x86, mce: Do not call del_timer_sync() in IRQ context
authorHidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Fri, 17 Jun 2011 08:40:36 +0000 (04:40 -0400)
committerBorislav Petkov <borislav.petkov@amd.com>
Wed, 14 Sep 2011 13:50:15 +0000 (15:50 +0200)
del_timer_sync() can cause a deadlock when called in interrupt context.
It is used with on_each_cpu() in some parts for sysfs files like bank*,
check_interval, cmci_disabled and ignore_ce.

However, use of on_each_cpu() results in calling the function passed
as the argument in interrupt context. This causes a flood of nested
warnings from del_timer_sync() (it runs on each CPU) caused even by a
simple file access like:

$ echo 300 > /sys/devices/system/machinecheck/machinecheck0/check_interval

Fortunately, these MCE-specific files are rarely used and AFAIK only few
MCE geeks experience this warning.

To remove the warning, move timer deletion outside of the interrupt
context.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
arch/x86/kernel/cpu/mcheck/mce.c

index 08363b0421222b92b5765bd2116cf9d4ecfca4d9..5b5cceec94c5e0966b231b7fb32dce8959fcdb54 100644 (file)
@@ -1140,6 +1140,15 @@ static void mce_start_timer(unsigned long data)
        add_timer_on(t, smp_processor_id());
 }
 
+/* Must not be called in IRQ context where del_timer_sync() can deadlock */
+static void mce_timer_delete_all(void)
+{
+       int cpu;
+
+       for_each_online_cpu(cpu)
+               del_timer_sync(&per_cpu(mce_timer, cpu));
+}
+
 static void mce_do_trigger(struct work_struct *work)
 {
        call_usermodehelper(mce_helper, mce_helper_argv, NULL, UMH_NO_WAIT);
@@ -1750,7 +1759,6 @@ static struct syscore_ops mce_syscore_ops = {
 
 static void mce_cpu_restart(void *data)
 {
-       del_timer_sync(&__get_cpu_var(mce_timer));
        if (!mce_available(__this_cpu_ptr(&cpu_info)))
                return;
        __mcheck_cpu_init_generic();
@@ -1760,16 +1768,15 @@ static void mce_cpu_restart(void *data)
 /* Reinit MCEs after user configuration changes */
 static void mce_restart(void)
 {
+       mce_timer_delete_all();
        on_each_cpu(mce_cpu_restart, NULL, 1);
 }
 
 /* Toggle features for corrected errors */
-static void mce_disable_ce(void *all)
+static void mce_disable_cmci(void *data)
 {
        if (!mce_available(__this_cpu_ptr(&cpu_info)))
                return;
-       if (all)
-               del_timer_sync(&__get_cpu_var(mce_timer));
        cmci_clear();
 }
 
@@ -1852,7 +1859,8 @@ static ssize_t set_ignore_ce(struct sys_device *s,
        if (mce_ignore_ce ^ !!new) {
                if (new) {
                        /* disable ce features */
-                       on_each_cpu(mce_disable_ce, (void *)1, 1);
+                       mce_timer_delete_all();
+                       on_each_cpu(mce_disable_cmci, NULL, 1);
                        mce_ignore_ce = 1;
                } else {
                        /* enable ce features */
@@ -1875,7 +1883,7 @@ static ssize_t set_cmci_disabled(struct sys_device *s,
        if (mce_cmci_disabled ^ !!new) {
                if (new) {
                        /* disable cmci */
-                       on_each_cpu(mce_disable_ce, NULL, 1);
+                       on_each_cpu(mce_disable_cmci, NULL, 1);
                        mce_cmci_disabled = 1;
                } else {
                        /* enable cmci */