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.
24 * Marek Olšák <maraeo@gmail.com>
41 #include "sid_tables.h"
42 #include "util/u_math.h"
43 #include "util/u_memory.h"
44 #include "util/u_string.h"
48 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
49 * read them, or use "aha -b -f file" to convert them to html.
51 #define COLOR_RESET "\033[0m"
52 #define COLOR_RED "\033[31m"
53 #define COLOR_GREEN "\033[1;32m"
54 #define COLOR_YELLOW "\033[1;33m"
55 #define COLOR_CYAN "\033[1;36m"
64 unsigned trace_id_count
;
65 enum chip_class chip_class
;
66 ac_debug_addr_callback addr_callback
;
67 void *addr_callback_data
;
72 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
);
74 static void print_spaces(FILE *f
, unsigned num
)
76 fprintf(f
, "%*s", num
, "");
79 static void print_value(FILE *file
, uint32_t value
, int bits
)
81 /* Guess if it's int or float */
82 if (value
<= (1 << 15)) {
84 fprintf(file
, "%u\n", value
);
86 fprintf(file
, "%u (0x%0*x)\n", value
, bits
/ 4, value
);
90 if (fabs(f
) < 100000 && f
*10 == floor(f
*10))
91 fprintf(file
, "%.1ff (0x%0*x)\n", f
, bits
/ 4, value
);
93 /* Don't print more leading zeros than there are bits. */
94 fprintf(file
, "0x%0*x\n", bits
/ 4, value
);
98 static void print_named_value(FILE *file
, const char *name
, uint32_t value
,
101 print_spaces(file
, INDENT_PKT
);
102 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ", name
);
103 print_value(file
, value
, bits
);
106 static const struct si_reg
*find_register(const struct si_reg
*table
,
110 for (unsigned i
= 0; i
< table_size
; i
++) {
111 const struct si_reg
*reg
= &table
[i
];
113 if (reg
->offset
== offset
)
120 void ac_dump_reg(FILE *file
, enum chip_class chip_class
, unsigned offset
,
121 uint32_t value
, uint32_t field_mask
)
123 const struct si_reg
*reg
= NULL
;
125 if (chip_class
>= GFX9
)
126 reg
= find_register(gfx9d_reg_table
, ARRAY_SIZE(gfx9d_reg_table
), offset
);
128 reg
= find_register(sid_reg_table
, ARRAY_SIZE(sid_reg_table
), offset
);
131 const char *reg_name
= sid_strings
+ reg
->name_offset
;
132 bool first_field
= true;
134 print_spaces(file
, INDENT_PKT
);
135 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ",
138 if (!reg
->num_fields
) {
139 print_value(file
, value
, 32);
143 for (unsigned f
= 0; f
< reg
->num_fields
; f
++) {
144 const struct si_field
*field
= sid_fields_table
+ reg
->fields_offset
+ f
;
145 const int *values_offsets
= sid_strings_offsets
+ field
->values_offset
;
146 uint32_t val
= (value
& field
->mask
) >>
147 (ffs(field
->mask
) - 1);
149 if (!(field
->mask
& field_mask
))
152 /* Indent the field. */
155 INDENT_PKT
+ strlen(reg_name
) + 4);
157 /* Print the field. */
158 fprintf(file
, "%s = ", sid_strings
+ field
->name_offset
);
160 if (val
< field
->num_values
&& values_offsets
[val
] >= 0)
161 fprintf(file
, "%s\n", sid_strings
+ values_offsets
[val
]);
163 print_value(file
, val
,
164 util_bitcount(field
->mask
));
171 print_spaces(file
, INDENT_PKT
);
172 fprintf(file
, COLOR_YELLOW
"0x%05x" COLOR_RESET
" <- 0x%08x\n", offset
, value
);
175 static uint32_t ac_ib_get(struct ac_ib_parser
*ib
)
179 if (ib
->cur_dw
< ib
->num_dw
) {
180 v
= ib
->ib
[ib
->cur_dw
];
182 /* Help figure out where garbage data is written to IBs.
184 * Arguably we should do this already when the IBs are written,
185 * see RADEON_VALGRIND. The problem is that client-requests to
186 * Valgrind have an overhead even when Valgrind isn't running,
187 * and radeon_emit is performance sensitive...
