diff options
Diffstat (limited to 'arch/powerpc/platforms/powermac/smp.c')
| -rw-r--r-- | arch/powerpc/platforms/powermac/smp.c | 822 |
1 files changed, 497 insertions, 325 deletions
diff --git a/arch/powerpc/platforms/powermac/smp.c b/arch/powerpc/platforms/powermac/smp.c index e1f9443cc87..5cbd4d67d5c 100644 --- a/arch/powerpc/platforms/powermac/smp.c +++ b/arch/powerpc/platforms/powermac/smp.c @@ -21,11 +21,9 @@ * as published by the Free Software Foundation; either version * 2 of the License, or (at your option) any later version. */ -#include <linux/config.h> #include <linux/kernel.h> #include <linux/sched.h> #include <linux/smp.h> -#include <linux/smp_lock.h> #include <linux/interrupt.h> #include <linux/kernel_stat.h> #include <linux/delay.h> @@ -34,9 +32,11 @@ #include <linux/errno.h> #include <linux/hardirq.h> #include <linux/cpu.h> +#include <linux/compiler.h> #include <asm/ptrace.h> -#include <asm/atomic.h> +#include <linux/atomic.h> +#include <asm/code-patching.h> #include <asm/irq.h> #include <asm/page.h> #include <asm/pgtable.h> @@ -51,6 +51,9 @@ #include <asm/cacheflush.h> #include <asm/keylargo.h> #include <asm/pmac_low_i2c.h> +#include <asm/pmac_pfunc.h> + +#include "pmac.h" #undef DEBUG @@ -61,11 +64,13 @@ #endif extern void __secondary_start_pmac_0(void); +extern int pmac_pfunc_base_install(void); -#ifdef CONFIG_PPC32 +static void (*pmac_tb_freeze)(int freeze); +static u64 timebase; +static int tb_req; -/* Sync flag for HW tb sync */ -static volatile int sec_tb_reset = 0; +#ifdef CONFIG_PPC_PMAC32_PSURGE /* * Powersurge (old powermac SMP) support. @@ -119,6 +124,10 @@ static volatile u32 __iomem *psurge_start; /* what sort of powersurge board we have */ static int psurge_type = PSURGE_NONE; +/* irq for secondary cpus to report */ +static struct irq_domain *psurge_host; +int psurge_secondary_virq; + /* * Set and clear IPIs for powersurge. */ @@ -151,53 +160,51 @@ static inline void psurge_clr_ipi(int cpu) /* * On powersurge (old SMP powermac architecture) we don't have * separate IPIs for separate messages like openpic does. Instead - * we have a bitmap for each processor, where a 1 bit means that - * the corresponding message is pending for that processor. - * Ideally each cpu's entry would be in a different cache line. + * use the generic demux helpers * -- paulus. */ -static unsigned long psurge_smp_message[NR_CPUS]; - -void psurge_smp_message_recv(struct pt_regs *regs) +static irqreturn_t psurge_ipi_intr(int irq, void *d) { - int cpu = smp_processor_id(); - int msg; - - /* clear interrupt */ - psurge_clr_ipi(cpu); + psurge_clr_ipi(smp_processor_id()); + smp_ipi_demux(); - if (num_online_cpus() < 2) - return; + return IRQ_HANDLED; +} - /* make sure there is a message there */ - for (msg = 0; msg < 4; msg++) - if (test_and_clear_bit(msg, &psurge_smp_message[cpu])) - smp_message_recv(msg, regs); +static void smp_psurge_cause_ipi(int cpu, unsigned long data) +{ + psurge_set_ipi(cpu); } -irqreturn_t psurge_primary_intr(int irq, void *d, struct pt_regs *regs) +static int psurge_host_map(struct irq_domain *h, unsigned int virq, + irq_hw_number_t hw) { - psurge_smp_message_recv(regs); - return IRQ_HANDLED; + irq_set_chip_and_handler(virq, &dummy_irq_chip, handle_percpu_irq); + + return 0; } -static void smp_psurge_message_pass(int target, int msg) +static const struct irq_domain_ops psurge_host_ops = { + .map = psurge_host_map, +}; + +static int psurge_secondary_ipi_init(void) { - int i; + int rc = -ENOMEM; - if (num_online_cpus() < 2) - return; + psurge_host = irq_domain_add_nomap(NULL, ~0, &psurge_host_ops, NULL); - for (i = 0; i < NR_CPUS; i++) { - if (!cpu_online(i)) - continue; - if (target == MSG_ALL - || (target == MSG_ALL_BUT_SELF && i != smp_processor_id()) - || target == i) { - set_bit(msg, &psurge_smp_message[i]); - psurge_set_ipi(i); - } - } + if (psurge_host) + psurge_secondary_virq = irq_create_direct_mapping(psurge_host); + + if (psurge_secondary_virq) + rc = request_irq(psurge_secondary_virq, psurge_ipi_intr, + IRQF_PERCPU | IRQF_NO_THREAD, "IPI", NULL); + + if (rc) + pr_err("Failed to setup secondary cpu IPI\n"); + + return rc; } /* @@ -264,6 +271,7 @@ static void __init psurge_quad_init(void) static int __init smp_psurge_probe(void) { int i, ncpus; + struct device_node *dn; /* We don't do SMP on the PPC601 -- paulus */ if (PVR_VER(mfspr(SPRN_PVR)) == 1) @@ -279,8 +287,10 @@ static int __init smp_psurge_probe(void) * in the hammerhead memory controller in the case of the * dual-cpu powersurge board. -- paulus. */ - if (find_devices("hammerhead") == NULL) + dn = of_find_node_by_name(NULL, "hammerhead"); + if (dn == NULL) return 1; + of_node_put(dn); hhead_base = ioremap(HAMMERHEAD_BASE, 0x800); quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024); @@ -291,6 +301,9 @@ static int __init smp_psurge_probe(void) psurge_quad_init(); /* All released cards using this HW design have 4 CPUs */ ncpus = 4; + /* No sure how timebase sync works on those, let's use SW */ + smp_ops->give_timebase = smp_generic_give_timebase; + smp_ops->take_timebase = smp_generic_take_timebase; } else { iounmap(quad_base); if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) { @@ -302,22 +315,38 @@ static int __init smp_psurge_probe(void) ncpus = 2; } + if (psurge_secondary_ipi_init()) + return 1; + psurge_start = ioremap(PSURGE_START, 4); psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4); - /* this is not actually strictly necessary -- paulus. */ - for (i = 1; i < ncpus; ++i) - smp_hw_index[i] = i; + /* This is necessary because OF doesn't know about the + * secondary cpu(s), and thus there aren't nodes in the + * device tree for them, and smp_setup_cpu_maps hasn't + * set their bits in cpu_present_mask. + */ + if (ncpus > NR_CPUS) + ncpus = NR_CPUS; + for (i = 1; i < ncpus ; ++i) + set_cpu_present(i, true); if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352); return ncpus; } -static void __init smp_psurge_kick_cpu(int nr) +static int __init smp_psurge_kick_cpu(int nr) { unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8; - unsigned long a; + unsigned long a, flags; + int i, j; + + /* Defining this here is evil ... but I prefer hiding that + * crap to avoid giving people ideas that they can do the + * same. + */ + extern volatile unsigned int cpu_callin_map[NR_CPUS]; /* may need to flush here if secondary bats aren't setup */ for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32) @@ -326,108 +355,159 @@ static void __init smp_psurge_kick_cpu(int nr) if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353); + /* This is going to freeze the timeebase, we disable interrupts */ + local_irq_save(flags); + out_be32(psurge_start, start); mb(); psurge_set_ipi(nr); - udelay(10); - psurge_clr_ipi(nr); - - if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354); -} -/* - * With the dual-cpu powersurge board, the decrementers and timebases - * of both cpus are frozen after the secondary cpu is started up, - * until we give the secondary cpu another interrupt. This routine - * uses this to get the timebases synchronized. - * -- paulus. - */ -static void __init psurge_dual_sync_tb(int cpu_nr) -{ - int t; - - set_dec(tb_ticks_per_jiffy); - set_tb(0, 0); - last_jiffy_stamp(cpu_nr) = 0; + /* + * We can't use udelay here because the timebase is now frozen. + */ + for (i = 0; i < 2000; ++i) + asm volatile("nop" : : : "memory"); + psurge_clr_ipi(nr); - if (cpu_nr > 0) { + /* + * Also, because the timebase is frozen, we must not return to the + * caller which will try to do udelay's etc... Instead, we wait -here- + * for the CPU to callin. + */ + for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) { + for (j = 1; j < 10000; j++) + asm volatile("nop" : : : "memory"); + asm volatile("sync" : : : "memory"); + } + if (!cpu_callin_map[nr]) + goto stuck; + + /* And we do the TB sync here too for standard dual CPU cards */ + if (psurge_type == PSURGE_DUAL) { + while(!tb_req) + barrier(); + tb_req = 0; + mb(); + timebase = get_tb(); + mb(); + while (timebase) + barrier(); mb(); - sec_tb_reset = 1; - return; } + stuck: + /* now interrupt the secondary, restarting both TBs */ + if (psurge_type == PSURGE_DUAL) + psurge_set_ipi(1); - /* wait for the secondary to have reset its TB before proceeding */ - for (t = 10000000; t > 0 && !sec_tb_reset; --t) - ; - - /* now interrupt the secondary, starting both TBs */ - psurge_set_ipi(1); + if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354); - smp_tb_synchronized = 1; + return 0; } static struct irqaction psurge_irqaction = { - .handler = psurge_primary_intr, - .flags = SA_INTERRUPT, - .mask = CPU_MASK_NONE, + .handler = psurge_ipi_intr, + .flags = IRQF_PERCPU | IRQF_NO_THREAD, .name = "primary IPI", }; static void __init smp_psurge_setup_cpu(int cpu_nr) { + if (cpu_nr != 0 || !psurge_start) + return; - if (cpu_nr == 0) { - /* If we failed to start the second CPU, we should still - * send it an IPI to start the timebase & DEC or we might - * have them stuck. - */ - if (num_online_cpus() < 2) { - if (psurge_type == PSURGE_DUAL) - psurge_set_ipi(1); - return; - } - /* reset the entry point so if we get another intr we won't - * try to startup again */ - out_be32(psurge_start, 0x100); - if (setup_irq(30, &psurge_irqaction)) - printk(KERN_ERR "Couldn't get primary IPI interrupt"); - } - - if (psurge_type == PSURGE_DUAL) - psurge_dual_sync_tb(cpu_nr); + /* reset the entry point so if we get another intr we won't + * try to startup again */ + out_be32(psurge_start, 0x100); + if (setup_irq(irq_create_mapping(NULL, 30), &psurge_irqaction)) + printk(KERN_ERR "Couldn't get primary IPI interrupt"); } void __init smp_psurge_take_timebase(void) { - /* Dummy implementation */ + if (psurge_type != PSURGE_DUAL) + return; + + tb_req = 1; + mb(); + while (!timebase) + barrier(); + mb(); + set_tb(timebase >> 32, timebase & 0xffffffff); + timebase = 0; + mb(); + set_dec(tb_ticks_per_jiffy/2); } void __init smp_psurge_give_timebase(void) { - /* Dummy implementation */ + /* Nothing to do here */ } /* PowerSurge-style Macs */ struct smp_ops_t psurge_smp_ops = { - .message_pass = smp_psurge_message_pass, + .message_pass = NULL, /* Use smp_muxed_ipi_message_pass */ + .cause_ipi = smp_psurge_cause_ipi, .probe = smp_psurge_probe, .kick_cpu = smp_psurge_kick_cpu, .setup_cpu = smp_psurge_setup_cpu, .give_timebase = smp_psurge_give_timebase, .take_timebase = smp_psurge_take_timebase, }; -#endif /* CONFIG_PPC32 - actually powersurge support */ +#endif /* CONFIG_PPC_PMAC32_PSURGE */ + +/* + * Core 99 and later support + */ + + +static void smp_core99_give_timebase(void) +{ + unsigned long flags; + + local_irq_save(flags); + + while(!tb_req) + barrier(); + tb_req = 0; + (*pmac_tb_freeze)(1); + mb(); + timebase = get_tb(); + mb(); + while (timebase) + barrier(); + mb(); + (*pmac_tb_freeze)(0); + mb(); + + local_irq_restore(flags); +} + + +static void smp_core99_take_timebase(void) +{ + unsigned long flags; + + local_irq_save(flags); + + tb_req = 1; + mb(); + while (!timebase) + barrier(); + mb(); + set_tb(timebase >> 32, timebase & 0xffffffff); + timebase = 0; + mb(); + + local_irq_restore(flags); +} #ifdef CONFIG_PPC64 /* * G5s enable/disable the timebase via an i2c-connected clock chip. */ -static struct device_node *pmac_tb_clock_chip_host; +static struct pmac_i2c_bus *pmac_tb_clock_chip_host; static u8 pmac_tb_pulsar_addr; -static void (*pmac_tb_freeze)(int freeze); -static DEFINE_SPINLOCK(timebase_lock); -static unsigned long timebase; static void smp_core99_cypress_tb_freeze(int freeze) { @@ -437,19 +517,20 @@ static void smp_core99_cypress_tb_freeze(int freeze) /* Strangely, the device-tree says address is 0xd2, but darwin * accesses 0xd0 ... */ - pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined); - rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host, - 0xd0 | pmac_low_i2c_read, - 0x81, &data, 1); + pmac_i2c_setmode(pmac_tb_clock_chip_host, + pmac_i2c_mode_combined); + rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, + 0xd0 | pmac_i2c_read, + 1, 0x81, &data, 1); if (rc != 0) goto bail; data = (data & 0xf3) | (freeze ? 0x00 : 0x0c); - pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub); - rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host, - 0xd0 | pmac_low_i2c_write, - 0x81, &data, 1); + pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub); + rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, + 0xd0 | pmac_i2c_write, + 1, 0x81, &data, 1); bail: if (rc != 0) { @@ -465,19 +546,20 @@ static void smp_core99_pulsar_tb_freeze(int freeze) u8 data; int rc; - pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_combined); - rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host, - pmac_tb_pulsar_addr | pmac_low_i2c_read, - 0x2e, &data, 1); + pmac_i2c_setmode(pmac_tb_clock_chip_host, + pmac_i2c_mode_combined); + rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, + pmac_tb_pulsar_addr | pmac_i2c_read, + 1, 0x2e, &data, 1); if (rc != 0) goto bail; data = (data & 0x88) | (freeze ? 0x11 : 0x22); - pmac_low_i2c_setmode(pmac_tb_clock_chip_host, pmac_low_i2c_mode_stdsub); - rc = pmac_low_i2c_xfer(pmac_tb_clock_chip_host, - pmac_tb_pulsar_addr | pmac_low_i2c_write, - 0x2e, &data, 1); + pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub); + rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, + pmac_tb_pulsar_addr | pmac_i2c_write, + 1, 0x2e, &data, 1); bail: if (rc != 0) { printk(KERN_ERR "Pulsar Timebase %s rc: %d\n", @@ -486,187 +568,110 @@ static void smp_core99_pulsar_tb_freeze(int freeze) } } - -static void smp_core99_give_timebase(void) -{ - /* Open i2c bus for synchronous access */ - if (pmac_low_i2c_open(pmac_tb_clock_chip_host, 0)) - panic("Can't open i2c for TB sync !\n"); - - spin_lock(&timebase_lock); - (*pmac_tb_freeze)(1); - mb(); - timebase = get_tb(); - spin_unlock(&timebase_lock); - - while (timebase) - barrier(); - - spin_lock(&timebase_lock); - (*pmac_tb_freeze)(0); - spin_unlock(&timebase_lock); - - /* Close i2c bus */ - pmac_low_i2c_close(pmac_tb_clock_chip_host); -} - - -static void __devinit smp_core99_take_timebase(void) -{ - while (!timebase) - barrier(); - spin_lock(&timebase_lock); - set_tb(timebase >> 32, timebase & 0xffffffff); - timebase = 0; - spin_unlock(&timebase_lock); -} - -static void __init smp_core99_setup(int ncpus) +static void __init smp_core99_setup_i2c_hwsync(int ncpus) { struct device_node *cc = NULL; struct device_node *p; - u32 *reg; + const char *name = NULL; + const u32 *reg; int ok; - /* HW sync only on these platforms */ - if (!machine_is_compatible("PowerMac7,2") && - !machine_is_compatible("PowerMac7,3") && - !machine_is_compatible("RackMac3,1")) - return; - /* Look for the clock chip */ while ((cc = of_find_node_by_name(cc, "i2c-hwclock")) != NULL) { p = of_get_parent(cc); - ok = p && device_is_compatible(p, "uni-n-i2c"); + ok = p && of_device_is_compatible(p, "uni-n-i2c"); of_node_put(p); if (!ok) continue; - reg = (u32 *)get_property(cc, "reg", NULL); + pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc); + if (pmac_tb_clock_chip_host == NULL) + continue; + reg = of_get_property(cc, "reg", NULL); if (reg == NULL) continue; - switch (*reg) { case 0xd2: - if (device_is_compatible(cc, "pulsar-legacy-slewing")) { + if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) { pmac_tb_freeze = smp_core99_pulsar_tb_freeze; pmac_tb_pulsar_addr = 0xd2; - printk(KERN_INFO "Timebase clock is Pulsar chip\n"); - } else if (device_is_compatible(cc, "cy28508")) { + name = "Pulsar"; + } else if (of_device_is_compatible(cc, "cy28508")) { pmac_tb_freeze = smp_core99_cypress_tb_freeze; - printk(KERN_INFO "Timebase clock is Cypress chip\n"); + name = "Cypress"; } break; case 0xd4: pmac_tb_freeze = smp_core99_pulsar_tb_freeze; pmac_tb_pulsar_addr = 0xd4; - printk(KERN_INFO "Timebase clock is Pulsar chip\n"); + name = "Pulsar"; break; } - if (pmac_tb_freeze != NULL) { - pmac_tb_clock_chip_host = of_get_parent(cc); - of_node_put(cc); + if (pmac_tb_freeze != NULL) break; - } } - if (pmac_tb_freeze == NULL) { - smp_ops->give_timebase = smp_generic_give_timebase; - smp_ops->take_timebase = smp_generic_take_timebase; + if (pmac_tb_freeze != NULL) { + /* Open i2c bus for synchronous access */ + if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) { + printk(KERN_ERR "Failed top open i2c bus for clock" + " sync, fallback to software sync !\n"); + goto no_i2c_sync; + } + printk(KERN_INFO "Processor timebase sync using %s i2c clock\n", + name); + return; } + no_i2c_sync: + pmac_tb_freeze = NULL; + pmac_tb_clock_chip_host = NULL; } -/* nothing to do here, caches are already set up by service processor */ -static inline void __devinit core99_init_caches(int cpu) + + +/* + * Newer G5s uses a platform function + */ + +static void smp_core99_pfunc_tb_freeze(int freeze) { + struct device_node *cpus; + struct pmf_args args; + + cpus = of_find_node_by_path("/cpus"); + BUG_ON(cpus == NULL); + args.count = 1; + args.u[0].v = !freeze; + pmf_call_function(cpus, "cpu-timebase", &args); + of_node_put(cpus); } #else /* CONFIG_PPC64 */ /* - * SMP G4 powermacs use a GPIO to enable/disable the timebase. + * SMP G4 use a GPIO to enable/disable the timebase. */ static unsigned int core99_tb_gpio; /* Timebase freeze GPIO */ -static unsigned int pri_tb_hi, pri_tb_lo; -static unsigned int pri_tb_stamp; - -/* not __init, called in sleep/wakeup code */ -void smp_core99_give_timebase(void) +static void smp_core99_gpio_tb_freeze(int freeze) { - unsigned long flags; - unsigned int t; - - /* wait for the secondary to be in take_timebase */ - for (t = 100000; t > 0 && !sec_tb_reset; --t) - udelay(10); - if (!sec_tb_reset) { - printk(KERN_WARNING "Timeout waiting sync on second CPU\n"); - return; - } - - /* freeze the timebase and read it */ - /* disable interrupts so the timebase is disabled for the - shortest possible time */ - local_irq_save(flags); - pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4); - pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0); - mb(); - pri_tb_hi = get_tbu(); - pri_tb_lo = get_tbl(); - pri_tb_stamp = last_jiffy_stamp(smp_processor_id()); - mb(); - - /* tell the secondary we're ready */ - sec_tb_reset = 2; - mb(); - - /* wait for the secondary to have taken it */ - for (t = 100000; t > 0 && sec_tb_reset; --t) - udelay(10); - if (sec_tb_reset) - printk(KERN_WARNING "Timeout waiting sync(2) on second CPU\n"); + if (freeze) + pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4); else - smp_tb_synchronized = 1; - - /* Now, restart the timebase by leaving the GPIO to an open collector */ - pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0); - pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0); - local_irq_restore(flags); + pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0); + pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0); } -/* not __init, called in sleep/wakeup code */ -void smp_core99_take_timebase(void) -{ - unsigned long flags; - - /* tell the primary we're here */ - sec_tb_reset = 1; - mb(); - - /* wait for the primary to set pri_tb_hi/lo */ - while (sec_tb_reset < 2) - mb(); - - /* set our stuff the same as the primary */ - local_irq_save(flags); - set_dec(1); - set_tb(pri_tb_hi, pri_tb_lo); - last_jiffy_stamp(smp_processor_id()) = pri_tb_stamp; - mb(); - /* tell the primary we're done */ - sec_tb_reset = 0; - mb(); - local_irq_restore(flags); -} +#endif /* !CONFIG_PPC64 */ /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */ volatile static long int core99_l2_cache; volatile static long int core99_l3_cache; -static void __devinit core99_init_caches(int cpu) +static void core99_init_caches(int cpu) { +#ifndef CONFIG_PPC64 if (!cpu_has_feature(CPU_FTR_L2CR)) return; @@ -692,30 +697,75 @@ static void __devinit core99_init_caches(int cpu) _set_L3CR(core99_l3_cache); printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache); } +#endif /* !CONFIG_PPC64 */ } static void __init smp_core99_setup(int ncpus) { - struct device_node *cpu; - u32 *tbprop = NULL; - int i; - - core99_tb_gpio = KL_GPIO_TB_ENABLE; /* default value */ - cpu = of_find_node_by_type(NULL, "cpu"); - if (cpu != NULL) { - tbprop = (u32 *)get_property(cpu, "timebase-enable", NULL); - if (tbprop) - core99_tb_gpio = *tbprop; - of_node_put(cpu); +#ifdef CONFIG_PPC64 + + /* i2c based HW sync on some G5s */ + if (of_machine_is_compatible("PowerMac7,2") || + of_machine_is_compatible("PowerMac7,3") || + of_machine_is_compatible("RackMac3,1")) + smp_core99_setup_i2c_hwsync(ncpus); + + /* pfunc based HW sync on recent G5s */ + if (pmac_tb_freeze == NULL) { + struct device_node *cpus = + of_find_node_by_path("/cpus"); + if (cpus && + of_get_property(cpus, "platform-cpu-timebase", NULL)) { + pmac_tb_freeze = smp_core99_pfunc_tb_freeze; + printk(KERN_INFO "Processor timebase sync using" + " platform function\n"); + } } - /* XXX should get this from reg properties */ - for (i = 1; i < ncpus; ++i) - smp_hw_index[i] = i; - powersave_nap = 0; -} +#else /* CONFIG_PPC64 */ + + /* GPIO based HW sync on ppc32 Core99 */ + if (pmac_tb_freeze == NULL && !of_machine_is_compatible("MacRISC4")) { + struct device_node *cpu; + const u32 *tbprop = NULL; + + core99_tb_gpio = KL_GPIO_TB_ENABLE; /* default value */ + cpu = of_find_node_by_type(NULL, "cpu"); + if (cpu != NULL) { + tbprop = of_get_property(cpu, "timebase-enable", NULL); + if (tbprop) + core99_tb_gpio = *tbprop; + of_node_put(cpu); + } + pmac_tb_freeze = smp_core99_gpio_tb_freeze; + printk(KERN_INFO "Processor timebase sync using" + " GPIO 0x%02x\n", core99_tb_gpio); + } + +#endif /* CONFIG_PPC64 */ + + /* No timebase sync, fallback to software */ + if (pmac_tb_freeze == NULL) { + smp_ops->give_timebase = smp_generic_give_timebase; + smp_ops->take_timebase = smp_generic_take_timebase; + printk(KERN_INFO "Processor timebase sync using software\n"); + } + +#ifndef CONFIG_PPC64 + { + int i; + + /* XXX should get this from reg properties */ + for (i = 1; i < ncpus; ++i) + set_hard_smp_processor_id(i, i); + } #endif + /* 32 bits SMP can't NAP */ + if (!of_machine_is_compatible("MacRISC4")) + powersave_nap = 0; +} + static int __init smp_core99_probe(void) { struct device_node *cpus; @@ -733,39 +783,46 @@ static int __init smp_core99_probe(void) if (ncpus <= 1) return 1; + /* We need to perform some early initialisations before we