2 * Copyright 2015 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
31 #define VG(x) ((void)0)
37 #include "sid_tables.h"
38 #include "util/u_math.h"
39 #include "util/u_memory.h"
40 #include "util/u_string.h"
44 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
45 * read them, or use "aha -b -f file" to convert them to html.
47 #define COLOR_RESET "\033[0m"
48 #define COLOR_RED "\033[31m"
49 #define COLOR_GREEN "\033[1;32m"
50 #define COLOR_YELLOW "\033[1;33m"
51 #define COLOR_CYAN "\033[1;36m"
60 unsigned trace_id_count
;
61 enum chip_class chip_class
;
62 ac_debug_addr_callback addr_callback
;
63 void *addr_callback_data
;
68 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
);
70 static void print_spaces(FILE *f
, unsigned num
)
72 fprintf(f
, "%*s", num
, "");
75 static void print_value(FILE *file
, uint32_t value
, int bits
)
77 /* Guess if it's int or float */
78 if (value
<= (1 << 15)) {
80 fprintf(file
, "%u\n", value
);
82 fprintf(file
, "%u (0x%0*x)\n", value
, bits
/ 4, value
);
86 if (fabs(f
) < 100000 && f
*10 == floor(f
*10))
87 fprintf(file
, "%.1ff (0x%0*x)\n", f
, bits
/ 4, value
);
89 /* Don't print more leading zeros than there are bits. */
90 fprintf(file
, "0x%0*x\n", bits
/ 4, value
);
94 static void print_named_value(FILE *file
, const char *name
, uint32_t value
,
97 print_spaces(file
, INDENT_PKT
);
98 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ", name
);
99 print_value(file
, value
, bits
);
102 static const struct si_reg
*find_register(enum chip_class chip_class
, unsigned offset
)
104 const struct si_reg
*table
;
107 switch (chip_class
) {
110 table
= gfx10_reg_table
;
111 table_size
= ARRAY_SIZE(gfx10_reg_table
);
114 table
= gfx9_reg_table
;
115 table_size
= ARRAY_SIZE(gfx9_reg_table
);
118 table
= gfx8_reg_table
;
119 table_size
= ARRAY_SIZE(gfx8_reg_table
);
122 table
= gfx7_reg_table
;
123 table_size
= ARRAY_SIZE(gfx7_reg_table
);
126 table
= gfx6_reg_table
;
127 table_size
= ARRAY_SIZE(gfx6_reg_table
);
133 for (unsigned i
= 0; i
< table_size
; i
++) {
134 const struct si_reg
*reg
= &table
[i
];
136 if (reg
->offset
== offset
)
143 const char *ac_get_register_name(enum chip_class chip_class
, unsigned offset
)
145 const struct si_reg
*reg
= find_register(chip_class
, offset
);
147 return reg
? sid_strings
+ reg
->name_offset
: "(no name)";
150 void ac_dump_reg(FILE *file
, enum chip_class chip_class
, unsigned offset
,
151 uint32_t value
, uint32_t field_mask
)
153 const struct si_reg
*reg
= find_register(chip_class
, offset
);
156 const char *reg_name
= sid_strings
+ reg
->name_offset
;
157 bool first_field
= true;
159 print_spaces(file
, INDENT_PKT
);
160 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ",
163 if (!reg
->num_fields
) {
164 print_value(file
, value
, 32);
168 for (unsigned f
= 0; f
< reg
->num_fields
; f
++) {
169 const struct si_field
*field
= sid_fields_table
+ reg
->fields_offset
+ f
;
170 const int *values_offsets
= sid_strings_offsets
+ field
->values_offset
;
171 uint32_t val
= (value
& field
->mask
) >>
172 (ffs(field
->mask
) - 1);
174 if (!(field
->mask
& field_mask
))
177 /* Indent the field. */
180 INDENT_PKT
+ strlen(reg_name
) + 4);
182 /* Print the field. */
183 fprintf(file
, "%s = ", sid_strings
+ field
->name_offset
);
185 if (val
< field
->num_values
&& values_offsets
[val
] >= 0)
186 fprintf(file
, "%s\n", sid_strings
+ values_offsets
[val
]);
188 print_value(file
, val
,
189 util_bitcount(field
->mask
));
196 print_spaces(file
, INDENT_PKT
);
197 fprintf(file
, COLOR_YELLOW
"0x%05x" COLOR_RESET
" <- 0x%08x\n", offset
, value
);
200 static uint32_t ac_ib_get(struct ac_ib_parser
*ib
)
204 if (ib
->cur_dw
< ib
->num_dw
) {
205 v
= ib
->ib
[ib
->cur_dw
];
207 /* Help figure out where garbage data is written to IBs.
