On 28/12/2017 19:14, Jason Gunthorpe wrote: > On Thu, Dec 28, 2017 at 07:10:31PM +0200, Mark Bloch wrote: > >> Well, it's not only that, moving the init srcu to the INIT stage >> will also make me wrap it around with CONFIG_INFINIBAND_ON_DEMAND_PAGING >> which I really don't like seeing in .c files, but we are already doing that >> so .... :) > > Don't micro-optimize so much. Who cares if the few struct members are > allocated when !CONFIG_INFINIBAND_ON_DEMAND_PAGING ? Always allocate > the SRCU and always initialize it. > Well, I didn't write the ODP code, so I don't know why they did it this way. Any place today that accesses that srcu is already under that CONFIG option, so the init srcu can also stay under that CONFIG. > Maybe even get rid of CONFIG_INFINIBAND_ON_DEMAND_PAGING ? > Never understood why we had it in the first place. > I'll talk with the guys that did/touch last the ODP code and see what they have to say, if there isn't a valid reason I'll post a patch that cleans this entire thing. >> If you feel that strongly about it, let me know and I'll change that, >> as well as the comment you had about the const in a different patch >> and we'll send v1 with the fixes. > > The function call back thing is really not nice, so I would be happy > to see it go away. > So I'll move the lock to INIT, but leave it under the ODP CONFIG option for now. If there isn't (It's already on my TODO list to ask around) a good reason for this CONFIG I'll send a patch that addresses that, sound good? > Jason > Mark -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
