Re: [BUG] Code reordering in swsusp breaks suspend on SMP systems

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Friday, 23 March 2007 00:30, Rafael J. Wysocki wrote:
> On Thursday, 22 March 2007 00:53, Rafael J. Wysocki wrote:
> > On Thursday, 22 March 2007 00:39, Maxim wrote:
> > > On Thursday 22 March 2007 01:24:25 Rafael J. Wysocki wrote:
> > > > On Thursday, 22 March 2007 00:09, Maxim wrote:
> > > > > On Thursday 22 March 2007 00:39:02 you wrote:
> > > > > > On Wednesday, 21 March 2007 23:21, Pavel Machek wrote:
> > > > > > > Hi!
> > > > > > > 
> > > > > > > > Starting with 2.6.21-rc1 suspend to ram and disk doesn't work anymore on my system.
> > > > > > > > 
> > > > > > > > I did a git-bisect and found that those commits break it:
> > > > > > > > 
> > > > > > > > e3c7db621bed4afb8e231cb005057f2feb5db557 - [PATCH] [PATCH] PM: Change code ordering in main.c
> > > > > > > > ed746e3b18f4df18afa3763155972c5835f284c5 - [PATCH] [PATCH] swsusp: Change code ordering in disk.c
> > > > > > > > 259130526c267550bc365d3015917d90667732f1 - [PATCH] [PATCH] swsusp: Change code ordering in user.c
> > > > > > > > 
> > > > > > > 
> > > > > > > (Yep, it was in my "to analyze" queue).
> > > > > > > 
> > > > > > > > I already reported about it, but now i know the reason why suspend breaks.
> > > > > > > > 
> > > > > > > > The problem is that both cpu_up/cpu_down were allowed to sleep until now, 
> > > > > > > > and it did work because those functions could be called only in process context
> > > > > > > > (the one that writes to /sys/devices/system/cpu/cpu*/online) or  idle thread  that does smp_init()).
> > > > > > > > 
> > > > > > > > But now they are called _after_ all tasks were suspended, so if cpu_down tries for example to take a lock
> > > > > > > > that is taken by different process, it can't since the different proccess is frozen and can't release the lock.
> > > > > > > > 
> > > > > > > 
> > > > > > > Thanks for detailed explanation.
> > > > > > > 
> > > > > > > ...but, on my machine suspend works ok in -rc4. I'm not seeing this.
> > > > > > > 
> > > > > > > ...by design, "frozen" tasks must not hold any locks. If frozen task
> > > > > > > holds a lock, that's a bug.
> > > > > > > 
> > > > > > > > Or, it is also possible to revert this change.
> > > > > > > 
> > > > > > > Are you using xfs?
> > > > > > 
> > > > > > Well, this is the only case that can trigger it.  There are no other freezable
> > > > > > workqueues.
> > > > > > 
> > > > > > Greetings,
> > > > > > Rafael
> > > > > > 
> > > > > 
> > > > > Hello,
> > > > > 
> > > > > 	Yes, you are right and it is XFS
> > > > > 
> > > > > 	System suspends and resumes with xfs and your patch correctly,
> > > > 
> > > > Could you please sent this information to the list?  I'd like it to reach all
> > > > of the CCed parites. ;-)
> > > 
> > > I did now ( sorry I just keep using this Answer command, instead of Answer to everybody)
> > > I didn't intend to send private email.
> > > > 
> > > > > 	Of course I need to mention that I had to unload microcode update driver because it prevented resume,
> > > > > 	because it calls firmware loader helper, and again sleeps on lock
> > > > 
> > > > This is interesting.  Did it happen before or is it a regression?
> > > 
> > > It is from the same group of bugs , I mean hang because cpu_up/down is called with frozen tasks
> > > Of course it didn't happen before those reordering commits were introduced
> > 
> > Well, we want cpu_up/down to be called after processes have been frozen, for
> > various reasons (one of them being that applications shouldn't see us playing
> > with the CPUs).
> > 
> > Thanks for reporting this, I'll have a look at the microcode update driver.
> 
> Well, I have invented the appended workaround, but I'm not sure how much
> sense it makes with respect to the microcode driver.  At least, it doesn't
> break my AMD64 SMP setup. ;-)

Modified version of the patch is appended.  Unfortunately I have no hardware
supporting the microcode updates.

Greetings,
Rafael


---
 arch/i386/kernel/microcode.c |   28 +++++++++++++++++++++++++---
 include/linux/cpu.h          |    2 ++
 kernel/cpu.c                 |   32 ++++++++++++++++----------------
 3 files changed, 43 insertions(+), 19 deletions(-)

Index: linux-2.6.21-rc4/arch/i386/kernel/microcode.c
===================================================================
--- linux-2.6.21-rc4.orig/arch/i386/kernel/microcode.c
+++ linux-2.6.21-rc4/arch/i386/kernel/microcode.c
@@ -567,6 +567,16 @@ static int cpu_request_microcode(int cpu
 	return error;
 }
 
