Re: [LAD] timers

On Sat, Nov 7, 2009 at 2:51 PM, Emanuel Rumpf <xbran(a)web.de> wrote: correct, to a large extent. Modern motherboads come with timers that are capable of massively higher precision than older ones, and allow for very precise scheduling of timer interrupts quite along time in advance of the timer going off. The existence of this kind of capability can be used to implement system scheduling in some moderately different ways. HR-Timer is the name given to some kernel patches (now in the mainstream kernel) that allowed the use of this timing hardware. The changes were a mix of new code that deals with the new capabilities, and patches that reimplement the existing system timer/scheduling design using the new hardware. So, HR-Timer (patches/code/design) requires a high res clock (like HPET). Note that one of the possible ways to use the HR Time infrastructure is to go totally tickless. In this design, the kernel is no longer woken at fixed, regular intervals as has been the case for the entire history of Linux. Instead, the kernel constantly adds new deadline times which, when expired, generate an interrupt. In the very, very unlikely case of there being only one process that is runnable, you can it without any timer deadlines at all (until some other process becomes runnable). Obviously, its rather complex, but also totally awesome stuff. This is separate from HR-Timer, but relies upon it, and it generally referred to as a "tickless kernel". the HR-Timer infrastructure is a MUCH better basis for scheduling MIDI. Even with RTC=10000 (10^4), you still don't have a clock that is running in sync with the nominal MIDI data rate. with HR-Timer in place, you can actually schedule the delivery of MIDI bytes precisely on time ("wake me up in another 0.3msecs") and not rely on a metronomic tick like RTC.