Re: Soft lockup regression since kernel 3.13

[David Cohen <david.a.cohen@linux.intel.com>]

On Wed, Apr 02, 2014 at 04:29:48PM -0700, David Cohen wrote:
> Hi,
> 
> I've detected a regression from upstream (using an Intel Merrifield
> device) since 3.13 (still exists in 3.14) which I never had much time
> to start to investigate until now. The symptoms are: the device boots
> and works fine for while until it silently hangs.
> 
> I finally bisected v3.12..v3.13 and found exactly which commit created
> the issue:
> 
> commit f27dde8deef33c9e58027df11ceab2198601d6a6
> Author: Peter Zijlstra <peterz@infradead.org>
> Date:   Wed Aug 14 14:55:31 2013 +0200
> 
>     sched: Add NEED_RESCHED to the preempt_count

FWIW I did further investigation and made the issue to go away with this
hack:

diff --git a/arch/x86/include/asm/preempt.h
b/arch/x86/include/asm/preempt.h
index c8b051933b1b..41744d99c88f 100644
--- a/arch/x86/include/asm/preempt.h
+++ b/arch/x86/include/asm/preempt.h
@@ -95,7 +95,7 @@ static __always_inline bool
__preempt_count_dec_and_test(void)
  */
   static __always_inline bool should_resched(void)
    {
    -       return unlikely(!__this_cpu_read_4(__preempt_count));
    +       return unlikely(tif_need_resched());
     }
      
       #ifdef CONFIG_PREEMPT
--

I extended part of this patch to x86:

commit ba1f14fbe70965ae0fb1655a5275a62723f65b77
Author: Peter Zijlstra <peterz@infradead.org>
Date:   Thu Nov 28 14:26:41 2013 +0100

    sched: Remove PREEMPT_NEED_RESCHED from generic code
--

Br, David Cohen