The goal of these 4 patches is to allow perf probing at absolute address
for uprobes and kprobes. During the development several small problem in
'perf probe -l' is found and fixed.
Patch 1 has already acked by Masami Hiramatsu.
Patch 2 fixes a problem about zero address in perf side.
Patch 3 is the main patch of these 4. It allows following command:
# perf probe /lib/x86_64-linux-gnu/libc-2.19.so 0xeb860
Different from v1, no extra '+' is required.
Patch 4 is a kernel side patch which fixes a small problem in uprobe_event
that, if address is 0, uprobe_events reports '0x (null)'. It is independent
from the other patches.
Wang Nan (4):
perf probe: Prevent segfault when reading probe point with absolute
address
perf probe: Fix list result when address is zero
perf probe: Support probing at absolute address
tracing/uprobes: Do not print '0x (null)' when offset is 0
kernel/trace/trace_uprobe.c | 17 +++-
tools/perf/util/probe-event.c | 210 +++++++++++++++++++++++++++++++++++++----
tools/perf/util/probe-event.h | 4 +
tools/perf/util/probe-finder.c | 21 +----
4 files changed, 216 insertions(+), 36 deletions(-)
--
1.8.3.4
'perf probe -l' panic if there is a manually inserted probing point
with absolute address. For example:
# echo 'p:probe/abs_ffffffff811e6615 0xffffffff811e6615' > /sys/kernel/debug/tracing/kprobe_events
# perf probe -l
Segmentation fault (core dumped)
This patch fixes this problem by considering the situation that
"tp->symbol == NULL" in find_perf_probe_point_from_dwarf() and
find_perf_probe_point_from_map().
After this patch:
# perf probe -l
probe:abs_ffffffff811e6615 (on SyS_write+5@fs/read_write.c)
And when debug info is missing:
# rm -rf ~/.debug
# mv /lib/modules/4.2.0-rc1+/build/vmlinux /lib/modules/4.2.0-rc1+/build/vmlinux.bak
# perf probe -l
probe:abs_ffffffff811e6615 (on sys_write+5)
Signed-off-by: Wang Nan <[email protected]>
Acked-by: Masami Hiramatsu <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Masami Hiramatsu <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
---
tools/perf/util/probe-event.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index f173372..ef325a2 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -517,7 +517,7 @@ static int find_perf_probe_point_from_dwarf(struct probe_trace_point *tp,
if (ret < 0)
goto error;
addr += stext;
- } else {
+ } else if (tp->symbol) {
addr = kernel_get_symbol_address_by_name(tp->symbol, false);
if (addr == 0)
goto error;
@@ -1817,17 +1817,17 @@ static int find_perf_probe_point_from_map(struct probe_trace_point *tp,
{
struct symbol *sym = NULL;
struct map *map;
- u64 addr;
+ u64 addr = tp->address;
int ret = -ENOENT;
if (!is_kprobe) {
map = dso__new_map(tp->module);
if (!map)
goto out;
- addr = tp->address;
sym = map__find_symbol(map, addr, NULL);
} else {
- addr = kernel_get_symbol_address_by_name(tp->symbol, true);
+ if (tp->symbol)
+ addr = kernel_get_symbol_address_by_name(tp->symbol, true);
if (addr) {
addr += tp->offset;
sym = __find_kernel_function(addr, &map);
--
1.8.3.4
When manually added uprobe point with zero address, 'perf probe -l'
reports error. For example:
# echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
/sys/kernel/debug/tracing/uprobe_events
# perf probe -l
Error: Failed to show event list.
Probing at 0x0 is possible and useful when lib.bin is not a normal
shared object but is manually mapped. However, in this case kernel
report:
# cat /sys/kernel/debug/tracing/uprobe_events
p:probe_libc/abs_0 /tmp/oxygen_root/lib64/libc-2.18.so:0x (null) arg1=%ax
This patch supports the above kernel output.
In addition, this patch also fixes convert_to_perf_probe_point() that,
if failed to find probe point from both of dwarf and map, in all cases
pp->function should be filled with a strdup() result, or the following
ENOMEM error code is incorrect. Remove !tp->module && !is_kprobe
condition so always use address to build function name if symbol is not
found.
Signed-off-by: Wang Nan <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Masami Hiramatsu <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
---
tools/perf/util/probe-event.c | 30 ++++++++++++++++++++++++++----
1 file changed, 26 insertions(+), 4 deletions(-)
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index ef325a2..3ce223d 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -1521,9 +1521,31 @@ int parse_probe_trace_command(const char *cmd, struct probe_trace_event *tev)
} else
p = argv[1];
fmt1_str = strtok_r(p, "+", &fmt);
- if (fmt1_str[0] == '0') /* only the address started with 0x */
- tp->address = strtoul(fmt1_str, NULL, 0);
- else {
+ /* only the address started with 0x */
+ if (fmt1_str[0] == '0') {
+ /*
+ * Fix a special case:
+ * if address == 0, kernel reports something like:
+ * p:probe_libc/abs_0 /lib/libc-2.18.so:0x (null) arg1=%ax
+ * Newer kernel may fix that, but we want to
+ * support old kernel also.
+ */
+ if (strcmp(fmt1_str, "0x") == 0) {
+ if (!argv[2] || strcmp(argv[2], "(null)")) {
+ ret = -EINVAL;
+ goto out;
+ }
+ tp->address = 0;
+
+ free(argv[2]);
+ for (i = 2; argv[i + 1] != NULL; i++)
+ argv[i] = argv[i + 1];
+
+ argv[i] = NULL;
+ argc -= 1;
+ } else
+ tp->address = strtoul(fmt1_str, NULL, 0);
+ } else {
/* Only the symbol-based probe has offset */
tp->symbol = strdup(fmt1_str);
if (tp->symbol == NULL) {
@@ -1868,7 +1890,7 @@ static int convert_to_perf_probe_point(struct probe_trace_point *tp,
if (tp->symbol) {
pp->function = strdup(tp->symbol);
pp->offset = tp->offset;
- } else if (!tp->module && !is_kprobe) {
+ } else {
ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
if (ret < 0)
return ret;
--
1.8.3.4
It should be useful to allow 'perf probe' probe at absolute offset of
a target. For example, when (u)probing at a instruction of a shared
object in a embedded system where debuginfo is not avaliable but we
know the offset of that instruction by manually digging.
