Date: Wed, 9 Jul 2008 23:12:56 +0200
From: Ingo Molnar <mingo@elte.hu>
To: Mike Travis <travis@sgi.com>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>,
       Andrew Morton <akpm@linux-foundation.org>,
       "Eric W. Biederman" <ebiederm@xmission.com>,
       "H. Peter Anvin" <hpa@zytor.com>,
       Christoph Lameter <cl@linux-foundation.org>,
       Jack Steiner <steiner@sgi.com>, linux-kernel@vger.kernel.org
Subject: Re: [RFC 00/15] x86_64: Optimize percpu accesses
Message-ID: <20080709211256.GD4298@elte.hu>
References: <20080709165129.292635000@polaris-admin.engr.sgi.com> <20080709192822.GB4804@elte.hu> <487525B3.90904@sgi.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <487525B3.90904@sgi.com>
User-Agent: Mutt/1.5.18 (2008-05-17)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1042
Lines: 24


* Mike Travis <travis@sgi.com> wrote:

> After that is taken care of, I'll start regression testing earlier 
> compilers. I think someone mentioned that gcc-2.something was the 
> minimum required...?

i think the current official minimum is around gcc-3.2 [2.x is out of 
question because we have a few feature dependencies on gcc-3.x] - but i 
stopped using it because it miscompiles the kernel so often. 4.0 was 
really bad due to large stack footprint. The 4.3.x series miscompiles 
the kernel too in certain situations - there was a high-rising 
kerneloops.org crash recently in ext3.

So in general, 'too new' is bad because it has new regressions, 'too 
old' is bad because it has unfixed old regressions. Somewhere in the 
middle, 4.2.x-ish, seems to be pretty robust in practice.

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/