Skip to contentSkip to conversation

kernel: 'now' may be used uninitialized in posix_cpu_timer_schedule function

Suggestion Chen Gang 2 participants 4 messages Mar 26, 2013 Arm Samsung Exynos

lore
Share & tools
lore

Chen Gang

author
#0 lore raw

Hello Maintainers:

compiling with EXTRA_CFLAGS=-W:
make V=1 EXTRA_CFLAGS=-W ARCH=arm s3c2410_defconfig make V=1 EXTRA_CFLAGS=-W ARCH=arm menuconfig
set 'arm-linux-gnu-' for cross chain prefix
make V=1 EXTRA_CFLAGS=-W ARCH=arm

it will report:
kernel/posix-cpu-timers.c:1065:19: warning: ‘now’ may be used uninitialized in this function [-Wuninitialized]

it seems it is really a bug.
can any member help to fix it ?
or provide additional suggestion ?
(it seems only "unsigned long long now = 0" is not enough).

:-)

in kernel/posix-cpu-timers.c:
for variable 'now' is defined without initialization (line 1029) it may be used without initialization (line 1066)

1026 void posix_cpu_timer_schedule(struct k_itimer *timer) 1027 {
1028 struct task_struct *p = timer->it.cpu.task;
1029 unsigned long long now;
1030
1031 if (unlikely(p == NULL))
1032 /*
1033 * The task was cleaned up already, no future firings.
1034 */
1035 goto out;
1036
1037 /*
1038 * Fetch the current sample and update the timer's expiry time.
1039 */
1040 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {

1041                 cpu_clock_sample(timer->it_clock, p, &now);
1042                 bump_cpu_timer(timer, now);
1043                 if (unlikely(p->exit_state)) {
1044                         clear_dead_task(timer, now);
1045                         goto out;
1046                 }
1047                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1048                 spin_lock(&p->sighand->siglock);

1049 } else {
1050 read_lock(&tasklist_lock);
1051 if (unlikely(p->sighand == NULL)) {
1052 /*

1053                          * The process has been reaped.
1054                          * We can't even collect a sample any more.
1055                          */
1056                         put_task_struct(p);
1057                         timer->it.cpu.task = p = NULL;
1058                         timer->it.cpu.expires = 0;
1059                         goto out_unlock;
1060                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1061                         /*
1062                          * We've noticed that the thread is dead, but
1063                          * not yet reaped.  Take this opportunity to
1064                          * drop our task ref.
1065                          */
1066                         clear_dead_task(timer, now);
1067                         goto out_unlock;
1068                 }
1069                 spin_lock(&p->sighand->siglock);
1070                 cpu_timer_sample_group(timer->it_clock, p, &now);
1071                 bump_cpu_timer(timer, now);
1072                 /* Leave the tasklist_lock locked for the call below.  */

1073 }
1074
1075 /*
1076 * Now re-arm for the new expiry time.
1077 */

1078         BUG_ON(!irqs_disabled());
1079         arm_timer(timer);
1080         spin_unlock(&p->sighand->siglock);

1081
1082 out_unlock:
1083 read_unlock(&tasklist_lock);
1084
1085 out:

1086         timer->it_overrun_last = timer->it_overrun;
1087         timer->it_overrun = -1;
1088         ++timer->it_requeue_pending;

1089 }

author
#1 lore raw

oh, sorry, it seems better to let ARM folks know about it.

;-)

Quoted 90 lines · Chen Gang

Hello Maintainers:

compiling with EXTRA_CFLAGS=-W:
make V=1 EXTRA_CFLAGS=-W ARCH=arm s3c2410_defconfig make V=1 EXTRA_CFLAGS=-W ARCH=arm menuconfig
set 'arm-linux-gnu-' for cross chain prefix
make V=1 EXTRA_CFLAGS=-W ARCH=arm

it will report:
kernel/posix-cpu-timers.c:1065:19: warning: ?now? may be used uninitialized in this function [-Wuninitialized]

it seems it is really a bug.
can any member help to fix it ?
or provide additional suggestion ?
(it seems only "unsigned long long now = 0" is not enough).

