Hi Hans, On Saturday 15 September 2012 09:41:59 Hans Verkuil wrote: > On Fri September 14 2012 23:05:45 Sakari Ailus wrote: > > Rémi Denis-Courmont wrote: > > > Le vendredi 14 septembre 2012 23:25:01, Sakari Ailus a écrit : > > >> I had a quick discussion with Laurent, and what he suggested was to use > > >> the kernel version to figure out the type of the timestamp. The drivers > > >> that use the monotonic time right now wouldn't be affected by the new > > >> flag on older kernels. If we've decided we're going to switch to > > >> monotonic time anyway, why not just change all the drivers now and > > >> forget the capability flag. > > > > > > That does not work In Real Life. > > > > > > People do port old drivers forward to new kernels. > > > People do port new drivers back to old kernels > > > > Why would you port a driver from an old kernel to a new kernel? Or are > > you talking about out-of-tree drivers? > > More likely the latter. > > > If you do port drivers across different kernel versions I guess you're > > supposed to use the appropriate interfaces for those kernels, too. Using > > a helper function helps here, so compiling a backported driver would > > fail w/o the helper function that produces the timestamp, forcing the > > backporter to notice the situation. > > > > Anyway, I don't have a very strict opinion on this, so I'm okay with the > > flag, too, but I personally simply don't think it's the best choice we > > can make now. Also, without converting the drivers now the user space > > must live with different kinds of timestamps much longer. > > There are a number of reasons why I want to go with a flag: > > - Out-of-tree drivers which are unlikely to switch to monotonic in practice > - Backporting drivers > - It makes it easy to verify in v4l2-compliance and enforce the use of > the monotonic clock. > - It's easy for apps to check. > > The third reason is probably the most important one for me. There tends to > be a great deal of inertia before changes like this are applied to new > drivers, and without being able to (easily) check this in v4l2-compliance > more drivers will be merged that keep using gettimeofday. It's all too easy > to miss in a review. If we switch all existing drivers to monotonic timestamps in kernel release 3.x, v4l2-compliance can just use the version it gets from VIDIOC_QUERYCAP and enforce monotonic timestamps verification if the version is >= 3.x. This isn't more difficult for apps to check than a dedicated flag (although it's less explicit). My concern is identical to Sakari's, I'm not very keen on introducing a flag that all drivers will set in the very near future and that we will have to keep around forever. > That doesn't mean that it isn't a good idea to convert existing drivers > asap. But it's not something I'm likely to take up myself. Sakari, are you volunteering for that ? ;-) > Creating a small helper function as you suggested elsewhere is a good idea > as well. I'll write something for that. -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
