On Mon, 2011-05-09 at 21:49 +0200, Karel Zak wrote:
> sched_getaffinity(pid, ....);
> print_affinity(...., tid, ....);
>
> Do you see the bug? You have to call sched_getaffinity() for each
> "tid", otherwise you print the same cpumask for all tasks.
>
> [...]
>
> > + if (all_tasks) {
> > + pid_t tid;
> > + struct proc_tasks *ts = proc_open_tasks(pid);
> > +
> > + if (!ts)
> > + err(EXIT_FAILURE, "cannot obtain the list of tasks");
> > + while (!proc_next_tid(ts, &tid))
> > + if (sched_setaffinity(tid, new_setsize, new_set) < 0)
> > + err(EXIT_FAILURE, _("failed to set tid %d's affinity"), tid);
> > + proc_close_tasks(ts);
> > + }
> > + else
> > + if (sched_setaffinity(pid, new_setsize, new_set) < 0)
> > + err(EXIT_FAILURE, _("failed to set pid %d's affinity"), pid);
>
> [...]
>
> The another problem is that the code uses two separate loops to print
> and set the masks, so it's possible that you will call sched_getaffinity()
> for completely different set of tasks than sched_setaffinity(). It would
> be better to use:
>
> do_taskset(pid_t tid, ...)
> {
> sched_getaffinity(tid, ...); /* current */
> print_affinity(tid, ...);
>
> sched_setaffinity(tid, ...); /* new */
> print_affinity(tid, ...);
> }
>
> main()
> {
> if (all_tasks) {
> struct proc_tasks *ts = proc_open_tasks(pid);
> while (!proc_next_tid(ts, &tid))
> do_taskset(tid, ...);
> proc_close_tasks(ts);
> } else
> do_taskset(pid, ...);
> }
>
> The output will be also more readable:
>
> pid 100's current affinity: FOO
> pid 100's new affinity: BAR
> pid 200's current affinity: FOO
> pid 200's new affinity: BAR
> ...
Ok, good observations. Below is version 2.
From: Davidlohr Bueso <dave@xxxxxxx>
Date: Tue, 10 May 2011 17:57:22 -0400
Subject: [PATCH] taskset: make threads aware
Add a new '-a' option to view/modify the CPU affinity for an entire group of threads belonging to a given PID.
We create two new functions, print_affinity() and do_taskset() for code simplification.
Example:
zeus@jilguero:~/src/util-linux/schedutils$ ./taskset -a -p 01 3142
pid 3142's current affinity mask: 2
pid 3142's new affinity mask: 1
pid 3164's current affinity mask: 2
pid 3164's new affinity mask: 1
pid 854's current affinity mask: 2
pid 854's new affinity mask: 1
Signed-off-by: Davidlohr Bueso <dave@xxxxxxx>
Tested-by: Jonathan Gonzalez <zeus@xxxxxxx>
---
schedutils/taskset.1 | 3 ++
schedutils/taskset.c | 90 +++++++++++++++++++++++++++++++++----------------
2 files changed, 63 insertions(+), 30 deletions(-)
diff --git a/schedutils/taskset.1 b/schedutils/taskset.1
index f374b62..4aa5e2c 100644
--- a/schedutils/taskset.1
+++ b/schedutils/taskset.1
@@ -72,6 +72,9 @@ returns, it is guaranteed that the given program has been scheduled to a legal
CPU.
.SH OPTIONS
.TP
+.BR \-a ,\ \-\-all-tasks
+set or retrieve the CPU affinity of all the tasks (threads) for a given PID.
