Bug 5062

Summary: kernel preemption breaks in several ways
Product: Process Management Reporter: Sanjoy Mahajan (sanjoy)
Component: PreemptionAssignee: Ingo Molnar (mingo)
Status: CLOSED CODE_FIX    
Severity: normal CC: linux, mingo, protasnb
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.13-rc6 Subsystem:
Regression: --- Bisected commit-id:
Bug Depends on: 5008, 5037, 5061    
Bug Blocks:    

Description Sanjoy Mahajan 2005-08-14 00:24:21 UTC
Distribution: Debian testing
Hardware Environment: TP 600X
Software Environment: 
Problem Description: I get the following errors/problems only with pre-emption
enabled in the kernel:

1. pcmcia 'unable to remove socket power" errors (Bug #5061).  Happens if full
preemption is enabled ('low-latency desktop' as preemption model + PREEMPT_BKL),
but not with mild preemption ('voluntary kernel preemption').

2. S3 hangs upon going to sleep (Bug #5037).  It happens regularly with full
pre-emption, not with 'no forced preemption', and probably not with mild
preemption (haven't been running that kernel for long enough to be sure).

3. errors like "os_wait_semaphore     : Failed to acquire" (Bug #5008).  Happens
with full or mild preemption, but not with voluntary preemption.

Steps to reproduce:
Comment 1 Sanjoy Mahajan 2005-08-15 07:29:29 UTC
> 3. errors like "os_wait_semaphore     : Failed to acquire" (Bug #5008). 
Happens > with full or mild preemption, but not with voluntary preemption.

I meant 'but not with no pre-emption'.

Comment 2 Dominik Brodowski 2006-01-10 09:36:13 UTC
What's the status of this bug? Is (at least) part 1 of it resolved?
Comment 3 Sanjoy Mahajan 2006-01-10 11:06:59 UTC
> 3. errors like "os_wait_semaphore     : Failed to acquire" (Bug  #5008). 

This part still happens with 2.6.15 (just started running it) with
'voluntary kernel preemption'.

> 2. S3 hangs upon going to sleep (Bug #5037).

With 2.6.15, I haven't seen this one using voluntary preemption, but I
haven't tried full preemption ('low-latency desktop') yet.

> 1. pcmcia 'unable to remove socket power" errors (Bug #5061). 

Also with 2.6.15, I haven't seen this one with voluntary preemption
but I need to test full preemption.  Daniel Ritz's patch reduced the
problem a lot with 2.6.13 (I think that's what I was running then)
with full preemption.  At first I thought it had solved it, but then I
it happened one more time.

-Sanjoy

`Never underestimate the evil of which men of power are capable.'
         --Bertrand Russell, _War Crimes in Vietnam_, chapter 1.

Comment 4 Natalie Protasevich 2007-07-06 09:26:52 UTC
Any update on this problem?
Thanks.
Comment 5 Ingo Molnar 2007-11-29 08:24:15 UTC
I'm closing this as fixed (all dependent bugs got closed AFAICS), please re-open it if you can still reproduce it.