On 11/19/2018 10:32 PM, Jason Gunthorpe wrote:
On Mon, Nov 19, 2018 at 10:25:37PM -0500, Tom Talpey wrote:
Absolutely, but if the offset walks into the high-order invariant bits,
they won't be invariant any longer. It would still work, but it would
defeat the wireshark trick that Chuck wants to play.
Sure, OK, if you care about the bits being invarient, and not simply
randomized :)
Well, that was Chuck's proposal, to use the XID. SMB Direct originally
used a formula, then changed to randomized, and in future will likely
change it again. The point is, it's the RDMA consumer's choice, not a
verb, or adapter, or protocol, requirement.
For many adaptors that means PAGE_SIZE or huge
page size level alignment..
I'm not sure I understand this. The FRWR hides the native page size, by
providing the page frames in the local work request. Their size isn't
known to the protocol.
Ah, some HW has limitations in how it can construct a VA. In many
implementation the low order bits of the VA must match the offset into
the page. If the page is 4K that is 12 bits, if it is 2M that is 21
bits..
So there should be restrictions on what start VA can be assigned to a
FRWR.
I think only the FBO (First Byte Offset) is really restricted. But
indeed yes, the number of possible FBO bits is a function of the page
size. So, if so, we agree - ok.
Tom.
