_common_if head;
682 uint64_t address;
683};
684
685struct ras_ih_if {
686 struct ras_common_if head;
687 ras_ih_cb cb;
688};
689
690struct ras_dispatch_if {
691 struct ras_common_if head;
692 struct amdgpu_iv_entry *entry;
693};
694
695struct ras_debug_if {
696 union {
697 struct ras_common_if head;
698 struct ras_inject_if inject;
699 };
700 int op;
701};
702
703struct amdgpu_ras_block_object {
704 struct ras_common_if ras_comm;
705
706 int (*ras_block_match)(struct amdgpu_ras_block_object *block_obj,
707 enum amdgpu_ras_block block, uint32_t sub_block_index);
708 int (*ras_late_init)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
709 void (*ras_fini)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
710 ras_ih_cb ras_cb;
711 const struct amdgpu_ras_block_hw_ops *hw_ops;
712};
713
714struct amdgpu_ras_block_hw_ops {
715 int (*ras_error_inject)(struct amdgpu_device *adev,
716 void *inject_if, uint32_t instance_mask);
717 void (*query_ras_error_count)(struct amdgpu_device *adev, void *ras_error_status);
718 void (*query_ras_error_status)(struct amdgpu_device *adev);
719 void (*query_ras_error_address)(struct amdgpu_device *adev, void *ras_error_status);
720 void (*reset_ras_error_count)(struct amdgpu_device *adev);
721 void (*reset_ras_error_status)(struct amdgpu_device *adev);
722 bool (*query_poison_status)(struct amdgpu_device *adev);
723 bool (*handle_poison_consumption)(struct amdgpu_device *adev);
724};
725
726/* work flow
727 * vbios
728 * 1: ras feature enable (enabled by default)
729 * psp
730 * 2: ras framework init (in ip_init)
731 * IP
732 * 3: IH add
733 * 4: debugfs/sysfs create
734 * 5: query/inject
735 * 6: debugfs/sysfs remove
736 * 7: IH remove
737 * 8: feature disable
738 */
739
740int amdgpu_ras_init_badpage_info(struct amdgpu_device *adev);
741int amdgpu_ras_recovery_init(struct amdgpu_device *adev, bool init_bp_info);
742
743void amdgpu_ras_resume(struct amdgpu_device *adev);
744void amdgpu_ras_suspend(struct amdgpu_device *adev);
745
746int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
747 unsigned long *ce_count,
748 unsigned long *ue_count,
749 struct ras_query_if *query_info);
750
751/* error handling functions */
752int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
753 struct eeprom_table_record *bps, int pages);
754
755int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
756 unsigned long *new_cnt);
757
758static inline enum ta_ras_block
759amdgpu_ras_block_to_ta(enum amdgpu_ras_block block) {
760 switch (block) {
761 case AMDGPU_RAS_BLOCK__UMC:
762 return TA_RAS_BLOCK__UMC;
763 case AMDGPU_RAS_BLOCK__SDMA:
764 return TA_RAS_BLOCK__SDMA;
765 case AMDGPU_RAS_BLOCK__GFX:
766 return TA_RAS_BLOCK__GFX;
767 case AMDGPU_RAS_BLOCK__MMHUB:
768 return TA_RAS_BLOCK__MMHUB;
769 case AMDGPU_RAS_BLOCK__ATHUB:
770 return TA_RAS_BLOCK__ATHUB;
771 case AMDGPU_RAS_BLOCK__PCIE_BIF:
772 return TA_RAS_BLOCK__PCIE_BIF;
773 case AMDGPU_RAS_BLOCK__HDP:
774 return TA_RAS_BLOCK__HDP;
775 case AMDGPU_RAS_BLOCK__XGMI_WAFL:
776 return TA_RAS_BLOCK__XGMI_WAFL;
777 case AMDGPU_RAS_BLOCK__DF:
778 return TA_RAS_BLOCK__DF;
779 case AMDGPU_RAS_BLOCK__SMN:
780 return TA_RAS_BLOCK__SMN;
781 case AMDGPU_RAS_BLOCK__SEM:
782 return TA_RAS_BLOCK__SEM;
783 case AMDGPU_RAS_BLOCK__MP0:
784 return TA_RAS_BLOCK__MP0;
785 case AMDGPU_RAS_BLOCK__MP1:
786 return TA_RAS_BLOCK__MP1;
787 case AMDGPU_RAS_BLOCK__FUSE:
788 return TA_RAS_BLOCK__FUSE;
789 case AMDGPU_RAS_BLOCK__MCA:
790 return TA_RAS_BLOCK__MCA;
