On Wed, Jun 23, 2010 at 10:11:47AM +0200, Henrik Rydberg wrote: > Dmitry Torokhov wrote: > > > Overall I am starting getting concerned about proper isolation between > > clients. Right now, if one client stops reading events and another one > > issues grab then the first client will only get events that were > > accumulated before grab tookm place. With the new shared buffer the > > first client may get "grabbed" events if it stop for long enough for > > buffer to wrap around. > > Doing some research, the semantics of ioctl have obviously been discussed > before, and I believe this points to another such issue. When grabbing a device, > are we guaranteeing that the device no longer sends events to other clients, or > are we guaranteeing that other clients can no longer read the device? If the > latter, clearing all client buffers in conjunction with a grab would be > appropriate, and would solve this issue. Yes, I think it would be acceptable approach. > > > Do we really same that much memory here? We normally do not have that > > many users connected to event devices at once... > > Ok, let's scratch this. Although I think the idea of multi-reader buffers is > sound, it is obviously sufficiently incompatible with the current approach to > produce distastefully complex patches. I will return with a new set which only > fixes the buffer resize problem, and leaves the rest for later. > Right, let's merge this and also MT slots and revisit this issue at some later point. Thanks. -- Dmitry -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
