]> Git Repo - linux.git/blob - drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
Merge tag '6.7-rc-ksmbd-server-fixes' of git://git.samba.org/ksmbd
[linux.git] / drivers / gpu / drm / amd / amdgpu / amdgpu_ras.h
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 #include "amdgpu_smuio.h"
32
33 struct amdgpu_iv_entry;
34
35 #define AMDGPU_RAS_FLAG_INIT_BY_VBIOS           (0x1 << 0)
36 /* position of instance value in sub_block_index of
37  * ta_ras_trigger_error_input, the sub block uses lower 12 bits
38  */
39 #define AMDGPU_RAS_INST_MASK 0xfffff000
40 #define AMDGPU_RAS_INST_SHIFT 0xc
41
42 enum amdgpu_ras_block {
43         AMDGPU_RAS_BLOCK__UMC = 0,
44         AMDGPU_RAS_BLOCK__SDMA,
45         AMDGPU_RAS_BLOCK__GFX,
46         AMDGPU_RAS_BLOCK__MMHUB,
47         AMDGPU_RAS_BLOCK__ATHUB,
48         AMDGPU_RAS_BLOCK__PCIE_BIF,
49         AMDGPU_RAS_BLOCK__HDP,
50         AMDGPU_RAS_BLOCK__XGMI_WAFL,
51         AMDGPU_RAS_BLOCK__DF,
52         AMDGPU_RAS_BLOCK__SMN,
53         AMDGPU_RAS_BLOCK__SEM,
54         AMDGPU_RAS_BLOCK__MP0,
55         AMDGPU_RAS_BLOCK__MP1,
56         AMDGPU_RAS_BLOCK__FUSE,
57         AMDGPU_RAS_BLOCK__MCA,
58         AMDGPU_RAS_BLOCK__VCN,
59         AMDGPU_RAS_BLOCK__JPEG,
60
61         AMDGPU_RAS_BLOCK__LAST
62 };
63
64 enum amdgpu_ras_mca_block {
65         AMDGPU_RAS_MCA_BLOCK__MP0 = 0,
66         AMDGPU_RAS_MCA_BLOCK__MP1,
67         AMDGPU_RAS_MCA_BLOCK__MPIO,
68         AMDGPU_RAS_MCA_BLOCK__IOHC,
69
70         AMDGPU_RAS_MCA_BLOCK__LAST
71 };
72
73 #define AMDGPU_RAS_BLOCK_COUNT  AMDGPU_RAS_BLOCK__LAST
74 #define AMDGPU_RAS_MCA_BLOCK_COUNT      AMDGPU_RAS_MCA_BLOCK__LAST
75 #define AMDGPU_RAS_BLOCK_MASK   ((1ULL << AMDGPU_RAS_BLOCK_COUNT) - 1)
76
77 enum amdgpu_ras_gfx_subblock {
78         /* CPC */
79         AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
80         AMDGPU_RAS_BLOCK__GFX_CPC_SCRATCH =
81                 AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_START,
82         AMDGPU_RAS_BLOCK__GFX_CPC_UCODE,
83         AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME1,
84         AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
85         AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME1,
86         AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME2,
87         AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
88         AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
89         AMDGPU_RAS_BLOCK__GFX_CPC_INDEX_END =
90                 AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2,
91         /* CPF */
92         AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
93         AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME2 =
94                 AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_START,
95         AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME1,
96         AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
97         AMDGPU_RAS_BLOCK__GFX_CPF_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPF_TAG,
98         /* CPG */
99         AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
100         AMDGPU_RAS_BLOCK__GFX_CPG_DMA_ROQ =
101                 AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_START,
102         AMDGPU_RAS_BLOCK__GFX_CPG_DMA_TAG,
103         AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
104         AMDGPU_RAS_BLOCK__GFX_CPG_INDEX_END = AMDGPU_RAS_BLOCK__GFX_CPG_TAG,
105         /* GDS */
106         AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
107         AMDGPU_RAS_BLOCK__GFX_GDS_MEM = AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_START,
108         AMDGPU_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
109         AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
110         AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
111         AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
112         AMDGPU_RAS_BLOCK__GFX_GDS_INDEX_END =
113                 AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
114         /* SPI */
115         AMDGPU_RAS_BLOCK__GFX_SPI_SR_MEM,
116         /* SQ */
117         AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
118         AMDGPU_RAS_BLOCK__GFX_SQ_SGPR = AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_START,
119         AMDGPU_RAS_BLOCK__GFX_SQ_LDS_D,
120         AMDGPU_RAS_BLOCK__GFX_SQ_LDS_I,
121         AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
122         AMDGPU_RAS_BLOCK__GFX_SQ_INDEX_END = AMDGPU_RAS_BLOCK__GFX_SQ_VGPR,
123         /* SQC (3 ranges) */
124         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
125         /* SQC range 0 */
126         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START =
127                 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_START,
128         AMDGPU_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
129                 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_START,