+.TP
.BR \-p ,\ \-\-pid
operate on an existing PID and not launch a new task
.TP
diff --git a/schedutils/taskset.c b/schedutils/taskset.c
index a861a90..357ab4c 100644
--- a/schedutils/taskset.c
+++ b/schedutils/taskset.c
@@ -28,6 +28,7 @@
#include "cpuset.h"
#include "nls.h"
#include "strutils.h"
+#include "procutils.h"
#include "c.h"
static void __attribute__((__noreturn__)) usage(FILE *out)
@@ -38,6 +39,7 @@ static void __attribute__((__noreturn__)) usage(FILE *out)
fprintf(out, _(
"Options:\n"
+ " -a, --all-tasks operate on all the tasks (threads) for a given pid\n"
" -p, --pid operate on existing given pid\n"
" -c, --cpu-list display and specify cpus in list format\n"
" -h, --help display this help\n"
@@ -61,16 +63,57 @@ static void __attribute__((__noreturn__)) usage(FILE *out)
exit(out == stderr ? EXIT_FAILURE : EXIT_SUCCESS);
}
+static void print_affinity(char *state, int c_opt, pid_t pid, char *buf, size_t buflen,
+ cpu_set_t *set, size_t setsize)
+{
+ char *cpuset, *str;
+
+ if (c_opt) {
+ cpuset = cpulist_create(buf, buflen, set, setsize);
+ str = "list";
+ } else {
+ cpuset = cpumask_create(buf, buflen, set, setsize);
+ str = "mask";
+ }
+
+ printf(_("pid %d's %s affinity %s: %s\n"), pid, state, str, cpuset);
+
+}
+
+static void do_taskset(pid_t pid, int get_only, cpu_set_t *new_set, size_t new_setsize,
+ cpu_set_t *cur_set, size_t cur_setsize, int c_opt, char *buf, size_t buflen)
+{
+ if (pid) {
+ /* current */
+ if (sched_getaffinity(pid, cur_setsize, cur_set) < 0)
+ err(EXIT_FAILURE, _("failed to get pid %d's affinity"), pid);
+ print_affinity("current", c_opt, pid, buf, buflen, cur_set, cur_setsize);
+ }
+
+ if (get_only)
+ return;
+
+ /* new */
+ if (sched_setaffinity(pid, new_setsize, new_set) < 0)
+ err(EXIT_FAILURE, _("failed to set pid %d's affinity"), pid);
+ if (pid) {
+ if (sched_getaffinity(pid, cur_setsize, cur_set) < 0)
+ err(EXIT_FAILURE, _("failed to get pid %d's affinity"), pid);
+ print_affinity("new", c_opt, pid, buf, buflen, cur_set, cur_setsize);
+ }
+}
+
int main(int argc, char *argv[])
{
cpu_set_t *new_set, *cur_set;
pid_t pid = 0;
- int opt, c_opt = 0, rc;
+ int opt, c_opt = 0, rc, get_only = 0, all_tasks = 0;
char *buf;
unsigned int ncpus;
size_t new_setsize, cur_setsize, cur_nbits, buflen;
static const struct option longopts[] = {
+ { "all-tasks", 0, NULL, 'a' },
{ "pid", 0, NULL, 'p' },
{ "cpu-list", 0, NULL, 'c' },
{ "help", 0, NULL, 'h' },
@@ -82,8 +125,11 @@ int main(int argc, char *argv[])
bindtextdomain(PACKAGE, LOCALEDIR);
textdomain(PACKAGE);
- while ((opt = getopt_long(argc, argv, "+pchV", longopts, NULL)) != -1) {
+ while ((opt = getopt_long(argc, argv, "+apchV", longopts, NULL)) != -1) {
switch (opt) {
+ case 'a':
+ all_tasks = 1;
+ break;
case 'p':
pid = strtol_or_err(argv[argc - 1], _("failed to parse pid"));
break;
@@ -133,20 +179,8 @@ int main(int argc, char *argv[])
if (!new_set)
err(EXIT_FAILURE, _("cpuset_alloc failed"));
- if (pid) {
- if (sched_getaffinity(pid, cur_setsize, cur_set) < 0)
- err(EXIT_FAILURE, _("failed to get pid %d's affinity"), pid);
-
- if (c_opt)
- printf(_("pid %d's current affinity list: %s\n"), pid,
- cpulist_create(buf, buflen, cur_set, cur_setsize));
- else
- printf(_("pid %d's current affinity mask: %s\n"), pid,
- cpumask_create(buf, buflen, cur_set, cur_setsize));
-
- if (argc - optind == 1)
- return EXIT_SUCCESS;
- }
+ if (argc - optind == 1)
+ get_only = 1;
rc = c_opt ? cpulist_parse(argv[optind], new_set, new_setsize) :
cpumask_parse(argv[optind], new_set, new_setsize);
@@ -156,20 +190,16 @@ int main(int argc, char *argv[])
c_opt ? _("CPU list") : _("CPU mask"),
argv[optind]);
- if (sched_setaffinity(pid, new_setsize, new_set) < 0)
- err(EXIT_FAILURE, _("failed to set pid %d's affinity"), pid);
-
- if (sched_getaffinity(pid, cur_setsize, cur_set) < 0)
- err(EXIT_FAILURE, _("failed to get pid %d's affinity"), pid);
-
- if (pid) {
- if (c_opt)
- printf(_("pid %d's new affinity list: %s\n"), pid,
- cpulist_create(buf, buflen, cur_set, cur_setsize));
- else
- printf(_("pid %d's new affinity mask: %s\n"), pid,
- cpumask_create(buf, buflen, cur_set, cur_setsize));
- }
+ if (all_tasks) {
+ pid_t tid;
+ struct proc_tasks *ts = proc_open_tasks(pid);
+ while (!proc_next_tid(ts, &tid))
+ do_taskset(tid, get_only, new_set, new_setsize, cur_set, cur_setsize,
+ c_opt, buf, buflen);
+ proc_close_tasks(ts);
+ } else
+ do_taskset(pid, get_only, new_set, new_setsize, cur_set, cur_setsize,
+ c_opt, buf, buflen);
free(buf);
cpuset_free(cur_set);
--
1.7.4.1
--
To unsubscribe from this list: send the line "unsubscribe util-linux" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
