Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull perf tooling fixes from Thomas Gleixner:
"Perf tooling fixes all over the place:
- Fix the selection of the main thread COMM in db-export
- Fix the disassemmbly display for BPF in annotate
- Fix cpumap mask setup in perf ftrace when only one CPU is present
- Add the missing 'cpu_clk_unhalted.core' event
- Fix CPU 0 bindings in NUMA benchmarks
- Fix the module size calculations for s390
- Handle the gap between kernel end and module start on s390
correctly
- Build and typo fixes"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf pmu-events: Fix missing "cpu_clk_unhalted.core" event
perf annotate: Fix s390 gap between kernel end and module start
perf record: Fix module size on s390
perf tools: Fix include paths in ui directory
perf tools: Fix a typo in a variable name in the Documentation Makefile
perf cpumap: Fix writing to illegal memory in handling cpumap mask
perf ftrace: Fix failure to set cpumask when only one cpu is present
perf db-export: Fix thread__exec_comm()
perf annotate: Fix printing of unaugmented disassembled instructions from BPF
perf bench numa: Fix cpu0 binding
+69
-16
+1
-1
tools/perf/Documentation/Makefile
+1
-1
tools/perf/Documentation/Makefile
+30
-1
tools/perf/arch/s390/util/machine.c
+30
-1
tools/perf/arch/s390/util/machine.c
···
6
6
#include "machine.h"
7
7
#include "api/fs/fs.h"
8
8
#include "debug.h"
9
+
#include "symbol.h"
9
10
10
-
int arch__fix_module_text_start(u64 *start, const char *name)
11
+
int arch__fix_module_text_start(u64 *start, u64 *size, const char *name)
11
12
{
12
13
u64 m_start = *start;
13
14
char path[PATH_MAX];
···
18
17
if (sysfs__read_ull(path, (unsigned long long *)start) < 0) {
19
18
pr_debug2("Using module %s start:%#lx\n", path, m_start);
20
19
*start = m_start;
20
+
} else {
21
+
/* Successful read of the modules segment text start address.
22
+
* Calculate difference between module start address
23
+
* in memory and module text segment start address.
24
+
* For example module load address is 0x3ff8011b000
25
+
* (from /proc/modules) and module text segment start
26
+
* address is 0x3ff8011b870 (from file above).
27
+
*
28
+
* Adjust the module size and subtract the GOT table
29
+
* size located at the beginning of the module.
30
+
*/
31
+
*size -= (*start - m_start);
21
32
}
22
33
23
34
return 0;
35
+
}
36
+
37
+
/* On s390 kernel text segment start is located at very low memory addresses,
38
+
* for example 0x10000. Modules are located at very high memory addresses,
39
+
* for example 0x3ff xxxx xxxx. The gap between end of kernel text segment
40
+
* and beginning of first module's text segment is very big.
41
+
* Therefore do not fill this gap and do not assign it to the kernel dso map.
42
+
*/
43
+
void arch__symbols__fixup_end(struct symbol *p, struct symbol *c)
44
+
{
45
+
if (strchr(p->name, '[') == NULL && strchr(c->name, '['))
46
+
/* Last kernel symbol mapped to end of page */
47
+
p->end = roundup(p->end, page_size);
48
+
else
49
+
p->end = c->start;
50
+
pr_debug4("%s sym:%s end:%#lx\n", __func__, p->name, p->end);
24
51
}
+4
-2
tools/perf/bench/numa.c
+4
-2
tools/perf/bench/numa.c
···
379
379
380
380
/* Allocate and initialize all memory on CPU#0: */
381
381
if (init_cpu0) {
382
-
orig_mask = bind_to_node(0);
383
-
bind_to_memnode(0);
382
+
int node = numa_node_of_cpu(0);
383
+
384
+
orig_mask = bind_to_node(node);
385
+
bind_to_memnode(node);
384
386
}
385
387
386
388
bytes = bytes0 + HPSIZE;
+1
-1
tools/perf/builtin-ftrace.c
+1
-1
tools/perf/builtin-ftrace.c
···
173
173
int last_cpu;
174
174
175
175
last_cpu = cpu_map__cpu(cpumap, cpumap->nr - 1);
176
-
mask_size = (last_cpu + 3) / 4 + 1;
176
+
mask_size = last_cpu / 4 + 2; /* one more byte for EOS */
177
177
mask_size += last_cpu / 32; /* ',' is needed for every 32th cpus */
178
178
179
179
cpumask = malloc(mask_size);
+1
tools/perf/pmu-events/jevents.c
+1
tools/perf/pmu-events/jevents.c
···
