in-kernel timer benchmark -v3
PATCH 1/1 in-kernel timer benchmark -v3
Happy New Year everyone!
Hope you all survived the turn of the year without complications. ;)
I would like to start with a new patch round. The first one is a revised and extended version of the earlier posted in-kernel timer benchmark based on a RTDM device. Changes since the first version:
* revised command line argument to set the test mode
* display test mode during startup
* display hint on missing benchmark device
* support of trace freeze on latency above specified threshold
* extended benchmark device to provide ipipe-tracing functions to userspace
* further minor fixes
I think the patch should be ready to merge now. This would also be helpful to provide a base on which Gilles can build his timer-scaling tests. BTW, we still need some RTDM abstraction for timers, but this is a different topic that will be addressed later.
Jan
rtdm/rtbenchmark.h +131 -0
@@ -0,0 +1,131 @@ +/** +* @file +* Real-Time Driver Model for Xenomai, benchmark device profile header +* +* @note Copyright (C) 2005 Jan Kiszka <[email protected]> +* +* Xenomai is free software; you can redistribute it and/or modify it +* under the terms of the GNU General Public License as published by +* the Free Software Foundation; either version 2 of the License, or +* (at your option) any later version. +* +* Xenomai is distributed in the hope that it will be useful, but +* WITHOUT ANY WARRANTY; without even the implied warranty of +* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +* General Public License for more details. +* +* You should have received a copy of the GNU General Public License +* along with Xenomai; if not, write to the Free Software Foundation, +* Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +* +* @ingroup rtbenchmark +*/ + +/*! +* @ingroup profiles +* @defgroup rtbenchmark Benchmark Devices +* +* This group of devices is intended to provide in-kernel benchmark results. +* Feel free to comment on this profile via the Xenomai mailing list +* ([email protected]) or directly to the author ([email protected]). @n +* @n +* +* @par Device Characteristics +* @ref rtdm_device.device_flags "Device Flags": @c RTDM_NAMED_DEVICE, @c RTDM_EXCLUSIVE @n +* @n +* @ref rtdm_device.device_name "Device Name": @c "rtbenchmark<N>", N >= 0 @n +* @n +* @ref rtdm_device.device_class "Device Class": @c RTDM_CLASS_BENCHMARK @n +* @n +* +* @par Supported Operations +* @b Open @n +* Environments: non-RT (RT optional)@n +* Specific return values: none @n +* @n +* @b Close @n +* Environments: non-RT (RT optional)@n +* Specific return values: none @n +* @n +* @b IOCTL @n +* Mandatory Environments: see @ref IOCTLs "below" @n +* Specific return values: see @ref IOCTLs "below" @n +* +* @{ +*/ + +#ifndef _RTBENCHMARK_H +#define _RTBENCHMARK_H + +#include <asm/types.h> +#include <xenomai/rtdm/rtdm.h> + +#define RTBNCH_TIMER_TASK 0 +#define RTBNCH_TIMER_HANDLER 1 + +typedef struct rtbnch_result { +long avg; +long min; +long max; +long overruns; +long test_loops; +}rtbnch_result_t ; + +typedef struct rtbnch_timerconfig { +int mode; +uint64_t period; +int warmup_loops; +int histogram_size; +int histogram_bucketsize; +long freeze_threshold; +}rtbnch_timerconfig_t ; + +typedef struct rtbnch_interm_result { +struct rtbnch_result last; +struct rtbnch_result overall; +}rtbnch_interm_result_t ; + +typedef struct rtbnch_overall_result { +struct rtbnch_result result; +long *histogram_avg; +long *histogram_min; +long *histogram_max; +}rtbnch_overall_result_t ; + + +#define RTIOC_TYPE_BENCHMARK RTDM_CLASS_BENCHMARK + + +/*! +* @name Sub-Classes of RTDM_CLASS_BENCHMARK +* @{ */ +#define RTDM_SUBCLASS_TIMER 0 +/** @} */ + + +/*! +* @anchor IOCTLs @name IOCTLs +* Benchmark device IOCTLs +* @{ */ + +#define RTBNCH_RTIOC_INTERM_RESULT \ + _IOWR(RTIOC_TYPE_BENCHMARK, 0x00, struct rtbnch_interm_result) + +#define RTBNCH_RTIOC_START_TMTEST \ + _IOW(RTIOC_TYPE_BENCHMARK, 0x10, struct rtbnch_timerconfig) + +#define RTBNCH_RTIOC_STOP_TMTEST \ + _IOWR(RTIOC_TYPE_BENCHMARK, 0x11, struct rtbnch_overall_result) + +#define RTBNCH_RTIOC_BEGIN_TRACE \ + _IOW(RTIOC_TYPE_BENCHMARK, 0x20, long) + +#define RTBNCH_RTIOC_END_TRACE \ + _IOW(RTIOC_TYPE_BENCHMARK, 0x21, long) + +#define RTBNCH_RTIOC_FREEZE_TRACE \ + _IOW(RTIOC_TYPE_BENCHMARK, 0x22, int) + +/** @} */ + +#endif /* _RTBENCHMARK_H */
rtdm/rtdm.h +1 -0
@@ -68,6 +68,7 @@#define RTDM_CLASS_CAN 3#define RTDM_CLASS_NETWORK 4#define RTDM_CLASS_RTMAC 5 +#define RTDM_CLASS_BENCHMARK 6/* #define RTDM_CLASS_USB ? #define RTDM_CLASS_FIREWIRE ?
