Re: [BUG 2.6.25-rc3] scheduler/hotplug: some processes are dealocked when cpu is set to offline

["Dmitry Adamushko" <dmitry.adamushko@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1999 bytes --]

On 02/03/2008, Yi Yang <yi.y.yang@intel.com> wrote:
> [ ... ]
>
>  Why isn't the kernel thread [watchdog/1] reaped by its parent?
> its state
>  is TASK_RUNNING with high priority (R< means this), why it isn't done?
>
>  Anyone ever met such a problem? Your thought?
>

iirc, Andrew had the same issue.

'watchdog's are supposed to be stopped with kthread_stop() from
softlockup.c :: cpu_callback() :: case CPU_DEAD.

The 'R' state of 'watchdog' is strange indeed.

would you please conduct a test with the patch [1] below and provide
me with additional output it generates?

(non-white-space-damaged versions are enclosed)

with your current set-up and also with
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs"

Then please try also the patch [2] (on top of [1]). With this 'magic'
patch applied, the issue 'seemed' to disappear on Andrew's set-up...

Thanks in advacne,


[1]

--- softlockup-prev.c   2008-03-03 12:35:01.000000000 +0100
+++ softlockup.c        2008-03-03 12:38:01.000000000 +0100
@@ -237,6 +237,8 @@ static int watchdog(void *__bind_cpu)

        }

+       printk(KERN_WARN "-> watchdog(cpu: %d) is done\n", this_cpu);
+
        return 0;
 }

@@ -249,6 +251,9 @@ cpu_callback(struct notifier_block *nfb,
        int hotcpu = (unsigned long)hcpu;
        struct task_struct *p;

+       printk(KERN_WARN "-> cpu_callback(cpu: %d, action: %lu,
check_cpu: %d)\n",
+               hotcpu, action, check_cpu);
+
        switch (action) {
        case CPU_UP_PREPARE:
        case CPU_UP_PREPARE_FROZEN:


[2]

--- softlockup-prev-2.c 2008-03-03 12:38:36.000000000 +0100
+++ softlockup.c        2008-03-03 12:39:02.000000000 +0100
@@ -294,6 +294,7 @@ cpu_callback(struct notifier_block *nfb,
        case CPU_DEAD_FROZEN:
                p = per_cpu(watchdog_task, hotcpu);
                per_cpu(watchdog_task, hotcpu) = NULL;
+               mlseep(1);
                kthread_stop(p);
                break;
 #endif /* CONFIG_HOTPLUG_CPU */


-- 
Best regards,
Dmitry Adamushko

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: softlockup-debug-1.patch --]
[-- Type: text/x-patch; name=softlockup-debug-1.patch, Size: 567 bytes --]

--- softlockup-prev.c	2008-03-03 12:35:01.000000000 +0100
+++ softlockup.c	2008-03-03 12:38:01.000000000 +0100
@@ -237,6 +237,8 @@ static int watchdog(void *__bind_cpu)
 
 	}
 
+	printk(KERN_WARN "-> watchdog(cpu: %d) is done\n", this_cpu);
+
 	return 0;
 }
 
@@ -249,6 +251,9 @@ cpu_callback(struct notifier_block *nfb,
 	int hotcpu = (unsigned long)hcpu;
 	struct task_struct *p;
 
+	printk(KERN_WARN "-> cpu_callback(cpu: %d, action: %lu, check_cpu: %d)\n",
+		hotcpu, action, check_cpu);
+
 	switch (action) {
 	case CPU_UP_PREPARE:
 	case CPU_UP_PREPARE_FROZEN:

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #3: softlockup-debug-2.patch --]
[-- Type: text/x-patch; name=softlockup-debug-2.patch, Size: 356 bytes --]

--- softlockup-prev-2.c	2008-03-03 12:38:36.000000000 +0100
+++ softlockup.c	2008-03-03 12:39:02.000000000 +0100
@@ -294,6 +294,7 @@ cpu_callback(struct notifier_block *nfb,
 	case CPU_DEAD_FROZEN:
 		p = per_cpu(watchdog_task, hotcpu);
 		per_cpu(watchdog_task, hotcpu) = NULL;
+		mlseep(1);
 		kthread_stop(p);
 		break;
 #endif /* CONFIG_HOTPLUG_CPU */