130         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
131         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
132         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
133         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
134         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
135         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
136         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX0_END =
137                 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
138         /* SQC range 1 */
139         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
140         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
141                 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_START,
142         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
143         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
144         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
145         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
146         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
147         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
148         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
149         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
150         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX1_END =
151                 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
152         /* SQC range 2 */
153         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
154         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
155                 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_START,
156         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
157         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
158         AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
159         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
160         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
161         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
162         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
163         AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
164         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END =
165                 AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
166         AMDGPU_RAS_BLOCK__GFX_SQC_INDEX_END =
167                 AMDGPU_RAS_BLOCK__GFX_SQC_INDEX2_END,
168         /* TA */
169         AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
170         AMDGPU_RAS_BLOCK__GFX_TA_FS_DFIFO =
171                 AMDGPU_RAS_BLOCK__GFX_TA_INDEX_START,
172         AMDGPU_RAS_BLOCK__GFX_TA_FS_AFIFO,
173         AMDGPU_RAS_BLOCK__GFX_TA_FL_LFIFO,
174         AMDGPU_RAS_BLOCK__GFX_TA_FX_LFIFO,
175         AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
176         AMDGPU_RAS_BLOCK__GFX_TA_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO,
177         /* TCA */
178         AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
179         AMDGPU_RAS_BLOCK__GFX_TCA_HOLE_FIFO =
180                 AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_START,
181         AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
182         AMDGPU_RAS_BLOCK__GFX_TCA_INDEX_END =
183                 AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO,
184         /* TCC (5 sub-ranges) */
185         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
186         /* TCC range 0 */
187         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START =
188                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_START,
189         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA =
190                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_START,
191         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
192         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
193         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
194         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
195         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
196         AMDGPU_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
197         AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
198         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX0_END =
199                 AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
200         /* TCC range 1 */
201         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
202         AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_DEC =
203                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_START,
204         AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
205         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX1_END =
206                 AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
207         /* TCC range 2 */
208         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
209         AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_DATA =
210                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_START,
211         AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
212         AMDGPU_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
213         AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
214         AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
215         AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO,
216         AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
217         AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
218         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX2_END =
219                 AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
220         /* TCC range 3 */
221         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
222         AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO =
223                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_START,
224         AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
225         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX3_END =
226                 AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
227         /* TCC range 4 */
228         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
229         AMDGPU_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
230                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_START,
231         AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
232         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END =
233                 AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
234         AMDGPU_RAS_BLOCK__GFX_TCC_INDEX_END =
235                 AMDGPU_RAS_BLOCK__GFX_TCC_INDEX4_END,
236         /* TCI */
237         AMDGPU_RAS_BLOCK__GFX_TCI_WRITE_RAM,
238         /* TCP */
239         AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
240         AMDGPU_RAS_BLOCK__GFX_TCP_CACHE_RAM =
241                 AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_START,
242         AMDGPU_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
243         AMDGPU_RAS_BLOCK__GFX_TCP_CMD_FIFO,
244         AMDGPU_RAS_BLOCK__GFX_TCP_VM_FIFO,
245         AMDGPU_RAS_BLOCK__GFX_TCP_DB_RAM,
246         AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
247         AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
248         AMDGPU_RAS_BLOCK__GFX_TCP_INDEX_END =
249                 AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
250         /* TD */
251         AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
252         AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_LO =
253                 AMDGPU_RAS_BLOCK__GFX_TD_INDEX_START,
254         AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
255         AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
256         AMDGPU_RAS_BLOCK__GFX_TD_INDEX_END = AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO,
257         /* EA (3 sub-ranges) */
258         AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
259         /* EA range 0 */
260         AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START =
261                 AMDGPU_RAS_BLOCK__GFX_EA_INDEX_START,
262         AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM =
263                 AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_START,
264         AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
265         AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
266         AMDGPU_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
267         AMDGPU_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
268         AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
269         AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
270         AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
271         AMDGPU_RAS_BLOCK__GFX_EA_INDEX0_END =
272                 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
273         /* EA range 1 */
274         AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
275         AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM =
276                 AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_START,
277         AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
278         AMDGPU_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
279         AMDGPU_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
280         AMDGPU_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
281         AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
282         AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
283         AMDGPU_RAS_BLOCK__GFX_EA_INDEX1_END =
284                 AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
285         /* EA range 2 */
286         AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
287         AMDGPU_RAS_BLOCK__GFX_EA_MAM_D0MEM =
288                 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_START,
289         AMDGPU_RAS_BLOCK__GFX_EA_MAM_D1MEM,
290         AMDGPU_RAS_BLOCK__GFX_EA_MAM_D2MEM,
291         AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
292         AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END =
293                 AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM,
294         AMDGPU_RAS_BLOCK__GFX_EA_INDEX_END =
295                 AMDGPU_RAS_BLOCK__GFX_EA_INDEX2_END,
296         /* UTC VM L2 bank */
297         AMDGPU_RAS_BLOCK__UTC_VML2_BANK_CACHE,
298         /* UTC VM walker */
299         AMDGPU_RAS_BLOCK__UTC_VML2_WALKER,
300         /* UTC ATC L2 2MB cache */
301         AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
302         /* UTC ATC L2 4KB cache */
303         AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
304         AMDGPU_RAS_BLOCK__GFX_MAX
305 };
306
307 enum amdgpu_ras_error_type {
308         AMDGPU_RAS_ERROR__NONE                                                  = 0,
309         AMDGPU_RAS_ERROR__PARITY                                                = 1,
310         AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE                                    = 2,
311         AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE                                   = 4,
312         AMDGPU_RAS_ERROR__POISON                                                = 8,
313 };
314
315 enum amdgpu_ras_ret {
316         AMDGPU_RAS_SUCCESS = 0,
317         AMDGPU_RAS_FAIL,
318         AMDGPU_RAS_UE,
319         AMDGPU_RAS_CE,
320         AMDGPU_RAS_PT,
321 };
322
323 /* ras error status reisger fields */
324 #define ERR_STATUS_LO__ERR_STATUS_VALID_FLAG__SHIFT     0x0
325 #define ERR_STATUS_LO__ERR_STATUS_VALID_FLAG_MASK       0x00000001L
326 #define ERR_STATUS_LO__MEMORY_ID__SHIFT                 0x18
327 #define ERR_STATUS_LO__MEMORY_ID_MASK                   0xFF000000L
328 #define ERR_STATUS_HI__ERR_INFO_VALID_FLAG__SHIFT       0x2
329 #define ERR_STATUS_HI__ERR_INFO_VALID_FLAG_MASK         0x00000004L
330 #define ERR_STATUS__ERR_CNT__SHIFT                      0x17
331 #define ERR_STATUS__ERR_CNT_MASK                        0x03800000L
332
333 #define AMDGPU_RAS_REG_ENTRY(ip, inst, reg_lo, reg_hi) \
334         ip##_HWIP, inst, reg_lo##_BASE_IDX, reg_lo, reg_hi##_BASE_IDX, reg_hi
335
336 #define AMDGPU_RAS_REG_ENTRY_OFFSET(hwip, ip_inst, segment, reg) \
337         (adev->reg_offset[hwip][ip_inst][segment] + (reg))
338
339 #define AMDGPU_RAS_ERR_INFO_VALID       (1 << 0)
340 #define AMDGPU_RAS_ERR_STATUS_VALID     (1 << 1)
341 #define AMDGPU_RAS_ERR_ADDRESS_VALID    (1 << 2)
342
343 #define AMDGPU_RAS_GPU_RESET_MODE2_RESET  (0x1 << 0)
344 #define AMDGPU_RAS_GPU_RESET_MODE1_RESET  (0x1 << 1)
345
346 struct amdgpu_ras_err_status_reg_entry {
347         uint32_t hwip;
348         uint32_t ip_inst;
349         uint32_t seg_lo;
350         uint32_t reg_lo;
351         uint32_t seg_hi;
352         uint32_t reg_hi;
353         uint32_t reg_inst;
354         uint32_t flags;
355         const char *block_name;
356 };
357
358 struct amdgpu_ras_memory_id_entry {
359         uint32_t memory_id;
360         const char *name;
361 };
362
363 struct ras_common_if {
364         enum amdgpu_ras_block block;
365         enum amdgpu_ras_error_type type;
366         uint32_t sub_block_index;
367         char name[32];
368 };
369
370 #define MAX_UMC_CHANNEL_NUM 32
371
372 struct ecc_info_per_ch {
373         uint16_t ce_count_lo_chip;
374         uint16_t ce_count_hi_chip;
375         uint64_t mca_umc_status;
376         uint64_t mca_umc_addr;
377         uint64_t mca_ceumc_addr;
378 };
379
380 struct umc_ecc_info {
381         struct ecc_info_per_ch ecc[MAX_UMC_CHANNEL_NUM];
382
383         /* Determine smu ecctable whether support
384          * record correctable error address
385          */
386         int record_ce_addr_supported;
387 };
388
389 struct amdgpu_ras {
390         /* ras infrastructure */
391         /* for ras itself. */
392         uint32_t features;
393         uint32_t schema;
394         struct list_head head;
395         /* sysfs */
396         struct device_attribute features_attr;
397         struct device_attribute version_attr;
398         struct device_attribute schema_attr;
399         struct bin_attribute badpages_attr;
400         struct dentry *de_ras_eeprom_table;
401         /* block array */
402         struct ras_manager *objs;
403
404         /* gpu recovery */
405         struct work_struct recovery_work;
406         atomic_t in_recovery;
407         struct amdgpu_device *adev;
408         /* error handler data */
409         struct ras_err_handler_data *eh_data;
410         struct mutex recovery_lock;
411
412         uint32_t flags;
413         bool reboot;
414         struct amdgpu_ras_eeprom_control eeprom_control;
415
