powerpc: Merge smp-tbsync.c (the generic timebase sync routine)

Signed-off-by: Paul Mackerras <paulus@samba.org>

3 files changed, 179 insertions, 0 deletions

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index 3cf03ab46113..f4e25c648fbb 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -300,6 +300,7 @@ config PPC_PMAC64
 	bool
 	depends on PPC_PMAC && POWER4
 	select U3_DART
+	select GENERIC_TBSYNC
 	default y
 
 config PPC_PREP
@@ -314,6 +315,7 @@ config PPC_MAPLE
 	bool "  Maple 970FX Evaluation Board"
 	select U3_DART
 	select MPIC_BROKEN_U3
+	select GENERIC_TBSYNC
 	default n
 	help
           This option enables support for the Maple 970FX Evaluation Board.
@@ -386,6 +388,11 @@ config PPC_MPC106
 	bool
 	default n
 
+config GENERIC_TBSYNC
+	bool
+	default y if CONFIG_PPC32 && CONFIG_SMP
+	default n
+
 source "drivers/cpufreq/Kconfig"
 
 config CPU_FREQ_PMAC
diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile
index 7a3e1155ac9a..631149ea93db 100644
--- a/arch/powerpc/kernel/Makefile
+++ b/arch/powerpc/kernel/Makefile
@@ -21,6 +21,7 @@ obj-$(CONFIG_PPC_RTAS)		+= rtas.o
 obj-$(CONFIG_RTAS_FLASH)	+= rtas_flash.o
 obj-$(CONFIG_RTAS_PROC)		+= rtas-proc.o
 obj-$(CONFIG_IBMVIO)		+= vio.o
+obj-$(CONFIG_GENERIC_TBSYNC)	+= smp-tbsync.o
 
 ifeq ($(CONFIG_PPC_MERGE),y)
 
diff --git a/arch/powerpc/kernel/smp-tbsync.c b/arch/powerpc/kernel/smp-tbsync.c
new file mode 100644
index 000000000000..9adef3bddad4
--- /dev/null
+++ b/arch/powerpc/kernel/smp-tbsync.c
@@ -0,0 +1,171 @@
+/*
+ * Smp timebase synchronization for ppc.
+ *
+ * Copyright (C) 2003 Samuel Rydh (samuel@ibrium.se)
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/smp.h>
+#include <linux/unistd.h>
+#include <linux/init.h>
+#include <asm/atomic.h>
+#include <asm/smp.h>
+#include <asm/time.h>
+
+#define NUM_ITER		300
+
+enum {
+	kExit=0, kSetAndTest, kTest
+};
+
+static struct {
+	volatile u64		tb;
+	volatile u64		mark;
+	volatile int		cmd;
+	volatile int		handshake;
+	int			filler[2];
+
+	volatile int		ack;
+	int			filler2[7];
+
+	volatile int		race_result;
+} *tbsync;
+
+static volatile int		running;
+
+static void __devinit enter_contest(u64 mark, long add)
+{
+	while (get_tb() < mark)
+		tbsync->race_result = add;
+}
+
+void __devinit smp_generic_take_timebase(void)
+{
+	int cmd;
+	u64 tb;
+
+	local_irq_disable();
+	while (!running)
+		barrier();
+	rmb();
+
+	for (;;) {
+		tbsync->ack = 1;
+		while (!tbsync->handshake)
+			barrier();
+		rmb();
+
+		cmd = tbsync->cmd;
+		tb = tbsync->tb;
+		mb();
+		tbsync->ack = 0;
+		if (cmd == kExit)
+			break;
+
+		while (tbsync->handshake)
+			barrier();
+		if (cmd == kSetAndTest)
+			set_tb(tb >> 32, tb & 0xfffffffful);
+		enter_contest(tbsync->mark, -1);
+	}
+	local_irq_enable();
+}
+
+static int __devinit start_contest(int cmd, long offset, int num)
+{
+	int i, score=0;
+	u64 tb;
+	long mark;
+
+	tbsync->cmd = cmd;
+
+	local_irq_disable();
+	for (i = -3; i < num; ) {
+		tb = get_tb() + 400;
+		tbsync->tb = tb + offset;
+		tbsync->mark = mark = tb + 400;
+
+		wmb();
+
+		tbsync->handshake = 1;
+		while (tbsync->ack)
+			barrier();
+
+		while (get_tb() <= tb)
+			barrier();
+		tbsync->handshake = 0;
+		enter_contest(mark, 1);
+
+		while (!tbsync->ack)
+			barrier();
+
+		if (i++ > 0)
+			score += tbsync->race_result;
+	}
+	local_irq_enable();
+	return score;
+}
+
+void __devinit smp_generic_give_timebase(void)
+{
+	int i, score, score2, old, min=0, max=5000, offset=1000;
+
+	printk("Synchronizing timebase\n");
+
+	/* if this fails then this kernel won't work anyway... */
+	tbsync = kmalloc( sizeof(*tbsync), GFP_KERNEL );
+	memset( tbsync, 0, sizeof(*tbsync) );
+	mb();
+	running = 1;
+
+	while (!tbsync->ack)
+		barrier();
+
+	printk("Got ack\n");
+
+	/* binary search */
+	for (old = -1; old != offset ; offset = (min+max) / 2) {
+		score = start_contest(kSetAndTest, offset, NUM_ITER);
+
+		printk("score %d, offset %d\n", score, offset );
+
+		if( score > 0 )
+			max = offset;
+		else
+			min = offset;
+		old = offset;
+	}
+	score = start_contest(kSetAndTest, min, NUM_ITER);
+	score2 = start_contest(kSetAndTest, max, NUM_ITER);
+
+	printk("Min %d (score %d), Max %d (score %d)\n",
+	       min, score, max, score2);
+	score = abs(score);
+	score2 = abs(score2);
+	offset = (score < score2) ? min : max;
+
+	/* guard against inaccurate mttb */
+	for (i = 0; i < 10; i++) {
+		start_contest(kSetAndTest, offset, NUM_ITER/10);
+
+		if ((score2 = start_contest(kTest, offset, NUM_ITER)) < 0)
+			score2 = -score2;
+		if (score2 <= score || score2 < 20)
+			break;
+	}
+	printk("Final offset: %d (%d/%d)\n", offset, score2, NUM_ITER );
+
+	/* exiting */
+	tbsync->cmd = kExit;
+	wmb();
+	tbsync->handshake = 1;
+	while (tbsync->ack)
+		barrier();
+	tbsync->handshake = 0;
+	kfree(tbsync);
+	tbsync = NULL;
+	running = 0;
+}