This patch enables following perf probe command syntax:
# perf probe 0xffffffff811e6615
And
# perf probe /lib/x86_64-linux-gnu/libc-2.19.so 0xeb860
In the above example, we don't need a anchor symbol, so it is possible
to compute absolute addresses using other methods and then use
'perf probe' to create the probing points.
This patch also fix a bug that, when offset is provided but function is
skipped, parse_perf_probe_point() will give function a "" string, so
the checking code at the end of this function is not used. For example:
# perf probe +0x1234
Failed to find symbol in kernel
Error: Failed to add events.
After this patch:
# perf probe +0x1234
Semantic error :Offset requires an entry function.
Error: Command Parse Error.
v1 -> v2:
Drop the leading '+' in cmdline;
Allow uprobing at offset 0x0;
Improve 'perf probe -l' result when uprobe at area without debuginfo.
Test result:
# perf probe 0xffffffff8119d175 %ax
# perf probe sys_write %ax
# perf probe /lib64/libc-2.18.so 0x0 %ax
# perf probe /lib64/libc-2.18.so 0x5 %ax
# perf probe /lib64/libc-2.18.so 0xd8e40 %ax
# perf probe /lib64/libc-2.18.so __write %ax
# perf probe /lib64/libc-2.18.so 0xd8e49 %ax
# cat /sys/kernel/debug/tracing/uprobe_events
p:probe_libc/abs_0 /lib64/libc-2.18.so:0x (null) arg1=%ax
p:probe_libc/abs_5 /lib64/libc-2.18.so:0x0000000000000005 arg1=%ax
p:probe_libc/abs_d8e40 /lib64/libc-2.18.so:0x00000000000d8e40 arg1=%ax
p:probe_libc/__write /lib64/libc-2.18.so:0x00000000000d8e40 arg1=%ax
p:probe_libc/abs_d8e49 /lib64/libc-2.18.so:0x00000000000d8e49 arg1=%ax
# cat /sys/kernel/debug/tracing/kprobe_events
p:probe/abs_ffffffff8119d175 0xffffffff8119d175 arg1=%ax
p:probe/sys_write _text+1692016 arg1=%ax
# perf probe -l
Failed to find debug information for address 5
probe:abs_ffffffff8119d175 (on sys_write+5 with arg1)
probe:sys_write (on sys_write with arg1)
probe_libc:__write (on @unix/syscall-template.S:81 in /lib64/libc-2.18.so with arg1)
probe_libc:abs_0 (on 0x0 in /lib64/libc-2.18.so with arg1)
probe_libc:abs_5 (on 0x5 in /lib64/libc-2.18.so with arg1)
probe_libc:abs_d8e40 (on @unix/syscall-template.S:81 in /lib64/libc-2.18.so with arg1)
probe_libc:abs_d8e49 (on __GI___libc_write+9 in /lib64/libc-2.18.so with arg1)
Signed-off-by: Wang Nan <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Masami Hiramatsu <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
---
tools/perf/util/probe-event.c | 172 ++++++++++++++++++++++++++++++++++++++---
tools/perf/util/probe-event.h | 4 +
tools/perf/util/probe-finder.c | 21 +----
3 files changed, 170 insertions(+), 27 deletions(-)
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index 3ce223d..b94a8d7 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -1196,15 +1196,37 @@ static int parse_perf_probe_point(char *arg, struct perf_probe_event *pev)
*ptr++ = '\0';
}
- tmp = strdup(arg);
- if (tmp == NULL)
- return -ENOMEM;
+ if (arg[0] == '\0')
+ tmp = NULL;
+ else {
+ tmp = strdup(arg);
+ if (tmp == NULL)
+ return -ENOMEM;
+ }
if (file_spec)
pp->file = tmp;
- else
+ else {
pp->function = tmp;
+ /*
+ * Keep pp->function even if this is absolute address,
+ * so it can mark whether abs_address is valid.
+ * Which make 'perf probe lib.bin 0x0' possible.
+ *
+ * Note that checking length of tmp is not needed
+ * because when we access tmp[1] we know tmp[0] is '0',
+ * so tmp[1] should always valid (but could be '\0').
+ */
+ if (tmp && !strncmp(tmp, "0x", 2)) {
+ pp->abs_address = strtoul(pp->function, &tmp, 0);
+ if (*tmp != '\0') {
+ semantic_error("Invalid absolute address.\n");
+ return -EINVAL;
+ }
+ }
+ }
+
/* Parse other options */
while (ptr) {
arg = ptr;
@@ -1802,14 +1824,29 @@ char *synthesize_probe_trace_command(struct probe_trace_event *tev)
if (len <= 0)
goto error;
- /* Uprobes must have tp->address and tp->module */
- if (tev->uprobes && (!tp->address || !tp->module))
+ /* Uprobes must have tp->module */
+ if (tev->uprobes && !tp->module)
goto error;
+ /*
+ * If tp->address == 0, then this point must be a
+ * absolute address uprobe.
+ * try_to_find_absolute_address() should have made
+ * tp->symbol to "0x0".
+ */
+ if (tev->uprobes && !tp->address) {
+ if (!tp->symbol || strcmp(tp->symbol, "0x0"))
+ goto error;
+ }
/* Use the tp->address for uprobes */
if (tev->uprobes)
ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s:0x%lx",
tp->module, tp->address);
+ else if (!strncmp(tp->symbol, "0x", 2))
+ /* Absolute address. See try_to_find_absolute_address() */
+ ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s%s0x%lx",
+ tp->module ?: "", tp->module ? ":" : "",
+ tp->address);
else
ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s%s%s+%lu",
tp->module ?: "", tp->module ? ":" : "",
@@ -1872,8 +1909,8 @@ out:
}
static int convert_to_perf_probe_point(struct probe_trace_point *tp,
- struct perf_probe_point *pp,
- bool is_kprobe)
+ struct perf_probe_point *pp,
+ bool is_kprobe)
{
char buf[128];
int ret;
@@ -2334,7 +2371,9 @@ static int probe_trace_event__set_name(struct probe_trace_event *tev,
if (pev->event)
event = pev->event;
else
- if (pev->point.function && !strisglob(pev->point.function))
+ if (pev->point.function &&
+ (strncmp(pev->point.function, "0x", 2) != 0) &&
+ !strisglob(pev->point.function))
event = pev->point.function;
else
event = tev->point.realname;
@@ -2604,6 +2643,98 @@ err_out:
goto out;
}
+static int try_to_find_absolute_address(struct perf_probe_event *pev,
+ struct probe_trace_event **tevs)
+{
+ struct perf_probe_point *pp = &pev->point;
+ struct probe_trace_event *tev;
+ struct probe_trace_point *tp;
+ int i, err;
+
+ if (!(pev->point.function && !strncmp(pev->point.function, "0x", 2)))
+ return -EINVAL;
+ if (perf_probe_event_need_dwarf(pev))
+ return -EINVAL;
+
+ /*
+ * This is 'perf probe /lib/libc.so 0xabcd'. Try to probe at
+ * absolute address.
