utils: enable @k as an abbreviated form of @kernel

[DanielTimLee]

this commit allows passing '@k' for kernel option as an abbreviated form of '@kernel'

created a util function for strstr(buf, '@k')

// before
$uftrace -K 2 -F sys_openat@kernel --force ./echo-sync
// after
$uftrace -K 2 -F sys_openat@k --force ./echo-sync

Signed-off-by: Daniel T. Lee [email protected]

[namhyung]

And you need to change setup_trigger() function to handle kernel filters/symbols separately.

[DanielTimLee]

I'm really sorry for the late response.

I might not following you, but the thing

change setup_trigger() function to handle kernel filters/symbols separately.

at here, the kernel filter/symbols you're mentioning about is it related to module in setup_trigger():utils/fitler.c?

uftrace/utils/filter.c

Lines 981 to 1017 in 2aac509

	if (module) {
	/* is it the main executable? */
	if (!strncmp(module, basename(symtabs->filename),
	strlen(module))) {
	ret += add_trigger_entry(root, &symtabs->symtab,
	&patt, &tr,
	&symtabs->dinfo,
	setting);
	ret += add_trigger_entry(root, &symtabs->dsymtab,
	&patt, &tr,
	&symtabs->dinfo,
	setting);
	}
	else if (!strcasecmp(module, "PLT")) {
	ret = add_trigger_entry(root, &symtabs->dsymtab,
	&patt, &tr,
	&symtabs->dinfo,
	setting);
	}
	else if (!strcasecmp(module, "kernel")) {
	ret = add_trigger_entry(root, get_kernel_symtab(),
	&patt, &tr,
	&symtabs->dinfo,
	setting);
	}
	else {
	map = find_map_by_name(symtabs, module);
	if (map) {
	ret = add_trigger_entry(root, &map->symtab,
	&patt, &tr,
	&map->dinfo,
	setting);
	}
	}
	free(module);
	}

(ps. I've updated code with 'comparing whole word' )

[namhyung]

Yes, module is a name of executable or library. The module name should come first after the @ sign but it can also have other parts like trigger actions.

[honggyukim]

It's not because of this change, but there is a different behavior between -F sys_write@kernel and -T sys_write@kernel,filter as follows:

$ uftrace -K 2 -F sys_write@kernel --force /usr/bin/uptime
 23:40:15 up 30 days, 12:51, 41 users,  load average: 0.34, 0.32, 0.19
# DURATION     TID     FUNCTION
 124.261 us [ 12374] | setlocale();
   2.400 us [ 12374] | bindtextdomain();
   1.667 us [ 12374] | textdomain();
   1.150 us [ 12374] | __cxa_atexit();
  15.070 us [ 12374] | getopt_long();
   2.416 ms [ 12374] | print_uptime();
   1.590 us [ 12374] | __fpending();
   1.086 us [ 12374] | ferror();
            [ 12374] | fclose() {
            [ 12374] |   sys_write() {
   0.870 us [ 12374] |     __fdget_pos();
  15.964 us [ 12374] |     vfs_write();
  33.307 us [ 12374] |   } /* sys_write */
  39.217 us [ 12374] | } /* fclose */
   0.207 us [ 12374] | __fpending();
   0.230 us [ 12374] | ferror();
   1.690 us [ 12374] | fclose();

Unlike -F option, -T sys_write@kernel,filter hides user functions as well.

$ uftrace -K 2 -T sys_write@kernel,filter --force /usr/bin/uptime
 23:40:28 up 30 days, 12:51, 41 users,  load average: 0.37, 0.32, 0.19
# DURATION     TID     FUNCTION
            [ 12409] | sys_write() {
   0.877 us [ 12409] |   __fdget_pos();
  12.154 us [ 12409] |   vfs_write();
  18.764 us [ 12409] | } /* sys_write */

[namhyung]

Yes, because the trigger also works for replay. Maybe it's better to remove reset_live_opts()?

[DanielTimLee]

updated.

[honggyukim]

Yes, because the trigger also works for replay. Maybe it's better to remove reset_live_opts()?

The replay output looks same if remove_live_opts() is not called as follows:

$ uftrace -K 2 -F sys_write@kernel --force /usr/bin/uptime
 14:54:18 up 34 days,  4:05, 46 users,  load average: 0.41, 0.18, 0.12
# DURATION     TID     FUNCTION
            [175126] | sys_write() {
   0.907 us [175126] |   __fdget_pos();
  20.483 us [175126] |   vfs_write();
  36.657 us [175126] | } /* sys_write */

$ uftrace -K 2 -T sys_write@kernel,filter --force /usr/bin/uptime
 14:54:26 up 34 days,  4:05, 46 users,  load average: 0.42, 0.19, 0.12
# DURATION     TID     FUNCTION
            [175156] | sys_write() {
   0.704 us [175156] |   __fdget_pos();
  22.287 us [175156] |   vfs_write();
  28.407 us [175156] | } /* sys_write */

[honggyukim]

I applied @DanielTimLee's update and the result looks good.

[namhyung]

LGTM