189 if (VALGRIND_CHECK_VALUE_IS_DEFINED(v
))
190 fprintf(ib
->f
, COLOR_RED
"Valgrind: The next DWORD is garbage"
193 fprintf(ib
->f
, "\n\035#%08x ", v
);
195 fprintf(ib
->f
, "\n\035#???????? ");
202 static void ac_parse_set_reg_packet(FILE *f
, unsigned count
, unsigned reg_offset
,
203 struct ac_ib_parser
*ib
)
205 unsigned reg_dw
= ac_ib_get(ib
);
206 unsigned reg
= ((reg_dw
& 0xFFFF) << 2) + reg_offset
;
207 unsigned index
= reg_dw
>> 28;
211 print_spaces(f
, INDENT_PKT
);
212 fprintf(f
, "INDEX = %u\n", index
);
215 for (i
= 0; i
< count
; i
++)
216 ac_dump_reg(f
, ib
->chip_class
, reg
+ i
*4, ac_ib_get(ib
), ~0);
219 static void ac_parse_packet3(FILE *f
, uint32_t header
, struct ac_ib_parser
*ib
,
220 int *current_trace_id
)
222 unsigned first_dw
= ib
->cur_dw
;
223 int count
= PKT_COUNT_G(header
);
224 unsigned op
= PKT3_IT_OPCODE_G(header
);
225 const char *predicate
= PKT3_PREDICATE(header
) ? "(predicate)" : "";
228 /* Print the name first. */
229 for (i
= 0; i
< ARRAY_SIZE(packet3_table
); i
++)
230 if (packet3_table
[i
].op
== op
)
233 if (i
< ARRAY_SIZE(packet3_table
)) {
234 const char *name
= sid_strings
+ packet3_table
[i
].name_offset
;
236 if (op
== PKT3_SET_CONTEXT_REG
||
237 op
== PKT3_SET_CONFIG_REG
||
238 op
== PKT3_SET_UCONFIG_REG
||
239 op
== PKT3_SET_SH_REG
)
240 fprintf(f
, COLOR_CYAN
"%s%s" COLOR_CYAN
":\n",
243 fprintf(f
, COLOR_GREEN
"%s%s" COLOR_RESET
":\n",
246 fprintf(f
, COLOR_RED
"PKT3_UNKNOWN 0x%x%s" COLOR_RESET
":\n",
249 /* Print the contents. */
251 case PKT3_SET_CONTEXT_REG
:
252 ac_parse_set_reg_packet(f
, count
, SI_CONTEXT_REG_OFFSET
, ib
);
254 case PKT3_SET_CONFIG_REG
:
255 ac_parse_set_reg_packet(f
, count
, SI_CONFIG_REG_OFFSET
, ib
);
257 case PKT3_SET_UCONFIG_REG
:
258 ac_parse_set_reg_packet(f
, count
, CIK_UCONFIG_REG_OFFSET
, ib
);
260 case PKT3_SET_SH_REG
:
261 ac_parse_set_reg_packet(f
, count
, SI_SH_REG_OFFSET
, ib
);
263 case PKT3_ACQUIRE_MEM
:
264 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
265 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
266 ac_dump_reg(f
, ib
->chip_class
, R_030230_CP_COHER_SIZE_HI
, ac_ib_get(ib
), ~0);
267 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
268 ac_dump_reg(f
, ib
->chip_class
, R_0301E4_CP_COHER_BASE_HI
, ac_ib_get(ib
), ~0);
269 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
271 case PKT3_SURFACE_SYNC
:
272 if (ib
->chip_class
>= CIK
) {
273 ac_dump_reg(f
, ib
->chip_class
, R_0301F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
274 ac_dump_reg(f
, ib
->chip_class
, R_0301F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
275 ac_dump_reg(f
, ib
->chip_class
, R_0301F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
277 ac_dump_reg(f
, ib
->chip_class
, R_0085F0_CP_COHER_CNTL
, ac_ib_get(ib
), ~0);
278 ac_dump_reg(f
, ib
->chip_class
, R_0085F4_CP_COHER_SIZE
, ac_ib_get(ib
), ~0);
279 ac_dump_reg(f
, ib
->chip_class
, R_0085F8_CP_COHER_BASE
, ac_ib_get(ib
), ~0);
281 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
283 case PKT3_EVENT_WRITE
: {
284 uint32_t event_dw
= ac_ib_get(ib
);
285 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
286 S_028A90_EVENT_TYPE(~0));
287 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
288 print_named_value(f
, "INV_L2", (event_dw
>> 20) & 0x1, 1);
290 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
