Christoph Hellwig wrote:
Specifically, does something like makes sense?
struct scatterlist my_sg; /* somewhere, but iser_send_command  stack */

if(!sc->use_sg) {
        my_sg.page   = virt_to_page(sc->request_buffer);
        my_sg.length = sc->request_bufflen;
        my_sg.offset = 0;
}

now continue as ususal to process my_sg (it can't be on the stack

Yes, that makes sense for now.  There's even a sg_init_one helper
to do the legwork for you.  But before iser support goes into mainline
that'll be onbsolete already and can be removed again.

Thanks for pointing me to sg_init_one (which saves me from my buggy code suggestion above which always sets zero offset...).

So the iser code will be cleaned in two phases with respect "single sc":

+1 assume everywhere the sc has SG and just SG-fy the singles.
+2 remove the SG-fy trick and maintain it as a patch for backports

Or.




_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to