[openib-general] [PATCH 1 of 5] IB/mthca: fix off-by-one in FMR handling on memfree

[Michael S. Tsirkin]

>  > There is a chance that this is the core of the 32-bit OS on 64-bit
>  > platforms bug we had.
>  > I run a certain test (copy 1G size file to the SRP disk) with and
>  > without this patch.
>  > With the patch I got a panic, and without it the test passed.
> 
> It's good to hear that this patch fixes the bug.  However I wonder why
> I never got a bug report for this problem. And I also wonder why this
> bug would only affect 32-bit OS on a 64-bit platform....

I have a theory for this: for the bug to trigger, you need your low memory
to be fragmented so that there are several entries in the chunk scatterlist
array.

I guess this is more likely on a 32 bit kernel machine with high memory present
than on a 64 bit kernel where all memory is low.

Whereas native 32 bit machines are low end so they likely either have PCI-X
(thus no memfree) or <1G RAM (thus no high memmory).

-- 
MST