login
Search
Hi Ingo,
Here are several updates of perf-probe. This series improves
data structure accessing. In this version, I added 'removing die()'
patches.
- Set the name of argument which traces a data structure member
as the last member of the data structure reference (e.g. f_mode
of file->f_mode). This allows us to use perf-trace for tracin
data-structure members.
- Add the basic type support. This allows us to fetch the memory
with specified bitwidth. Usually, data-structure members are
packed on the memory, this means if we want to read a member
from memory, we have to access it with type casting.
kprobe-tracer now support tracing argument with basic types
(u8,u16,u32,u64,s8,s16,s32,s64), and perf-probe decodes the type
information of the members.
- Support canonical frame address, which is used for refering
frame-base on the kernel built without CONFIG_FRAME_POINTER.
- Remove almost all die()s from util code, according to acme's
advice. Thanks!
TODOs (possible features):
- Support array element (var[N])
- Support string/dynamic arrays (*var, var[N..M])
- Support tracing static variables (non global)
- Support dynamic array-indexing (var[var2])
- Support force type-casting ((type)var)
- Show what deta-structure member is assigned to each argument.
- Better support for probes on modules
- More debugger like enhancements(%next, --disasm, etc.)
Thank you,
---
Masami Hiramatsu (8):
perf probe: Remove die() from probe-event code
perf probe: Remove die() from probe-finder code
perf probe: Support DW_OP_call_frame_cfa in debuginfo
perf probe: Support basic type casting
perf probe: Query basic types from debuginfo
tracing/kprobes: Support basic types on dynamic events
perf probe: Use the last field name as the argument name
perf probe: Support argument name
Documentation/trace/kprobetrace.txt | 4
kernel/trace/trace.h | 16 -
...Set given names to event arguments. The syntax is same as kprobe-tracer, you can add 'NAME=' right before each argument. e.g. ./perf probe vfs_read foo=file then, 'foo' is set to the argument name as below. ./perf probe -l probe:vfs_read (on vfs_read@linux-2.6-tip/fs/read_write.c with foo) Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> --- tools/perf/Documentation/perf-probe.txt | 10 ++++++++- tools/perf/builtin-probe.c | 4 ++-- tools/perf/util/probe-event.c | 34 ++++++++++++++++++++++--------- tools/perf/util/probe-event.h | 1 + tools/perf/util/probe-finder.c | 27 +++++++++++++++---------- 5 files changed, 52 insertions(+), 24 deletions(-) diff --git a/tools/perf/Documentation/perf-probe.txt b/tools/perf/Documentation/perf-probe.txt index bb671b3..e36ed4d 100644 --- a/tools/perf/Documentation/perf-probe.txt +++ b/tools/perf/Documentation/perf-probe.txt @@ -79,7 +79,15 @@ Probe points are defined by following syntax. 'EVENT' specifies the name of new event, if omitted, it will be set the name of the probed function. Currently, event group name is set as 'probe'. 'FUNC' specifies a probed function name, and it may have one of the following options; '+OFFS' is the offset from function entry address in bytes, ':RLN' is the relative-line number from function entry line, and '%return' means that it probes function return. And ';PTN' means lazy matching pattern (see LAZY MATCHING). Note that ';PTN' must be the end of the probe point definition. In addition, '@SRC' specifies a source file which has that function. It is also possible to specify a probe point by the source line number or lazy matching by using 'SRC:ALN' or 'SRC;PTN' syntax, where 'SRC' is the ...
Set the last field name to the argument name when the argument is refering a data-structure member. e.g. ./perf probe --add 'vfs_read file->f_mode' Add new event: probe:vfs_read (on vfs_read with f_mode=file->f_mode) This probe records file->f_mode, but the argument name becomes "f_mode". This enables perf-trace command to parse trace event format correctly. Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> --- tools/perf/Documentation/perf-probe.txt | 2 +- tools/perf/util/probe-event.c | 4 ++++ 2 files changed, 5 insertions(+), 1 deletions(-) diff --git a/tools/perf/Documentation/perf-probe.txt b/tools/perf/Documentation/perf-probe.txt index e36ed4d..441324f 100644 --- a/tools/perf/Documentation/perf-probe.txt +++ b/tools/perf/Documentation/perf-probe.txt @@ -87,7 +87,7 @@ Each probe argument follows below syntax. [NAME=]LOCALVAR|$retval|%REG|@SYMBOL -'NAME' specifies the name of this argument (optional). You can use the name of local variable, local data structure member (e.g. var->field, var.field2), or kprobe-tracer argument format (e.g. $retval, %ax, etc). +'NAME' specifies the name of this argument (optional). You can use the name of local variable, local data structure member (e.g. var->field, var.field2), or kprobe-tracer argument format (e.g. $retval, %ax, etc). Note that the name of this argument will be set as the last member name if you specify a local data structure member (e.g. field2 for 'var->field1.field2'.) LINE SYNTAX ----------- diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c index ab6f53d..19de8b7 100644 --- a/tools/perf/util/probe-event.c +++ b/tools/perf/util/probe-event.c @@ -481,6 +481,10 @@ static void parse_perf_probe_arg(const ...