416         bool error_query_ready;
417
418         /* bad page count threshold */
419         uint32_t bad_page_cnt_threshold;
420
421         /* disable ras error count harvest in recovery */
422         bool disable_ras_err_cnt_harvest;
423
424         /* is poison mode supported */
425         bool poison_supported;
426
427         /* RAS count errors delayed work */
428         struct delayed_work ras_counte_delay_work;
429         atomic_t ras_ue_count;
430         atomic_t ras_ce_count;
431
432         /* record umc error info queried from smu */
433         struct umc_ecc_info umc_ecc;
434
435         /* Indicates smu whether need update bad channel info */
436         bool update_channel_flag;
437         /* Record status of smu mca debug mode */
438         bool is_mca_debug_mode;
439
440         /* Record special requirements of gpu reset caller */
441         uint32_t  gpu_reset_flags;
442 };
443
444 struct ras_fs_data {
445         char sysfs_name[48];
446         char debugfs_name[32];
447 };
448
449 struct ras_err_info {
450         struct amdgpu_smuio_mcm_config_info mcm_info;
451         u64 ce_count;
452         u64 ue_count;
453 };
454
455 struct ras_err_node {
456         struct list_head node;
457         struct ras_err_info err_info;
458 };
459
460 struct ras_err_data {
461         unsigned long ue_count;
462         unsigned long ce_count;
463         unsigned long err_addr_cnt;
464         struct eeprom_table_record *err_addr;
465         u32 err_list_count;
466         struct list_head err_node_list;
467 };
468
469 #define for_each_ras_error(err_node, err_data) \
470         list_for_each_entry(err_node, &(err_data)->err_node_list, node)
471
472 struct ras_err_handler_data {
473         /* point to bad page records array */
474         struct eeprom_table_record *bps;
475         /* the count of entries */
476         int count;
477         /* the space can place new entries */
478         int space_left;
479 };
480
481 typedef int (*ras_ih_cb)(struct amdgpu_device *adev,
482                 void *err_data,
483                 struct amdgpu_iv_entry *entry);
484
485 struct ras_ih_data {
486         /* interrupt bottom half */
487         struct work_struct ih_work;
488         int inuse;
489         /* IP callback */
490         ras_ih_cb cb;
491         /* full of entries */
492         unsigned char *ring;
493         unsigned int ring_size;
494         unsigned int element_size;
495         unsigned int aligned_element_size;
496         unsigned int rptr;
497         unsigned int wptr;
498 };
499
500 struct ras_manager {
501         struct ras_common_if head;
502         /* reference count */
503         int use;
504         /* ras block link */
505         struct list_head node;
506         /* the device */
507         struct amdgpu_device *adev;
508         /* sysfs */
509         struct device_attribute sysfs_attr;
510         int attr_inuse;
511
512         /* fs node name */
513         struct ras_fs_data fs_data;
514
515         /* IH data */
516         struct ras_ih_data ih_data;
517
518         struct ras_err_data err_data;
519 };
520
521 struct ras_badpage {
522         unsigned int bp;
523         unsigned int size;
524         unsigned int flags;
525 };
526
527 /* interfaces for IP */
528 struct ras_fs_if {
529         struct ras_common_if head;
530         const char* sysfs_name;
531         char debugfs_name[32];
532 };
533
534 struct ras_query_if {
535         struct ras_common_if head;
536         unsigned long ue_count;
537         unsigned long ce_count;
538 };
539
540 struct ras_inject_if {
541         struct ras_common_if head;
542         uint64_t address;
543         uint64_t value;
544         uint32_t instance_mask;
545 };
546
547 struct ras_cure_if {
548         struct ras_common_if head;
549         uint64_t address;
550 };
551
552 struct ras_ih_if {
553         struct ras_common_if head;
554         ras_ih_cb cb;
555 };
556
557 struct ras_dispatch_if {
558         struct ras_common_if head;
559         struct amdgpu_iv_entry *entry;
560 };
561
562 struct ras_debug_if {
563         union {
564                 struct ras_common_if head;
565                 struct ras_inject_if inject;
566         };
567         int op;
568 };
569
570 struct amdgpu_ras_block_object {
571         struct ras_common_if  ras_comm;
572
573         int (*ras_block_match)(struct amdgpu_ras_block_object *block_obj,
574                                 enum amdgpu_ras_block block, uint32_t sub_block_index);
575         int (*ras_late_init)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
576         void (*ras_fini)(struct amdgpu_device *adev, struct ras_common_if *ras_block);
577         ras_ih_cb ras_cb;
578         const struct amdgpu_ras_block_hw_ops *hw_ops;
579 };
580
581 struct amdgpu_ras_block_hw_ops {
582         int  (*ras_error_inject)(struct amdgpu_device *adev,
583                         void *inject_if, uint32_t instance_mask);
