[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] ARM: xehheap_megabytes less than 32

On Mon, 2015-07-20 at 20:00 +0000, Chris (Christopher) Brand wrote:
> Thanks, Ian.
> >That was for x86, but no I don't think it is inherently ridiculous to try 
> >and reduce the Xen heap quite substantially.
> I thought that might have been for x86.
> >One thing you should check though is what happens to domheap allocations 
> >when the domheap is empty but the xenheap still has pages. I have a feeling 
> >>that such allocations can be satisfied from the xenheap, at least under 
> >some circumstances (depending on CONFIG_SEPARATE_XENHEAP?).
> >
> >Rather than trying to unpick xen/common/page_alloc.c and the various 
> >heaps/nodes it might be easier to just confirm experimentally...
> I saw an older email of yours relating to this. Will see if I can try that.
> This did prompt me to try building without CONFIG_SEPARATE_XENHEAP,
> which seems like it should also solve my problem, but the result
> didn't boot (which is a bit surprising, given that it looks like that
> is not set for ARM64).

CONFIG_SEPARATE_XENHEAP is not a user facing option, it is something
which the architecture port can opt in or out of. Turning it off
requires a large 1:1 mapping covering most of RAM, which cannot be done
on arm32 since the 32-bit address space is too limited. Even if it made
sense for arm32 changing this option would also require reworking of the
start of day heap setup code.


> >WRT the 32mb limit, I wonder if that is the same underlying issue as you 
> >tripped over for the frametable mapping?
> I do now see that setup_xenheap_mappings() calls create_32mb_mappings(), 
> which presumably imposes a granularity of 32MB on xenheap_megabytes.
> Chris

Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.