Re: [PATCH] IB/ipoib: Change To max_cm_mtu when changing mode to connected

[Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1650 bytes --]

On Tue, 2015-06-09 at 14:17 +0300, Erez Shitrit wrote:
> On Tue, Jun 9, 2015 at 12:04 AM, Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> > On Mon, 2015-06-08 at 10:42 -0600, Jason Gunthorpe wrote:
> >> On Sun, Jun 07, 2015 at 01:36:11PM +0300, Erez Shitrit wrote:
> >> >
> >> > When switching between modes (datagram / connected) change the MTU
> >> > accordingly.
> >> > datagram mode up to 4K, connected mode up to (64K - 0x10).
> >>
> >> Is this a bug fix (describe the user visible impact)? Should it go to
> >> -stable? Add a Fixes: line?
> >
> > I'm not sure I would call it a bug.  Setting the MTU to max
> > automatically is a policy decision more than anything else.  Currently,
> > you have to enable connected mode, then set the MTU.  Both initscripts
> > and NetworkManager do this for you on Red Hat, so the user sees no
> > difference here.  I can't speak to other OSes.  I'm OK with setting
> > connected mode MTU to max by default once we get the scatter/gather
> > support for IPoIB added.
> 
> Hi Doug,
> There is no connection to the scatter/gather patch in that fix.

Yes, I'm well aware of that.  However, 64k MTU in connected mode does
order 5 page allocations.  That has been the source of a number of bugs
in our bug tracking system in the past.  I wouldn't make it default to
turning on 64k MTU without user intervention for that reason.  Once the
SG patch for IPoIB is included, that problem goes away and I'm OK making
IPoIB turn on the large MTU by default.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
              GPG KeyID: 0E572FDD


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]