> -----Original Message----- > From: Richard Cochran <richardcoch...@gmail.com> > Sent: Saturday, October 12, 2019 4:27 PM > To: Keller, Jacob E <jacob.e.kel...@intel.com> > Cc: netdev@vger.kernel.org; Intel Wired LAN > <intel-wired-...@lists.osuosl.org>; > Kirsher, Jeffrey T <jeffrey.t.kirs...@intel.com>; Brandon Streiff > <brandon.stre...@ni.com> > Subject: Re: [net-next v3 3/7] mv88e6xxx: reject unsupported external > timestamp flags > > On Sat, Oct 12, 2019 at 07:36:31PM +0000, Keller, Jacob E wrote: > > Right, so in practice, unless it supports both edges, it should reject > > setting both > RISING and FALLING together. > > Enforcing that now *could* break existing user space, but I wonder > whether any programs would actually be affected. > > Maybe we can add a STRICT flag than requests strict checking. If user > space uses the "2" ioctl, then we would add this flag before invoking > the driver callback. > > Thanks, > Richard
That could work. I don't know how much it's worth fixing it, but that would be the right way to fix it, I think. I think the strict flag should do the following: a) enforce that you must set at least one of the two edge flags (ensuring that a request to timestamp without one of the edges is rejected) b) drivers *must* honor the edge flags or exit with a rejection if they can't support it. (unlike without 'strict', which would be like today and driver defined). c) possibly: reject both flags at once since it doesn't really make sense to create a timestamp for each edge. At least, I think that makes it harder to actually use the timestamps since you need to be more careful to separate the two timestamps. I'm not 100% sure on (c) but (b) can only be implemented in drivers, since the stack wouldn't know which modes the driver supports. Thanks, Jake
