Skip to contentSkip to conversation

don't let printk unconditionally turn on interrupts

PATCH In review Steven Rostedt 2 participants 3 messages Jan 21, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 don't let printk unconditionally turn on interrupts

Ingo,

My first try at booting -rt8 on my machine crashed immediately. No nmi, no nothing. Just a lockup at the registration of the ACPI_PM timer. This would happen every time, and after struggling for a while, I finally found out why!

The printk in timeofday_periodic_hook that is called holding the write_lock of system_time_lock (a raw_seq_lock) was causing lots of havoc. The printk would turn on interrupts, and then I would get a deadlock when the interrupt would do a read on system_time_lock.

So here's the patch:

-- Steve

kernel/printk.c +1 -1
Index: linux-2.6.15-rt8/kernel/printk.c
===================================================================
@@ -772,7 +772,7 @@
 		 */
 #if defined(CONFIG_PREEMPT_RT) && !defined(CONFIG_PRINTK_IGNORE_LOGLEVEL) && !defined(CONFIG_PPC) \
     && !defined(CONFIG_PARANOID_GENERIC_TIME)
-		spin_unlock_irq(&logbuf_lock);
+		spin_unlock_irqrestore(&logbuf_lock, flags);
 #else
 		spin_unlock(&logbuf_lock);
 #endif
#1 lore raw
Quoted 11 lines · Steven Rostedt

Ingo,

My first try at booting -rt8 on my machine crashed immediately. No nmi, no nothing. Just a lockup at the registration of the ACPI_PM timer. This would happen every time, and after struggling for a while, I finally found out why!

The printk in timeofday_periodic_hook that is called holding the write_lock of system_time_lock (a raw_seq_lock) was causing lots of havoc. The printk would turn on interrupts, and then I would get a deadlock when the interrupt would do a read on system_time_lock.

argh. This piece of code has caused so many problems already. Some people want console drivers to be preemptible, but i guess being able to boot trumps things ;)

could you check out the variant in 2.6.15-rt11 - i tweaked it slightly over yours, hopefully making the behavior more consistent.

Ingo

#2 lore raw
Quoted 2 lines · Ingo Molnar

could you check out the variant in 2.6.15-rt11 - i tweaked it slightly over yours, hopefully making the behavior more consistent.

I'll take a look Monday.

Thanks,

-- Steve