Date: Mon, 7 Jun 2010 21:24:29 -0800 From: Tsuyoshi Ozawa <ozawa@t-oza.net> To: Alexander Motin <mav@freebsd.org> Cc: FreeBSD-Current <freebsd-current@freebsd.org>, freebsd-arch@freebsd.org Subject: Re: RFC: New event timers infrastructure Message-ID: <AANLkTikyWXHYezBW1mAlTU0Oi_Z1qlPbAyZllux_wFSB@mail.gmail.com> In-Reply-To: <AANLkTimfyn0aCs7PFRHL8RONvg03_QVXK_90oy084IIx@mail.gmail.com> References: <4C0C1AE4.8050807@FreeBSD.org> <AANLkTimfyn0aCs7PFRHL8RONvg03_QVXK_90oy084IIx@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
2010/6/6, Alexander Motin <mav@freebsd.org>: > Hi. > > Most of x86 systems now has at least 4 types of event timers: i8254, > RTC, LAPIC and HPET. Respective code in kernel is very tangled, heavily > hardcoded and absolutely not scalable. I have reimplemented it, trying > to solve these issues. > > I did such things: > - created unified timer driver's API (sys/timeet.h, kernel/kern_et.c). > It supports global and per-CPU timers, periodic and one-shot. Provides > driver and consumer interfaces for choosing timers and operating them; > - cleaned existing x86 event timer driver's code and modified it for > new API (x86/isa/atrtc.c, x86/isa/clock.c, x86/x86/local_apic.c). LAPIC > timer is now per-CPU and supports both periodic and one-shot modes; > - extended HPET driver to support it's event timers in periodic and > one-shot mode (dev/acpica/acpi_hpet.c). Support for per-CPU operation > and FSB interrupts planned for later; > - written mostly machine-independent mid-layer for managing any present > timers to provide clocks needed for kernel (x86/x86/timeevents.c). It > supports both global and per-CPU timers. Now it supports only periodic > mode, but one-shot mode support planned for later. > > All this stuff deeply configurable via both loader tunables on boot and > sysctls in real time: > > %sysctl kern.eventtimer > kern.eventtimer.choice: LAPIC(500) HPET(400) HPET1(390) HPET2(390) > i8254(100) RTC(0) > kern.eventtimer.et.LAPIC.flags: 7 > kern.eventtimer.et.LAPIC.frequency: 99752386 > kern.eventtimer.et.LAPIC.quality: 500 > kern.eventtimer.et.HPET.flags: 3 > kern.eventtimer.et.HPET.frequency: 14318180 > kern.eventtimer.et.HPET.quality: 400 > kern.eventtimer.et.HPET1.flags: 3 > kern.eventtimer.et.HPET1.frequency: 14318180 > kern.eventtimer.et.HPET1.quality: 390 > kern.eventtimer.et.HPET2.flags: 3 > kern.eventtimer.et.HPET2.frequency: 14318180 > kern.eventtimer.et.HPET2.quality: 390 > kern.eventtimer.et.RTC.flags: 1 > kern.eventtimer.et.RTC.frequency: 32768 > kern.eventtimer.et.RTC.quality: 0 > kern.eventtimer.et.i8254.flags: 1 > kern.eventtimer.et.i8254.frequency: 1193182 > kern.eventtimer.et.i8254.quality: 100 > kern.eventtimer.timer2: NONE > kern.eventtimer.timer1: i8254 > kern.eventtimer.singlemul: 2 > > By default system chooses two timers with highest "quality" for > hardclock and statclock/profclock. User may affect that choice via > disabling unwanted drivers and/or via direct specification of wanted > ones. It is possible to change timers on-flight via sysctls: > > %sysctl kern.eventtimer.timer1=hpet > kern.eventtimer.timer1: i8254 -> HPET > %sysctl kern.eventtimer.timer2=hpet1 > kern.eventtimer.timer2: NONE -> HPET1 > > After every timer change, if two timers available, mid-layer > cross-checks them, and if one of them is not functional - replaces it. > > If there is no second timer available, or user specified to not use it - > mid-layer automatically increases rate of the first timer and divide > it's frequency to satisfy system needs as good as possible. User may > specify how fast he wish to run fist timer relative to hz by setting > kern.eventtimer.singlemul tunable/sysctl. > > When profiling is active, mid-layer automatically rises respective timer > frequency to about 8KHz (was 1KHz previously) and decreases it back on > profiling end. > > All above was tested on i386 and amd64. XEN was not affected and builds > fine. pc98 was slightly touched. It wasn't tested, but builds fine. It's > pc98/cbus/clock.c needs respective rewrite to use new features. Other > architectures are untouched, but if any of them may benefit from this > functionality - it should be possible to share most of the code. > > Latest patches can be found here: > http://people.freebsd.org/~mav/et.20100606.patch > > Known issues: > - i8254 timer generates 18Hz interrupt rate when not used and not > disabled. I haven't found a way to disable it's interrupt