+static void apply_microcode_on_cpu(int cpu)
+{
+	cpumask_t old;
+
+	old = current->cpus_allowed;
+	set_cpus_allowed(current, cpumask_of_cpu(cpu));
+	apply_microcode(cpu);
+	set_cpus_allowed(current, old);
+}
+
 static void microcode_init_cpu(int cpu)
 {
 	cpumask_t old;
@@ -663,13 +673,21 @@ static int mc_sysdev_add(struct sys_devi
 		return 0;
 
 	pr_debug("Microcode:CPU %d added\n", cpu);
-	memset(uci, 0, sizeof(*uci));
+	/* If suspend_cpu_hotplug is set, the system is resuming and we should
+	 * use the data from before the suspend.
+	 */
+	if (!suspend_cpu_hotplug)
+		memset(uci, 0, sizeof(*uci));
 
 	err = sysfs_create_group(&sys_dev->kobj, &mc_attr_group);
 	if (err)
 		return err;
 
-	microcode_init_cpu(cpu);
+	if (suspend_cpu_hotplug && uci->valid)
+		apply_microcode_on_cpu(cpu);
+	else
+		microcode_init_cpu(cpu);
+
 	return 0;
 }
 
@@ -680,7 +698,11 @@ static int mc_sysdev_remove(struct sys_d
 	if (!cpu_online(cpu))
 		return 0;
 	pr_debug("Microcode:CPU %d removed\n", cpu);
-	microcode_fini_cpu(cpu);
+	/* If suspend_cpu_hotplug is set, the system is suspending and we should
+	 * keep the microcode in memory for the resume.
+	 */
+	if (!suspend_cpu_hotplug)
+		microcode_fini_cpu(cpu);
 	sysfs_remove_group(&sys_dev->kobj, &mc_attr_group);
 	return 0;
 }
Index: linux-2.6.21-rc4/include/linux/cpu.h
===================================================================
--- linux-2.6.21-rc4.orig/include/linux/cpu.h
+++ linux-2.6.21-rc4/include/linux/cpu.h
@@ -127,6 +127,8 @@ static inline int cpu_is_offline(int cpu
 #endif		/* CONFIG_HOTPLUG_CPU */
 
 #ifdef CONFIG_SUSPEND_SMP
+extern int suspend_cpu_hotplug;
+
 extern int disable_nonboot_cpus(void);
 extern void enable_nonboot_cpus(void);
 #else
Index: linux-2.6.21-rc4/kernel/cpu.c
===================================================================
--- linux-2.6.21-rc4.orig/kernel/cpu.c
+++ linux-2.6.21-rc4/kernel/cpu.c
@@ -254,6 +254,12 @@ int __cpuinit cpu_up(unsigned int cpu)
 }
 
 #ifdef CONFIG_SUSPEND_SMP
+/* Needed to prevent the microcode driver from requesting firmware in its CPU
+ * hotplug notifier during the suspend/resume.
+ */
+int suspend_cpu_hotplug;
+EXPORT_SYMBOL(suspend_cpu_hotplug);
+
 static cpumask_t frozen_cpus;
 
 int disable_nonboot_cpus(void)
@@ -261,16 +267,8 @@ int disable_nonboot_cpus(void)
 	int cpu, first_cpu, error = 0;
 
 	mutex_lock(&cpu_add_remove_lock);
-	first_cpu = first_cpu(cpu_present_map);
-	if (!cpu_online(first_cpu)) {
-		error = _cpu_up(first_cpu);
-		if (error) {
-			printk(KERN_ERR "Could not bring CPU%d up.\n",
-				first_cpu);
-			goto out;
-		}
-	}
-
+	suspend_cpu_hotplug = 1;
+	first_cpu = first_cpu(cpu_online_map);
 	/* We take down all of the non-boot CPUs in one shot to avoid races
 	 * with the userspace trying to use the CPU hotplug at the same time
 	 */
@@ -296,7 +294,7 @@ int disable_nonboot_cpus(void)
 	} else {
 		printk(KERN_ERR "Non-boot CPUs are not disabled\n");
 	}
-out:
+	suspend_cpu_hotplug = 0;
 	mutex_unlock(&cpu_add_remove_lock);
 	return error;
 }
@@ -308,20 +306,22 @@ void enable_nonboot_cpus(void)
 	/* Allow everyone to use the CPU hotplug again */
 	mutex_lock(&cpu_add_remove_lock);
 	cpu_hotplug_disabled = 0;
-	mutex_unlock(&cpu_add_remove_lock);
 	if (cpus_empty(frozen_cpus))
-		return;
+		goto out;
 
+	suspend_cpu_hotplug = 1;
 	printk("Enabling non-boot CPUs ...\n");
 	for_each_cpu_mask(cpu, frozen_cpus) {
-		error = cpu_up(cpu);
+		error = _cpu_up(cpu);
 		if (!error) {
 			printk("CPU%d is up\n", cpu);
 			continue;
 		}
-		printk(KERN_WARNING "Error taking CPU%d up: %d\n",
-			cpu, error);
+		printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error);
 	}
 	cpus_clear(frozen_cpus);
+	suspend_cpu_hotplug = 0;
+out:
+	mutex_unlock(&cpu_add_remove_lock);
 }
 #endif