Re: 2.4.25-rc1: BUG: wrong zone alignment, it will crash

[Michael Frank <mhf@linuxmail.org>]

On Saturday 07 February 2004 01:07, Randy.Dunlap wrote:
> On Sat, 7 Feb 2004 00:46:31 +0800 Michael Frank <mhf@linuxmail.org> wrote:
> 
> | On Friday 06 February 2004 23:36, Rik van Riel wrote:
> | > On Fri, 6 Feb 2004, Michael Frank wrote:
> | > 
> | > > > > 300MB HIGHMEM available.
> | > > > > 195MB LOWMEM available.
> | > > > > On node 0 totalpages: 126960
> | > > > > zone(0): 4096 pages.
> | > > > > zone(1): 46064 pages.
> | > > > > zone(2): 76800 pages.
> | > > > > BUG: wrong zone alignment, it will crash
> | > 
> | > > It is supposed to work, just a bug in the zone alignment code.
> | > 
> | > The error isn't in the kernel, it's between the chair and the keyboard.
> | > You have created a lowmem zone of a size that doesn't correctly
> | > align with the largest blocks used by the buddy allocator.
> | > 
> | > > I have have to use HIGHMEM emulation for testing.
> | > 
> | > Then you'll need to choose a different size for the highmem=
> | > parameter, one that doesn't cause an unaligned boundary.
> | 
> | Which is not user friendly and does not match the documentation.
> 
> Interesting boot option...  but what doc. are you referring to?

Line 863 of 2.4.2[45]/arch/i386/setup.c

/*
 * highmem=size forces highmem to be exactly 'size' bytes.
 * This works even on boxes that have no highmem otherwise.
 * This also works to reduce highmem size on bigger boxes.
 */
else if (!memcmp(from, "highmem=", 8))
	highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;

> 
> 
> | > Alternatively, you could submit a patch so the highmem= boot
> | > option parsing code does the aligning for you.
> | 
> | OK, will do. I'll produce and test a patch.
> 
> Please include an update to Documentation/kernel-parameters.txt .

Sure.

> 
> | > However, that would simply be an improvement to the kernel and
> | > nothing like a bug you can demand to get fixed now.
> | 
> | OK, Please note that I only passed on the message produced by the kernel 
> |   BUG: wrong zone alignment, it will crash 
> | 
> | Perhaps the kernel should have reported it as "Invalid value for highmem" 
> | instead of "BUG" ;)
> 
> How does this option work?  Does it just fake highmem_pages of
> low memory as being in high memory?

Sort of, it maps some pages into highmem zone instead of lowmem zone.

Regards
Michael