+ *
+ * Only one tev can be generated by this.
+ */
+ *tevs = zalloc(sizeof(*tev));
+ if (!*tevs)
+ return -ENOMEM;
+
+ tev = *tevs;
+ tp = &tev->point;
+
+ /*
+ * Don't use tp->offset, use address directly, because
+ * in synthesize_probe_trace_command() address cannot be
+ * zero.
+ */
+ tp->address = pev->point.abs_address;
+ tp->retprobe = pp->retprobe;
+ tev->uprobes = pev->uprobes;
+
+ err = -ENOMEM;
+ /*
+ * Give it a '0x' leading symbol name.
+ * In __add_probe_trace_events, a NULL symbol is interpreted as
+ * invalud.
+ */
+ if (asprintf(&tp->symbol, "0x%lx", tp->address) < 0)
+ goto errout;
+
+ /* For kprobe, check range */
+ if ((!tev->uprobes) &&
+ (kprobe_warn_out_range(tev->point.symbol,
+ tev->point.address))) {
+ err = -EACCES;
+ goto errout;
+ }
+
+ if (asprintf(&tp->realname, "abs_%lx", tp->address) < 0)
+ goto errout;
+
+ if (pev->target) {
+ tp->module = strdup(pev->target);
+ if (!tp->module)
+ goto errout;
+ }
+
+ if (tev->group) {
+ tev->group = strdup(pev->group);
+ if (!tev->group)
+ goto errout;
+ }
+
+ if (pev->event) {
+ tev->event = strdup(pev->event);
+ if (!tev->event)
+ goto errout;
+ }
+
+ tev->nargs = pev->nargs;
+ tev->args = zalloc(sizeof(struct probe_trace_arg) * tev->nargs);
+ if (!tev->args) {
+ err = -ENOMEM;
+ goto errout;
+ }
+ for (i = 0; i < tev->nargs; i++)
+ copy_to_probe_trace_arg(&tev->args[i], &pev->args[i]);
+
+ return 1;
+
+errout:
+ if (*tevs) {
+ clear_probe_trace_events(*tevs, 1);
+ *tevs = NULL;
+ }
+ return err;
+}
+
bool __weak arch__prefers_symtab(void) { return false; }
static int convert_to_probe_trace_events(struct perf_probe_event *pev,
@@ -2620,6 +2751,10 @@ static int convert_to_probe_trace_events(struct perf_probe_event *pev,
}
}
+ ret = try_to_find_absolute_address(pev, tevs);
+ if (ret > 0)
+ return ret;
+
if (arch__prefers_symtab() && !perf_probe_event_need_dwarf(pev)) {
ret = find_probe_trace_events_from_map(pev, tevs);
if (ret > 0)
@@ -2787,3 +2922,22 @@ end:
return ret;
}
+int copy_to_probe_trace_arg(struct probe_trace_arg *tvar,
+ struct perf_probe_arg *pvar)
+{
+ tvar->value = strdup(pvar->var);
+ if (tvar->value == NULL)
+ return -ENOMEM;
+ if (pvar->type) {
+ tvar->type = strdup(pvar->type);
+ if (tvar->type == NULL)
+ return -ENOMEM;
+ }
+ if (pvar->name) {
+ tvar->name = strdup(pvar->name);
+ if (tvar->name == NULL)
+ return -ENOMEM;
+ } else
+ tvar->name = NULL;
+ return 0;
+}
diff --git a/tools/perf/util/probe-event.h b/tools/perf/util/probe-event.h
index 9c7b99e..de0dd13 100644
--- a/tools/perf/util/probe-event.h
+++ b/tools/perf/util/probe-event.h
@@ -61,6 +61,7 @@ struct perf_probe_point {
bool retprobe; /* Return probe flag */
char *lazy_line; /* Lazy matching pattern */
unsigned long offset; /* Offset from function entry */
+ unsigned long abs_address; /* Absolute address of the point */
};
/* Perf probe probing argument field chain */
@@ -162,4 +163,7 @@ int e_snprintf(char *str, size_t size, const char *format, ...)
