Or Gerlitz wrote:
Eli Cohen wrote:
I think you're right and that's how it should have been done. However
querying device cap will cause quite a few changes in other patches so
I think we should better leave this as it is right now. I will however
re-generate the patches for 2.6.25 .
you can not leave this as is for ofed 1.3, b/c as Roland said, no one
sets the value of ca->flags in the NON mellanox hw drivers (eg ehca,
ipath) so ipoib basically tests a random value, do you think it can
remain that way?
I think the flags member in clear when the object is allocated so there should
be no problem with this. I will re-check though.
Basically, I think you can query the device once and have the attributes
stored at the priv structure, so you can use it for deciding on the S/G,
checksum, LSO (RSS...) capabilities, etc.
Yes, that's what I have in mind to do.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