209 * Arguably we should do this already when the IBs are written,
210 * see RADEON_VALGRIND. The problem is that client-requests to
211 * Valgrind have an overhead even when Valgrind isn't running,
212 * and radeon_emit is performance sensitive...
214 if (VALGRIND_CHECK_VALUE_IS_DEFINED(v
))
215 fprintf(ib
->f
, COLOR_RED
"Valgrind: The next DWORD is garbage"
218 fprintf(ib
->f
, "\n\035#%08x ", v
);
220 fprintf(ib
->f
, "\n\035#???????? ");
227 static void ac_parse_set_reg_packet(FILE *f
, unsigned count
, unsigned reg_offset
,
228 struct ac_ib_parser
*ib
)
230 unsigned reg_dw
= ac_ib_get(ib
);
231 unsigned reg
= ((reg_dw
& 0xFFFF) << 2) + reg_offset
;
232 unsigned index
= reg_dw
>> 28;
236 print_spaces(f
, INDENT_PKT
);
237 fprintf(f
, "INDEX = %u\n", index
);
240 for (i
= 0; i
< count
; i
++)
241 ac_dump_reg(f
, ib
->chip_class
, reg
+ i
*4, ac_ib_get(ib
), ~0);
244 static void ac_parse_packet3(FILE *f
, uint32_t header
, struct ac_ib_parser
*ib
,
245 int *current_trace_id
)
247 unsigned first_dw
= ib
->cur_dw
;
248 int count
= PKT_COUNT_G(header
);
249 unsigned op
= PKT3_IT_OPCODE_G(header
);
250 const char *predicate
= PKT3_PREDICATE(header
) ? "(predicate)" : "";
253 /* Print the name first. */
254 for (i
= 0; i
< ARRAY_SIZE(packet3_table
); i
++)
255 if (packet3_table
[i
].op
== op
)
258 if (i
< ARRAY_SIZE(packet3_table
)) {
259 const char *name
= sid_strings
+ packet3_table
[i
].name_offset
;
261 if (op
== PKT3_SET_CONTEXT_REG
||
262 op
== PKT3_SET_CONFIG_REG
||
263 op
== PKT3_SET_UCONFIG_REG
||
264 op
== PKT3_SET_UCONFIG_REG_INDEX
||
265 op
== PKT3_SET_SH_REG
)
266 fprintf(f
, COLOR_CYAN
"%s%s" COLOR_CYAN
":\n",
269 fprintf(f
, COLOR_GREEN
"%s%s" COLOR_RESET
":\n",
272 fprintf(f
, COLOR_RED
"PKT3_UNKNOWN 0x%x%s" COLOR_RESET
":\n",
275 /* Print the contents. */
277 case PKT3_SET_CONTEXT_REG
:
278 ac_parse_set_reg_packet(f
, count
, SI_CONTEXT_REG_OFFSET
, ib
);
280 case PKT3_SET_CONFIG_REG
:
281 ac_parse_set_reg_packet(f
, count
, SI_CONFIG_REG_OFFSET
, ib
);
283 case PKT3_SET_UCONFIG_REG
:
284 case PKT3_SET_UCONFIG_REG_INDEX
:
285 ac_parse_set_reg_packet(f
, count
, CIK_UCONFIG_REG_OFFSET
, ib
);
287 case PKT3_SET_SH_REG
:
288 ac_parse_set_reg_packet(f
, count
, SI_SH_REG_OFFSET
, ib
);
290 case PKT3_ACQUIRE_MEM
:
291 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
292 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
293 ac_dump_reg(f
, ib
->chip_class
, R_030230_CP_COHER_SIZE_HI
, ac_ib_get(ib
), ~0);
294 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
295 ac_dump_reg(f
, ib
->chip_class
, R_0301E4_CP_COHER_BASE_HI
, ac_ib_get(ib
), ~0);
296 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
297 if (ib
->chip_class
>= GFX10
)
298 ac_dump_reg(f
, ib
->chip_class
, R_586_GCR_CNTL
, ac_ib_get(ib
), ~0);
300 case PKT3_SURFACE_SYNC
:
301 if (ib
->chip_class
>= GFX7
) {
302 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
303 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
304 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
306 ac_dump_reg(f
, ib
->chip_class
, R_0085F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
307 ac_dump_reg(f
, ib
->chip_class
, R_0085F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
308 ac_dump_reg(f
, ib
->chip_class
, R_0085F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
310 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
312 case PKT3_EVENT_WRITE
: {
313 uint32_t event_dw
= ac_ib_get(ib
);
314 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
