From 7ccbe504b5ee766d33211a507189a06f3079b29b Mon Sep 17 00:00:00 2001 From: Benjamin Herrenschmidt Date: Thu, 18 Jun 2009 23:30:07 +0000 Subject: powerpc/pmac: Fix issues with PowerMac "PowerSurge" SMP The old PowerSurge SMP (ie, dual or quad 604 machines) code has numerous issues in modern world. One is cpu_possible_map is set too late (the device-tree is bogus) so we fail to allocate the interrupt stacks and crash. Another problem is the fact the timebase is frozen by the bringup of the second CPU so the delays in the generic code will hang, we need to move some of the calling procedure to inside the powermac code. This makes it boot again for me Signed-off-by: Benjamin Herrenschmidt --- arch/powerpc/kernel/smp.c | 3 +- arch/powerpc/platforms/powermac/setup.c | 41 ++------ arch/powerpc/platforms/powermac/smp.c | 166 ++++++++++++++++++++------------ 3 files changed, 112 insertions(+), 98 deletions(-) diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c index 65484b2200b..0b47de07302 100644 --- a/arch/powerpc/kernel/smp.c +++ b/arch/powerpc/kernel/smp.c @@ -68,7 +68,8 @@ EXPORT_PER_CPU_SYMBOL(cpu_core_map); /* SMP operations for this machine */ struct smp_ops_t *smp_ops; -static volatile unsigned int cpu_callin_map[NR_CPUS]; +/* Can't be static due to PowerMac hackery */ +volatile unsigned int cpu_callin_map[NR_CPUS]; int smt_enabled_at_boot = 1; diff --git a/arch/powerpc/platforms/powermac/setup.c b/arch/powerpc/platforms/powermac/setup.c index 86f69a4eb49..c2052265636 100644 --- a/arch/powerpc/platforms/powermac/setup.c +++ b/arch/powerpc/platforms/powermac/setup.c @@ -103,11 +103,6 @@ unsigned long smu_cmdbuf_abs; EXPORT_SYMBOL(smu_cmdbuf_abs); #endif -#ifdef CONFIG_SMP -extern struct smp_ops_t psurge_smp_ops; -extern struct smp_ops_t core99_smp_ops; -#endif /* CONFIG_SMP */ - static void pmac_show_cpuinfo(struct seq_file *m) { struct device_node *np; @@ -341,34 +336,6 @@ static void __init pmac_setup_arch(void) ROOT_DEV = DEFAULT_ROOT_DEVICE; #endif -#ifdef CONFIG_SMP - /* Check for Core99 */ - ic = of_find_node_by_name(NULL, "uni-n"); - if (!ic) - ic = of_find_node_by_name(NULL, "u3"); - if (!ic) - ic = of_find_node_by_name(NULL, "u4"); - if (ic) { - of_node_put(ic); - smp_ops = &core99_smp_ops; - } -#ifdef CONFIG_PPC32 - else { - /* - * We have to set bits in cpu_possible_map here since the - * secondary CPU(s) aren't in the device tree, and - * setup_per_cpu_areas only allocates per-cpu data for - * CPUs in the cpu_possible_map. - */ - int cpu; - - for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu) - cpu_set(cpu, cpu_possible_map); - smp_ops = &psurge_smp_ops; - } -#endif -#endif /* CONFIG_SMP */ - #ifdef CONFIG_ADB if (strstr(cmd_line, "adb_sync")) { extern int __adb_probe_sync; @@ -512,6 +479,14 @@ static void __init pmac_init_early(void) #ifdef CONFIG_PPC64 iommu_init_early_dart(); #endif + + /* SMP Init has to be done early as we need to patch up + * cpu_possible_map before interrupt stacks are allocated + * or kaboom... + */ +#ifdef CONFIG_SMP + pmac_setup_smp(); +#endif } static int __init pmac_declare_of_platform_devices(void) diff --git a/arch/powerpc/platforms/powermac/smp.c b/arch/powerpc/platforms/powermac/smp.c index cf1dbe75889..6d4da7b46b4 100644 --- a/arch/powerpc/platforms/powermac/smp.c +++ b/arch/powerpc/platforms/powermac/smp.c @@ -64,10 +64,11 @@ 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_PPC32 /* * Powersurge (old powermac SMP) support. @@ -294,6 +295,9 @@ 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) { @@ -308,18 +312,15 @@ static int __init smp_psurge_probe(void) psurge_start = ioremap(PSURGE_START, 4); psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4); - /* - * This is necessary because OF doesn't know about the + /* 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_possible_map and cpu_present_map. + * set their bits in cpu_present_map. */ if (ncpus > NR_CPUS) ncpus = NR_CPUS; - for (i = 1; i < ncpus ; ++i) { + for (i = 1; i < ncpus ; ++i) cpu_set(i, cpu_present_map); - set_hard_smp_processor_id(i, i); - } if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352); @@ -329,8 +330,14 @@ static int __init smp_psurge_probe(void) static void __init smp_psurge_kick_cpu(int nr) { unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8; - unsigned long a; - int i; + 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) @@ -339,47 +346,52 @@ 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); + /* * We can't use udelay here because the timebase is now frozen. */ for (i = 0; i < 2000; ++i) - barrier(); + asm volatile("nop" : : : "memory"); 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); - /* XXX fixme */ - set_tb(0, 0); - - 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); } static struct irqaction psurge_irqaction = { @@ -390,36 +402,35 @@ static struct irqaction psurge_irqaction = { static void __init smp_psurge_setup_cpu(int cpu_nr) { + if (cpu_nr != 0) + 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(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 */ @@ -437,9 +448,6 @@ struct smp_ops_t psurge_smp_ops = { * Core 99 and later support */ -static void (*pmac_tb_freeze)(int freeze); -static u64 timebase; -static int tb_req; static void smp_core99_give_timebase(void) { @@ -478,7 +486,6 @@ static void __devinit smp_core99_take_timebase(void) set_tb(timebase >> 32, timebase & 0xffffffff); timebase = 0; mb(); - set_dec(tb_ticks_per_jiffy/2); local_irq_restore(flags); } @@ -920,3 +927,34 @@ struct smp_ops_t core99_smp_ops = { # endif #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_PPC32 + else { + /* We have to set bits in cpu_possible_map 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) + cpu_set(cpu, cpu_possible_map); + smp_ops = &psurge_smp_ops; + } +#endif /* CONFIG_PPC32 */ +} + -- cgit v1.2.3-18-g5258