/* Maximum index number of event-name postfix */
#define MAX_EVENT_INDEX 1024
+int copy_to_probe_trace_arg(struct probe_trace_arg *tvar,
+ struct perf_probe_arg *pvar);
+
#endif /*_PROBE_EVENT_H */
diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
index 4b1e074..5ab9cd6 100644
--- a/tools/perf/util/probe-finder.c
+++ b/tools/perf/util/probe-finder.c
@@ -553,24 +553,9 @@ static int find_variable(Dwarf_Die *sc_die, struct probe_finder *pf)
char buf[32], *ptr;
int ret = 0;
- if (!is_c_varname(pf->pvar->var)) {
- /* Copy raw parameters */
- pf->tvar->value = strdup(pf->pvar->var);
- if (pf->tvar->value == NULL)
- return -ENOMEM;
- if (pf->pvar->type) {
- pf->tvar->type = strdup(pf->pvar->type);
- if (pf->tvar->type == NULL)
- return -ENOMEM;
- }
- if (pf->pvar->name) {
- pf->tvar->name = strdup(pf->pvar->name);
- if (pf->tvar->name == NULL)
- return -ENOMEM;
- } else
- pf->tvar->name = NULL;
- return 0;
- }
+ /* Copy raw parameters */
+ if (!is_c_varname(pf->pvar->var))
+ return copy_to_probe_trace_arg(pf->tvar, pf->pvar);
if (pf->pvar->name)
pf->tvar->name = strdup(pf->pvar->name);
--
1.8.3.4
When manually added uprobe point with zero address, 'uprobe_events' output
'(null)' instead of 0x00000000:
# echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
/sys/kernel/debug/tracing/uprobe_events
# cat /sys/kernel/debug/tracing/uprobe_events
p:probe_libc/abs_0 /path/to/lib.bin:0x (null) arg1=%ax
This patch fixes this behavior:
# cat /sys/kernel/debug/tracing/uprobe_events
p:probe_libc/abs_0 /path/to/lib.bin:0x0000000000000000
Signed-off-by: Wang Nan <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Masami Hiramatsu <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
---
kernel/trace/trace_uprobe.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
index f97479f..d2f6d0b 100644
--- a/kernel/trace/trace_uprobe.c
+++ b/kernel/trace/trace_uprobe.c
@@ -601,7 +601,22 @@ static int probes_seq_show(struct seq_file *m, void *v)
seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
trace_event_name(&tu->tp.call));
- seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
+ seq_printf(m, " %s:", tu->filename);
+
+ /* Don't print "0x (null)" when offset is 0 */
+ if (tu->offset) {
+ seq_printf(m, "0x%p", (void *)tu->offset);
+ } else {
+ switch (sizeof(void *)) {
+ case 4:
+ seq_printf(m, "0x00000000");
+ break;
+ case 8:
+ default:
+ seq_printf(m, "0x0000000000000000");
+ break;
+ }
+ }
for (i = 0; i < tu->tp.nr_args; i++)
seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
--
1.8.3.4
> From: Wang Nan [mailto:[email protected]]
>
> The goal of these 4 patches is to allow perf probing at absolute address
> for uprobes and kprobes. During the development several small problem in
> 'perf probe -l' is found and fixed.
>
> Patch 1 has already acked by Masami Hiramatsu.
>
> Patch 2 fixes a problem about zero address in perf side.
>
> Patch 3 is the main patch of these 4. It allows following command:
>
> # perf probe /lib/x86_64-linux-gnu/libc-2.19.so 0xeb860
>
> Different from v1, no extra '+' is required.
Yeah, this seems better for me. Since '+' has double meanings "positive value" and
"add something", I think it's not good for this context.
>
> Patch 4 is a kernel side patch which fixes a small problem in uprobe_event
> that, if address is 0, uprobe_events reports '0x (null)'. It is independent
> from the other patches.
Oops, right.
Thanks!
>
> Wang Nan (4):
> perf probe: Prevent segfault when reading probe point with absolute
> address
> perf probe: Fix list result when address is zero
> perf probe: Support probing at absolute address
> tracing/uprobes: Do not print '0x (null)' when offset is 0
>
> kernel/trace/trace_uprobe.c | 17 +++-
> tools/perf/util/probe-event.c | 210 +++++++++++++++++++++++++++++++++++++----
> tools/perf/util/probe-event.h | 4 +
> tools/perf/util/probe-finder.c | 21 +----
> 4 files changed, 216 insertions(+), 36 deletions(-)
>
> --
> 1.8.3.4
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
> From: Wang Nan [mailto:[email protected]]
>
> When manually added uprobe point with zero address, 'perf probe -l'
> reports error. For example:
>
> # echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
> /sys/kernel/debug/tracing/uprobe_events
>
> # perf probe -l
> Error: Failed to show event list.
>
> Probing at 0x0 is possible and useful when lib.bin is not a normal
> shared object but is manually mapped. However, in this case kernel
> report:
>
> # cat /sys/kernel/debug/tracing/uprobe_events
> p:probe_libc/abs_0 /tmp/oxygen_root/lib64/libc-2.18.so:0x (null) arg1=%ax
>
> This patch supports the above kernel output.
Former part is OK to me.
>
> In addition, this patch also fixes convert_to_perf_probe_point() that,
> if failed to find probe point from both of dwarf and map, in all cases
> pp->function should be filled with a strdup() result, or the following
> ENOMEM error code is incorrect. Remove !tp->module && !is_kprobe
> condition so always use address to build function name if symbol is not
> found.
No, I think this should be included in 3/4, since until that perf probe
doesn't support absolute address on kernel.
Thank you,
>
> Signed-off-by: Wang Nan <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Masami Hiramatsu <[email protected]>
> Cc: Namhyung Kim <[email protected]>
> Cc: Steven Rostedt <[email protected]>
> ---
> tools/perf/util/probe-event.c | 30 ++++++++++++++++++++++++++----
> 1 file changed, 26 insertions(+), 4 deletions(-)
>
> diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
> index ef325a2..3ce223d 100644
> --- a/tools/perf/util/probe-event.c
> +++ b/tools/perf/util/probe-event.c
> @@ -1521,9 +1521,31 @@ int parse_probe_trace_command(const char *cmd, struct probe_trace_event *tev)
> } else
> p = argv[1];
> fmt1_str = strtok_r(p, "+", &fmt);
> - if (fmt1_str[0] == '0') /* only the address started with 0x */
> - tp->address = strtoul(fmt1_str, NULL, 0);
> - else {
> + /* only the address started with 0x */
> + if (fmt1_str[0] == '0') {
> + /*
> + * Fix a special case:
> + * if address == 0, kernel reports something like:
> + * p:probe_libc/abs_0 /lib/libc-2.18.so:0x (null) arg1=%ax
> + * Newer kernel may fix that, but we want to
> + * support old kernel also.
> + */
> + if (strcmp(fmt1_str, "0x") == 0) {
> + if (!argv[2] || strcmp(argv[2], "(null)")) {
> + ret = -EINVAL;
> + goto out;
> + }
> + tp->address = 0;
> +
> + free(argv[2]);
> + for (i = 2; argv[i + 1] != NULL; i++)
> + argv[i] = argv[i + 1];
> +
> + argv[i] = NULL;
> + argc -= 1;
> + } else
> + tp->address = strtoul(fmt1_str, NULL, 0);
> + } else {
> /* Only the symbol-based probe has offset */
> tp->symbol = strdup(fmt1_str);
> if (tp->symbol == NULL) {
> @@ -1868,7 +1890,7 @@ static int convert_to_perf_probe_point(struct probe_trace_point *tp,
> if (tp->symbol) {
> pp->function = strdup(tp->symbol);
> pp->offset = tp->offset;
> - } else if (!tp->module && !is_kprobe) {
> + } else {
> ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
> if (ret < 0)
> return ret;
> --
> 1.8.3.4
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
On 2015/8/26 17:29, 平松雅巳 / HIRAMATU，MASAMI wrote:
>> From: Wang Nan [mailto:[email protected]]
>>
>> When manually added uprobe point with zero address, 'perf probe -l'
>> reports error. For example:
>>
>> # echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
>> /sys/kernel/debug/tracing/uprobe_events
>>
>> # perf probe -l
>> Error: Failed to show event list.