315 S_028A90_EVENT_TYPE(~0));
316 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
317 print_named_value(f
, "INV_L2", (event_dw
>> 20) & 0x1, 1);
319 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
320 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 16);
324 case PKT3_EVENT_WRITE_EOP
: {
325 uint32_t event_dw
= ac_ib_get(ib
);
326 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
327 S_028A90_EVENT_TYPE(~0));
328 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
329 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
330 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
331 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
332 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
333 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
334 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
335 uint32_t addr_hi_dw
= ac_ib_get(ib
);
336 print_named_value(f
, "ADDRESS_HI", addr_hi_dw
, 16);
337 print_named_value(f
, "DST_SEL", (addr_hi_dw
>> 16) & 0x3, 2);
338 print_named_value(f
, "INT_SEL", (addr_hi_dw
>> 24) & 0x7, 3);
339 print_named_value(f
, "DATA_SEL", addr_hi_dw
>> 29, 3);
340 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
341 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
344 case PKT3_RELEASE_MEM
: {
345 uint32_t event_dw
= ac_ib_get(ib
);
346 if (ib
->chip_class
>= GFX10
) {
347 ac_dump_reg(f
, ib
->chip_class
, R_490_RELEASE_MEM_OP
, event_dw
, ~0u);
349 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
350 S_028A90_EVENT_TYPE(~0));
351 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
352 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
353 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
354 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
355 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
356 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
357 print_named_value(f
, "TC_NC_ACTION_ENA", (event_dw
>> 19) & 0x1, 1);
358 print_named_value(f
, "TC_WC_ACTION_ENA", (event_dw
>> 20) & 0x1, 1);
359 print_named_value(f
, "TC_MD_ACTION_ENA", (event_dw
>> 21) & 0x1, 1);
361 uint32_t sel_dw
= ac_ib_get(ib
);
362 print_named_value(f
, "DST_SEL", (sel_dw
>> 16) & 0x3, 2);
363 print_named_value(f
, "INT_SEL", (sel_dw
>> 24) & 0x7, 3);
364 print_named_value(f
, "DATA_SEL", sel_dw
>> 29, 3);
365 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
366 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
367 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
368 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
369 print_named_value(f
, "CTXID", ac_ib_get(ib
), 32);
372 case PKT3_WAIT_REG_MEM
:
373 print_named_value(f
, "OP", ac_ib_get(ib
), 32);
374 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
375 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
376 print_named_value(f
, "REF", ac_ib_get(ib
), 32);
377 print_named_value(f
, "MASK", ac_ib_get(ib
), 32);
378 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
380 case PKT3_DRAW_INDEX_AUTO
:
381 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
382 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
384 case PKT3_DRAW_INDEX_2
:
385 ac_dump_reg(f
, ib
->chip_class
, R_028A78_VGT_DMA_MAX_SIZE
, ac_ib_get(ib
), ~0);
386 ac_dump_reg(f
, ib
->chip_class
, R_0287E8_VGT_DMA_BASE
, ac_ib_get(ib
), ~0);
387 ac_dump_reg(f
, ib
->chip_class
, R_0287E4_VGT_DMA_BASE_HI
, ac_ib_get(ib
), ~0);
388 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
389 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
391 case PKT3_INDEX_TYPE
:
392 ac_dump_reg(f
, ib
->chip_class
, R_028A7C_VGT_DMA_INDEX_TYPE
, ac_ib_get(ib
), ~0);
394 case PKT3_NUM_INSTANCES
:
395 ac_dump_reg(f
, ib
->chip_class
, R_030934_VGT_NUM_INSTANCES
, ac_ib_get(ib
), ~0);
397 case PKT3_WRITE_DATA
:
398 ac_dump_reg(f
, ib
->chip_class
, R_370_CONTROL
, ac_ib_get(ib
), ~0);
399 ac_dump_reg(f
, ib
->chip_class
, R_371_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
400 ac_dump_reg(f
, ib
->chip_class
, R_372_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
401 /* The payload is written automatically */
404 ac_dump_reg(f
, ib
->chip_class
, R_410_CP_DMA_WORD0
, ac_ib_get(ib
), ~0);
405 ac_dump_reg(f
, ib
->chip_class
, R_411_CP_DMA_WORD1
, ac_ib_get(ib
), ~0);
406 ac_dump_reg(f
, ib
->chip_class
, R_412_CP_DMA_WORD2
, ac_ib_get(ib
), ~0);
407 ac_dump_reg(f
, ib
->chip_class
, R_413_CP_DMA_WORD3
, ac_ib_get(ib
), ~0);
408 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
411 ac_dump_reg(f
, ib
->chip_class
, R_500_DMA_DATA_WORD0
, ac_ib_get(ib
), ~0);
412 ac_dump_reg(f
, ib
->chip_class
, R_501_SRC_ADDR_LO
, ac_ib_get(ib
), ~0);
413 ac_dump_reg(f
, ib
->chip_class
, R_502_SRC_ADDR_HI
, ac_ib_get(ib
), ~0);
414 ac_dump_reg(f
, ib
->chip_class
, R_503_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
415 ac_dump_reg(f
, ib
->chip_class
, R_504_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
416 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
418 case PKT3_INDIRECT_BUFFER_SI
:
419 case PKT3_INDIRECT_BUFFER_CONST
:
420 case PKT3_INDIRECT_BUFFER_CIK
: {
421 uint32_t base_lo_dw
= ac_ib_get(ib
);
422 ac_dump_reg(f
, ib
->chip_class
, R_3F0_IB_BASE_LO
, base_lo_dw
, ~0);
423 uint32_t base_hi_dw
= ac_ib_get(ib
);
424 ac_dump_reg(f
, ib
->chip_class
, R_3F1_IB_BASE_HI
, base_hi_dw
, ~0);
425 uint32_t control_dw
= ac_ib_get(ib
);
426 ac_dump_reg(f
, ib
->chip_class
, R_3F2_IB_CONTROL
, control_dw
, ~0);
428 if (!ib
->addr_callback
)
431 uint64_t addr
= ((uint64_t)base_hi_dw
<< 32) | base_lo_dw
;
432 void *data
= ib
->addr_callback(ib
->addr_callback_data
, addr
);
436 if (G_3F2_CHAIN(control_dw
)) {
438 ib
->num_dw
= G_3F2_IB_SIZE(control_dw
);
443 struct ac_ib_parser ib_recurse
;
444 memcpy(&ib_recurse
, ib
, sizeof(ib_recurse
));
445 ib_recurse
.ib
= data
;
446 ib_recurse
.num_dw
= G_3F2_IB_SIZE(control_dw
);
447 ib_recurse
.cur_dw
= 0;
448 if(ib_recurse
.trace_id_count
) {
449 if (*current_trace_id
== *ib
->trace_ids
) {
450 ++ib_recurse
.trace_ids
;
451 --ib_recurse
.trace_id_count
;
453 ib_recurse
.trace_id_count
= 0;
457 fprintf(f
, "\n\035>------------------ nested begin ------------------\n");
458 ac_do_parse_ib(f
, &ib_recurse
);
459 fprintf(f
, "\n\035<------------------- nested end -------------------\n");
462 case PKT3_CLEAR_STATE
:
463 case PKT3_INCREMENT_DE_COUNTER
:
464 case PKT3_PFP_SYNC_ME
:
467 if (header
== PKT3_NOP_PAD
) {
468 count
= -1; /* One dword NOP. */
469 } else if (count
== 0 && ib
->cur_dw
< ib
->num_dw
&&
470 AC_IS_TRACE_POINT(ib
->ib
[ib
->cur_dw
])) {
471 unsigned packet_id
= AC_GET_TRACE_POINT_ID(ib
->ib
[ib
->cur_dw
]);
473 print_spaces(f
, INDENT_PKT
);
474 fprintf(f
, COLOR_RED
"Trace point ID: %u\n", packet_id
);
476 if (!ib
->trace_id_count
)
477 break; /* tracing was disabled */
479 *current_trace_id
= packet_id
;
481 print_spaces(f
, INDENT_PKT
);
482 if (packet_id
< *ib
->trace_ids
)
484 "This trace point was reached by the CP."
