From 43a0d76a65e6c9f278774c6164e6ec2295f7248e Mon Sep 17 00:00:00 2001 From: Benjamin Herrenschmidt Date: Fri, 6 Feb 2004 13:51:37 +1100 Subject: [PATCH] Add Samuel Rydth improved software CPU timebase synchronisation used on machines that don't have a HW facility (or we don't drive it yet like G5s) --- arch/ppc/kernel/Makefile | 2 +- arch/ppc/kernel/smp-tbsync.c | 181 +++++++++++++++++++++++++++++++++++ arch/ppc/kernel/smp.c | 39 -------- 3 files changed, 182 insertions(+), 40 deletions(-) create mode 100644 arch/ppc/kernel/smp-tbsync.c diff --git a/arch/ppc/kernel/Makefile b/arch/ppc/kernel/Makefile index f343fd3f21a4f..33b6ca87127d0 100644 --- a/arch/ppc/kernel/Makefile +++ b/arch/ppc/kernel/Makefile @@ -27,7 +27,7 @@ obj-$(CONFIG_MODULES) += module.o ppc_ksyms.o obj-$(CONFIG_PCI) += pci.o obj-$(CONFIG_PCI) += pci-dma.o obj-$(CONFIG_KGDB) += ppc-stub.o -obj-$(CONFIG_SMP) += smp.o +obj-$(CONFIG_SMP) += smp.o smp-tbsync.o obj-$(CONFIG_TAU) += temp.o ifdef CONFIG_MATH_EMULATION diff --git a/arch/ppc/kernel/smp-tbsync.c b/arch/ppc/kernel/smp-tbsync.c new file mode 100644 index 0000000000000..2c9cd95bcea6d --- /dev/null +++ b/arch/ppc/kernel/smp-tbsync.c @@ -0,0 +1,181 @@ +/* + * Smp timebase synchronization for ppc. + * + * Copyright (C) 2003 Samuel Rydh (samuel@ibrium.se) + * + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define NUM_ITER 300 + +enum { + kExit=0, kSetAndTest, kTest +}; + +static struct { + volatile int tbu; + volatile int tbl; + volatile int mark; + volatile int cmd; + volatile int handshake; + int filler[3]; + + volatile int ack; + int filler2[7]; + + volatile int race_result; +} *tbsync; + +static volatile int running; + +static void __devinit +enter_contest( int mark, int add ) +{ + while( (int)(get_tbl() - mark) < 0 ) + tbsync->race_result = add; +} + +void __devinit +smp_generic_take_timebase( void ) +{ + int cmd, tbl, tbu; + + local_irq_disable(); + while( !running ) + ; + rmb(); + + for( ;; ) { + tbsync->ack = 1; + while( !tbsync->handshake ) + ; + rmb(); + + cmd = tbsync->cmd; + tbl = tbsync->tbl; + tbu = tbsync->tbu; + tbsync->ack = 0; + if( cmd == kExit ) + return; + + if( cmd == kSetAndTest ) { + while( tbsync->handshake ) + ; + asm volatile ("mttbl %0" :: "r" (tbl) ); + asm volatile ("mttbu %0" :: "r" (tbu) ); + } else { + while( tbsync->handshake ) + ; + } + enter_contest( tbsync->mark, -1 ); + } + local_irq_enable(); +} + +static int __devinit +start_contest( int cmd, int offset, int num ) +{ + int i, tbu, tbl, mark, score=0; + + tbsync->cmd = cmd; + + local_irq_disable(); + for( i=-3; itbu = tbu = get_tbu(); + tbsync->tbl = tbl + offset; + tbsync->mark = mark = tbl + 400; + + wmb(); + + tbsync->handshake = 1; + while( tbsync->ack ) + ; + + while( (int)(get_tbl() - tbl) <= 0 ) + ; + tbsync->handshake = 0; + enter_contest( mark, 1 ); + + while( !tbsync->ack ) + ; + + if( tbsync->tbu != get_tbu() || ((tbsync->tbl ^ get_tbl()) & 0x80000000) ) + continue; + 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 ) + ; + + /* 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 ) + ; + tbsync->handshake = 0; + kfree( tbsync ); + tbsync = NULL; + running = 0; + + /* all done */ + smp_tb_synchronized = 1; +} diff --git a/arch/ppc/kernel/smp.c b/arch/ppc/kernel/smp.c index 9d8e34c4702b1..2574f567148ab 100644 --- a/arch/ppc/kernel/smp.c +++ b/arch/ppc/kernel/smp.c @@ -61,10 +61,6 @@ static struct smp_ops_t *smp_ops; /* all cpu mappings are 1-1 -- Cort */ volatile unsigned long cpu_callin_map[NR_CPUS]; -#define TB_SYNC_PASSES 4 -volatile unsigned long __initdata tb_sync_flag = 0; -volatile unsigned long __initdata tb_offset = 0; - int start_secondary(void *); extern int cpu_idle(void *unused); void smp_call_function_interrupt(void); @@ -291,41 +287,6 @@ void smp_call_function_interrupt(void) atomic_inc(&call_data->finished); } -/* FIXME: Do this properly for all archs --RR */ -static spinlock_t timebase_lock = SPIN_LOCK_UNLOCKED; -static unsigned int timebase_upper = 0, timebase_lower = 0; - -void __devinit -smp_generic_give_timebase(void) -{ - spin_lock(&timebase_lock); - do { - timebase_upper = get_tbu(); - timebase_lower = get_tbl(); - } while (timebase_upper != get_tbu()); - spin_unlock(&timebase_lock); - - while (timebase_upper || timebase_lower) - rmb(); -} - -void __devinit -smp_generic_take_timebase(void) -{ - int done = 0; - - while (!done) { - spin_lock(&timebase_lock); - if (timebase_upper || timebase_lower) { - set_tb(timebase_upper, timebase_lower); - timebase_upper = 0; - timebase_lower = 0; - done = 1; - } - spin_unlock(&timebase_lock); - } -} - static void __devinit smp_store_cpu_info(int id) { struct cpuinfo_PPC *c = &cpu_data[id]; -- 2.53.0