RE: [PATCH] sched: Make nr_uninterruptible count a signed value
From: Diwakar Tundlam
Date: Wed May 09 2012 - 15:04:15 EST
The issue is because when promoting from 32-bit unsigned to 64-bit signed, the compiler first does SIZE increase and then the sign change. If it did SIGN change first and then size increase (with sign-extension) and it will work correctly.
But then, the compiler is probably doing the right thing here because the starting point is an unsigned value and when it is assigned a negative value, the compiler is perfectly justified in converting it to a large positive value which is what happened here.
The theoretical correct fix is to change the declaration to signed long nr_uninterruptible simply because it can sometimes become -ve.
The other fix is to fix the print statement - but it feels wrong.
It would be good to fix the declaration...
From: Peter Zijlstra [mailto:peterz@xxxxxxxxxxxxx]
Sent: Wednesday, May 09, 2012 1:12 AM
To: Diwakar Tundlam
Cc: 'Ingo Molnar'; 'David Rientjes'; 'linux-kernel@xxxxxxxxxxxxxxx'; Peter De Schrijver
Subject: RE: [PATCH] sched: Make nr_uninterruptible count a signed value
On Tue, 2012-05-08 at 15:46 -0700, Diwakar Tundlam wrote:
> Maybe it is an artifact of 32-bit machine displaying 64-bit print format.
> An (unsigned long)(-24) promoted to (signed long long) ends up as 4294967272.
> As seen in my output of sched_debug.
Ah, quite possible. %Ld is indeed %lld and the value is long, not long long. So the proper fix is to fudge that printk statement somehow.
> Your machine is probably natively 64-bit.
Yeah, I gave up on 32bit computing a while ago..