>>
>> Probing at 0x0 is possible and useful when lib.bin is not a normal
>> shared object but is manually mapped. However, in this case kernel
>> report:
>>
>> # cat /sys/kernel/debug/tracing/uprobe_events
>> p:probe_libc/abs_0 /tmp/oxygen_root/lib64/libc-2.18.so:0x (null) arg1=%ax
>>
>> This patch supports the above kernel output.
> Former part is OK to me.
>
>> In addition, this patch also fixes convert_to_perf_probe_point() that,
>> if failed to find probe point from both of dwarf and map, in all cases
>> pp->function should be filled with a strdup() result, or the following
>> ENOMEM error code is incorrect. Remove !tp->module && !is_kprobe
>> condition so always use address to build function name if symbol is not
>> found.
> No, I think this should be included in 3/4, since until that perf probe
> doesn't support absolute address on kernel.
This is for manually introduced probing points.
convert_to_perf_probe_point is called by:
perf probe -l -> show_perf_probe_events -> __show_perf_probe_events --> convert_to_perf_probe_event
If there is a manually inserted probing point there, 'perf probe -l' reports an error.
Here is the result if I remove the second part:
$ git log --oneline # I'm in 2/4
aae2c2c perf probe: Fix list result when address is zero
931dff5 perf probe: Prevent segfault when reading probe point with
absolute address
...
$ git diff # remove the second part
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index 926bcec..de2c27e 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -1888,7 +1888,7 @@ static int convert_to_perf_probe_point(struct
probe_trace_point *tp,
if (tp->symbol) {
pp->function = strdup(tp->symbol);
pp->offset = tp->offset;
- } else {
+ } else if (!tp->module && !is_kprobe) {
ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
if (ret < 0)
return ret;
...
# echo 'p:probe_libc/abs_5 /lib64/libc.so.6:0x5' >
/sys/kernel/debug/tracing/uprobe_events
# cat /sys/kernel/debug/tracing/uprobe_events
p:probe_libc/abs_5 /lib64/libc.so.6:0x0000000000000005
# perf probe -l
Error: Failed to show event list
Then add it back:
$ git reset --hard HEAD
...
# perf probe -l
probe_libc:abs_5 (on 0x5 in /lib64/libc.so.6)
Thank you.
> Thank you,
>
>> Signed-off-by: Wang Nan <[email protected]>
>> Cc: Arnaldo Carvalho de Melo <[email protected]>
>> Cc: Ingo Molnar <[email protected]>
>> Cc: Masami Hiramatsu <[email protected]>
>> Cc: Namhyung Kim <[email protected]>
>> Cc: Steven Rostedt <[email protected]>
>> ---
>> tools/perf/util/probe-event.c | 30 ++++++++++++++++++++++++++----
>> 1 file changed, 26 insertions(+), 4 deletions(-)
>>
>> diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
>> index ef325a2..3ce223d 100644
>> --- a/tools/perf/util/probe-event.c
>> +++ b/tools/perf/util/probe-event.c
>> @@ -1521,9 +1521,31 @@ int parse_probe_trace_command(const char *cmd, struct probe_trace_event *tev)
>> } else
>> p = argv[1];
>> fmt1_str = strtok_r(p, "+", &fmt);
>> - if (fmt1_str[0] == '0') /* only the address started with 0x */
>> - tp->address = strtoul(fmt1_str, NULL, 0);
>> - else {
>> + /* only the address started with 0x */
>> + if (fmt1_str[0] == '0') {
>> + /*
>> + * Fix a special case:
>> + * if address == 0, kernel reports something like:
>> + * p:probe_libc/abs_0 /lib/libc-2.18.so:0x (null) arg1=%ax
>> + * Newer kernel may fix that, but we want to
>> + * support old kernel also.
>> + */
>> + if (strcmp(fmt1_str, "0x") == 0) {
>> + if (!argv[2] || strcmp(argv[2], "(null)")) {
>> + ret = -EINVAL;
>> + goto out;
>> + }
>> + tp->address = 0;
>> +
>> + free(argv[2]);
>> + for (i = 2; argv[i + 1] != NULL; i++)
>> + argv[i] = argv[i + 1];
>> +
>> + argv[i] = NULL;
>> + argc -= 1;
>> + } else
>> + tp->address = strtoul(fmt1_str, NULL, 0);
>> + } else {
>> /* Only the symbol-based probe has offset */
>> tp->symbol = strdup(fmt1_str);
>> if (tp->symbol == NULL) {
>> @@ -1868,7 +1890,7 @@ static int convert_to_perf_probe_point(struct probe_trace_point *tp,
>> if (tp->symbol) {
>> pp->function = strdup(tp->symbol);
>> pp->offset = tp->offset;
>> - } else if (!tp->module && !is_kprobe) {
>> + } else {
>> ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
>> if (ret < 0)
>> return ret;
>> --
>> 1.8.3.4
> From: Wang Nan [mailto:[email protected]]
>
> It should be useful to allow 'perf probe' probe at absolute offset of
> a target. For example, when (u)probing at a instruction of a shared
> object in a embedded system where debuginfo is not avaliable but we
> know the offset of that instruction by manually digging.
>
> This patch enables following perf probe command syntax:
>
> # perf probe 0xffffffff811e6615
>
> And
>
> # perf probe /lib/x86_64-linux-gnu/libc-2.19.so 0xeb860
>
> In the above example, we don't need a anchor symbol, so it is possible
> to compute absolute addresses using other methods and then use
> 'perf probe' to create the probing points.
