Date: Sun, 06 Apr 2003 23:32:40 +0200 From: Dag-Erling Smorgrav <des@ofug.org> To: Nate Lawson <nate@root.org> Cc: Tor.Egge@cvsup.no.freebsd.org Subject: Re: cvs commit: src/sys/conf options.i386 src/sys/i386/i386 tsc.c src/sys/i386/conf NOTES Message-ID: <xzpof3js45j.fsf@flood.ping.uio.no> In-Reply-To: <Pine.BSF.4.21.0304061338460.23366-100000@root.org> (Nate Lawson's message of "Sun, 6 Apr 2003 13:39:27 -0700 (PDT)") References: <Pine.BSF.4.21.0304061338460.23366-100000@root.org>
next in thread | previous in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
There wouldn't be much point in that unless the SMP_TSC option also
forced the TSC to be selected at boot time. On most SMP systems, the
PIIX timecounter is automatically selected by virtue of being
discovered last.
On a related note, the attached patch converts the SMP_TSC option into
a boot-time tunable. It also enables the TSC on single-CPU systems
running an SMP kernel, since there is no synchronization problem with
a single CPU.
I must have bad benchmark karma, BTW - I can see no reduction of
context switch time or any other significant performance boost when
using the TSC on my dual Celeron system.
DES
--
Dag-Erling Smorgrav - des@ofug.org
[-- Attachment #2 --]
Index: conf/options.i386
===================================================================
RCS file: /home/ncvs/src/sys/conf/options.i386,v
retrieving revision 1.189
diff -u -r1.189 options.i386
--- conf/options.i386 4 Apr 2003 23:54:46 -0000 1.189
+++ conf/options.i386 6 Apr 2003 21:16:49 -0000
@@ -41,7 +41,6 @@
CLK_CALIBRATION_LOOP opt_clock.h
CLK_USE_I8254_CALIBRATION opt_clock.h
CLK_USE_TSC_CALIBRATION opt_clock.h
-SMP_TSC opt_clock.h
TIMER_FREQ opt_clock.h
NO_F00F_HACK opt_cpu.h
Index: i386/conf/NOTES
===================================================================
RCS file: /home/ncvs/src/sys/i386/conf/NOTES,v
retrieving revision 1.1083
diff -u -r1.1083 NOTES
--- i386/conf/NOTES 4 Apr 2003 23:54:46 -0000 1.1083
+++ i386/conf/NOTES 6 Apr 2003 21:18:01 -0000
@@ -247,11 +247,6 @@
options CLK_USE_I8254_CALIBRATION
options CLK_USE_TSC_CALIBRATION
-# One some SMP mainboards, the TSCs can be used in SMP mode due to
-# them being synchronized. This can significantly reduce the context
-# switch cost.
-options SMP_TSC
-
#####################################################################
# MISCELLANEOUS DEVICES AND OPTIONS
Index: i386/i386/tsc.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/tsc.c,v
retrieving revision 1.198
diff -u -r1.198 tsc.c
--- i386/i386/tsc.c 4 Apr 2003 23:54:46 -0000 1.198
+++ i386/i386/tsc.c 6 Apr 2003 21:30:43 -0000
@@ -30,11 +30,12 @@
#include <sys/param.h>
#include <sys/systm.h>
+#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/timetc.h>
#include <sys/kernel.h>
-#include <sys/sysctl.h>
#include <sys/power.h>
+#include <sys/smp.h>
#include <machine/clock.h>
#include <machine/md_var.h>
#include <machine/specialreg.h>
@@ -43,6 +44,13 @@
int tsc_is_broken;
u_int tsc_present;
+#ifdef SMP
+static int smp_tsc;
+SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RD, &smp_tsc, 0,
+ "Indicates whether the TSC is safe to use in SMP mode");
+TUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc);
+#endif
+
static unsigned tsc_get_timecount(struct timecounter *tc);
static struct timecounter tsc_timecounter = {
@@ -77,14 +85,17 @@
if (bootverbose)
printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq);
-#if defined(SMP) && !defined(SMP_TSC)
+#ifdef SMP
/*
- * We can not use the TSC in SMP mode, until we figure out a
- * cheap (impossible), reliable and precise (yeah right!) way
- * to synchronize the TSCs of all the CPUs.
- * Modern SMP hardware has the ACPI timer and we use that.
+ * We can not use the TSC in SMP mode unless the TSCs on all CPUs
+ * are somehow synchronized. Some hardware configurations do
+ * this, but we have no way of determining whether this is the
+ * case, so we do not use the TSC in multi-processor systems
+ * unless the user indicated (by setting kern.timecounter.smp_tsc
+ * to 1) that he believes that his TSCs are synchronized.
*/
- return;
+ if (mp_ncpus > 1 && !smp_tsc)
+ return;
#endif
/*
Index: sys/timetc.h
===================================================================
RCS file: /home/ncvs/src/sys/sys/timetc.h,v
retrieving revision 1.56
diff -u -r1.56 timetc.h
--- sys/timetc.h 29 Jan 2003 11:29:22 -0000 1.56
+++ sys/timetc.h 6 Apr 2003 21:23:57 -0000
@@ -64,4 +64,8 @@
void tc_setclock(struct timespec *ts);
void tc_ticktock(void);
+#ifdef SYSCTL_DECL
+SYSCTL_DECL(_kern_timecounter);
+#endif
+
#endif /* !_SYS_TIMETC_H_ */
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?xzpof3js45j.fsf>