can start + * setting up SMP as we are running before initcalls + */ + pmac_pfunc_base_install(); + pmac_i2c_init(); + + /* Setup various bits like timebase sync method, ability to nap, ... */ smp_core99_setup(ncpus); + + /* Install IPIs */ mpic_request_ipis(); + + /* Collect l2cr and l3cr values from CPU 0 */ core99_init_caches(0); return ncpus; } -static void __devinit smp_core99_kick_cpu(int nr) +static int smp_core99_kick_cpu(int nr) { unsigned int save_vector; - unsigned long new_vector; - unsigned long flags; - volatile unsigned int *vector - = ((volatile unsigned int *)(KERNELBASE+0x100)); + unsigned long target, flags; + unsigned int *vector = (unsigned int *)(PAGE_OFFSET+0x100); if (nr < 0 || nr > 3) - return; - if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu", 0x346); + return -ENOENT; + + if (ppc_md.progress) + ppc_md.progress("smp_core99_kick_cpu", 0x346); local_irq_save(flags); - local_irq_disable(); /* Save reset vector */ save_vector = *vector; - /* Setup fake reset vector that does - * b __secondary_start_pmac_0 + nr*8 - KERNELBASE + /* Setup fake reset vector that does + * b __secondary_start_pmac_0 + nr*8 */ - new_vector = (unsigned long) __secondary_start_pmac_0 + nr * 8; - *vector = 0x48000002 + new_vector - KERNELBASE; - - /* flush data cache and inval instruction cache */ - flush_icache_range((unsigned long) vector, (unsigned long) vector + 4); + target = (unsigned long) __secondary_start_pmac_0 + nr * 8; + patch_branch(vector, target, BRANCH_SET_LINK); /* Put some life in our friend */ pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0); @@ -783,9 +840,11 @@ static void __devinit smp_core99_kick_cpu(int nr) local_irq_restore(flags); if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347); + + return 0; } -static void __devinit smp_core99_setup_cpu(int cpu_nr) +static void smp_core99_setup_cpu(int cpu_nr) { /* Setup L2/L3 */ if (cpu_nr != 0) @@ -793,73 +852,186 @@ static void __devinit smp_core99_setup_cpu(int cpu_nr) /* Setup openpic */ mpic_setup_this_cpu(); +} - if (cpu_nr == 0) { -#ifdef CONFIG_POWER4 - extern void g5_phy_disable_cpu1(void); +#ifdef CONFIG_PPC64 +#ifdef CONFIG_HOTPLUG_CPU +static int smp_core99_cpu_notify(struct notifier_block *self, + unsigned long action, void *hcpu) +{ + int rc; - /* If we didn't start the second CPU, we must take - * it off the bus - */ - if (machine_is_compatible("MacRISC4") && - num_online_cpus() < 2) - g5_phy_disable_cpu1(); -#endif /* CONFIG_POWER4 */ - if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349); + switch(action) { + case CPU_UP_PREPARE: + case CPU_UP_PREPARE_FROZEN: + /* Open i2c bus if it was used for tb sync */ + if (pmac_tb_clock_chip_host) { + rc = pmac_i2c_open(pmac_tb_clock_chip_host, 1); + if (rc) { + pr_err("Failed to open i2c bus for time sync\n"); + return notifier_from_errno(rc); + } + } + break; + case CPU_ONLINE: + case CPU_UP_CANCELED: + /* Close i2c bus if it was used for tb sync */ + if (pmac_tb_clock_chip_host) + pmac_i2c_close(pmac_tb_clock_chip_host); + break; + default: + break; } + return NOTIFY_OK; } - -/* Core99 Macs (dual G4s and G5s) */ -struct smp_ops_t core99_smp_ops = { - .message_pass = smp_mpic_message_pass, - .probe = smp_core99_probe, - .kick_cpu = smp_core99_kick_cpu, - .setup_cpu = smp_core99_setup_cpu, - .give_timebase = smp_core99_give_timebase, - .take_timebase = smp_core99_take_timebase, +static struct notifier_block smp_core99_cpu_nb = { + .notifier_call = smp_core99_cpu_notify, }; +#endif /* CONFIG_HOTPLUG_CPU */ -#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PPC32) +static void __init smp_core99_bringup_done(void) +{ + extern void g5_phy_disable_cpu1(void); + + /* Close i2c bus if it was used for tb sync */ + if (pmac_tb_clock_chip_host) + pmac_i2c_close(pmac_tb_clock_chip_host); + + /* If we didn't start the second CPU, we must take + * it off the bus. + */ + if (of_machine_is_compatible("MacRISC4") && + num_online_cpus() < 2) { + set_cpu_present(1, false); + g5_phy_disable_cpu1(); + } +#ifdef CONFIG_HOTPLUG_CPU + register_cpu_notifier(&smp_core99_cpu_nb); +#endif + + if (ppc_md.progress) + ppc_md.progress("smp_core99_bringup_done", 0x349); +} +#endif /* CONFIG_PPC64 */ + +#ifdef CONFIG_HOTPLUG_CPU -int __cpu_disable(void) +static int smp_core99_cpu_disable(void) { - cpu_clear(smp_processor_id(), cpu_online_map); + int rc = generic_cpu_disable(); + if (rc) + return rc; - /* XXX reset cpu affinity here */ mpic_cpu_set_priority(0xf); - asm volatile("mtdec %0" : : "r" (0x7fffffff)); - mb(); - udelay(20); - asm volatile("mtdec %0" : : "r" (0x7fffffff)); + return 0; } -extern void low_cpu_die(void) __attribute__((noreturn)); /* in sleep.S */ -static int cpu_dead[NR_CPUS]; +#ifdef CONFIG_PPC32 -void cpu_die(void) +static void pmac_cpu_die(void) { + int cpu = smp_processor_id(); + local_irq_disable(); - cpu_dead[smp_processor_id()] = 1; + idle_task_exit(); + pr_debug("CPU%d offline\n", cpu); + generic_set_cpu_dead(cpu); + smp_wmb(); mb(); low_cpu_die(); } -void __cpu_die(unsigned int cpu) +#else /* CONFIG_PPC32 */ + +static void pmac_cpu_die(void) { - int timeout; + int cpu = smp_processor_id(); - timeout = 1000; - while (!cpu_dead[cpu]) { - if (--timeout == 0) { - printk("CPU %u refused to die!\n", cpu); - break; - } - msleep(1); + local_irq_disable(); + idle_task_exit(); + + /* + * turn off as much as possible, we'll be + * kicked out as this will only be invoked + * on core99 platforms for now ... + */ + + printk(KERN_INFO "CPU#%d offline\n", cpu); + generic_set_cpu_dead(cpu); + smp_wmb(); + + /* + * Re-enable interrupts. The NAP code needs to enable them + * anyways, do it now so we deal with the case where one already + * happened while soft-disabled. + * We shouldn't get any external interrupts, only decrementer, and the + * decrementer handler is safe for use on offline CPUs + */ + local_irq_enable(); + + while (1) { + /* let's not take timer interrupts too often ... */ + set_dec(0x7fffffff); + + /* Enter NAP mode */ + power4_idle(); } - cpu_callin_map[cpu] = 0; - cpu_dead[cpu] = 0; } +#endif /* else CONFIG_PPC32 */ +#endif /* CONFIG_HOTPLUG_CPU */ + +/* Core99 Macs (dual G4s and G5s) */ +struct smp_ops_t core99_smp_ops = { + .message_pass = smp_mpic_message_pass, + .probe = smp_core99_probe, +#ifdef CONFIG_PPC64 + .bringup_done = smp_core99_bringup_done, +#endif + .kick_cpu = smp_core99_kick_cpu, + .setup_cpu = smp_core99_setup_cpu, + .give_timebase = smp_core99_give_timebase, + .take_timebase = smp_core99_take_timebase, +#if defined(CONFIG_HOTPLUG_CPU) + .cpu_disable = smp_core99_cpu_disable, + .cpu_die = generic_cpu_die, +#endif +}; + +void __init pmac_setup_smp(void) +{ + struct device_node *np; + + /* Check for Core99 */ + np = of_find_node_by_name(NULL, "uni-n"); + if (!np) + np = of_find_node_by_name(NULL, "u3"); + if (!np) + np = of_find_node_by_name(NULL, "u4"); + if (np) { + of_node_put(np); + smp_ops = &core99_smp_ops; + } +#ifdef CONFIG_PPC_PMAC32_PSURGE + else { + /* We have to set bits in cpu_possible_mask here since the + * secondary CPU(s) aren't in the device tree. Various + * things won't be initialized for CPUs not in the possible + * map, so we really need to fix it up here. + */ + int cpu; + + for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu) + set_cpu_possible(cpu, true); + smp_ops = &psurge_smp_ops; + } +#endif /* CONFIG_PPC_PMAC32_PSURGE */ + +#ifdef CONFIG_HOTPLUG_CPU + ppc_md.cpu_die = pmac_cpu_die; #endif +} + + |