testsuite/latency/latency.c +175 -46
335 lines not on this page · raw message
testsuite/latency/runinfo +1 -1
@@ -1 +1 @@ -latency:native:!./latency;popall:control_c +latency:native+rtdm+timerbench:!./latency;popall:control_c
testsuite/latency/Makefile.am +1 -0
@@ -10,6 +10,7 @@ latency_LDADD = \ ../../skins/native/libnative.la \ + ../../skins/rtdm/librtdm.la \ -lpthread -lm install-data-local:
drivers/Kconfig +1 -0
@@ -1 +1,2 @@ source "drivers/xenomai/16550A/Kconfig" +source "drivers/xenomai/benchmark/Kconfig"
drivers/Config.in +1 -0
@@ -8,5 +8,6 @@ mainmenu_option next_comment comment 'Real-time drivers' source drivers/xenomai/16550A/Config.in + source drivers/xenomai/benchmark/Config.in endmenu fi
drivers/benchmark/Kconfig +7 -0
@@ -0,0 +1,7 @@ +config XENO_DRIVERS_TIMERBENCH + depends on XENO_SKIN_RTDM + tristate "Timer benchmark driver" + default n + help + Kernel-based benchmark driver for timer latency evaluation. + See testsuite/latency for a possible front-end.
drivers/benchmark/timerbench.c +538 -0
538 lines not on this page · raw message
drivers/benchmark/Config.in +5 -0
@@ -0,0 +1,5 @@ +# +# Xenomai configuration for Linux v2.4 +# + +dep_tristate 'Timer benchmark driver' CONFIG_XENO_DRIVERS_TIMERBENCH $CONFIG_XENO_SKIN_RTDM
drivers/benchmark/Makefile +30 -0
@@ -0,0 +1,30 @@ +ifeq ($(PATCHLEVEL),6) + +# Makefile frag for Linux v2.6 + +obj-$(CONFIG_XENO_DRIVERS_TIMERBENCH) += xeno_timerbench.o + +xeno_timerbench-y := timerbench.o + +else + +# Makefile frag for Linux v2.4 + +O_TARGET := built-in.o + +obj-$(CONFIG_XENO_DRIVERS_TIMERBENCH) := xeno_timerbench.o + +list-multi := xeno_timerbench.o + +xeno_timerbench-objs := timerbench.o + +export-objs := $(xeno_timerbench-objs) + +EXTRA_CFLAGS += -I$(TOPDIR)/include/xenomai/compat + +include $(TOPDIR)/Rules.make + +xeno_timerbench.o: $(xeno_timerbench-objs) + $(LD) -r -o $@ $(xeno_timerbench-objs) + +endif
drivers/Makefile +3 -1
@@ -2,7 +2,7 @@ # Makefile frag for Linux v2.6 -obj-$(CONFIG_XENOMAI) += 16550A/ +obj-$(CONFIG_XENOMAI) += 16550A/ benchmark/ else
@@ -10,6 +10,8 @@ subdir-$(CONFIG_XENO_DRIVERS_16550A) += 16550A +subdir-$(CONFIG_XENO_DRIVERS_TIMERBENCH) += benchmark + include $(TOPDIR)/Rules.make endif
Quoted 4 lines · Jan Kiszka
Happy New Year everyone!
Hope you all survived the turn of the year without complications. ;)
Nearly buried alive in patches, but this is hardly a complication.