291 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 16);
295 case PKT3_EVENT_WRITE_EOP
: {
296 uint32_t event_dw
= ac_ib_get(ib
);
297 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
298 S_028A90_EVENT_TYPE(~0));
299 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
300 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
301 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
302 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
303 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
304 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
305 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
306 uint32_t addr_hi_dw
= ac_ib_get(ib
);
307 print_named_value(f
, "ADDRESS_HI", addr_hi_dw
, 16);
308 print_named_value(f
, "DST_SEL", (addr_hi_dw
>> 16) & 0x3, 2);
309 print_named_value(f
, "INT_SEL", (addr_hi_dw
>> 24) & 0x7, 3);
310 print_named_value(f
, "DATA_SEL", addr_hi_dw
>> 29, 3);
311 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
312 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
315 case PKT3_RELEASE_MEM
: {
316 uint32_t event_dw
= ac_ib_get(ib
);
317 ac_dump_reg(f
, ib
->chip_class
, R_028A90_VGT_EVENT_INITIATOR
, event_dw
,
318 S_028A90_EVENT_TYPE(~0));
319 print_named_value(f
, "EVENT_INDEX", (event_dw
>> 8) & 0xf, 4);
320 print_named_value(f
, "TCL1_VOL_ACTION_ENA", (event_dw
>> 12) & 0x1, 1);
321 print_named_value(f
, "TC_VOL_ACTION_ENA", (event_dw
>> 13) & 0x1, 1);
322 print_named_value(f
, "TC_WB_ACTION_ENA", (event_dw
>> 15) & 0x1, 1);
323 print_named_value(f
, "TCL1_ACTION_ENA", (event_dw
>> 16) & 0x1, 1);
324 print_named_value(f
, "TC_ACTION_ENA", (event_dw
>> 17) & 0x1, 1);
325 print_named_value(f
, "TC_NC_ACTION_ENA", (event_dw
>> 19) & 0x1, 1);
326 print_named_value(f
, "TC_WC_ACTION_ENA", (event_dw
>> 20) & 0x1, 1);
327 print_named_value(f
, "TC_MD_ACTION_ENA", (event_dw
>> 21) & 0x1, 1);
328 uint32_t sel_dw
= ac_ib_get(ib
);
329 print_named_value(f
, "DST_SEL", (sel_dw
>> 16) & 0x3, 2);
330 print_named_value(f
, "INT_SEL", (sel_dw
>> 24) & 0x7, 3);
331 print_named_value(f
, "DATA_SEL", sel_dw
>> 29, 3);
332 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
333 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
334 print_named_value(f
, "DATA_LO", ac_ib_get(ib
), 32);
335 print_named_value(f
, "DATA_HI", ac_ib_get(ib
), 32);
336 print_named_value(f
, "CTXID", ac_ib_get(ib
), 32);
339 case PKT3_WAIT_REG_MEM
:
340 print_named_value(f
, "OP", ac_ib_get(ib
), 32);
341 print_named_value(f
, "ADDRESS_LO", ac_ib_get(ib
), 32);
342 print_named_value(f
, "ADDRESS_HI", ac_ib_get(ib
), 32);
343 print_named_value(f
, "REF", ac_ib_get(ib
), 32);
344 print_named_value(f
, "MASK", ac_ib_get(ib
), 32);
345 print_named_value(f
, "POLL_INTERVAL", ac_ib_get(ib
), 16);
347 case PKT3_DRAW_INDEX_AUTO
:
348 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
349 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
351 case PKT3_DRAW_INDEX_2
:
352 ac_dump_reg(f
, ib
->chip_class
, R_028A78_VGT_DMA_MAX_SIZE
, ac_ib_get(ib
), ~0);
353 ac_dump_reg(f
, ib
->chip_class
, R_0287E8_VGT_DMA_BASE
, ac_ib_get(ib
), ~0);
354 ac_dump_reg(f
, ib
->chip_class
, R_0287E4_VGT_DMA_BASE_HI
, ac_ib_get(ib
), ~0);
355 ac_dump_reg(f
, ib
->chip_class
, R_030930_VGT_NUM_INDICES
, ac_ib_get(ib
), ~0);
356 ac_dump_reg(f
, ib
->chip_class
, R_0287F0_VGT_DRAW_INITIATOR
, ac_ib_get(ib
), ~0);
358 case PKT3_INDEX_TYPE
