/*
* SMP boot-related support
*
- * Copyright (C) 1998-2003 Hewlett-Packard Co
+ * Copyright (C) 1998-2003, 2005 Hewlett-Packard Co
* David Mosberger-Tang <davidm@hpl.hp.com>
+ * Copyright (C) 2001, 2004-2005 Intel Corp
+ * Rohit Seth <rohit.seth@intel.com>
+ * Suresh Siddha <suresh.b.siddha@intel.com>
+ * Gordon Jin <gordon.jin@intel.com>
+ * Ashok Raj <ashok.raj@intel.com>
*
* 01/05/16 Rohit Seth <rohit.seth@intel.com> Moved SMP booting functions from smp.c to here.
* 01/04/27 David Mosberger <davidm@hpl.hp.com> Added ITC synching code.
* 02/07/31 David Mosberger <davidm@hpl.hp.com> 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 <ashok.raj@intel.com> Added CPU Hotplug Support
+ * 04/12/26 Jin Gordon <gordon.jin@intel.com>
+ * 04/12/26 Rohit Seth <rohit.seth@intel.com>
+ * Add multi-threading and multi-core detection
+ * 05/01/30 Suresh Siddha <suresh.b.siddha@intel.com>
+ * Setup cpu_sibling_map and cpu_core_map
*/
#include <linux/config.h>
#define Dprintk(x...)
#endif
+#ifdef CONFIG_HOTPLUG_CPU
+/*
+ * 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:
cpumask_t cpu_possible_map;
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);
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();
}
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();
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);
{
{
/* 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();
.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 (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);
/*
{
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
*/
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!
cpu_set(smp_processor_id(), cpu_callin_map);
}
+/*
+ * mt_info[] is a temporary store for all info returned by
+ * PAL_LOGICAL_TO_PHYSICAL, to be copied into cpuinfo_ia64 when the
+ * specific cpu comes.
+ */
+static struct {
+ __u32 socket_id;
+ __u16 core_id;
+ __u16 thread_id;
+ __u16 proc_fixed_addr;
+ __u8 valid;
+} mt_info[NR_CPUS] __devinitdata;
+
#ifdef CONFIG_HOTPLUG_CPU
-extern void fixup_irqs(void);
-/* must be called with cpucontrol mutex held */
-static int __devinit cpu_enable(unsigned int cpu)
+static inline void
+remove_from_mtinfo(int cpu)
{
- per_cpu(cpu_state,cpu) = CPU_UP_PREPARE;
- wmb();
+ int i;
- while (!cpu_online(cpu))
- cpu_relax();
- return 0;
+ for_each_cpu(i)
+ if (mt_info[i].valid && mt_info[i].socket_id ==
+ cpu_data(cpu)->socket_id)
+ mt_info[i].valid = 0;
+}
+
+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);
+
+ /* if this cpu is the last in the core group, remove all its info
+ * from mt_info structure
+ */
+ if (last)
+ remove_from_mtinfo(cpu);
}
+extern void fixup_irqs(void);
+/* must be called with cpucontrol mutex held */
int __cpu_disable(void)
{
int cpu = smp_processor_id();
if (cpu == 0)
return -EBUSY;
+ remove_siblinginfo(cpu);
fixup_irqs();
local_flush_tlb_all();
- printk ("Disabled cpu %u\n", smp_processor_id());
+ cpu_clear(cpu, cpu_callin_map);
return 0;
}
/* 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);
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;
(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)
{
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;
+
/* 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;
}
ia64_sal_strerror(sal_ret));
}
+static inline int __devinit
+check_for_mtinfo_index(void)
+{
+ int i;
+
+ for_each_cpu(i)
+ if (!mt_info[i].valid)
+ return i;
+
+ return -1;
+}
+
+/*
+ * Search the mt_info to find out if this socket's cid/tid information is
+ * cached or not. If the socket exists, fill in the core_id and thread_id
+ * in cpuinfo
+ */
+static int __devinit
+check_for_new_socket(__u16 logical_address, struct cpuinfo_ia64 *c)
+{
+ int i;
+ __u32 sid = c->socket_id;
+
+ for_each_cpu(i) {
+ if (mt_info[i].valid && mt_info[i].proc_fixed_addr == logical_address
+ && mt_info[i].socket_id == sid) {
+ c->core_id = mt_info[i].core_id;
+ c->thread_id = mt_info[i].thread_id;
+ return 1; /* not a new socket */
+ }
+ }
+ return 0;
+}
+
+/*
+ * 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;
+ u64 proc_fixed_addr;
+ int count, i;
+ pal_logical_to_physical_t info;
+
+ if (smp_num_cpucores == 1 && smp_num_siblings == 1)
+ return;
+
+ if ((status = ia64_pal_logical_to_phys(0, &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;
+ }
+ if ((status = ia64_pal_fixed_addr(&proc_fixed_addr)) != PAL_STATUS_SUCCESS) {
+ printk(KERN_ERR "ia64_pal_fixed_addr 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;
+ count = c->num_log = info.overview_num_log;
+
+ /* If the thread and core id information is already cached, then
+ * we will simply update cpu_info and return. Otherwise, we will
+ * do the PAL calls and cache core and thread id's of all the siblings.
+ */
+ if (check_for_new_socket(proc_fixed_addr, c))
+ return;
+
+ for (i = 0; i < count; i++) {
+ int index;
+
+ if (i && (status = ia64_pal_logical_to_phys(i, &info))
+ != PAL_STATUS_SUCCESS) {
+ printk(KERN_ERR "ia64_pal_logical_to_phys failed"
+ " with %ld\n", status);
+ return;
+ }
+ if (info.log2_la == proc_fixed_addr) {
+ c->core_id = info.log1_cid;
+ c->thread_id = info.log1_tid;
+ }
+
+ index = check_for_mtinfo_index();
+ /* We will not do the mt_info caching optimization in this case.
+ */
+ if (index < 0)
+ continue;
+
+ mt_info[index].valid = 1;
+ mt_info[index].socket_id = c->socket_id;
+ mt_info[index].core_id = info.log1_cid;
+ mt_info[index].thread_id = info.log1_tid;
+ mt_info[index].proc_fixed_addr = info.log2_la;
+ }
+}
git://git.onelab.eu / linux-2.6.git / blobdiff