OK, this looks good :)
>
> This patch also fix a bug that, when offset is provided but function is
> skipped, parse_perf_probe_point() will give function a "" string, so
> the checking code at the end of this function is not used. For example:
>
> # perf probe +0x1234
> Failed to find symbol in kernel
> Error: Failed to add events.
>
> After this patch:
>
> # perf probe +0x1234
> Semantic error :Offset requires an entry function.
> Error: Command Parse Error.
Hmm, could you please split the bugfix from feature addition patch?
Thank you,
>
> v1 -> v2:
> Drop the leading '+' in cmdline;
> Allow uprobing at offset 0x0;
> Improve 'perf probe -l' result when uprobe at area without debuginfo.
>
> Test result:
>
> # perf probe 0xffffffff8119d175 %ax
> # perf probe sys_write %ax
> # perf probe /lib64/libc-2.18.so 0x0 %ax
> # perf probe /lib64/libc-2.18.so 0x5 %ax
> # perf probe /lib64/libc-2.18.so 0xd8e40 %ax
> # perf probe /lib64/libc-2.18.so __write %ax
> # perf probe /lib64/libc-2.18.so 0xd8e49 %ax
> # cat /sys/kernel/debug/tracing/uprobe_events
>
> p:probe_libc/abs_0 /lib64/libc-2.18.so:0x (null) arg1=%ax
> p:probe_libc/abs_5 /lib64/libc-2.18.so:0x0000000000000005 arg1=%ax
> p:probe_libc/abs_d8e40 /lib64/libc-2.18.so:0x00000000000d8e40 arg1=%ax
> p:probe_libc/__write /lib64/libc-2.18.so:0x00000000000d8e40 arg1=%ax
> p:probe_libc/abs_d8e49 /lib64/libc-2.18.so:0x00000000000d8e49 arg1=%ax
>
> # cat /sys/kernel/debug/tracing/kprobe_events
>
> p:probe/abs_ffffffff8119d175 0xffffffff8119d175 arg1=%ax
> p:probe/sys_write _text+1692016 arg1=%ax
>
> # perf probe -l
>
> Failed to find debug information for address 5
> probe:abs_ffffffff8119d175 (on sys_write+5 with arg1)
> probe:sys_write (on sys_write with arg1)
> probe_libc:__write (on @unix/syscall-template.S:81 in /lib64/libc-2.18.so with arg1)
> probe_libc:abs_0 (on 0x0 in /lib64/libc-2.18.so with arg1)
> probe_libc:abs_5 (on 0x5 in /lib64/libc-2.18.so with arg1)
> probe_libc:abs_d8e40 (on @unix/syscall-template.S:81 in /lib64/libc-2.18.so with arg1)
> probe_libc:abs_d8e49 (on __GI___libc_write+9 in /lib64/libc-2.18.so with arg1)
>
> Signed-off-by: Wang Nan <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Masami Hiramatsu <[email protected]>
> Cc: Namhyung Kim <[email protected]>
> Cc: Steven Rostedt <[email protected]>
> ---
> tools/perf/util/probe-event.c | 172 ++++++++++++++++++++++++++++++++++++++---
> tools/perf/util/probe-event.h | 4 +
> tools/perf/util/probe-finder.c | 21 +----
> 3 files changed, 170 insertions(+), 27 deletions(-)
>
> diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
> index 3ce223d..b94a8d7 100644
> --- a/tools/perf/util/probe-event.c
> +++ b/tools/perf/util/probe-event.c
> @@ -1196,15 +1196,37 @@ static int parse_perf_probe_point(char *arg, struct perf_probe_event *pev)
> *ptr++ = '\0';
> }
>
> - tmp = strdup(arg);
> - if (tmp == NULL)
> - return -ENOMEM;
> + if (arg[0] == '\0')
> + tmp = NULL;
> + else {
> + tmp = strdup(arg);
> + if (tmp == NULL)
> + return -ENOMEM;
> + }
>
> if (file_spec)
> pp->file = tmp;
> - else
> + else {
> pp->function = tmp;
>
> + /*
> + * Keep pp->function even if this is absolute address,
> + * so it can mark whether abs_address is valid.
> + * Which make 'perf probe lib.bin 0x0' possible.
> + *
> + * Note that checking length of tmp is not needed
> + * because when we access tmp[1] we know tmp[0] is '0',
> + * so tmp[1] should always valid (but could be '\0').
> + */
> + if (tmp && !strncmp(tmp, "0x", 2)) {
> + pp->abs_address = strtoul(pp->function, &tmp, 0);
> + if (*tmp != '\0') {
> + semantic_error("Invalid absolute address.\n");
> + return -EINVAL;
> + }
> + }
> + }
> +
> /* Parse other options */
> while (ptr) {
> arg = ptr;
> @@ -1802,14 +1824,29 @@ char *synthesize_probe_trace_command(struct probe_trace_event *tev)
> if (len <= 0)
> goto error;
>
> - /* Uprobes must have tp->address and tp->module */
> - if (tev->uprobes && (!tp->address || !tp->module))
> + /* Uprobes must have tp->module */
> + if (tev->uprobes && !tp->module)
> goto error;
> + /*
> + * If tp->address == 0, then this point must be a
> + * absolute address uprobe.
> + * try_to_find_absolute_address() should have made
> + * tp->symbol to "0x0".
> + */
> + if (tev->uprobes && !tp->address) {
> + if (!tp->symbol || strcmp(tp->symbol, "0x0"))
> + goto error;
> + }
>
> /* Use the tp->address for uprobes */
> if (tev->uprobes)
> ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s:0x%lx",
> tp->module, tp->address);
> + else if (!strncmp(tp->symbol, "0x", 2))
> + /* Absolute address. See try_to_find_absolute_address() */
> + ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s%s0x%lx",
> + tp->module ?: "", tp->module ? ":" : "",
> + tp->address);
> else
> ret = e_snprintf(buf + len, MAX_CMDLEN - len, "%s%s%s+%lu",
> tp->module ?: "", tp->module ? ":" : "",
> @@ -1872,8 +1909,8 @@ out:
> }
>
> static int convert_to_perf_probe_point(struct probe_trace_point *tp,
> - struct perf_probe_point *pp,
> - bool is_kprobe)
> + struct perf_probe_point *pp,
> + bool is_kprobe)
> {
> char buf[128];
> int ret;
> @@ -2334,7 +2371,9 @@ static int probe_trace_event__set_name(struct probe_trace_event *tev,
> if (pev->event)
> event = pev->event;
> else
> - if (pev->point.function && !strisglob(pev->point.function))
> + if (pev->point.function &&
> + (strncmp(pev->point.function, "0x", 2) != 0) &&
> + !strisglob(pev->point.function))
> event = pev->point.function;
> else
> event = tev->point.realname;
> @@ -2604,6 +2643,98 @@ err_out:
> goto out;
> }
>
> +static int try_to_find_absolute_address(struct perf_probe_event *pev,
> + struct probe_trace_event **tevs)
> +{
> + struct perf_probe_point *pp = &pev->point;
> + struct probe_trace_event *tev;
> + struct probe_trace_point *tp;
> + int i, err;
> +
> + if (!(pev->point.function && !strncmp(pev->point.function, "0x", 2)))
> + return -EINVAL;
> + if (perf_probe_event_need_dwarf(pev))
> + return -EINVAL;
> +
> + /*
> + * This is 'perf probe /lib/libc.so 0xabcd'. Try to probe at
> + * absolute address.
