Messages in this thread | | | Date | Sun, 23 Nov 2008 20:57:39 -0800 (PST) | From | Davide Libenzi <> | Subject | Re: poll: allow f_op->poll to sleep, take #3 |
| |
On Mon, 24 Nov 2008, Tejun Heo wrote:
> Ingo Molnar wrote: > >> And @sync might be useful depending on who's waking it up, so we > >> either need to change the wake interface or give it an easier way to > >> pass those parameters as received. The callback function isn't the > >> right place to ignore those parameters. It simply doesn't know why > >> the caller is passing them in or what they mean under the > >> circumstances. > > > > We'll likely eliminate the 'sync' parameter from the scheduler. It's > > not a flag that should be proliferated. > > But it's still being used in quite hot paths (pipe, splice, socket) > and I don't really wanna mix up a change which can cause subtle > scheduling related performance regression into this patch. How about > using the dummy waitqueue hack for now and when removing the @sync > param, switch it to one of wakeup APIs? I'll be happy to add big /* > TODO */ comment in the function.
I was noticing that wake ups on the poll queues do not use sync=1, AFAICS. So using wake_up_state() and ignoring 'sync' should be fine. And 'key' is already ignored by default by default_wake_function().
- Davide
| |