791 case AMDGPU_RAS_BLOCK__VCN:
792 return TA_RAS_BLOCK__VCN;
793 case AMDGPU_RAS_BLOCK__JPEG:
794 return TA_RAS_BLOCK__JPEG;
795 default:
796 WARN_ONCE(1, "RAS ERROR: unexpected block id %d\n", block);
797 return TA_RAS_BLOCK__UMC;
798 }
799}
800
801static inline enum ta_ras_error_type
802amdgpu_ras_error_to_ta(enum amdgpu_ras_error_type error) {
803 switch (error) {
804 case AMDGPU_RAS_ERROR__NONE:
805 return TA_RAS_ERROR__NONE;
806 case AMDGPU_RAS_ERROR__PARITY:
807 return TA_RAS_ERROR__PARITY;
808 case AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE:
809 return TA_RAS_ERROR__SINGLE_CORRECTABLE;
810 case AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE:
811 return TA_RAS_ERROR__MULTI_UNCORRECTABLE;
812 case AMDGPU_RAS_ERROR__POISON:
813 return TA_RAS_ERROR__POISON;
814 default:
815 WARN_ONCE(1, "RAS ERROR: unexpected error type %d\n", error);
816 return TA_RAS_ERROR__NONE;
817 }
818}
819
820/* called in ip_init and ip_fini */
821int amdgpu_ras_init(struct amdgpu_device *adev);
822int amdgpu_ras_late_init(struct amdgpu_device *adev);
823int amdgpu_ras_fini(struct amdgpu_device *adev);
824int amdgpu_ras_pre_fini(struct amdgpu_device *adev);
825
826int amdgpu_ras_block_late_init(struct amdgpu_device *adev,
827 struct ras_common_if *ras_block);
828
829void amdgpu_ras_block_late_fini(struct amdgpu_device *adev,
830 struct ras_common_if *ras_block);
831
832int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
833 struct ras_common_if *head, bool enable);
834
835int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev,
836 struct ras_common_if *head, bool enable);
837
838int amdgpu_ras_sysfs_create(struct amdgpu_device *adev,
839 struct ras_common_if *head);
840
841int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev,
842 struct ras_common_if *head);
843
844void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev);
845
846int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
847 struct ras_query_if *info);
848
849int amdgpu_ras_reset_error_count(struct amdgpu_device *adev,
850 enum amdgpu_ras_block block);
851int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
852 enum amdgpu_ras_block block);
853
854int amdgpu_ras_error_inject(struct amdgpu_device *adev,
855 struct ras_inject_if *info);
856
857int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev,
858 struct ras_common_if *head);
859
860int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev,
861 struct ras_common_if *head);
862
863int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev,
864 struct ras_dispatch_if *info);
865
866struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
867 struct ras_common_if *head);
868
869extern atomic_t amdgpu_ras_in_intr;
870
871static inline bool amdgpu_ras_intr_triggered(void)
872{
873 return !!atomic_read(&amdgpu_ras_in_intr);
874}
875
876static inline void amdgpu_ras_intr_cleared(void)
877{
878 atomic_set(&amdgpu_ras_in_intr, 0);
879}
880
881void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev);
882
883void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready);
884
885bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev);
886
887void amdgpu_release_ras_context(struct amdgpu_device *adev);
888
889int amdgpu_persistent_edc_harvesting_supported(struct amdgpu_device *adev);
890
891const char *get_ras_block_str(struct ras_common_if *ras_block);
892
893bool amdgpu_ras_is_poison_mode_supported(struct amdgpu_device *adev);
894
895int amdgpu_ras_is_supported(struct amdgpu_device *adev, unsigned int block);