source while > holding spinlock. > - timer drivers code will need some more cleaning after interrupt > handler will be able to return both argument and frame same time. > > Feedback is very appreciated. > > -- > Alexander Motin > This is excellent! I'll try this to apply this patch and rewrite my old dynamic ticks code to fit this event timer API . Thank you, Alexander ! -- Tsuyoshi Ozawa <ozawa@t-oza.net> 2010/06/07 7:03 "Alexander Motin" <mav@freebsd.org>: Hi. Most of x86 systems now has at least 4 types of event timers: i8254, RTC, LAPIC and HPET. Respective code in kernel is very tangled, heavily hardcoded and absolutely not scalable. I have reimplemented it, trying to solve these issues. I did such things: - created unified timer driver's API (sys/timeet.h, kernel/kern_et.c). It supports global and per-CPU timers, periodic and one-shot. Provides driver and consumer interfaces for choosing timers and operating them; - cleaned existing x86 event timer driver's code and modified it for new API (x86/isa/atrtc.c, x86/isa/clock.c, x86/x86/local_apic.c). LAPIC timer is now per-CPU and supports both periodic and one-shot modes; - extended HPET driver to support it's event timers in periodic and one-shot mode (dev/acpica/acpi_hpet.c). Support for per-CPU operation and FSB interrupts planned for later; - written mostly machine-independent mid-layer for managing any present timers to provide clocks needed for kernel (x86/x86/timeevents.c). It supports both global and per-CPU timers. Now it supports only periodic mode, but one-shot mode support planned for later. All this stuff deeply configurable via both loader tunables on boot and sysctls in real time: %sysctl kern.eventtimer kern.eventtimer.choice: LAPIC(500) HPET(400) HPET1(390) HPET2(390) i8254(100) RTC(0) kern.eventtimer.et.LAPIC.flags: 7 kern.eventtimer.et.LAPIC.frequency: 99752386 kern.eventtimer.et.LAPIC.quality: 500 kern.eventtimer.et.HPET.flags: 3 kern.eventtimer.et.HPET.frequency: 14318180 kern.eventtimer.et.HPET.quality: 400 kern.eventtimer.et.HPET1.flags: 3 kern.eventtimer.et.HPET1.frequency: 14318180 kern.eventtimer.et.HPET1.quality: 390 kern.eventtimer.et.HPET2.flags: 3 kern.eventtimer.et.HPET2.frequency: 14318180 kern.eventtimer.et.HPET2.quality: 390 kern.eventtimer.et.RTC.flags: 1 kern.eventtimer.et.RTC.frequency: 32768 kern.eventtimer.et.RTC.quality: 0 kern.eventtimer.et.i8254.flags: 1 kern.eventtimer.et.i8254.frequency: 1193182 kern.eventtimer.et.i8254.quality: 100 kern.eventtimer.timer2: NONE kern.eventtimer.timer1: i8254 kern.eventtimer.singlemul: 2 By default system chooses two timers with highest "quality" for hardclock and statclock/profclock. User may affect that choice via disabling unwanted drivers and/or via direct specification of wanted ones. It is possible to change timers on-flight via sysctls: %sysctl kern.eventtimer.timer1=hpet kern.eventtimer.timer1: i8254 -> HPET %sysctl kern.eventtimer.timer2=hpet1 kern.eventtimer.timer2: NONE -> HPET1 After every timer change, if two timers available, mid-layer cross-checks them, and if one of them is not functional - replaces it. If there is no second timer available, or user specified to not use it - mid-layer automatically increases rate of the first timer and divide it's frequency to satisfy system needs as good as possible. User may specify how fast he wish to run fist timer relative to hz by setting kern.eventtimer.singlemul tunable/sysctl. When profiling is active, mid-layer automatically rises respective timer frequency to about 8KHz (was 1KHz previously) and decreases it back on profiling end. All above was tested on i386 and amd64. XEN was not affected and builds fine. pc98 was slightly touched. It wasn't tested, but builds fine. It's pc98/cbus/clock.c needs respective rewrite to use new features. Other architectures are untouched, but if any of them may benefit from this functionality - it should be possible to share most of the code. Latest patches can be found here: http://people.freebsd.org/~mav/et.20100606.patch Known issues: - i8254 timer generates 18Hz interrupt rate when not used and not disabled. I haven't found a way to disable it's interrupt source while holding spinlock. - timer drivers code will need some more cleaning after interrupt handler will be able to return both argument and frame same time. Feedback is very appreciated. -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTikyWXHYezBW1mAlTU0Oi_Z1qlPbAyZllux_wFSB>