Re: [RFC PATCH 2/3] arch, arm32: add the XEN_DOMCTL_memory_mapping hypercall

[Julien Grall <julien.grall@linaro.org>]

On 05/03/14 21:59, Arianna Avanzini wrote:
> Hello,

Hello Arianna,

>> The comment "1:1 iomem mapping (gfn == mfn)" seems wrong here. The
>> implementation you wrote allow gfn != mfn.
>>
>
> Right, thank you.
> Sorry if I bother you with one more question about this hunk; aside from the
> error pointed out by Eric with the usage of pfn instead of paddr_t, is it OK to
> map the address range all at once with map_mmio_regions() as proposed? I noticed
> that the x86-related code seems to map the range one mfn at a time with
> set_mmio_p2m_entry(), and I wonder if this different approach has any benefit I
> didn't think about.

map_mmio_regions will take p2m lock and do MMU management (e.g TLB 
flush...). For the latter the code  has been written to only do TLB 
flush, if needed, once per call. More the range is large, better it is.

So, it's better to call the function only once for the whole range.

Regards,

-- 
Julien Grall