> + *
> + * Only one tev can be generated by this.
> + */
> + *tevs = zalloc(sizeof(*tev));
> + if (!*tevs)
> + return -ENOMEM;
> +
> + tev = *tevs;
> + tp = &tev->point;
> +
> + /*
> + * Don't use tp->offset, use address directly, because
> + * in synthesize_probe_trace_command() address cannot be
> + * zero.
> + */
> + tp->address = pev->point.abs_address;
> + tp->retprobe = pp->retprobe;
> + tev->uprobes = pev->uprobes;
> +
> + err = -ENOMEM;
> + /*
> + * Give it a '0x' leading symbol name.
> + * In __add_probe_trace_events, a NULL symbol is interpreted as
> + * invalud.
> + */
> + if (asprintf(&tp->symbol, "0x%lx", tp->address) < 0)
> + goto errout;
> +
> + /* For kprobe, check range */
> + if ((!tev->uprobes) &&
> + (kprobe_warn_out_range(tev->point.symbol,
> + tev->point.address))) {
> + err = -EACCES;
> + goto errout;
> + }
> +
> + if (asprintf(&tp->realname, "abs_%lx", tp->address) < 0)
> + goto errout;
> +
> + if (pev->target) {
> + tp->module = strdup(pev->target);
> + if (!tp->module)
> + goto errout;
> + }
> +
> + if (tev->group) {
> + tev->group = strdup(pev->group);
> + if (!tev->group)
> + goto errout;
> + }
> +
> + if (pev->event) {
> + tev->event = strdup(pev->event);
> + if (!tev->event)
> + goto errout;
> + }
> +
> + tev->nargs = pev->nargs;
> + tev->args = zalloc(sizeof(struct probe_trace_arg) * tev->nargs);
> + if (!tev->args) {
> + err = -ENOMEM;
> + goto errout;
> + }
> + for (i = 0; i < tev->nargs; i++)
> + copy_to_probe_trace_arg(&tev->args[i], &pev->args[i]);
> +
> + return 1;
> +
> +errout:
> + if (*tevs) {
> + clear_probe_trace_events(*tevs, 1);
> + *tevs = NULL;
> + }
> + return err;
> +}
> +
> bool __weak arch__prefers_symtab(void) { return false; }
>
> static int convert_to_probe_trace_events(struct perf_probe_event *pev,
> @@ -2620,6 +2751,10 @@ static int convert_to_probe_trace_events(struct perf_probe_event *pev,
> }
> }
>
> + ret = try_to_find_absolute_address(pev, tevs);
> + if (ret > 0)
> + return ret;
> +
> if (arch__prefers_symtab() && !perf_probe_event_need_dwarf(pev)) {
> ret = find_probe_trace_events_from_map(pev, tevs);
> if (ret > 0)
> @@ -2787,3 +2922,22 @@ end:
> return ret;
> }
>
> +int copy_to_probe_trace_arg(struct probe_trace_arg *tvar,
> + struct perf_probe_arg *pvar)
> +{
> + tvar->value = strdup(pvar->var);
> + if (tvar->value == NULL)
> + return -ENOMEM;
> + if (pvar->type) {
> + tvar->type = strdup(pvar->type);
> + if (tvar->type == NULL)
> + return -ENOMEM;
> + }
> + if (pvar->name) {
> + tvar->name = strdup(pvar->name);
> + if (tvar->name == NULL)
> + return -ENOMEM;
> + } else
> + tvar->name = NULL;
> + return 0;
> +}
> diff --git a/tools/perf/util/probe-event.h b/tools/perf/util/probe-event.h
> index 9c7b99e..de0dd13 100644
> --- a/tools/perf/util/probe-event.h
> +++ b/tools/perf/util/probe-event.h
> @@ -61,6 +61,7 @@ struct perf_probe_point {
> bool retprobe; /* Return probe flag */
> char *lazy_line; /* Lazy matching pattern */
> unsigned long offset; /* Offset from function entry */
> + unsigned long abs_address; /* Absolute address of the point */
> };
>
> /* Perf probe probing argument field chain */
> @@ -162,4 +163,7 @@ int e_snprintf(char *str, size_t size, const char *format, ...)