486 else if (packet_id
== *ib
->trace_ids
)
488 "!!!!! This is the last trace point that "
489 "was reached by the CP !!!!!"
491 else if (packet_id
+1 == *ib
->trace_ids
)
493 "!!!!! This is the first trace point that "
494 "was NOT been reached by the CP !!!!!"
498 "!!!!! This trace point was NOT reached "
506 /* print additional dwords */
507 while (ib
->cur_dw
<= first_dw
+ count
)
510 if (ib
->cur_dw
> first_dw
+ count
+ 1)
511 fprintf(f
, COLOR_RED
"\n!!!!! count in header too low !!!!!"
516 * Parse and print an IB into a file.
518 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
)
520 int current_trace_id
= -1;
522 while (ib
->cur_dw
< ib
->num_dw
) {
523 uint32_t header
= ac_ib_get(ib
);
524 unsigned type
= PKT_TYPE_G(header
);
528 ac_parse_packet3(f
, header
, ib
, ¤t_trace_id
);
532 if (header
== 0x80000000) {
533 fprintf(f
, COLOR_GREEN
"NOP (type 2)" COLOR_RESET
"\n");
538 fprintf(f
, "Unknown packet type %i\n", type
);
544 static void format_ib_output(FILE *f
, char *out
)
551 if (out
[0] == '\n' && out
[1] == '\035')
553 if (out
[0] == '\035') {
561 unsigned indent
= 4 * depth
;
566 print_spaces(f
, indent
);
568 char *end
= strchrnul(out
, '\n');
569 fwrite(out
, end
- out
, 1, f
);
570 fputc('\n', f
); /* always end with a new line */
582 * Parse and print an IB into a file.
586 * \param num_dw size of the IB
587 * \param chip_class chip class
588 * \param trace_ids the last trace IDs that are known to have been reached
589 * and executed by the CP, typically read from a buffer
590 * \param trace_id_count The number of entries in the trace_ids array.
591 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
593 * \param addr_callback_data user data for addr_callback
595 void ac_parse_ib_chunk(FILE *f
, uint32_t *ib_ptr
, int num_dw
, const int *trace_ids
,
596 unsigned trace_id_count
, enum chip_class chip_class
,
597 ac_debug_addr_callback addr_callback
, void *addr_callback_data
)
599 struct ac_ib_parser ib
= {};
602 ib
.trace_ids
= trace_ids
;
603 ib
.trace_id_count
= trace_id_count
;
604 ib
.chip_class
= chip_class
;
605 ib
.addr_callback
= addr_callback
;
606 ib
.addr_callback_data
= addr_callback_data
;
610 FILE *memf
= open_memstream(&out
, &outsize
);
612 ac_do_parse_ib(memf
, &ib
);
616 format_ib_output(f
, out
);
620 if (ib
.cur_dw
> ib
.num_dw
) {
621 printf("\nPacket ends after the end of IB.\n");
627 * Parse and print an IB into a file.
631 * \param num_dw size of the IB
632 * \param chip_class chip class
633 * \param trace_ids the last trace IDs that are known to have been reached
634 * and executed by the CP, typically read from a buffer
635 * \param trace_id_count The number of entries in the trace_ids array.
636 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
638 * \param addr_callback_data user data for addr_callback
640 void ac_parse_ib(FILE *f
, uint32_t *ib
, int num_dw
, const int *trace_ids
,
641 unsigned trace_id_count
, const char *name
,
642 enum chip_class chip_class
, ac_debug_addr_callback addr_callback
,
643 void *addr_callback_data
)
645 fprintf(f
, "------------------ %s begin ------------------\n", name
);
647 ac_parse_ib_chunk(f
, ib
, num_dw
, trace_ids
, trace_id_count
,
648 chip_class
, addr_callback
, addr_callback_data
);
650 fprintf(f
, "------------------- %s end -------------------\n\n", name
);
654 * Parse dmesg and return TRUE if a VM fault has been detected.
