I have a couple of SMP i386 boxes with 2GB RAM. They suffer from poor
performance due to block IO page bouncing with highmem enabled. I have
tried the block-highmem patch but this causes occasional oopses and even
panics under high IO load.
Will there be any ill effects if I change __PAGE_OFFSET to 0x7000000 or
thereabouts so that all RAM is mapped?
I presume I only have to change this is page.h and vmlinux.lds.S as in 2.2
kernels.
--
Alistair Riddell - BOFH
IT Manager, George Watson's College, Edinburgh
Tel: +44 131 446 6070 Fax: +44 131 452 8594
Microsoft - because god hates us
On Sat, Jan 05 2002, Alistair Riddell wrote:
> I have a couple of SMP i386 boxes with 2GB RAM. They suffer from poor
> performance due to block IO page bouncing with highmem enabled. I have
> tried the block-highmem patch but this causes occasional oopses and even
> panics under high IO load.
Well thanks for sending in a bug report on that. Mind doing so?
--
Jens Axboe
On Sun, 6 Jan 2002, Jens Axboe wrote:
> On Sat, Jan 05 2002, Alistair Riddell wrote:
> > I have a couple of SMP i386 boxes with 2GB RAM. They suffer from poor
> > performance due to block IO page bouncing with highmem enabled. I have
> > tried the block-highmem patch but this causes occasional oopses and even
> > panics under high IO load.
>
> Well thanks for sending in a bug report on that. Mind doing so?
I did send in a report a while back to the list and yourself but it was
ignored... I will try to gather together the details and/or reproduce it
and send details again.
--
Alistair Riddell - BOFH
IT Manager, George Watson's College, Edinburgh
Tel: +44 131 446 6070 Fax: +44 131 452 8594
Microsoft - because god hates us
On Sun, 6 Jan 2002, Jens Axboe wrote:
> On Sat, Jan 05 2002, Alistair Riddell wrote:
> > I have a couple of SMP i386 boxes with 2GB RAM. They suffer from poor
> > performance due to block IO page bouncing with highmem enabled. I have
> > tried the block-highmem patch but this causes occasional oopses and even
> > panics under high IO load.
>
> Well thanks for sending in a bug report on that. Mind doing so?
See below message, please let me know if I can provide any further
information.
Date: Thu, 29 Nov 2001 09:04:26 +0000 (GMT)
From: Alistair Riddell <[email protected]>
To: Jens Axboe <[email protected]>
Cc: Linux Kernel <[email protected]>
Subject: Re: [patch] block-highmem-all-18b
Erk... I applied this patch to 2.2.16pre1 on a 2GB SMP machine with an
onboard aic7899 scsi controller. This morning I came in to find the
machine had panicked with "Ththththaats all folks. Too dangerous to
continue."
There was nothing else in syslog either remotely or on disk, and nothing
else on the console apart from the panic.
Any ideas what might cause this?
Let me know if I can provide any further information.
On Thu, 22 Nov 2001, Jens Axboe wrote:
> A minor update version, nothing major. Changes:
>
> - Megaraid highmem I/O enabled again, since 2.4.14 this should be safe.
> Verified by Arjan. (me)
> - Change can_dma_32 to highmem_io to make the meaning more clear (me).
> - Drop discontig pfn change for now (me)
>
> Find it here:
>
> *.kernel.org/pub/linux/kernel/people/axboe/patches/2.4.15-pre9/block-highmem-all-18b.bz2
--
Alistair Riddell - BOFH
IT Manager, George Watson's College, Edinburgh
Tel: +44 131 446 6070 Fax: +44 131 452 8594
Microsoft - because god hates us