Re: [BUG FIX] [PATCH] fork_init() max_low_pfn fixes potential OOM bug on big highmem machine

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Tue, Jun 29, 2004 at 06:48:14PM +0800, Coywolf Qi Hunt wrote:
> Russell King wrote:
> Actually there's physical DRAM offset: PHY_OFFSET, defined on ARM only. 
> max_low_pfn happens to be the same as `num_lowpages'.
> These assignments seems illogical in naming. But just happen to let this 
> patch work.  Other platforms may still break.

That may be a bug actually.  Looking at ll_rw_blk.c:

        unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
        if (bounce_pfn < blk_max_low_pfn) {

        blk_max_low_pfn = max_low_pfn;

dma_addr are physical addresses, so bounce_pfn is referenced to a PFN0
equal to physical address 0.  This implies that blk_max_low_pfn is
likewise, as is max_low_pfn.

> [coywolf@everest ~/linux-2.6.7/arch]$ grep max_low_pfn arm* -rn
> arm/mm/init.c:235:      max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);

However, here, max_low_pfn of zero corresponds with the PFN of
PHYS_OFFSET.  We have something with two different origins being
compared, which is nonsense.  So something is wrong somewhere,
and my money is on max_low_pfn.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core