On Mon, 2018-12-03 at 17:33 +0000, Jason Gunthorpe wrote:
> On Mon, Dec 03, 2018 at 12:26:10PM -0500, Doug Ledford wrote:
>
> > > > A per-char put_user() sounds pretty expensive since you are having to do
> > > > a __uaccess_begin();__uaccess_end() around every put_user(). I took it
> > > > as is, but it might be worth considering doing something like a 32
> > > > byte
> > >
> > > This patch requries the last series of write() handlers before it will
> > > work properly, don't apply it just yet - will create a crazy security
> > > problem..
> >
> > Yes, I thought I saw you say you applied it already? And I started from
> > your wip/jgg-for-next that's currently pushed. Was I incorrect in that?
>
> That is the right branch, but there are three series reworking write,
> only the first two are applied, the last one still in patchworks
> didn't get applied yet..
Ahh, see, the cover page for this series states:
Based on "Have the core handler dispatch code create the driver udata"
series:
And I saw that series was applied. No big deal, nothing pushed, so I'll
leave this in its current branch. The final one needed is this one then
(a single patch, not a series), yes?
[rdma-next] RDMA/uverbs: Implement an ioctl that can call write and
write_ex handlers
> And this patch has to go after the prior three for all the bits to
> work properly, and as a uapi change it should ideally wait for the
> corresponding rdma-core patch to be posted..
Fair enough. Is that coming soon?
--
Doug Ledford <dledford@xxxxxxxxxx>
GPG KeyID: B826A3330E572FDD
Key fingerprint = AE6B 1BDA 122B 23B4 265B 1274 B826 A333 0E57 2FDD
Attachment:
signature.asc
Description: This is a digitally signed message part
