Re: [Xen-devel] [PATCH 03 of 38] swiotlb: allow architectures tooverrideswiotlb pool allocation

["Jan Beulich" <jbeulich@xxxxxxxxxx>]

>>> Jeremy Fitzhardinge <jeremy@xxxxxxxx> 14.11.08 20:33 >>>
>Jan Beulich wrote:
>> Not directly related to this patch alone, but to the combined set of changes
>> to swiotlb: I don't see any handling of CONFIG_HIGHMEM here (or at least
>> a note that this a known limitation needing work). I mention this because
>> this was the largest part of the changes I had posted long ago to make
>> lib/swiotlb.c Xen-ready, and which got rejected due to their ugliness.
>>   
>
>Was that Andi's objection on the grounds that he didn't think that Xen 
>should need swiotlb at all?

No, Tony Luck actually merged it, but someone else (I don't recall who it
was) requested it to be reverted again.

>I have to admit I didn't follow that thread very closely (or threads, as 
>I seem to remember).  Do you have a pointer to the pertinent bits?

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=51099005ab8e09d68a13fea8d55bc739c1040ca6

>> While perhaps less intrusive to take care of, I also didn't see an equivalent
>> of the range_straddles_page_boundary() logic, without which I can't see
>> how this would work in the common case.
>>   
>Could you be more specific?  The swiotlb allocation should be machine 
>contiguous and so there's no stradding required, but I think I'm missing 
>your point.

The question is whether a multi-page piece of memory must be funneled
through the swiotlb in the first place. In native code, checking whether
the first/last byte satisfies the address_needs_mapping() check is
sufficient, but in Xen you also need to check whether the known to be
physically contiguous pages are also machine-contiguous.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel