1
0
Fork 0
alistair23-linux/tools/perf/tests/dwarf-unwind.c

181 lines
3.8 KiB
C
Raw Normal View History

#include <linux/compiler.h>
#include <linux/types.h>
#include <unistd.h>
#include "tests.h"
#include "debug.h"
#include "machine.h"
#include "event.h"
#include "unwind.h"
#include "perf_regs.h"
#include "map.h"
#include "thread.h"
#include "callchain.h"
perf test: Fix dwarf unwind using libunwind. Perf tool fails to unwind user stack if the event raises in a shared object. This patch improves tests/dwarf-unwind.c to demonstrate the problem by utilizing commonly used glibc function "bsearch". If perf is not statically linked, the testcase will try to unwind a mixed call trace. By debugging libunwind I found that there is a bug in unwind-libunwind: it always passes 0 as segbase to libunwind, cause libunwind unable to locate debug_frame entry fir first level ip address (I add some more debugging output into libunwind to make things clear): >_Uarm_dwarf_find_debug_frame: start_ip = 10be98, end_ip = 10c2a4 >_Uarm_dwarf_find_debug_frame: found debug_frame table `/lib/libc-2.18.so': segbase=0x0, len=7, gp=0x0, table_data=0x449388 >_Uarm_dwarf_search_unwind_table: call lookup:ip = b6cd3bcc, segbase = 0, rel_ip = b6cd3bcc >lookup: e->start_ip_offset = bcf18 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 6d314 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 33d0c (rel_ip = b6cd3bcc) ... >lookup: e->start_ip_offset = 15d0c (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 15c40 (rel_ip = b6cd3bcc) >_Uarm_dwarf_search_unwind_table: IP b6cd3bcc inside range b6c12000-b6d4c000, but no explicit unwind info found >put_rs_cache: unmasking signals/interrupts and releasing lock >_Uarm_dwarf_step: returning -10 >_Uarm_step: dwarf_step()=-10 This patch passes map->start as segbase to dwarf_find_debug_frame(), so di will be initialized correctly. In addition, dso and executable are different when setting segbase. This patch first check whether the elf is executable, and pass segbase only for shared object. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1421203007-75799-1-git-send-email-wangnan0@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-01-13 19:36:47 -07:00
/* For bsearch. We try to unwind functions in shared object. */
#include <stdlib.h>
static int mmap_handler(struct perf_tool *tool __maybe_unused,
union perf_event *event,
struct perf_sample *sample __maybe_unused,
struct machine *machine)
{
return machine__process_mmap2_event(machine, event, NULL);
}
static int init_live_machine(struct machine *machine)
{
union perf_event event;
pid_t pid = getpid();
return perf_event__synthesize_mmap_events(NULL, &event, pid, pid,
mmap_handler, machine, true, 500);
}
perf test: Fix dwarf unwind using libunwind. Perf tool fails to unwind user stack if the event raises in a shared object. This patch improves tests/dwarf-unwind.c to demonstrate the problem by utilizing commonly used glibc function "bsearch". If perf is not statically linked, the testcase will try to unwind a mixed call trace. By debugging libunwind I found that there is a bug in unwind-libunwind: it always passes 0 as segbase to libunwind, cause libunwind unable to locate debug_frame entry fir first level ip address (I add some more debugging output into libunwind to make things clear): >_Uarm_dwarf_find_debug_frame: start_ip = 10be98, end_ip = 10c2a4 >_Uarm_dwarf_find_debug_frame: found debug_frame table `/lib/libc-2.18.so': segbase=0x0, len=7, gp=0x0, table_data=0x449388 >_Uarm_dwarf_search_unwind_table: call lookup:ip = b6cd3bcc, segbase = 0, rel_ip = b6cd3bcc >lookup: e->start_ip_offset = bcf18 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 6d314 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 33d0c (rel_ip = b6cd3bcc) ... >lookup: e->start_ip_offset = 15d0c (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 15c40 (rel_ip = b6cd3bcc) >_Uarm_dwarf_search_unwind_table: IP b6cd3bcc inside range b6c12000-b6d4c000, but no explicit unwind info found >put_rs_cache: unmasking signals/interrupts and releasing lock >_Uarm_dwarf_step: returning -10 >_Uarm_step: dwarf_step()=-10 This patch passes map->start as segbase to dwarf_find_debug_frame(), so di will be initialized correctly. In addition, dso and executable are different when setting segbase. This patch first check whether the elf is executable, and pass segbase only for shared object. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1421203007-75799-1-git-send-email-wangnan0@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-01-13 19:36:47 -07:00
