Hey Roland, Why don't I use the same scheme as Heiko for now. Then I won't hold up the release. At this point, I can't say for sure what else we might suggest to change in the core APIs anyway. I'll submit a patch down the road to fix the core to handle this ala create_cq(), but it can wait until 1.1.0.
Sound like a plan? Steve. On Wed, 2006-02-15 at 22:18 +0100, Heiko J Schick wrote: > Hello Steven, > > we had this problem, too. For ibv_cmd_create_qp we pass via our cmd > struct > the address of the response block to the kernel.The kernel can copies > then > all values into the response block (from kernel space to user space). > > The provider library can then the provider specific information which > are > passed from the kernel. > > See ehcau_create_qp in ehca_umain.c for more information. > > Regards, > Heiko > > On Feb 15, 2006, at 8:00 PM, Steve Wise wrote: > > > I have a provider in the works that needs provider-specific > > information > > passed from the kernel create_qp verb back to the provider > > library. It > > seems that ibv_cmd_create_qp() doesn't allow for provider-specific > > data > > to be passed out of the kernel back to the provider lib. > > > > ibv_cmd_create_cq() does support this. > > > > I can come up with a patch to the code to support this, but I > > wanted to > > query the group to make sure I'm not missing something. > > > > Is it a reasonable extension to add this to create_qp()? > > > > Thanks, > > > > Steve. > > > > _______________________________________________ > > openib-general mailing list > > [email protected] > > http://openib.org/mailman/listinfo/openib-general > > > > To unsubscribe, please visit http://openib.org/mailman/listinfo/ > > openib-general > > _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