584         void (*query_ras_error_count)(struct amdgpu_device *adev, void *ras_error_status);
585         void (*query_ras_error_status)(struct amdgpu_device *adev);
586         void (*query_ras_error_address)(struct amdgpu_device *adev, void *ras_error_status);
587         void (*reset_ras_error_count)(struct amdgpu_device *adev);
588         void (*reset_ras_error_status)(struct amdgpu_device *adev);
589         bool (*query_poison_status)(struct amdgpu_device *adev);
590         bool (*handle_poison_consumption)(struct amdgpu_device *adev);
591 };
592
593 /* work flow
594  * vbios
595  * 1: ras feature enable (enabled by default)
596  * psp
597  * 2: ras framework init (in ip_init)
598  * IP
599  * 3: IH add
600  * 4: debugfs/sysfs create
601  * 5: query/inject
602  * 6: debugfs/sysfs remove
603  * 7: IH remove
604  * 8: feature disable
605  */
606
607
608 int amdgpu_ras_recovery_init(struct amdgpu_device *adev);
609
610 void amdgpu_ras_resume(struct amdgpu_device *adev);
611 void amdgpu_ras_suspend(struct amdgpu_device *adev);
612
613 int amdgpu_ras_query_error_count(struct amdgpu_device *adev,
614                                  unsigned long *ce_count,
615                                  unsigned long *ue_count,
616                                  struct ras_query_if *query_info);
617
618 /* error handling functions */
619 int amdgpu_ras_add_bad_pages(struct amdgpu_device *adev,
620                 struct eeprom_table_record *bps, int pages);
621
622 int amdgpu_ras_save_bad_pages(struct amdgpu_device *adev,
623                 unsigned long *new_cnt);
624
625 static inline enum ta_ras_block
626 amdgpu_ras_block_to_ta(enum amdgpu_ras_block block) {
627         switch (block) {
628         case AMDGPU_RAS_BLOCK__UMC:
629                 return TA_RAS_BLOCK__UMC;
630         case AMDGPU_RAS_BLOCK__SDMA:
631                 return TA_RAS_BLOCK__SDMA;
632         case AMDGPU_RAS_BLOCK__GFX:
633                 return TA_RAS_BLOCK__GFX;
634         case AMDGPU_RAS_BLOCK__MMHUB:
635                 return TA_RAS_BLOCK__MMHUB;
636         case AMDGPU_RAS_BLOCK__ATHUB:
637                 return TA_RAS_BLOCK__ATHUB;
638         case AMDGPU_RAS_BLOCK__PCIE_BIF:
639                 return TA_RAS_BLOCK__PCIE_BIF;
640         case AMDGPU_RAS_BLOCK__HDP:
641                 return TA_RAS_BLOCK__HDP;
642         case AMDGPU_RAS_BLOCK__XGMI_WAFL:
643                 return TA_RAS_BLOCK__XGMI_WAFL;
644         case AMDGPU_RAS_BLOCK__DF:
645                 return TA_RAS_BLOCK__DF;
646         case AMDGPU_RAS_BLOCK__SMN:
647                 return TA_RAS_BLOCK__SMN;
648         case AMDGPU_RAS_BLOCK__SEM:
649                 return TA_RAS_BLOCK__SEM;
650         case AMDGPU_RAS_BLOCK__MP0:
651                 return TA_RAS_BLOCK__MP0;
652         case AMDGPU_RAS_BLOCK__MP1:
653                 return TA_RAS_BLOCK__MP1;
654         case AMDGPU_RAS_BLOCK__FUSE:
655                 return TA_RAS_BLOCK__FUSE;
656         case AMDGPU_RAS_BLOCK__MCA:
657                 return TA_RAS_BLOCK__MCA;
658         case AMDGPU_RAS_BLOCK__VCN:
659                 return TA_RAS_BLOCK__VCN;
660         case AMDGPU_RAS_BLOCK__JPEG:
661                 return TA_RAS_BLOCK__JPEG;
662         default:
663                 WARN_ONCE(1, "RAS ERROR: unexpected block id %d\n", block);
664                 return TA_RAS_BLOCK__UMC;
665         }
666 }
667
668 static inline enum ta_ras_error_type
669 amdgpu_ras_error_to_ta(enum amdgpu_ras_error_type error) {
670         switch (error) {
671         case AMDGPU_RAS_ERROR__NONE:
672                 return TA_RAS_ERROR__NONE;
673         case AMDGPU_RAS_ERROR__PARITY:
674                 return TA_RAS_ERROR__PARITY;
675         case AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE:
676                 return TA_RAS_ERROR__SINGLE_CORRECTABLE;
677         case AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE:
678                 return TA_RAS_ERROR__MULTI_UNCORRECTABLE;
679         case AMDGPU_RAS_ERROR__POISON:
680                 return TA_RAS_ERROR__POISON;
681         default:
682                 WARN_ONCE(1, "RAS ERROR: unexpected error type %d\n", error);
683                 return TA_RAS_ERROR__NONE;
684         }
685 }
686
687 /* called in ip_init and ip_fini */
688 int amdgpu_ras_init(struct amdgpu_device *adev);
689 int amdgpu_ras_late_init(struct amdgpu_device *adev);
690 int amdgpu_ras_fini(struct amdgpu_device *adev);
691 int amdgpu_ras_pre_fini(struct amdgpu_device *adev);
692
693 int amdgpu_ras_block_late_init(struct amdgpu_device *adev,
694                         struct ras_common_if *ras_block);
695
696 void amdgpu_ras_block_late_fini(struct amdgpu_device *adev,
697                           struct ras_common_if *ras_block);
698
699 int amdgpu_ras_feature_enable(struct amdgpu_device *adev,
700                 struct ras_common_if *head, bool enable);
701
702 int amdgpu_ras_feature_enable_on_boot(struct amdgpu_device *adev,
703                 struct ras_common_if *head, bool enable);
704
705 int amdgpu_ras_sysfs_create(struct amdgpu_device *adev,
706                 struct ras_common_if *head);
707
708 int amdgpu_ras_sysfs_remove(struct amdgpu_device *adev,
709                 struct ras_common_if *head);
710
711 void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev);
712
713 int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
714                 struct ras_query_if *info);
715
716 int amdgpu_ras_reset_error_count(struct amdgpu_device *adev,
717                 enum amdgpu_ras_block block);
718 int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
719                 enum amdgpu_ras_block block);
720
721 int amdgpu_ras_error_inject(struct amdgpu_device *adev,
722                 struct ras_inject_if *info);
723
724 int amdgpu_ras_interrupt_add_handler(struct amdgpu_device *adev,
725                 struct ras_common_if *head);
726
727 int amdgpu_ras_interrupt_remove_handler(struct amdgpu_device *adev,
728                 struct ras_common_if *head);
729
730 int amdgpu_ras_interrupt_dispatch(struct amdgpu_device *adev,
731                 struct ras_dispatch_if *info);
732
733 struct ras_manager *amdgpu_ras_find_obj(struct amdgpu_device *adev,
734                 struct ras_common_if *head);
735
736 extern atomic_t amdgpu_ras_in_intr;
737
738 static inline bool amdgpu_ras_intr_triggered(void)
739 {
740         return !!atomic_read(&amdgpu_ras_in_intr);
741 }
742
743 static inline void amdgpu_ras_intr_cleared(void)
744 {
745         atomic_set(&amdgpu_ras_in_intr, 0);
746 }
747
748 void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev);
749
750 void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready);
751
752 bool amdgpu_ras_need_emergency_restart(struct amdgpu_device *adev);
753
754 void amdgpu_release_ras_context(struct amdgpu_device *adev);
755
756 int amdgpu_persistent_edc_harvesting_supported(struct amdgpu_device *adev);
757
758 const char *get_ras_block_str(struct ras_common_if *ras_block);
759
760 bool amdgpu_ras_is_poison_mode_supported(struct amdgpu_device *adev);
761
762 int amdgpu_ras_is_supported(struct amdgpu_device *adev, unsigned int block);
763
764 int amdgpu_ras_reset_gpu(struct amdgpu_device *adev);
765
766 struct amdgpu_ras* amdgpu_ras_get_context(struct amdgpu_device *adev);
767
768 int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras *ras_con);
769
770 void amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable);
771 bool amdgpu_ras_get_mca_debug_mode(struct amdgpu_device *adev);
772
773 int amdgpu_ras_register_ras_block(struct amdgpu_device *adev,
774                                 struct amdgpu_ras_block_object *ras_block_obj);
775 void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev);
776 void amdgpu_ras_get_error_type_name(uint32_t err_type, char *err_type_name);
777 bool amdgpu_ras_inst_get_memory_id_field(struct amdgpu_device *adev,
778                                          const struct amdgpu_ras_err_status_reg_entry *reg_entry,
779                                          uint32_t instance,
780                                          uint32_t *memory_id);
781 bool amdgpu_ras_inst_get_err_cnt_field(struct amdgpu_device *adev,
782                                        const struct amdgpu_ras_err_status_reg_entry *reg_entry,
783                                        uint32_t instance,
784                                        unsigned long *err_cnt);
785 void amdgpu_ras_inst_query_ras_error_count(struct amdgpu_device *adev,
786                                            const struct amdgpu_ras_err_status_reg_entry *reg_list,
787                                            uint32_t reg_list_size,
788                                            const struct amdgpu_ras_memory_id_entry *mem_list,
789                                            uint32_t mem_list_size,
790                                            uint32_t instance,
791                                            uint32_t err_type,
792                                            unsigned long *err_count);
793 void amdgpu_ras_inst_reset_ras_error_count(struct amdgpu_device *adev,
794                                            const struct amdgpu_ras_err_status_reg_entry *reg_list,
795                                            uint32_t reg_list_size,
796                                            uint32_t instance);
797
798 int amdgpu_ras_error_data_init(struct ras_err_data *err_data);
799 void amdgpu_ras_error_data_fini(struct ras_err_data *err_data);
800 int amdgpu_ras_error_statistic_ce_count(struct ras_err_data *err_data,
801                                         struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count);
802 int amdgpu_ras_error_statistic_ue_count(struct ras_err_data *err_data,
803                                         struct amdgpu_smuio_mcm_config_info *mcm_info, u64 count);
804
805 #endif
This page took 0.083211 seconds and 4 git commands to generate.