Bug 6787

Summary: suspend to ram fails when process is being debugged
Product: Power Management Reporter: Jeremy Fitzhardinge (jeremy)
Component: OtherAssignee: Rafael J. Wysocki (rjwysocki)
Status: CLOSED CODE_FIX    
Severity: normal CC: pavel, rjwysocki
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.17-mm6 Subsystem:
Regression: --- Bisected commit-id:
Attachments: Debug patch
Fix candidate
Better fix

Description Jeremy Fitzhardinge 2006-07-03 14:49:44 UTC 
Most recent kernel where this bug did not occur: not sure
Distribution:FC6 (devel)
Hardware Environment:Lenovo Thinkpad X60
Software Environment:
Problem Description:
When a process is being debugged (stopped in a breakpoint, for example), suspend
fails.

Steps to reproduce:
Start program under gdb.  Set breakpoint.  Run program until it hits breakpoint.
  Attempt to suspend.
Comment 1 Pavel Machek 2006-07-03 15:44:36 UTC 
It would be nice to get rjw@sisk.pl to Cc, but he apparently does not have
bugzilla account (or has it under different name?)
Comment 2 Pavel Machek 2006-07-03 15:49:35 UTC 
Okay, I can reproduce it on fairly recent 2.6.17-git.
Comment 3 Rafael J. Wysocki 2006-07-04 02:55:37 UTC 
I think the "freeze" signal is catched due to ptrace() before the process
has a chance to go to the refrigerator.  Then the parent is supposed to do
something, but if it's already frozen, it can't.

If that's what really happens, the signal-handling code will need a fix.
Comment 4 Rafael J. Wysocki 2006-07-14 14:45:12 UTC 
Created attachment 8555 [details]
Debug patch

More precisely, if the ptraced task has already hit a breakpoint, it is in the
TASK_TRACED state and we can't deliver any signals directly to it until the
parent decides what to do with it.  However, if the parent is frozen, it can't
do anything and the ptraced task becomes unfreezeable.	Thus it seems we should
treat it as such in kernel/power/process.c#freezeable().  If that really is the
case, this patch should help, so please try it.
Comment 5 Rafael J. Wysocki 2006-07-16 04:20:05 UTC 
Created attachment 8561 [details]
Fix candidate

This patch should fix the problem (works on my box).

I believe it won't break any other corner cases related to the tracing of
tasks. :-)
Comment 6 Rafael J. Wysocki 2006-07-27 00:02:04 UTC 
Created attachment 8625 [details]
Better fix

This patch fixes the problem on my box.