:
359 ac_dump_reg(f
, ib
->chip_class
, R_028A7C_VGT_DMA_INDEX_TYPE
, ac_ib_get(ib
), ~0);
361 case PKT3_NUM_INSTANCES
:
362 ac_dump_reg(f
, ib
->chip_class
, R_030934_VGT_NUM_INSTANCES
, ac_ib_get(ib
), ~0);
364 case PKT3_WRITE_DATA
:
365 ac_dump_reg(f
, ib
->chip_class
, R_370_CONTROL
, ac_ib_get(ib
), ~0);
366 ac_dump_reg(f
, ib
->chip_class
, R_371_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
367 ac_dump_reg(f
, ib
->chip_class
, R_372_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
368 /* The payload is written automatically */
371 ac_dump_reg(f
, ib
->chip_class
, R_410_CP_DMA_WORD0
, ac_ib_get(ib
), ~0);
372 ac_dump_reg(f
, ib
->chip_class
, R_411_CP_DMA_WORD1
, ac_ib_get(ib
), ~0);
373 ac_dump_reg(f
, ib
->chip_class
, R_412_CP_DMA_WORD2
, ac_ib_get(ib
), ~0);
374 ac_dump_reg(f
, ib
->chip_class
, R_413_CP_DMA_WORD3
, ac_ib_get(ib
), ~0);
375 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
378 ac_dump_reg(f
, ib
->chip_class
, R_500_DMA_DATA_WORD0
, ac_ib_get(ib
), ~0);
379 ac_dump_reg(f
, ib
->chip_class
, R_501_SRC_ADDR_LO
, ac_ib_get(ib
), ~0);
380 ac_dump_reg(f
, ib
->chip_class
, R_502_SRC_ADDR_HI
, ac_ib_get(ib
), ~0);
381 ac_dump_reg(f
, ib
->chip_class
, R_503_DST_ADDR_LO
, ac_ib_get(ib
), ~0);
382 ac_dump_reg(f
, ib
->chip_class
, R_504_DST_ADDR_HI
, ac_ib_get(ib
), ~0);
383 ac_dump_reg(f
, ib
->chip_class
, R_414_COMMAND
, ac_ib_get(ib
), ~0);
385 case PKT3_INDIRECT_BUFFER_SI
:
386 case PKT3_INDIRECT_BUFFER_CONST
:
387 case PKT3_INDIRECT_BUFFER_CIK
: {
388 uint32_t base_lo_dw
= ac_ib_get(ib
);
389 ac_dump_reg(f
, ib
->chip_class
, R_3F0_IB_BASE_LO
, base_lo_dw
, ~0);
390 uint32_t base_hi_dw
= ac_ib_get(ib
);
391 ac_dump_reg(f
, ib
->chip_class
, R_3F1_IB_BASE_HI
, base_hi_dw
, ~0);
392 uint32_t control_dw
= ac_ib_get(ib
);
393 ac_dump_reg(f
, ib
->chip_class
, R_3F2_CONTROL
, control_dw
, ~0);
395 if (!ib
->addr_callback
)
398 uint64_t addr
= ((uint64_t)base_hi_dw
<< 32) | base_lo_dw
;
399 void *data
= ib
->addr_callback(ib
->addr_callback_data
, addr
);
403 if (G_3F2_CHAIN(control_dw
)) {
405 ib
->num_dw
= G_3F2_IB_SIZE(control_dw
);
410 struct ac_ib_parser ib_recurse
;
411 memcpy(&ib_recurse
, ib
, sizeof(ib_recurse
));
412 ib_recurse
.ib
= data
;
413 ib_recurse
.num_dw
= G_3F2_IB_SIZE(control_dw
);
414 ib_recurse
.cur_dw
= 0;
415 if(ib_recurse
.trace_id_count
) {
416 if (*current_trace_id
== *ib
->trace_ids
) {
417 ++ib_recurse
.trace_ids
;
418 --ib_recurse
.trace_id_count
;
420 ib_recurse
.trace_id_count
= 0;
424 fprintf(f
, "\n\035>------------------ nested begin ------------------\n");
425 ac_do_parse_ib(f
, &ib_recurse
);
426 fprintf(f
, "\n\035<------------------- nested end -------------------\n");
429 case PKT3_CLEAR_STATE
:
430 case PKT3_INCREMENT_DE_COUNTER
:
431 case PKT3_PFP_SYNC_ME
:
434 if (header
== 0xffff1000) {
435 count
= -1; /* One dword NOP. */
436 } else if (count
== 0 && ib
->cur_dw
< ib
->num_dw
&&
437 AC_IS_TRACE_POINT(ib
->ib
[ib
->cur_dw
])) {
438 unsigned packet_id
= AC_GET_TRACE_POINT_ID(ib
->ib
[ib
->cur_dw
]);
440 print_spaces(f
, INDENT_PKT
);
441 fprintf(f
, COLOR_RED
"Trace point ID: %u\n", packet_id
);
443 if (!ib
->trace_id_count
)
444 break; /* tracing was disabled */
446 *current_trace_id
= packet_id
;
448 print_spaces(f
, INDENT_PKT
);
449 if (packet_id
< *ib
->trace_ids
)
451 "This trace point was reached by the CP."
