login
Header Space

 
 

[BUG] CONFIG_IRQBALANCE=y broken?

Score:
Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]
To: <linux-kernel@...>
Cc: Ingo Molnar <mingo@...>, Thomas Gleixner <tglx@...>, Stephen Hemminger <shemminger@...>
Date: Monday, April 7, 2008 - 1:17 pm

Hi Ingo, Thomas,

When kernel-2.6.24.x is compiled with 

CONFIG_IRQBALANCE=y, 

and heavy LAN (outbound) traffic, I experience LAN dying due to
IRQ# masked at the IO_APIC (bit 16 is 1, for the corresponding irq
vector) (SMP x86_32,  AMD X2), as described in my previous post:

http://lkml.org/lkml/2008/3/25/165

and links to previous posts there in. When

CONFIG_IRQBALANCE is not set

I am not able to reproduce the problem. That makes me think that 
there is a race/bug somewhere in the irq balancing code that 
leaves (rarely) the corresponding IRQ# masked at the IO-APIC.

Any comments, suggestions, patches?

Additional info?

Thank you in advance for your answer.

Regards

Marin Mitov

--
Previous message: [thread] [date] [author]
Next message: [thread] [date] [author]

Messages in current thread:
[BUG] CONFIG_IRQBALANCE=y broken?, Marin Mitov, (Mon Apr 7, 1:17 pm)
speck-geostationary