> /* Maximum index number of event-name postfix */
> #define MAX_EVENT_INDEX 1024
>
> +int copy_to_probe_trace_arg(struct probe_trace_arg *tvar,
> + struct perf_probe_arg *pvar);
> +
> #endif /*_PROBE_EVENT_H */
> diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
> index 4b1e074..5ab9cd6 100644
> --- a/tools/perf/util/probe-finder.c
> +++ b/tools/perf/util/probe-finder.c
> @@ -553,24 +553,9 @@ static int find_variable(Dwarf_Die *sc_die, struct probe_finder *pf)
> char buf[32], *ptr;
> int ret = 0;
>
> - if (!is_c_varname(pf->pvar->var)) {
> - /* Copy raw parameters */
> - pf->tvar->value = strdup(pf->pvar->var);
> - if (pf->tvar->value == NULL)
> - return -ENOMEM;
> - if (pf->pvar->type) {
> - pf->tvar->type = strdup(pf->pvar->type);
> - if (pf->tvar->type == NULL)
> - return -ENOMEM;
> - }
> - if (pf->pvar->name) {
> - pf->tvar->name = strdup(pf->pvar->name);
> - if (pf->tvar->name == NULL)
> - return -ENOMEM;
> - } else
> - pf->tvar->name = NULL;
> - return 0;
> - }
> + /* Copy raw parameters */
> + if (!is_c_varname(pf->pvar->var))
> + return copy_to_probe_trace_arg(pf->tvar, pf->pvar);
>
> if (pf->pvar->name)
> pf->tvar->name = strdup(pf->pvar->name);
> --
> 1.8.3.4
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
> From: Wang Nan [mailto:[email protected]]
>
> When manually added uprobe point with zero address, 'uprobe_events' output
> '(null)' instead of 0x00000000:
>
> # echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
> /sys/kernel/debug/tracing/uprobe_events
>
> # cat /sys/kernel/debug/tracing/uprobe_events
> p:probe_libc/abs_0 /path/to/lib.bin:0x (null) arg1=%ax
>
> This patch fixes this behavior:
>
> # cat /sys/kernel/debug/tracing/uprobe_events
> p:probe_libc/abs_0 /path/to/lib.bin:0x0000000000000000
Looks Good to me :)
Acked-by: Masami Hiramatsu <[email protected]>
Thanks!
>
> Signed-off-by: Wang Nan <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> Cc: Masami Hiramatsu <[email protected]>
> Cc: Namhyung Kim <[email protected]>
> Cc: Steven Rostedt <[email protected]>
> ---
> kernel/trace/trace_uprobe.c | 17 ++++++++++++++++-
> 1 file changed, 16 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/trace/trace_uprobe.c b/kernel/trace/trace_uprobe.c
> index f97479f..d2f6d0b 100644
> --- a/kernel/trace/trace_uprobe.c
> +++ b/kernel/trace/trace_uprobe.c
> @@ -601,7 +601,22 @@ static int probes_seq_show(struct seq_file *m, void *v)
>
> seq_printf(m, "%c:%s/%s", c, tu->tp.call.class->system,
> trace_event_name(&tu->tp.call));
> - seq_printf(m, " %s:0x%p", tu->filename, (void *)tu->offset);
> + seq_printf(m, " %s:", tu->filename);
> +
> + /* Don't print "0x (null)" when offset is 0 */
> + if (tu->offset) {
> + seq_printf(m, "0x%p", (void *)tu->offset);
> + } else {
> + switch (sizeof(void *)) {
> + case 4:
> + seq_printf(m, "0x00000000");
> + break;
> + case 8:
> + default:
> + seq_printf(m, "0x0000000000000000");
> + break;
> + }
> + }
>
> for (i = 0; i < tu->tp.nr_args; i++)
> seq_printf(m, " %s=%s", tu->tp.args[i].name, tu->tp.args[i].comm);
> --
> 1.8.3.4
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?
> From: Wangnan (F) [mailto:[email protected]]
> >>
> >> When manually added uprobe point with zero address, 'perf probe -l'
> >> reports error. For example:
> >>
> >> # echo p:probe_libc/abs_0 /path/to/lib.bin:0x0 arg1=%ax > \
> >> /sys/kernel/debug/tracing/uprobe_events
> >>
> >> # perf probe -l
> >> Error: Failed to show event list.
> >>
> >> Probing at 0x0 is possible and useful when lib.bin is not a normal
> >> shared object but is manually mapped. However, in this case kernel
> >> report:
> >>
> >> # cat /sys/kernel/debug/tracing/uprobe_events
> >> p:probe_libc/abs_0 /tmp/oxygen_root/lib64/libc-2.18.so:0x (null) arg1=%ax
> >>
> >> This patch supports the above kernel output.
> > Former part is OK to me.
> >
> >> In addition, this patch also fixes convert_to_perf_probe_point() that,
> >> if failed to find probe point from both of dwarf and map, in all cases
> >> pp->function should be filled with a strdup() result, or the following
> >> ENOMEM error code is incorrect. Remove !tp->module && !is_kprobe
> >> condition so always use address to build function name if symbol is not
> >> found.
> > No, I think this should be included in 3/4, since until that perf probe
> > doesn't support absolute address on kernel.
>
> This is for manually introduced probing points.
>
> convert_to_perf_probe_point is called by:
>
> perf probe -l -> show_perf_probe_events -> __show_perf_probe_events --> convert_to_perf_probe_event
>
> If there is a manually inserted probing point there, 'perf probe -l' reports an error.
>
> Here is the result if I remove the second part:
>
> $ git log --oneline # I'm in 2/4
> aae2c2c perf probe: Fix list result when address is zero
> 931dff5 perf probe: Prevent segfault when reading probe point with
> absolute address
> ...
>
> $ git diff # remove the second part
> diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
> index 926bcec..de2c27e 100644
> --- a/tools/perf/util/probe-event.c
> +++ b/tools/perf/util/probe-event.c
> @@ -1888,7 +1888,7 @@ static int convert_to_perf_probe_point(struct
> probe_trace_point *tp,
> if (tp->symbol) {
> pp->function = strdup(tp->symbol);
> pp->offset = tp->offset;
> - } else {
> + } else if (!tp->module && !is_kprobe) {
> ret = e_snprintf(buf, 128, "0x%" PRIx64, (u64)tp->address);
> if (ret < 0)
> return ret;
>
>
> ...
> # echo 'p:probe_libc/abs_5 /lib64/libc.so.6:0x5' >
> /sys/kernel/debug/tracing/uprobe_events
> # cat /sys/kernel/debug/tracing/uprobe_events
> p:probe_libc/abs_5 /lib64/libc.so.6:0x0000000000000005
> # perf probe -l
> Error: Failed to show event list
>
> Then add it back:
>
> $ git reset --hard HEAD
> ...
> # perf probe -l
> probe_libc:abs_5 (on 0x5 in /lib64/libc.so.6)
Hmm, OK, then this should be in the 1/4 or splitted as an independent patch,
because this part it is not "when address is zero", just support showing
absolute address probes on user applications.
Thank you,
????{.n?+???????+%?????ݶ??w??{.n?+????{??G?????{ay? ʇڙ?,j??f???h?????????z_??(?階?ݢj"???m??????G????????????&???~???iO???z??v?^?m???? ????????I?