:-)

in kernel/posix-cpu-timers.c:
for variable 'now' is defined without initialization (line 1029) it may be used without initialization (line 1066)

1026 void posix_cpu_timer_schedule(struct k_itimer *timer) 1027 {
1028 struct task_struct *p = timer->it.cpu.task;
1029 unsigned long long now;
1030
1031 if (unlikely(p == NULL))
1032 /*
1033 * The task was cleaned up already, no future firings.
1034 */
1035 goto out;
1036
1037 /*
1038 * Fetch the current sample and update the timer's expiry time.
1039 */
1040 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {

1041                 cpu_clock_sample(timer->it_clock, p, &now);
1042                 bump_cpu_timer(timer, now);
1043                 if (unlikely(p->exit_state)) {
1044                         clear_dead_task(timer, now);
1045                         goto out;
1046                 }
1047                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1048                 spin_lock(&p->sighand->siglock);

1049 } else {
1050 read_lock(&tasklist_lock);
1051 if (unlikely(p->sighand == NULL)) {
1052 /*

1053                          * The process has been reaped.
1054                          * We can't even collect a sample any more.
1055                          */
1056                         put_task_struct(p);
1057                         timer->it.cpu.task = p = NULL;
1058                         timer->it.cpu.expires = 0;
1059                         goto out_unlock;
1060                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1061                         /*
1062                          * We've noticed that the thread is dead, but
1063                          * not yet reaped.  Take this opportunity to
1064                          * drop our task ref.
1065                          */
1066                         clear_dead_task(timer, now);
1067                         goto out_unlock;
1068                 }
1069                 spin_lock(&p->sighand->siglock);
1070                 cpu_timer_sample_group(timer->it_clock, p, &now);
1071                 bump_cpu_timer(timer, now);
1072                 /* Leave the tasklist_lock locked for the call below.  */

1073 }
1074
1075 /*
1076 * Now re-arm for the new expiry time.
1077 */

1078         BUG_ON(!irqs_disabled());
1079         arm_timer(timer);
1080         spin_unlock(&p->sighand->siglock);

1081
1082 out_unlock:
1083 read_unlock(&tasklist_lock);
1084
1085 out:

1086         timer->it_overrun_last = timer->it_overrun;
1087         timer->it_overrun = -1;
1088         ++timer->it_requeue_pending;

1089 }

Signature · 3 lines
Chen Gang

Asianux Corporation
#2 lore raw
In reply to Chen Gang

2013/3/26 Chen Gang <[email protected]>:

Quoted 8 lines

Hello Maintainers:

compiling with EXTRA_CFLAGS=-W:
make V=1 EXTRA_CFLAGS=-W ARCH=arm s3c2410_defconfig make V=1 EXTRA_CFLAGS=-W ARCH=arm menuconfig
set 'arm-linux-gnu-' for cross chain prefix
make V=1 EXTRA_CFLAGS=-W ARCH=arm

Quoted 7 lines

it will report:
kernel/posix-cpu-timers.c:1065:19: warning: ‘now’ may be used uninitialized in this function [-Wuninitialized]

it seems it is really a bug.
can any member help to fix it ?
or provide additional suggestion ?
(it seems only "unsigned long long now = 0" is not enough).

Yeah it's missing a call to cpu_timer_sample_group() before clear_dead_task(). Andrew Morton reported the warning and I have a pending patch to fix that. I'm just checking a few other things before sending it. These clear_dead_task() calls seem to also conflict with cleanup_timers(). I'm fixing that too.

Thanks for your report!

author
#3 lore raw
Quoted 16 lines · Frederic Weisbecker

2013/3/26 Chen Gang <[email protected]>:

Quoted 8 lines
Quoted 8 lines

Hello Maintainers:

compiling with EXTRA_CFLAGS=-W:
make V=1 EXTRA_CFLAGS=-W ARCH=arm s3c2410_defconfig make V=1 EXTRA_CFLAGS=-W ARCH=arm menuconfig
set 'arm-linux-gnu-' for cross chain prefix
make V=1 EXTRA_CFLAGS=-W ARCH=arm

Quoted 7 lines
Quoted 7 lines

it will report:
kernel/posix-cpu-timers.c:1065:19: warning: �now� may be used uninitialized in this function [-Wuninitialized]

it seems it is really a bug.
can any member help to fix it ?
or provide additional suggestion ?
(it seems only "unsigned long long now = 0" is not enough).

Quoted 9 lines

Yeah it's missing a call to cpu_timer_sample_group() before clear_dead_task(). Andrew Morton reported the warning and I have a pending patch to fix that. I'm just checking a few other things before sending it. These clear_dead_task() calls seem to also conflict with cleanup_timers(). I'm fixing that too.

Thanks for your report!

thank you, too.

:-)

Signature · 3 lines
Chen Gang

Asianux Corporation