Quoted 4 lines
I would like to start with a new patch round. The first one is a revised and extended version of the earlier posted in-kernel timer benchmark based on a RTDM device. Changes since the first version:
Quoted 8 lines
* revised command line argument to set the test mode
* display test mode during startup
* display hint on missing benchmark device
* support of trace freeze on latency above specified threshold
* extended benchmark device to provide ipipe-tracing functions to userspace
* further minor fixes
Ok, merged.
Quoted 5 lines
I think the patch should be ready to merge now. This would also be helpful to provide a base on which Gilles can build his timer-scaling tests. BTW, we still need some RTDM abstraction for timers, but this is a different topic that will be addressed later.
Btw, since -rc1 is out, I've deprecated the Adeos oldgen support for all archs (ia64 is now I-pipe based too), so that I can merge Dmitry's work later on, exclusively into the newgen codebase. So if anyone still keeps any venerable pre-Ipipe patch in some hideout of his hdd, just let it go now; I know it's heartbreaking, but Xenomai is thankless.
Signature · 1 line
Philippe.
...as a follow-up, here is a patch to fix some bugs in the new trace triggering and to improve (from my POV) the output of this tool in some regards.
Jan
testsuite/latency/latency.c +25 -15
@@ -39,6 +39,11 @@#define TIMER_HANDLER 2int test_mode =USER_TASK ; +const char *test_mode_names[] = { +"periodic user-mode task" , +"in-kernel periodic task" , +"in-kernel timer handler" +};time_t test_start, test_end;/* report test duration */ int test_loops =0 ;/* outer loop count */
@@ -78,8 +83,7 @@return ; } -if (freeze_threshold >0 ) - freeze_threshold =rt_timer_ns2tsc (freeze_threshold*1000 ); + freeze_threshold =rt_timer_ns2tsc (freeze_threshold); nsamples =ONE_BILLION / period_ns; period_tsc =rt_timer_ns2tsc (period_ns);
@@ -122,11 +126,15 @@if (dt < minj) minj = dt; sumj += dt; -if ((freeze_threshold >0 ) && (dt > freeze_threshold)) -rt_dev_ioctl (benchdev,RTBNCH_RTIOC_FREEZE_TRACE , freeze_threshold); +if (!(finished || warmup)) + { +if ((freeze_threshold >0 ) && (dt > freeze_threshold)) +rt_dev_ioctl (benchdev,RTBNCH_RTIOC_FREEZE_TRACE , + freeze_threshold); -if (!(finished || warmup) && (do_histogram || do_stats)) -add_histogram (histogram_avg, dt); +if (do_histogram || do_stats) +add_histogram (histogram_avg, dt); + } }if (!warmup)
@@ -173,8 +181,6 @@return ; } -printf ("== Test mode: periodic userspace task\n" ); - }else {struct rtbnch_timerconfig config;
@@ -196,13 +202,13 @@fprintf (stderr,"latency: failed to start in-kernel timer benchmark, code %d\n" ,err);return ; } - -printf ("== Test mode: in-kernel %s\n" , - (test_mode ==TIMER_HANDLER ) ?"timer handler" :"periodic task" ); }time (&start); +if (WARMUP_TIME ) +printf ("warming up...\n" ); +if (quiet)fprintf (stderr,"running quietly for %d seconds\n" , test_duration);
@@ -256,8 +262,9 @@time_t now, dt;time (&now); dt = now - start -WARMUP_TIME ; -printf ("RTT| %.2ld:%.2ld:%.2ld\n" , - dt /3600 ,(dt /60 ) %60 ,dt %60 ); +printf ("RTT| %.2ld:%.2ld:%.2ld (%s, %Ld us period)\n" , + dt /3600 ,(dt /60 ) %60 ,dt %60 , + test_mode_names[test_mode],period_ns /1000 );printf ("RTH|%12s|%12s|%12s|%8s|%12s|%12s\n" ,"-----lat min" ,"-----lat avg" ,"-----lat max" ,"-overrun" ,"----lat best" ,"---lat worst" );
@@ -469,7 +476,7 @@case 'f' : - freeze_threshold =atoi (optarg); + freeze_threshold =atoi (optarg)*1000 ;break ;default :
@@ -520,7 +527,10 @@setlinebuf (stdout); -printf ("== Sampling period: %Ld us\n" ,period_ns /1000 ); +printf ("== Sampling period: %Ld us\n" +"== Test mode: %s\n" , + period_ns /1000 , + test_mode_names[test_mode]);mlockall (MCL_CURRENT |MCL_FUTURE );
Quoted 3 lines · Jan Kiszka
...as a follow-up, here is a patch to fix some bugs in the new trace triggering and to improve (from my POV) the output of this tool in some regards.
Applied on my own.
Jan