Re: [patch] Make slab allocator work with SLAB_MUST_HWCACHE_ALIGN

From: Arjan van de Ven <arjanv@redhat.com>
To: Ravikiran G Thirumalai <kiran@in.ibm.com>
Cc: Manfred Spraul <manfred@colorfullife.com>,
	akpm@osdl.org, linux-kernel@vger.kernel.org,
	Robert Love <rml@tech9.net>,
	dipankar@in.ibm.com
Subject: Re: [patch] Make slab allocator work with SLAB_MUST_HWCACHE_ALIGN
Date: Fri, 12 Sep 2003 11:10:37 +0200	[thread overview]
Message-ID: <1063357837.5021.3.camel@laptop.fenrus.com> (raw)
In-Reply-To: <20030912085921.GB1128@llm08.in.ibm.com>

[-- Attachment #1: Type: text/plain, Size: 707 bytes --]

On Fri, 2003-09-12 at 10:59, Ravikiran G Thirumalai wrote:

> > I have two concerns:
> > - what about users that compile a kernel for a pIII and then run it on a 
> > p4? L1_CACHE_BYTES is 32 bytes, but the real cache line size is 128 bytes.
> 
> If the target cpu is chosen properly during compile time, L1_CACHE_BYTES
> is set accordingly (CONFIG_X86_L1_CACHE_SHIFT on x86) so this should
> not be an issue right?

In my opinion there is no excuse to make such an assumption at compile
time if its basically free to do it at runtime instead...
for structure padding, using the config setting is obviously the only
way to go; but for kmalloc you can use the runtime, actual, cacheline
size/

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]