896
897int amdgpu_ras_reset_gpu(struct amdgpu_device *adev);
898
899struct amdgpu_ras* amdgpu_ras_get_context(struct amdgpu_device *adev);
900
901int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_con);
902
903int amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable);
904int amdgpu_ras_set_aca_debug_mode(struct amdgpu_device *adev, bool enable);
905bool amdgpu_ras_get_aca_debug_mode(struct amdgpu_device *adev);
906bool amdgpu_ras_get_error_query_mode(struct amdgpu_device *adev,
907 unsigned int *mode);
908
909int amdgpu_ras_register_ras_block(struct amdgpu_device *adev,
910 struct amdgpu_ras_block_object *ras_block_obj);
911void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev);
912void amdgpu_ras_get_error_type_name(uint32_t err_type, char *err_type_name);
913bool amdgpu_ras_inst_get_memory_id_field(struct amdgpu_device *adev,
914 const struct amdgpu_ras_err_status_reg_entry *reg_entry,
915 uint32_t instance,
916 uint32_t *memory_id);
917bool amdgpu_ras_inst_get_err_cnt_field(struct amdgpu_device *adev,
918 const struct amdgpu_ras_err_status_reg_entry *reg_entry,
919 uint32_t instance,
920 unsigned long *err_cnt);
921void amdgpu_ras_inst_query_ras_error_count(struct amdgpu_device *adev,
922 const struct amdgpu_ras_err_status_reg_entry *reg_list,
923 uint32_t reg_list_size,
924 const struct amdgpu_ras_memory_id_entry *mem_list,
925 uint32_t mem_list_size,
926 uint32_t instance,
927 uint32_t err_type,
928 unsigned long *err_count);
929void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev,
930 const struct amdgpu_ras_err_status_reg_entry *reg_list,
931 uint32_t reg_list_size,
932 uint32_t instance);
933
934int amdgpu_ras_error_data_init(struct ras_err_data *err_data);
935void amdgpu_ras_error_data_fini(struct ras_err_data *err_data);
936int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
937 struct amdgpu_smuio_mcm_config_info *mcm_info,
938 u64 count);
939int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
940 struct amdgpu_smuio_mcm_config_info *mcm_info,
941 u64 count);
942int amdgpu_ras_error_statistic_de_count(struct ras_err_data *err_data,
943 struct amdgpu_smuio_mcm_config_info *mcm_info,
944 u64 count);
945void amdgpu_ras_query_boot_status(struct amdgpu_device *adev, u32 num_instances);
946int amdgpu_ras_bind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk,
947 const struct aca_info *aca_info, void *data);
948int amdgpu_ras_unbind_aca(struct amdgpu_device *adev, enum amdgpu_ras_block blk);
949
950ssize_t amdgpu_ras_aca_sysfs_read(struct device *dev, struct device_attribute *attr,
951 struct aca_handle *handle, char *buf, void *data);
952
953void amdgpu_ras_set_fed(struct amdgpu_device *adev, bool status);
954bool amdgpu_ras_get_fed_status(struct amdgpu_device *adev);
955
956u64 amdgpu_ras_acquire_event_id(struct amdgpu_device *adev, enum ras_event_type type);
957int amdgpu_ras_mark_ras_event_caller(struct amdgpu_device *adev, enum ras_event_type type,
958 const void *caller);
959
960int amdgpu_ras_reserve_page(struct amdgpu_device *adev, uint64_t pfn);
961
962int amdgpu_ras_put_poison_req(struct amdgpu_device *adev,
963 enum amdgpu_ras_block block, uint16_t pasid,
964 pasid_notify pasid_fn, void *data, uint32_t reset);
965
966bool amdgpu_ras_in_recovery(struct amdgpu_device *adev);
967
968__printf(3, 4)
969void amdgpu_ras_event_log_print(struct amdgpu_device *adev, u64 event_id,
970 const char *fmt, ...);
971
972bool amdgpu_ras_is_rma(struct amdgpu_device *adev);
973#endif