[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-arm-kernel
Subject:    Re: [CFT] Mark IO as bufferable
From:       Russell King - ARM Linux <linux () arm ! linux ! org ! uk>
Date:       2005-06-28 16:16:53
Message-ID: 20050628171653.E10022 () flint ! arm ! linux ! org ! uk
[Download RAW message or body]

On Tue, Jun 28, 2005 at 05:59:20PM +0200, Matthias Welwarsky wrote:
> On Monday 27 June 2005 18:37, Russell King - ARM Linux wrote:
> > Can people please test out the effect of this change.  Potentially it
> > should get slightly better IO performance.  Or it could prevent the
> > kernel booting.
> 
> I wonder what this change does in detail. Does bufferable mean "use write-
> back cache" or is it just to use the write buffer?

It'll use the write buffer.  What this means is that individual writes
should end up in the WB without stalling the CPU pipeline, and should
appear on the processor bus shortly thereafter.

> Either way, I think it's quite a dangerous change. Imagine an IDE driver:
> The ATAPI4 spec says (not literally): "After writing to the command
> register, wait for at least 400ns before reading the status register".

Bear in mind that later ARM CPUs will hit these issues, so they'll need
to eventually get fixed one way or another.

However, at present I'm just vaguely interested in what the effect is
on these older CPUs, and whether it looks like being a big issue.

We may have to start integrating the memory barriers into our collective
thinking like other Linux already have to - in fact we most likely will
once ARMv6 becomes mainstream.

-------------------------------------------------------------------
List admin: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:        http://www.arm.linux.org.uk/mailinglists/faq.php
Etiquette:  http://www.arm.linux.org.uk/mailinglists/etiquette.php
[prev in list] [next in list] [prev in thread] [next in thread]