656 * \param chip_class chip class
657 * \param old_dmesg_timestamp previous dmesg timestamp parsed at init time
658 * \param out_addr detected VM fault addr
660 bool ac_vm_fault_occured(enum chip_class chip_class
,
661 uint64_t *old_dmesg_timestamp
, uint64_t *out_addr
)
666 uint64_t dmesg_timestamp
= 0;
669 FILE *p
= popen("dmesg", "r");
673 while (fgets(line
, sizeof(line
), p
)) {
676 if (!line
[0] || line
[0] == '\n')
679 /* Get the timestamp. */
680 if (sscanf(line
, "[%u.%u]", &sec
, &usec
) != 2) {
681 static bool hit
= false;
683 fprintf(stderr
, "%s: failed to parse line '%s'\n",
689 dmesg_timestamp
= sec
* 1000000ull + usec
;
691 /* If just updating the timestamp. */
695 /* Process messages only if the timestamp is newer. */
696 if (dmesg_timestamp
<= *old_dmesg_timestamp
)
699 /* Only process the first VM fault. */
703 /* Remove trailing \n */
705 if (len
&& line
[len
-1] == '\n')
708 /* Get the message part. */
709 msg
= strchr(line
, ']');
714 const char *header_line
, *addr_line_prefix
, *addr_line_format
;
716 if (chip_class
>= GFX9
) {
718 * ..: [gfxhub] VMC page fault (src_id:0 ring:158 vm_id:2 pas_id:0)
719 * ..: at page 0x0000000219f8f000 from 27
720 * ..: VM_L2_PROTECTION_FAULT_STATUS:0x0020113C
722 header_line
= "VMC page fault";
723 addr_line_prefix
= " at page";
724 addr_line_format
= "%"PRIx64
;
726 header_line
= "GPU fault detected:";
727 addr_line_prefix
= "VM_CONTEXT1_PROTECTION_FAULT_ADDR";
728 addr_line_format
= "%"PRIX64
;
733 if (strstr(msg
, header_line
))
737 msg
= strstr(msg
, addr_line_prefix
);
739 msg
= strstr(msg
, "0x");
742 if (sscanf(msg
, addr_line_format
, out_addr
) == 1)
754 if (dmesg_timestamp
> *old_dmesg_timestamp
)
755 *old_dmesg_timestamp
= dmesg_timestamp
;
760 static int compare_wave(const void *p1
, const void *p2
)
762 struct ac_wave_info
*w1
= (struct ac_wave_info
*)p1
;
763 struct ac_wave_info
*w2
= (struct ac_wave_info
*)p2
;
765 /* Sort waves according to PC and then SE, SH, CU, etc. */
782 if (w1
->simd
< w2
->simd
)
784 if (w1
->simd
> w2
->simd
)
786 if (w1
->wave
< w2
->wave
)
788 if (w1
->wave
> w2
->wave
)
794 /* Return wave information. "waves" should be a large enough array. */
795 unsigned ac_get_wave_info(enum chip_class chip_class
,
796 struct ac_wave_info waves
[AC_MAX_WAVES_PER_CHIP
])
798 char line
[2000], cmd
[128];
799 unsigned num_waves
= 0;
801 sprintf(cmd
, "umr -O halt_waves -wa %s", chip_class
>= GFX10
? "gfx_0.0.0" : "gfx");
803 FILE *p
= popen(cmd
, "r");
807 if (!fgets(line
, sizeof(line
), p
) ||
808 strncmp(line
, "SE", 2) != 0) {
813 while (fgets(line
, sizeof(line
), p
)) {
814 struct ac_wave_info
*w
;
815 uint32_t pc_hi
, pc_lo
, exec_hi
, exec_lo
;
817 assert(num_waves
< AC_MAX_WAVES_PER_CHIP
);
818 w
= &waves
[num_waves
];
820 if (sscanf(line
, "%u %u %u %u %u %x %x %x %x %x %x %x",
821 &w
->se
, &w
->sh
, &w
->cu
, &w
->simd
, &w
->wave
,
822 &w
->status
, &pc_hi
, &pc_lo
, &w
->inst_dw0
,
823 &w
->inst_dw1
, &exec_hi
, &exec_lo
) == 12) {
824 w
->pc
= ((uint64_t)pc_hi
<< 32) | pc_lo
;
825 w
->exec
= ((uint64_t)exec_hi
<< 32) | exec_lo
;
831 qsort(waves
, num_waves
, sizeof(struct ac_wave_info
), compare_wave
);