#define MAX_STACK 8
static int unwind_entry(struct unwind_entry *entry, void *arg)
{
unsigned long *cnt = (unsigned long *) arg;
char *symbol = entry->sym ? entry->sym->name : NULL;
static const char *funcs[MAX_STACK] = {
"test__arch_unwind_sample",
"unwind_thread",
perf test: Fix dwarf unwind using libunwind. Perf tool fails to unwind user stack if the event raises in a shared object. This patch improves tests/dwarf-unwind.c to demonstrate the problem by utilizing commonly used glibc function "bsearch". If perf is not statically linked, the testcase will try to unwind a mixed call trace. By debugging libunwind I found that there is a bug in unwind-libunwind: it always passes 0 as segbase to libunwind, cause libunwind unable to locate debug_frame entry fir first level ip address (I add some more debugging output into libunwind to make things clear): >_Uarm_dwarf_find_debug_frame: start_ip = 10be98, end_ip = 10c2a4 >_Uarm_dwarf_find_debug_frame: found debug_frame table `/lib/libc-2.18.so': segbase=0x0, len=7, gp=0x0, table_data=0x449388 >_Uarm_dwarf_search_unwind_table: call lookup:ip = b6cd3bcc, segbase = 0, rel_ip = b6cd3bcc >lookup: e->start_ip_offset = bcf18 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 6d314 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 33d0c (rel_ip = b6cd3bcc) ... >lookup: e->start_ip_offset = 15d0c (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 15c40 (rel_ip = b6cd3bcc) >_Uarm_dwarf_search_unwind_table: IP b6cd3bcc inside range b6c12000-b6d4c000, but no explicit unwind info found >put_rs_cache: unmasking signals/interrupts and releasing lock >_Uarm_dwarf_step: returning -10 >_Uarm_step: dwarf_step()=-10 This patch passes map->start as segbase to dwarf_find_debug_frame(), so di will be initialized correctly. In addition, dso and executable are different when setting segbase. This patch first check whether the elf is executable, and pass segbase only for shared object. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1421203007-75799-1-git-send-email-wangnan0@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-01-13 19:36:47 -07:00
"compare",
"bsearch",
"krava_3",
"krava_2",
"krava_1",
"test__dwarf_unwind"
};
if (*cnt >= MAX_STACK) {
pr_debug("failed: crossed the max stack value %d\n", MAX_STACK);
return -1;
}
if (!symbol) {
pr_debug("failed: got unresolved address 0x%" PRIx64 "\n",
entry->ip);
return -1;
}
pr_debug("got: %s 0x%" PRIx64 "\n", symbol, entry->ip);
return strcmp((const char *) symbol, funcs[(*cnt)++]);
}
__attribute__ ((noinline))
static int unwind_thread(struct thread *thread)
{
struct perf_sample sample;
unsigned long cnt = 0;
int err = -1;
memset(&sample, 0, sizeof(sample));
if (test__arch_unwind_sample(&sample, thread)) {
pr_debug("failed to get unwind sample\n");
goto out;
}
err = unwind__get_entries(unwind_entry, &cnt, thread,
&sample, MAX_STACK);
if (err)
pr_debug("unwind failed\n");
else if (cnt != MAX_STACK) {
pr_debug("got wrong number of stack entries %lu != %d\n",
cnt, MAX_STACK);
err = -1;
}
out:
free(sample.user_stack.data);
free(sample.user_regs.regs);
return err;
}
perf test: Fix dwarf unwind using libunwind. Perf tool fails to unwind user stack if the event raises in a shared object. This patch improves tests/dwarf-unwind.c to demonstrate the problem by utilizing commonly used glibc function "bsearch". If perf is not statically linked, the testcase will try to unwind a mixed call trace. By debugging libunwind I found that there is a bug in unwind-libunwind: it always passes 0 as segbase to libunwind, cause libunwind unable to locate debug_frame entry fir first level ip address (I add some more debugging output into libunwind to make things clear): >_Uarm_dwarf_find_debug_frame: start_ip = 10be98, end_ip = 10c2a4 >_Uarm_dwarf_find_debug_frame: found debug_frame table `/lib/libc-2.18.so': segbase=0x0, len=7, gp=0x0, table_data=0x449388 >_Uarm_dwarf_search_unwind_table: call lookup:ip = b6cd3bcc, segbase = 0, rel_ip = b6cd3bcc >lookup: e->start_ip_offset = bcf18 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 6d314 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 33d0c (rel_ip = b6cd3bcc) ... >lookup: e->start_ip_offset = 15d0c (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 15c40 (rel_ip = b6cd3bcc) >_Uarm_dwarf_search_unwind_table: IP b6cd3bcc inside range b6c12000-b6d4c000, but no explicit unwind info found >put_rs_cache: unmasking signals/interrupts and releasing lock >_Uarm_dwarf_step: returning -10 >_Uarm_step: dwarf_step()=-10 This patch passes map->start as segbase to dwarf_find_debug_frame(), so di will be initialized correctly. In addition, dso and executable are different when setting segbase. This patch first check whether the elf is executable, and pass segbase only for shared object. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1421203007-75799-1-git-send-email-wangnan0@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-01-13 19:36:47 -07:00
static int global_unwind_retval = -INT_MAX;
__attribute__ ((noinline))
static int compare(void *p1, void *p2)
{
/* Any possible value should be 'thread' */
struct thread *thread = *(struct thread **)p1;
if (global_unwind_retval == -INT_MAX)
global_unwind_retval = unwind_thread(thread);
return p1 - p2;
}
__attribute__ ((noinline))
static int krava_3(struct thread *thread)
{
perf test: Fix dwarf unwind using libunwind. Perf tool fails to unwind user stack if the event raises in a shared object. This patch improves tests/dwarf-unwind.c to demonstrate the problem by utilizing commonly used glibc function "bsearch". If perf is not statically linked, the testcase will try to unwind a mixed call trace. By debugging libunwind I found that there is a bug in unwind-libunwind: it always passes 0 as segbase to libunwind, cause libunwind unable to locate debug_frame entry fir first level ip address (I add some more debugging output into libunwind to make things clear): >_Uarm_dwarf_find_debug_frame: start_ip = 10be98, end_ip = 10c2a4 >_Uarm_dwarf_find_debug_frame: found debug_frame table `/lib/libc-2.18.so': segbase=0x0, len=7, gp=0x0, table_data=0x449388 >_Uarm_dwarf_search_unwind_table: call lookup:ip = b6cd3bcc, segbase = 0, rel_ip = b6cd3bcc >lookup: e->start_ip_offset = bcf18 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 6d314 (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 33d0c (rel_ip = b6cd3bcc) ... >lookup: e->start_ip_offset = 15d0c (rel_ip = b6cd3bcc) >lookup: e->start_ip_offset = 15c40 (rel_ip = b6cd3bcc) >_Uarm_dwarf_search_unwind_table: IP b6cd3bcc inside range b6c12000-b6d4c000, but no explicit unwind info found >put_rs_cache: unmasking signals/interrupts and releasing lock >_Uarm_dwarf_step: returning -10 >_Uarm_step: dwarf_step()=-10 This patch passes map->start as segbase to dwarf_find_debug_frame(), so di will be initialized correctly. In addition, dso and executable are different when setting segbase. This patch first check whether the elf is executable, and pass segbase only for shared object. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1421203007-75799-1-git-send-email-wangnan0@huawei.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-01-13 19:36:47 -07:00
struct thread *array[2] = {thread, thread};
void *fp = &bsearch;
/*
* make _bsearch a volatile function pointer to
* prevent potential optimization, which may expand
* bsearch and call compare directly from this function,
* instead of libc shared object.
*/
void *(*volatile _bsearch)(void *, void *, size_t,
size_t, int (*)(void *, void *));
_bsearch = fp;
_bsearch(array, &thread, 2, sizeof(struct thread **), compare);
return global_unwind_retval;
}
__attribute__ ((noinline))
static int krava_2(struct thread *thread)
{
return krava_3(thread);
}
__attribute__ ((noinline))
static int krava_1(struct thread *thread)
{
return krava_2(thread);
}
int test__dwarf_unwind(void)
{
struct machines machines;
struct machine *machine;
struct thread *thread;
int err = -1;
machines__init(&machines);
machine = machines__find(&machines, HOST_KERNEL_ID);
if (!machine) {
pr_err("Could not get machine\n");
return -1;
}
callchain_param.record_mode = CALLCHAIN_DWARF;
if (init_live_machine(machine)) {
pr_err("Could not init machine\n");
goto out;
}
if (verbose > 1)
machine__fprintf(machine, stderr);
thread = machine__find_thread(machine, getpid(), getpid());
if (!thread) {
pr_err("Could not get thread\n");
goto out;
}
err = krava_1(thread);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-06 17:43:22 -06:00
thread__put(thread);
out:
machine__delete_threads(machine);
machine__exit(machine);
machines__exit(&machines);
return err;
}