[openib-general][patch review] srp: fmr implementation,

[Or Gerlitz]

On 3/22/06, Vu Pham <vuhuong at mellanox.com> wrote:

>> Is there any reason to make the IO virtual address of the FMR equal to
>> the local bus address?

> No

I kind of sure to remember that the Mellanox HW requires that

(F)MR IO VA (mod page_shift) === the SG first Bus Address (mod page_shift)

this causes code that uses FMR to have IO VA = SG first bus addr

>> If we made the IO address 0 all the time, then the code would be
>> simpler and there wouldn't even be a need for the io_addr member of
the structure.

> It's a good idea. Here is the patch (this also does not have
> reformatting issue from previous patch)

If I am correct then this would work only for SG whose first element
has offset 0

Vu/Dror ???

Or.