Support basic types of integer (u8, u16, u32, u64, s8, s16, s32, s64) in
kprobe tracer. With this patch, users can specify above basic types on
each arguments after ':'. If omitted, the argument type is set as
unsigned long (u32 or u64, arch-dependent).
e.g.
echo 'p account_system_time+0 hardirq_offset=%si:s32' > kprobe_events
adds a probe recording hardirq_offset in signed-32bits value on the
entry of account_system_time.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
---
Documentation/trace/kprobetrace.txt | 4
kernel/trace/trace.h | 16 -
kernel/trace/trace_kprobe.c | 535 ++++++++++++++++++++++-------------
3 files changed, 334 insertions(+), 221 deletions(-)
diff --git a/Documentation/trace/kprobetrace.txt b/Documentation/trace/kprobetrace.txt
index a9100b2..ec94748 100644
--- a/Documentation/trace/kprobetrace.txt
+++ b/Documentation/trace/kprobetrace.txt
@@ -40,7 +40,9 @@ Synopsis of kprobe_events
$stack : Fetch stack address.
$retval : Fetch return value.(*)
+|-offs(FETCHARG) : Fetch memory at FETCHARG +|- offs address.(**)
- NAME=FETCHARG: Set NAME as the argument name of FETCHARG.
+ NAME=FETCHARG : Set NAME as the argument name of FETCHARG.
+ FETCHARG:TYPE : Set TYPE as the type of FETCHARG. Currently, basic types
+ (u8/u16/u32/u64/s8/s16/s32/s64) are supported.
(*) only for return probe.
(**) this is useful for fetching a field of data structures.
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index bec2c97..3ebdb6b 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -102,29 +102,17 @@ struct syscall_trace_exit {
long ret;
};
-struct kprobe_trace_entry {
+struct ...Add basic type casting for arguments to perf probe. This allows users to specify the actual type of arguments. Of course, if user sets invalid types, kprobe-tracer rejects that. Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Mike Galbraith <efault@gmx.de> Cc: Frederic Weisbecker <fweisbec@gmail.com> --- tools/perf/Documentation/perf-probe.txt | 3 ++- tools/perf/util/probe-event.c | 23 ++++++++++++++++++++++- tools/perf/util/probe-event.h | 1 + tools/perf/util/probe-finder.c | 10 ++++++++-- 4 files changed, 33 insertions(+), 4 deletions(-) diff --git a/tools/perf/Documentation/perf-probe.txt b/tools/perf/Documentation/perf-probe.txt index 441324f..63c25d3 100644 --- a/tools/perf/Documentation/perf-probe.txt +++ b/tools/perf/Documentation/perf-probe.txt @@ -85,9 +85,10 @@ PROBE ARGUMENT -------------- Each probe argument follows below syntax. - [NAME=]LOCALVAR|$retval|%REG|@SYMBOL + [NAME=]LOCALVAR|$retval|%REG|@SYMBOL[:TYPE] 'NAME' specifies the name of this argument (optional). You can use the name of local variable, local data structure member (e.g. var->field, var.field2), or kprobe-tracer argument format (e.g. $retval, %ax, etc). Note that the name of this argument will be set as the last member name if you specify a local data structure member (e.g. field2 for 'var->field1.field2'.) +'TYPE' casts the type of this argument (optional). If omitted, perf probe automatically set the type based on debuginfo. LINE SYNTAX ----------- diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c index 05ca4a9..bef2805 100644 --- a/tools/perf/util/probe-event.c +++ b/tools/perf/util/probe-event.c @@ -435,7 +435,7 @@ static void parse_perf_probe_point(char *arg, struct perf_probe_event *pev) } /* Parse perf-probe event argument ...
