[PATCH] gettimeoffset for swarm

[PATCH] gettimeoffset for swarm

[Jun Sun]

[-- Attachment #1: Type: text/plain, Size: 553 bytes --]


By default, swarm will use calibrate_div64_gettimeoffset().  That does
not work in SMP mode because the two cores have different counter register
value.  This patch gives swarm its own and accurate gettimeoffset().

The symptom without this patch is that cpu core 1 does not have micro-second
resolution of gettimeofday().

It applies to 2.4 branch with 1 or 2 line offset warnings.  It does not apply 
to 2.5 branch. 2.5 branch appears to be outdated.

http://linux.junsun.net/patches/oss.sgi.com/submitted/020515.swarm-own-gettimeoffset.patch

Jun



[-- Attachment #2: 020515.swarm-own-gettimeoffset.patch --]
[-- Type: text/plain, Size: 2963 bytes --]


By default, swarm will use calibrate_div64_gettimeoffset().  That does
not work in SMP mode because the two cores have different counter register
value.  This patch gives swarm its own and accurate gettimeoffset().

The symptom without this patch is that cpu core 1 does not have micro-second
resolution of gettimeofday().

Jun

diff -Nru linux/arch/mips/sibyte/sb1250/time.c.orig linux/arch/mips/sibyte/sb1250/time.c
--- linux/arch/mips/sibyte/sb1250/time.c.orig	Tue Apr 30 17:29:25 2002
+++ linux/arch/mips/sibyte/sb1250/time.c	Fri May 10 14:53:39 2002
@@ -34,6 +34,7 @@
 #include <asm/ptrace.h>
 #include <asm/addrspace.h>
 #include <asm/time.h>
+#include <asm/debug.h>
 
 #include <asm/sibyte/sb1250.h>
 #include <asm/sibyte/sb1250_regs.h>
@@ -107,3 +108,19 @@
 	 */
 	ll_local_timer_interrupt(0, regs);
 }
+
+/*
+ * We use our own do_gettimeoffset() instead of the generic one,
+ * because the generic one does not work for SMP case.
+ * In addition, since we use general timer 0 for system time,
+ * we can get accurate intra-jiffy offset without calibration.
+ */
+unsigned long sb1250_gettimeoffset(void)
+{
+	unsigned long count =
+		in64(KSEG1 + A_SCD_TIMER_REGISTER(0, R_SCD_TIMER_CNT));
+
+	db_assert(count <= 1000000 / HZ);
+	return 1000000/HZ - count;
+}
+
diff -Nru linux/arch/mips/sibyte/swarm/setup.c.orig linux/arch/mips/sibyte/swarm/setup.c
--- linux/arch/mips/sibyte/swarm/setup.c.orig	Tue Apr 30 17:38:09 2002
+++ linux/arch/mips/sibyte/swarm/setup.c	Fri May 10 14:54:36 2002
@@ -237,12 +237,15 @@
          * interrupts through low-level (direct) meachanism.
          */
 
+	/* Use our own gettimeoffset() routine */
+	do_gettimeoffset = sb1250_gettimeoffset;
+
         /* We only need to setup the generic timer */
         sb1250_time_init();
 }
 
 extern int xicor_set_time(unsigned long);
-extern unsigned int xicor_get_time(void);
+extern unsigned long xicor_get_time(void);
 
 void __init swarm_setup(void)
 {
diff -Nru linux/include/asm-mips/sibyte/sb1250.h.orig linux/include/asm-mips/sibyte/sb1250.h
--- linux/include/asm-mips/sibyte/sb1250.h.orig	Tue Jan 15 13:25:45 2002
+++ linux/include/asm-mips/sibyte/sb1250.h	Fri May 10 14:51:50 2002
@@ -20,6 +20,7 @@
 #define _ASM_SIBYTE_SB1250_H
 
 extern void sb1250_time_init(void);
+extern unsigned long sb1250_gettimeoffset(void);
 extern void sb1250_mask_irq(int cpu, int irq);
 extern void sb1250_unmask_irq(int cpu, int irq);
 extern void sb1250_smp_finish(void);
diff -Nru linux/include/asm-mips64/sibyte/sb1250.h.orig linux/include/asm-mips64/sibyte/sb1250.h
--- linux/include/asm-mips64/sibyte/sb1250.h.orig	Tue Apr 30 17:30:44 2002
+++ linux/include/asm-mips64/sibyte/sb1250.h	Fri May 10 14:52:16 2002
@@ -20,6 +20,7 @@
 #define _ASM_SIBYTE_SB1250_H
 
 extern void sb1250_time_init(void);
+extern unsigned long sb1250_gettimeoffset(void);
 extern void sb1250_mask_irq(int cpu, int irq);
 extern void sb1250_unmask_irq(int cpu, int irq);
 extern void sb1250_smp_finish(void);

Re: [PATCH] gettimeoffset for swarm

[Maciej W. Rozycki]

On Mon, 13 May 2002, Jun Sun wrote:

> By default, swarm will use calibrate_div64_gettimeoffset().  That does
> not work in SMP mode because the two cores have different counter register
> value.  This patch gives swarm its own and accurate gettimeoffset().

 You may consider synchronising the counters as it's done for the i386,
instead.  See synchronize_tsc_ap().  It may be useful for other purposes
as well. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +