X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=arch%2Fia64%2Fkernel%2Fsmpboot.c;h=b21ddecea943a63f51130b20fd8372322cec29e2;hb=refs%2Fheads%2Fvserver;hp=02dc1fc08a974cf311541ddce75b7b1e7927b359;hpb=6a77f38946aaee1cd85eeec6cf4229b204c15071;p=linux-2.6.git diff --git a/arch/ia64/kernel/smpboot.c b/arch/ia64/kernel/smpboot.c index 02dc1fc08..b21ddecea 100644 --- a/arch/ia64/kernel/smpboot.c +++ b/arch/ia64/kernel/smpboot.c @@ -1,16 +1,26 @@ /* * SMP boot-related support * - * Copyright (C) 1998-2003 Hewlett-Packard Co + * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co * David Mosberger-Tang + * Copyright (C) 2001, 2004-2005 Intel Corp + * Rohit Seth + * Suresh Siddha + * Gordon Jin + * Ashok Raj * * 01/05/16 Rohit Seth Moved SMP booting functions from smp.c to here. * 01/04/27 David Mosberger Added ITC synching code. * 02/07/31 David Mosberger Switch over to hotplug-CPU boot-sequence. * smp_boot_cpus()/smp_commence() is replaced by * smp_prepare_cpus()/__cpu_up()/smp_cpus_done(). + * 04/06/21 Ashok Raj Added CPU Hotplug Support + * 04/12/26 Jin Gordon + * 04/12/26 Rohit Seth + * Add multi-threading and multi-core detection + * 05/01/30 Suresh Siddha + * Setup cpu_sibling_map and cpu_core_map */ -#include #include #include @@ -58,11 +68,48 @@ #define Dprintk(x...) #endif +#ifdef CONFIG_HOTPLUG_CPU +#ifdef CONFIG_PERMIT_BSP_REMOVE +#define bsp_remove_ok 1 +#else +#define bsp_remove_ok 0 +#endif + +/* + * Store all idle threads, this can be reused instead of creating + * a new thread. Also avoids complicated thread destroy functionality + * for idle threads. + */ +struct task_struct *idle_thread_array[NR_CPUS]; + +/* + * Global array allocated for NR_CPUS at boot time + */ +struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS]; + +/* + * start_ap in head.S uses this to store current booting cpu + * info. + */ +struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0]; + +#define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]); + +#define get_idle_for_cpu(x) (idle_thread_array[(x)]) +#define set_idle_for_cpu(x,p) (idle_thread_array[(x)] = (p)) + +#else + +#define get_idle_for_cpu(x) (NULL) +#define set_idle_for_cpu(x,p) +#define set_brendez_area(x) +#endif + /* * ITC synchronization related stuff: */ -#define MASTER 0 +#define MASTER (0) #define SLAVE (SMP_CACHE_BYTES/8) #define NUM_ROUNDS 64 /* magic value */ @@ -77,7 +124,7 @@ extern void __devinit calibrate_delay (void); extern void start_ap (void); extern unsigned long ia64_iobase; -task_t *task_for_booting_cpu; +struct task_struct *task_for_booting_cpu; /* * State for each CPU @@ -87,9 +134,14 @@ DEFINE_PER_CPU(int, cpu_state); /* Bitmasks of currently online, and possible CPUs */ cpumask_t cpu_online_map; EXPORT_SYMBOL(cpu_online_map); -cpumask_t cpu_possible_map; +cpumask_t cpu_possible_map = CPU_MASK_NONE; EXPORT_SYMBOL(cpu_possible_map); +cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned; +cpumask_t cpu_sibling_map[NR_CPUS] __cacheline_aligned; +int smp_num_siblings = 1; +int smp_num_cpucores = 1; + /* which logical CPU number maps to which CPU (physical APIC ID) */ volatile int ia64_cpu_to_sapicid[NR_CPUS]; EXPORT_SYMBOL(ia64_cpu_to_sapicid); @@ -104,6 +156,27 @@ char __initdata no_int_routing; unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */ +#ifdef CONFIG_FORCE_CPEI_RETARGET +#define CPEI_OVERRIDE_DEFAULT (1) +#else +#define CPEI_OVERRIDE_DEFAULT (0) +#endif + +unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT; + +static int __init +cmdl_force_cpei(char *str) +{ + int value=0; + + get_option (&str, &value); + force_cpei_retarget = value; + + return 1; +} + +__setup("force_cpei=", cmdl_force_cpei); + static int __init nointroute (char *str) { @@ -114,6 +187,27 @@ nointroute (char *str) __setup("nointroute", nointroute); +static void fix_b0_for_bsp(void) +{ +#ifdef CONFIG_HOTPLUG_CPU + int cpuid; + static int fix_bsp_b0 = 1; + + cpuid = smp_processor_id(); + + /* + * Cache the b0 value on the first AP that comes up + */ + if (!(fix_bsp_b0 && cpuid)) + return; + + sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0]; + printk ("Fixed BSP b0 value from CPU %d\n", cpuid); + + fix_bsp_b0 = 0; +#endif +} + void sync_master (void *arg) { @@ -124,7 +218,8 @@ sync_master (void *arg) local_irq_save(flags); { for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) { - while (!go[MASTER]); + while (!go[MASTER]) + cpu_relax(); go[MASTER] = 0; go[SLAVE] = ia64_get_itc(); } @@ -147,7 +242,8 @@ get_delta (long *rt, long *master) for (i = 0; i < NUM_ITERS; ++i) { t0 = ia64_get_itc(); go[MASTER] = 1; - while (!(tm = go[SLAVE])); + while (!(tm = go[SLAVE])) + cpu_relax(); go[SLAVE] = 0; t1 = ia64_get_itc(); @@ -226,7 +322,8 @@ ia64_sync_itc (unsigned int master) return; } - while (go[MASTER]); /* wait for master to be ready */ + while (go[MASTER]) + cpu_relax(); /* wait for master to be ready */ spin_lock_irqsave(&itc_sync_lock, flags); { @@ -277,8 +374,9 @@ smp_setup_percpu_timer (void) static void __devinit smp_callin (void) { - int cpuid, phys_id; + int cpuid, phys_id, itc_master; extern void ia64_init_itm(void); + extern volatile int time_keeper_id; #ifdef CONFIG_PERFMON extern void pfm_init_percpu(void); @@ -286,6 +384,7 @@ smp_callin (void) cpuid = smp_processor_id(); phys_id = hard_smp_processor_id(); + itc_master = time_keeper_id; if (cpu_online(cpuid)) { printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n", @@ -293,9 +392,12 @@ smp_callin (void) BUG(); } + fix_b0_for_bsp(); + lock_ipi_calllock(); cpu_set(cpuid, cpu_online_map); unlock_ipi_calllock(); + per_cpu(cpu_state, cpuid) = CPU_ONLINE; smp_setup_percpu_timer(); @@ -314,8 +416,8 @@ smp_callin (void) * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls * local_bh_enable(), which bugs out if irqs are not enabled... */ - Dprintk("Going to syncup ITC with BP.\n"); - ia64_sync_itc(0); + Dprintk("Going to syncup ITC with ITC Master.\n"); + ia64_sync_itc(itc_master); } /* @@ -345,10 +447,10 @@ start_secondary (void *unused) { /* Early console may use I/O ports */ ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase)); - Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id()); efi_map_pal_code(); cpu_init(); + preempt_disable(); smp_callin(); cpu_idle(); @@ -361,15 +463,17 @@ struct pt_regs * __devinit idle_regs(struct pt_regs *regs) } struct create_idle { + struct work_struct work; struct task_struct *idle; struct completion done; int cpu; }; void -do_fork_idle(void *_c_idle) +do_fork_idle(struct work_struct *work) { - struct create_idle *c_idle = _c_idle; + struct create_idle *c_idle = + container_of(work, struct create_idle, work); c_idle->idle = fork_idle(c_idle->cpu); complete(&c_idle->done); @@ -380,26 +484,38 @@ do_boot_cpu (int sapicid, int cpu) { int timeout; struct create_idle c_idle = { + .work = __WORK_INITIALIZER(c_idle.work, do_fork_idle), .cpu = cpu, .done = COMPLETION_INITIALIZER(c_idle.done), }; - DECLARE_WORK(work, do_fork_idle, &c_idle); + + c_idle.idle = get_idle_for_cpu(cpu); + if (c_idle.idle) { + init_idle(c_idle.idle, cpu); + goto do_rest; + } + /* * We can't use kernel_thread since we must avoid to reschedule the child. */ if (!keventd_up() || current_is_keventd()) - work.func(work.data); + c_idle.work.func(&c_idle.work); else { - schedule_work(&work); + schedule_work(&c_idle.work); wait_for_completion(&c_idle.done); } if (IS_ERR(c_idle.idle)) panic("failed fork for CPU %d", cpu); + + set_idle_for_cpu(cpu, c_idle.idle); + +do_rest: task_for_booting_cpu = c_idle.idle; Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid); + set_brendez_area(cpu); platform_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0); /* @@ -427,25 +543,11 @@ decay (char *str) { int ticks; get_option (&str, &ticks); - cache_decay_ticks = ticks; return 1; } __setup("decay=", decay); -/* - * # of ticks an idle task is considered cache-hot. Highly application-dependent. There - * are apps out there which are known to suffer significantly with values >= 4. - */ -unsigned long cache_decay_ticks = 10; /* equal to MIN_TIMESLICE */ - -static void -smp_tune_scheduling (void) -{ - printk(KERN_INFO "task migration cache decay timeout: %ld msecs.\n", - (cache_decay_ticks + 1) * 1000 / HZ); -} - /* * Initialize the logical CPU number to SAPICID mapping */ @@ -457,9 +559,6 @@ smp_build_cpu_map (void) for (cpu = 0; cpu < NR_CPUS; cpu++) { ia64_cpu_to_sapicid[cpu] = -1; -#ifdef CONFIG_HOTPLUG_CPU - cpu_set(cpu, cpu_possible_map); -#endif } ia64_cpu_to_sapicid[0] = boot_cpu_id; @@ -477,47 +576,6 @@ smp_build_cpu_map (void) } } -#ifdef CONFIG_NUMA - -/* on which node is each logical CPU (one cacheline even for 64 CPUs) */ -u8 cpu_to_node_map[NR_CPUS] __cacheline_aligned; -EXPORT_SYMBOL(cpu_to_node_map); -/* which logical CPUs are on which nodes */ -cpumask_t node_to_cpu_mask[MAX_NUMNODES] __cacheline_aligned; - -/* - * Build cpu to node mapping and initialize the per node cpu masks. - */ -void __init -build_cpu_to_node_map (void) -{ - int cpu, i, node; - - for(node=0; node= 0) ? node : 0; - if (node >= 0) - cpu_set(cpu, node_to_cpu_mask[node]); - } -} - -#endif /* CONFIG_NUMA */ - /* * Cycle through the APs sending Wakeup IPIs to boot each. */ @@ -544,7 +602,6 @@ smp_prepare_cpus (unsigned int max_cpus) printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id); current_thread_info()->cpu = 0; - smp_tune_scheduling(); /* * If SMP should be disabled, then really disable it! @@ -565,21 +622,84 @@ void __devinit smp_prepare_boot_cpu(void) { cpu_set(smp_processor_id(), cpu_online_map); cpu_set(smp_processor_id(), cpu_callin_map); + per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE; } #ifdef CONFIG_HOTPLUG_CPU +static inline void +clear_cpu_sibling_map(int cpu) +{ + int i; + + for_each_cpu_mask(i, cpu_sibling_map[cpu]) + cpu_clear(cpu, cpu_sibling_map[i]); + for_each_cpu_mask(i, cpu_core_map[cpu]) + cpu_clear(cpu, cpu_core_map[i]); + + cpu_sibling_map[cpu] = cpu_core_map[cpu] = CPU_MASK_NONE; +} + +static void +remove_siblinginfo(int cpu) +{ + int last = 0; + + if (cpu_data(cpu)->threads_per_core == 1 && + cpu_data(cpu)->cores_per_socket == 1) { + cpu_clear(cpu, cpu_core_map[cpu]); + cpu_clear(cpu, cpu_sibling_map[cpu]); + return; + } + + last = (cpus_weight(cpu_core_map[cpu]) == 1 ? 1 : 0); + + /* remove it from all sibling map's */ + clear_cpu_sibling_map(cpu); +} + extern void fixup_irqs(void); -/* must be called with cpucontrol mutex held */ -static int __devinit cpu_enable(unsigned int cpu) + +int migrate_platform_irqs(unsigned int cpu) { - per_cpu(cpu_state,cpu) = CPU_UP_PREPARE; - wmb(); + int new_cpei_cpu; + irq_desc_t *desc = NULL; + cpumask_t mask; + int retval = 0; - while (!cpu_online(cpu)) - cpu_relax(); - return 0; + /* + * dont permit CPEI target to removed. + */ + if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) { + printk ("CPU (%d) is CPEI Target\n", cpu); + if (can_cpei_retarget()) { + /* + * Now re-target the CPEI to a different processor + */ + new_cpei_cpu = any_online_cpu(cpu_online_map); + mask = cpumask_of_cpu(new_cpei_cpu); + set_cpei_target_cpu(new_cpei_cpu); + desc = irq_desc + ia64_cpe_irq; + /* + * Switch for now, immediatly, we need to do fake intr + * as other interrupts, but need to study CPEI behaviour with + * polling before making changes. + */ + if (desc) { + desc->chip->disable(ia64_cpe_irq); + desc->chip->set_affinity(ia64_cpe_irq, mask); + desc->chip->enable(ia64_cpe_irq); + printk ("Re-targetting CPEI to cpu %d\n", new_cpei_cpu); + } + } + if (!desc) { + printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu); + retval = -EBUSY; + } + } + return retval; } +/* must be called with cpucontrol mutex held */ int __cpu_disable(void) { int cpu = smp_processor_id(); @@ -587,12 +707,23 @@ int __cpu_disable(void) /* * dont permit boot processor for now */ - if (cpu == 0) - return -EBUSY; + if (cpu == 0 && !bsp_remove_ok) { + printk ("Your platform does not support removal of BSP\n"); + return (-EBUSY); + } + + cpu_clear(cpu, cpu_online_map); + if (migrate_platform_irqs(cpu)) { + cpu_set(cpu, cpu_online_map); + return (-EBUSY); + } + + remove_siblinginfo(cpu); + cpu_clear(cpu, cpu_online_map); fixup_irqs(); local_flush_tlb_all(); - printk ("Disabled cpu %u\n", smp_processor_id()); + cpu_clear(cpu, cpu_callin_map); return 0; } @@ -604,12 +735,7 @@ void __cpu_die(unsigned int cpu) /* They ack this in play_dead by setting CPU_DEAD */ if (per_cpu(cpu_state, cpu) == CPU_DEAD) { - /* - * TBD: Enable this when physical removal - * or when we put the processor is put in - * SAL_BOOT_RENDEZ mode - * cpu_clear(cpu, cpu_callin_map); - */ + printk ("CPU %d is now offline\n", cpu); return; } msleep(100); @@ -617,11 +743,6 @@ void __cpu_die(unsigned int cpu) printk(KERN_ERR "CPU %u didn't die...\n", cpu); } #else /* !CONFIG_HOTPLUG_CPU */ -static int __devinit cpu_enable(unsigned int cpu) -{ - return 0; -} - int __cpu_disable(void) { return -ENOSYS; @@ -644,14 +765,31 @@ smp_cpus_done (unsigned int dummy) * Allow the user to impress friends. */ - for (cpu = 0; cpu < NR_CPUS; cpu++) - if (cpu_online(cpu)) - bogosum += cpu_data(cpu)->loops_per_jiffy; + for_each_online_cpu(cpu) { + bogosum += cpu_data(cpu)->loops_per_jiffy; + } printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n", (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100); } +static inline void __devinit +set_cpu_sibling_map(int cpu) +{ + int i; + + for_each_online_cpu(i) { + if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) { + cpu_set(i, cpu_core_map[cpu]); + cpu_set(cpu, cpu_core_map[i]); + if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) { + cpu_set(i, cpu_sibling_map[cpu]); + cpu_set(cpu, cpu_sibling_map[i]); + } + } + } +} + int __devinit __cpu_up (unsigned int cpu) { @@ -663,21 +801,27 @@ __cpu_up (unsigned int cpu) return -EINVAL; /* - * Already booted.. just enable and get outa idle lool + * Already booted cpu? not valid anymore since we dont + * do idle loop tightspin anymore. */ if (cpu_isset(cpu, cpu_callin_map)) - { - cpu_enable(cpu); - local_irq_enable(); - while (!cpu_isset(cpu, cpu_online_map)) - mb(); - return 0; - } + return -EINVAL; + + per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; /* Processor goes to start_secondary(), sets online flag */ ret = do_boot_cpu(sapicid, cpu); if (ret < 0) return ret; + if (cpu_data(cpu)->threads_per_core == 1 && + cpu_data(cpu)->cores_per_socket == 1) { + cpu_set(cpu, cpu_sibling_map[cpu]); + cpu_set(cpu, cpu_core_map[cpu]); + return 0; + } + + set_cpu_sibling_map(cpu); + return 0; } @@ -705,3 +849,59 @@ init_smp_config(void) ia64_sal_strerror(sal_ret)); } +/* + * identify_siblings(cpu) gets called from identify_cpu. This populates the + * information related to logical execution units in per_cpu_data structure. + */ +void __devinit +identify_siblings(struct cpuinfo_ia64 *c) +{ + s64 status; + u16 pltid; + pal_logical_to_physical_t info; + + if (smp_num_cpucores == 1 && smp_num_siblings == 1) + return; + + if ((status = ia64_pal_logical_to_phys(-1, &info)) != PAL_STATUS_SUCCESS) { + printk(KERN_ERR "ia64_pal_logical_to_phys failed with %ld\n", + status); + return; + } + if ((status = ia64_sal_physical_id_info(&pltid)) != PAL_STATUS_SUCCESS) { + printk(KERN_ERR "ia64_sal_pltid failed with %ld\n", status); + return; + } + + c->socket_id = (pltid << 8) | info.overview_ppid; + c->cores_per_socket = info.overview_cpp; + c->threads_per_core = info.overview_tpc; + c->num_log = info.overview_num_log; + + c->core_id = info.log1_cid; + c->thread_id = info.log1_tid; +} + +/* + * returns non zero, if multi-threading is enabled + * on at least one physical package. Due to hotplug cpu + * and (maxcpus=), all threads may not necessarily be enabled + * even though the processor supports multi-threading. + */ +int is_multithreading_enabled(void) +{ + int i, j; + + for_each_present_cpu(i) { + for_each_present_cpu(j) { + if (j == i) + continue; + if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) { + if (cpu_data(j)->core_id == cpu_data(i)->core_id) + return 1; + } + } + } + return 0; +} +EXPORT_SYMBOL_GPL(is_multithreading_enabled);