Remove die() and DIE_IF() code from util/probe-finder.c since
these 'sudden death' in utility functions make reusing it from
other code (especially tui/gui) difficult.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
---
tools/perf/util/probe-event.c | 4
tools/perf/util/probe-finder.c | 517 +++++++++++++++++++++++++---------------
2 files changed, 322 insertions(+), 199 deletions(-)
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index bef2805..7893b32 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -151,10 +151,10 @@ static int try_to_find_kprobe_trace_events(struct perf_probe_event *pev,
/* Error path */
if (need_dwarf) {
- if (ntevs == -ENOENT)
+ if (ntevs == -EBADF)
pr_warning("No dwarf info found in the vmlinux - "
"please rebuild with CONFIG_DEBUG_INFO=y.\n");
- die("Could not analyze debuginfo.");
+ die("Failed to analyze debuginfo.");
}
pr_debug("An error occurred in debuginfo analysis."
" Try to use symbols.\n");
diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
index 1f45285..4a57504 100644
--- a/tools/perf/util/probe-finder.c
+++ b/tools/perf/util/probe-finder.c
@@ -196,19 +196,7 @@ static bool die_compare_name(Dwarf_Die *dw_die, const char *tname)
{
const char *name;
name = dwarf_diename(dw_die);
- DIE_IF(name == NULL);
- return strcmp(tname, name);
-}
-
-/* Get entry pc(or low pc, 1st entry of ranges) of the die */
-static Dwarf_Addr die_get_entrypc(Dwarf_Die *dw_die)
-{
- Dwarf_Addr epc;
- int ret;
-
- ret = dwarf_entrypc(dw_die, &epc);
- DIE_IF(ret == -1);
- return epc;
+ return name ? strcmp(tname, name) : -1;
}
/* Get type die, but skip qualifiers and typedef ...Hi Arnaldo, Has this code done what you suggested? :) I'd like to have your comment. Thank you, -- Masami Hiramatsu e-mail: mhiramat@redhat.com --
It improves the current situation, yes, but there are still cases there where die() is called, I assume that is left for later, right? Like here, the TUI/GUI can try to add a probe but if it fails it can still continue providing things like a "perf top" window, analysing Yeah, there are still lots of places, in other areas that don't return a status, just printing a warning and silently failing, like in the trace Correct, if you look at tools/perf/util/debug.c, eprintf() looks if we're in TUI mode, use_browser is true and either fprintfs(stderr) if not or calls a routine to put it at the "status line" (bottom) in the TUI. Having a list with the last messages so that we can have a log window, or keeping it in a log file that would then be browser is an enhancement We have to kill those xzcalloc, etc, too they are die() in disguise :-) - Arnaldo --
With the next (8/8) patch, all die()s are removed at least from Sure, this die() is removed by next (8/8) patch. Sorry, I've split it because Hmm, I think that will cost high, because only failing to allocate memory, which theoretically means we can't continue to operate it. In that case, we'd better just use backtrace() and die(). Thank you, -- Masami Hiramatsu e-mail: mhiramat@redhat.com --
Consider a situation where we are trying to allocate lots of objects allocated for some big operation (adding probes for all functions in all threads, whatever), we can just say to the user "hey, you don't have memory to do this" but other operations are possible, so calling panic(), oops, die() is not the right thing to do. - Arnaldo --
Hmm, OK, so we'd better remove wrapper.c... Thank you, -- Masami Hiramatsu e-mail: mhiramat@redhat.com --
Yeah, that came from git, we'll eventually get rid of it, its just that I wanted to make clear the rationale behind avoiding spreading its use. - Arnaldo --
When building kernel without CONFIG_FRAME_POINTER, gcc uses
CFA (canonical frame address) for frame base. With this patch,
perf probe just gets CFI (call frame information) from debuginfo
and search corresponding CFA from the CFI. IOW, this allows
perf probe works correctly on the kernel without CONFIG_FRAME_POINTER.
<Before>
./perf probe -fn sched_slice:12 lw.weight
Fatal: DW_OP 156 is not supported.
(^^^ DW_OP_call_frame_cfa)
<After>
./perf probe -fn sched_slice:12 lw.weight
Add new event:
probe:sched_slice (on sched_slice:12 with weight=lw.weight)
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
---
tools/perf/util/probe-finder.c | 14 +++++++++++---
tools/perf/util/probe-finder.h | 1 +
2 files changed, 12 insertions(+), 3 deletions(-)
diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
index ab47673..1f45285 100644
--- a/tools/perf/util/probe-finder.c
+++ b/tools/perf/util/probe-finder.c
@@ -398,7 +398,6 @@ static void convert_location(Dwarf_Op *op, struct probe_finder *pf)
const char *regs;
struct kprobe_trace_arg *tvar = pf->tvar;
- /* TODO: support CFA */
/* If this is based on frame buffer, set the offset */
if (op->atom == DW_OP_fbreg) {
if (pf->fb_ops == NULL)
@@ -629,11 +628,17 @@ static void convert_probe_point(Dwarf_Die *sp_die, struct probe_finder *pf)
/* Get the frame base attribute/ops */
dwarf_attr(sp_die, DW_AT_frame_base, &fb_attr);
ret = dwarf_getlocation_addr(&fb_attr, pf->addr, &pf->fb_ops, &nops, 1);
- if (ret <= 0 || nops == 0)
+ if (ret <= 0 || nops == 0) {
pf->fb_ops = NULL;
+ } else if (nops == 1 && pf->fb_ops[0].atom == DW_OP_call_frame_cfa &&
+ pf->cfi != NULL) {
+ Dwarf_Frame *frame;
+ ret = ...Remove die() and DIE_IF() code from util/probe-event.c since
these 'sudden death' in utility functions make reusing it from
other code (especially tui/gui) difficult.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
---
tools/perf/builtin-probe.c | 70 +++--
tools/perf/util/probe-event.c | 615 ++++++++++++++++++++++++++++-------------
tools/perf/util/probe-event.h | 24 +-
3 files changed, 480 insertions(+), 229 deletions(-)
diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
index dafaf76..5259c5a 100644
--- a/tools/perf/builtin-probe.c
+++ b/tools/perf/builtin-probe.c
@@ -59,23 +59,25 @@ static struct {
/* Parse an event definition. Note that any error must die. */
-static void parse_probe_event(const char *str)
+static int parse_probe_event(const char *str)
{
struct perf_probe_event *pev = &params.events[params.nevents];
+ int ret;
pr_debug("probe-definition(%d): %s\n", params.nevents, str);
if (++params.nevents == MAX_PROBES)
die("Too many probes (> %d) are specified.", MAX_PROBES);
/* Parse a perf-probe command into event */
- parse_perf_probe_command(str, pev);
-
+ ret = parse_perf_probe_command(str, pev);
pr_debug("%d arguments\n", pev->nargs);
+
+ return ret;
}
-static void parse_probe_event_argv(int argc, const char **argv)
+static int parse_probe_event_argv(int argc, const char **argv)
{
- int i, len;
+ int i, len, ret;
char *buf;
/* Bind up rest arguments */
@@ -86,16 +88,18 @@ static void parse_probe_event_argv(int argc, const char **argv)
len = 0;
for (i = 0; i < argc; i++)
len += sprintf(&buf[len], "%s ", argv[i]);
- parse_probe_event(buf);
+ ret = parse_probe_event(buf);
free(buf);
+ return ret;
}
static int ...Query the basic type information (byte-size and signed-flag) from
debuginfo and pass that to kprobe-tracer. This is especially useful
for tracing the members of data structure, because each member has
different byte-size on the memory.
Signed-off-by: Masami Hiramatsu <mhiramat@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
---
tools/perf/util/probe-event.c | 9 +++++
tools/perf/util/probe-event.h | 1 +
tools/perf/util/probe-finder.c | 78 ++++++++++++++++++++++++++++++++++++----
3 files changed, 80 insertions(+), 8 deletions(-)
diff --git a/tools/perf/util/probe-event.c b/tools/perf/util/probe-event.c
index 19de8b7..05ca4a9 100644
--- a/tools/perf/util/probe-event.c
+++ b/tools/perf/util/probe-event.c
@@ -740,6 +740,13 @@ static int synthesize_kprobe_trace_arg(struct kprobe_trace_arg *arg,
buf += ret;
buflen -= ret;
}
+ /* Print argument type */
+ if (arg->type) {
+ ret = e_snprintf(buf, buflen, ":%s", arg->type);
+ if (ret <= 0)
+ return ret;
+ buf += ret;
+ }
return buf - tmp;
}
@@ -848,6 +855,8 @@ void clear_kprobe_trace_event(struct kprobe_trace_event *tev)
free(tev->args[i].name);
if (tev->args[i].value)
free(tev->args[i].value);
+ if (tev->args[i].type)
+ free(tev->args[i].type);
ref = tev->args[i].ref;
while (ref) {
next = ref->next;
diff --git a/tools/perf/util/probe-event.h b/tools/perf/util/probe-event.h
index a73ede6..0759db6 100644
--- a/tools/perf/util/probe-event.h
+++ b/tools/perf/util/probe-event.h
@@ -23,6 +23,7 @@ struct kprobe_trace_arg_ref {
struct kprobe_trace_arg {
char *name; /* Argument name */
char *value; /* Base value */
+ char *type; /* Type name */
struct kprobe_trace_arg_ref *ref; /* Referencing offset */
};
diff --git ...