453
453
{ "inst_retired.any_p", "event=0xc0" },
454
454
{ "cpu_clk_unhalted.ref", "event=0x0,umask=0x03" },
455
455
{ "cpu_clk_unhalted.thread", "event=0x3c" },
456
+
{ "cpu_clk_unhalted.core", "event=0x3c" },
456
457
{ "cpu_clk_unhalted.thread_any", "event=0x3c,any=1" },
457
458
{ NULL, NULL},
458
459
};
+5
-4
tools/perf/ui/browser.c
+5
-4
tools/perf/ui/browser.c
···
1
1
// SPDX-License-Identifier: GPL-2.0
2
-
#include "../string2.h"
3
-
#include "../config.h"
4
-
#include "../../perf.h"
2
+
#include "../util/util.h"
3
+
#include "../util/string2.h"
4
+
#include "../util/config.h"
5
+
#include "../perf.h"
5
6
#include "libslang.h"
6
7
#include "ui.h"
7
8
#include "util.h"
···
15
14
#include "browser.h"
16
15
#include "helpline.h"
17
16
#include "keysyms.h"
18
-
#include "../color.h"
17
+
#include "../util/color.h"
19
18
#include <linux/ctype.h>
20
19
#include <linux/zalloc.h>
21
20
+1
-1
tools/perf/ui/tui/progress.c
+1
-1
tools/perf/ui/tui/progress.c
+1
-1
tools/perf/util/annotate.c
+1
-1
tools/perf/util/annotate.c
+4
-1
tools/perf/util/cpumap.c
+4
-1
tools/perf/util/cpumap.c
+2
-1
tools/perf/util/machine.c
+2
-1
tools/perf/util/machine.c
···
1378
1378
return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1379
1379
}
1380
1380
int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1381
+
u64 *size __maybe_unused,
1381
1382
const char *name __maybe_unused)
1382
1383
{
1383
1384
return 0;
···
1390
1389
struct machine *machine = arg;
1391
1390
struct map *map;
1392
1391
1393
-
if (arch__fix_module_text_start(&start, name) < 0)
1392
+
if (arch__fix_module_text_start(&start, &size, name) < 0)
1394
1393
return -1;
1395
1394
1396
1395
map = machine__findnew_module_map(machine, start, name);
+1
-1
tools/perf/util/machine.h
+1
-1
tools/perf/util/machine.h
···
222
222
223
223
struct map *machine__findnew_module_map(struct machine *machine, u64 start,
224
224
const char *filename);
225
-
int arch__fix_module_text_start(u64 *start, const char *name);
225
+
int arch__fix_module_text_start(u64 *start, u64 *size, const char *name);
226
226
227
227
int machine__load_kallsyms(struct machine *machine, const char *filename);
228
228
+6
-1
tools/perf/util/symbol.c
+6
-1
tools/perf/util/symbol.c
···
92
92
return tail - str;
93
93
}
94
94
95
+
void __weak arch__symbols__fixup_end(struct symbol *p, struct symbol *c)
96
+
{
97
+
p->end = c->start;
98
+
}
99
+
95
100
const char * __weak arch__normalize_symbol_name(const char *name)
96
101
{
97
102
return name;
···
223
218
curr = rb_entry(nd, struct symbol, rb_node);
224
219
225
220
if (prev->end == prev->start && prev->end != curr->start)
226
-
prev->end = curr->start;
221
+
arch__symbols__fixup_end(prev, curr);
227
222
}
228
223
229
224
/* Last entry */
+1
tools/perf/util/symbol.h
+1
tools/perf/util/symbol.h
···
288
288
#define SYMBOL_A 0
289
289
#define SYMBOL_B 1
290
290
291
+
void arch__symbols__fixup_end(struct symbol *p, struct symbol *c);
291
292
int arch__compare_symbol_names(const char *namea, const char *nameb);
292
293
int arch__compare_symbol_names_n(const char *namea, const char *nameb,
293
294
unsigned int n);
+11
-1
tools/perf/util/thread.c
+11
-1
tools/perf/util/thread.c
···
214
214
215
215
struct comm *thread__exec_comm(const struct thread *thread)
216
216
{
217
-
struct comm *comm, *last = NULL;
217
+
struct comm *comm, *last = NULL, *second_last = NULL;
218
218
219
219
list_for_each_entry(comm, &thread->comm_list, list) {
220
220
if (comm->exec)
221
221
return comm;
222
+
second_last = last;
222
223
last = comm;
223
224
}
225
+
226
+
/*
227
+
* 'last' with no start time might be the parent's comm of a synthesized
228
+
* thread (created by processing a synthesized fork event). For a main
229
+
* thread, that is very probably wrong. Prefer a later comm to avoid
230
+
* that case.
231
+
*/
232
+
if (second_last && !last->start && thread->pid_ == thread->tid)
233
+
return second_last;
224
234
225
235
return last;
226
236
}