453 else if (packet_id
== *ib
->trace_ids
)
455 "!!!!! This is the last trace point that "
456 "was reached by the CP !!!!!"
458 else if (packet_id
+1 == *ib
->trace_ids
)
460 "!!!!! This is the first trace point that "
461 "was NOT been reached by the CP !!!!!"
465 "!!!!! This trace point was NOT reached "
473 /* print additional dwords */
474 while (ib
->cur_dw
<= first_dw
+ count
)
477 if (ib
->cur_dw
> first_dw
+ count
+ 1)
478 fprintf(f
, COLOR_RED
"\n!!!!! count in header too low !!!!!"
483 * Parse and print an IB into a file.
485 static void ac_do_parse_ib(FILE *f
, struct ac_ib_parser
*ib
)
487 int current_trace_id
= -1;
489 while (ib
->cur_dw
< ib
->num_dw
) {
490 uint32_t header
= ac_ib_get(ib
);
491 unsigned type
= PKT_TYPE_G(header
);
495 ac_parse_packet3(f
, header
, ib
, ¤t_trace_id
);
499 if (header
== 0x80000000) {
500 fprintf(f
, COLOR_GREEN
"NOP (type 2)" COLOR_RESET
"\n");
505 fprintf(f
, "Unknown packet type %i\n", type
);
511 static void format_ib_output(FILE *f
, char *out
)
518 if (out
[0] == '\n' && out
[1] == '\035')
520 if (out
[0] == '\035') {
528 unsigned indent
= 4 * depth
;
533 print_spaces(f
, indent
);
535 char *end
= util_strchrnul(out
, '\n');
536 fwrite(out
, end
- out
, 1, f
);
537 fputc('\n', f
); /* always end with a new line */
549 * Parse and print an IB into a file.
553 * \param num_dw size of the IB
554 * \param chip_class chip class
555 * \param trace_ids the last trace IDs that are known to have been reached
556 * and executed by the CP, typically read from a buffer
557 * \param trace_id_count The number of entries in the trace_ids array.
558 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
560 * \param addr_callback_data user data for addr_callback
562 void ac_parse_ib_chunk(FILE *f
, uint32_t *ib_ptr
, int num_dw
, const int *trace_ids
,
563 unsigned trace_id_count
, enum chip_class chip_class
,
564 ac_debug_addr_callback addr_callback
, void *addr_callback_data
)
566 struct ac_ib_parser ib
= {};
569 ib
.trace_ids
= trace_ids
;
570 ib
.trace_id_count
= trace_id_count
;
571 ib
.chip_class
= chip_class
;
572 ib
.addr_callback
= addr_callback
;
573 ib
.addr_callback_data
= addr_callback_data
;
577 FILE *memf
= open_memstream(&out
, &outsize
);
579 ac_do_parse_ib(memf
, &ib
);
583 format_ib_output(f
, out
);
587 if (ib
.cur_dw
> ib
.num_dw
) {
588 printf("\nPacket ends after the end of IB.\n");
594 * Parse and print an IB into a file.
598 * \param num_dw size of the IB
599 * \param chip_class chip class
600 * \param trace_ids the last trace IDs that are known to have been reached
601 * and executed by the CP, typically read from a buffer
602 * \param trace_id_count The number of entries in the trace_ids array.
603 * \param addr_callback Get a mapped pointer of the IB at a given address. Can
605 * \param addr_callback_data user data for addr_callback
607 void ac_parse_ib(FILE *f
, uint32_t *ib
, int num_dw
, const int *trace_ids
,
608 unsigned trace_id_count
, const char *name
,
609 enum chip_class chip_class
, ac_debug_addr_callback addr_callback
,
610 void *addr_callback_data
)
612 fprintf(f
, "------------------ %s begin ------------------\n", name
);
614 ac_parse_ib_chunk(f
, ib
, num_dw
, trace_ids
, trace_id_count
,
615 chip_class
, addr_callback
, addr_callback_data
);
617 fprintf(f
, "------------------- %s end -------------------\n\n", name
);
621 * Parse dmesg and return TRUE if a VM fault has been detected.
623 * \param chip_class chip class
624 * \param old_dmesg_timestamp previous dmesg timestamp parsed at init time
625 * \param out_addr detected VM fault addr
627 bool ac_vm_fault_occured(enum chip_class chip_class
,
628 uint64_t *old_dmesg_timestamp
, uint64_t *out_addr
)
633 uint64_t dmesg_timestamp
= 0;
636 FILE *p
= popen("dmesg", "r");
640 while (fgets(line
, sizeof(line
), p
)) {
643 if (!line
[0] || line
[0] == '\n')
646 /* Get the timestamp. */
647 if (sscanf(line
, "[%u.%u]", &sec
, &usec
) != 2) {
648 static bool hit
= false;
650 fprintf(stderr
, "%s: failed to parse line '%s'\n",
656 dmesg_timestamp
= sec
* 1000000ull + usec
;
658 /* If just updating the timestamp. */
662 /* Process messages only if the timestamp is newer. */
663 if (dmesg_timestamp
<= *old_dmesg_timestamp
)
666 /* Only process the first VM fault. */
670 /* Remove trailing \n */
672 if (len
&& line
[len
-1] == '\n')
675 /* Get the message part. */
676 msg
= strchr(line
, ']');
681 const char *header_line
, *addr_line_prefix
, *addr_line_format
;
683 if (chip_class
>= GFX9
) {
685 * ..: [gfxhub] VMC page fault (src_id:0 ring:158 vm_id:2 pas_id:0)
686 * ..: at page 0x0000000219f8f000 from 27
687 * ..: VM_L2_PROTECTION_FAULT_STATUS:0x0020113C
689 header_line
= "VMC page fault";
690 addr_line_prefix
= " at page";
691 addr_line_format
= "%"PRIx64
;
693 header_line
= "GPU fault detected:";
694 addr_line_prefix
= "VM_CONTEXT1_PROTECTION_FAULT_ADDR";
695 addr_line_format
= "%"PRIX64
;
700 if (strstr(msg
, header_line
))
704 msg
= strstr(msg
, addr_line_prefix
);
706 msg
= strstr(msg
, "0x");
709 if (sscanf(msg
, addr_line_format
, out_addr
) == 1)
721 if (dmesg_timestamp
> *old_dmesg_timestamp
)
722 *old_dmesg_timestamp
= dmesg_timestamp
;
727 static int compare_wave(const void *p1
, const void *p2
)
729 struct ac_wave_info
*w1
= (struct ac_wave_info
*)p1
;
730 struct ac_wave_info
*w2
= (struct ac_wave_info
*)p2
;
732 /* Sort waves according to PC and then SE, SH, CU, etc. */
749 if (w1
->simd
< w2
->simd
)
751 if (w1
->simd
> w2
->simd
)
753 if (w1
->wave
< w2
->wave
)
755 if (w1
->wave
> w2
->wave
)
761 /* Return wave information. "waves" should be a large enough array. */
762 unsigned ac_get_wave_info(struct ac_wave_info waves
[AC_MAX_WAVES_PER_CHIP
])
765 unsigned num_waves
= 0;
767 FILE *p
= popen("umr -wa", "r");
771 if (!fgets(line
, sizeof(line
), p
) ||
772 strncmp(line
, "SE", 2) != 0) {
777 while (fgets(line
, sizeof(line
), p
)) {
778 struct ac_wave_info
*w
;
779 uint32_t pc_hi
, pc_lo
, exec_hi
, exec_lo
;
781 assert(num_waves
< AC_MAX_WAVES_PER_CHIP
);
782 w
= &waves
[num_waves
];
784 if (sscanf(line
, "%u %u %u %u %u %x %x %x %x %x %x %x",
785 &w
->se
, &w
->sh
, &w
->cu
, &w
->simd
, &w
->wave
,
786 &w
->status
, &pc_hi
, &pc_lo
, &w
->inst_dw0
,
787 &w
->inst_dw1
, &exec_hi
, &exec_lo
) == 12) {
788 w
->pc
= ((uint64_t)pc_hi
<< 32) | pc_lo
;
789 w
->exec
= ((uint64_t)exec_hi
<< 32) | exec_lo
;
795 qsort(waves
, num_waves
, sizeof(struct ac_wave_info
), compare_wave
);