Messages in this thread | | | Subject | Re: Initial bits to help pull jiffies out of drivers | From | Arjan van de Ven <> | Date | Tue, 03 Aug 2004 10:27:16 +0200 |
| |
On Tue, 2004-07-27 at 21:59, Alan Cox wrote: > This is really for comment, the basic idea is to add some relative > timer functionality. This gives us timeout objects as well as pulling > jiffies use into one place in the timer code. The need for the old > interfaces never goes away however because some code uses a previous > event base to construct timeouts to avoid sliding due to the latency > between service and re-addition. > > (please cc me on comments)
My gripe with this is that the interface still is relative-to-HZ time. I'm convinced that driver(writers) are better off with an absolute time interface, eg add_timeout_ms(), add_timeout_us() etc. (which btw also give a hint about the accuracy required, so that the kernel can group milisecond delays together even when they got scheduled at different usecs, once we get timers that accurate)
[unhandled content-type:application/pgp-signature] | |