is_xen_swiotlb_buffer() vs CONFIG_HIGHMEM

is_xen_swiotlb_buffer() vs CONFIG_HIGHMEM

[Jan Beulich]

Jeremy,

shouldn't is_xen_swiotlb_buffer() also check against e.g. highstart_pfn
to deal with wrap-around potentially leading to aliasing of
[xen_io_tlb_start,xen_io_tlb_end)?

Jan

Re: is_xen_swiotlb_buffer() vs CONFIG_HIGHMEM

[Jeremy Fitzhardinge]

On 12/14/2009 07:02 AM, Jan Beulich wrote:
> shouldn't is_xen_swiotlb_buffer() also check against e.g. highstart_pfn
> to deal with wrap-around potentially leading to aliasing of
> [xen_io_tlb_start,xen_io_tlb_end)?
>    

So what you're saying is that if the pfn is for a high page, then 
"__va(pfn << PAGE_SHIFT);" is simply invalid.  The resulting addr is 
undefined, and may end up erroneously aliasing with the xen tlb buffer.

Do I understand your concern properly?

     J

Re: is_xen_swiotlb_buffer() vs CONFIG_HIGHMEM

[Jan Beulich]

>>> Jeremy Fitzhardinge <jeremy@goop.org> 15.12.09 19:34 >>>
>On 12/14/2009 07:02 AM, Jan Beulich wrote:
>> shouldn't is_xen_swiotlb_buffer() also check against e.g. highstart_pfn
>> to deal with wrap-around potentially leading to aliasing of
>> [xen_io_tlb_start,xen_io_tlb_end)?
>>    
>
>So what you're saying is that if the pfn is for a high page, then 
>"__va(pfn << PAGE_SHIFT);" is simply invalid.  The resulting addr is 
>undefined, and may end up erroneously aliasing with the xen tlb buffer.
>
>Do I understand your concern properly?

Yes.

Jan