2001-10-03 00:58:46

by Justin A

[permalink] [raw]
Subject: 2.4.10/Preemt STOP bug

It seems there is a bug in 2.4.10 or the preemptable patch that causes the STOP
signal to not work right 19 times out of 20 or so.

This is most easily seen by running 'seq 1 100000' in a terminal and pressing
control Z or using kill -STOP the proccess stops, but the parent process never
comes back, kill -CONT has to be used in order to get the proccess back.

CPU usage my be a factor in this, yes(1) backgrounds correctly more often then
seq does.

There are no problems with fg/CONT.

-Justin


2001-10-03 01:19:07

by Robert Love

[permalink] [raw]
Subject: Re: 2.4.10/Preemt STOP bug

On Tue, 2001-10-02 at 20:58, Justin A wrote:
> It seems there is a bug in 2.4.10 or the preemptable patch that causes the STOP
> signal to not work right 19 times out of 20 or so.
>
> This is most easily seen by running 'seq 1 100000' in a terminal and pressing
> control Z or using kill -STOP the proccess stops, but the parent process never
> comes back, kill -CONT has to be used in order to get the proccess back.
>
> CPU usage my be a factor in this, yes(1) backgrounds correctly more often then
> seq does.
>
> There are no problems with fg/CONT.

The problem is with the preemptible kernel patch...

You need to be using a recent patch, this bug was fixed awhile back in
an explicit patch (patch-rml-2.4.xx-preempt-ptrace-signal-fix-1).

It was merged into the 2.4.10 patch with revision 5, and it is in all
newer patches.

You can get patches for 2.4.10, 2.4.11-pre2, and 2.4.10-ac3 (soon -ac4)
at http://tech9.net/rml/linux

They all fix this problem.

--
Robert M. Love
rml at ufl.edu
rml at tech9.net