[PATCH] x86_64: CPU hotplug support

  Experimental CPU hotplug patch for x86_64
  -----------------------------------------
This supports logical CPU online and offline.
- Test with maxcpus=1, and then kick other cpu's off to test if init code
  is all cleaned up. CONFIG_SCHED_SMT works as well.
- idle threads are forked on demand from keventd threads for clean startup

TBD:
1. Not tested on a real NUMA machine (tested with numa=fake=2)
2. Handle ACPI pieces for physical hotplug support.

Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Acked-by: Andi Kleen <ak@muc.de>
Acked-by: Zwane Mwaikambo <zwane@arm.linux.org.uk>
Signed-off-by: Shaohua.li<shaohua.li@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

1 files changed, 190 insertions, 15 deletions

diff --git a/arch/x86_64/kernel/smpboot.c b/arch/x86_64/kernel/smpboot.c
index bc98a6722cba..5a3f955b6576 100644
--- a/arch/x86_64/kernel/smpboot.c
+++ b/arch/x86_64/kernel/smpboot.c
@@ -34,6 +34,7 @@
  *      Andi Kleen              :       Converted to new state machine.
  *                                      Various cleanups.
  *                                      Probably mostly hotplug CPU ready now.
+ *      Ashok Raj                       : CPU hotplug support
  */
 
 
@@ -98,6 +99,37 @@ EXPORT_SYMBOL(cpu_core_map);
 extern unsigned char trampoline_data[];
 extern unsigned char trampoline_end[];
 
+/* State of each CPU */
+DEFINE_PER_CPU(int, cpu_state) = { 0 };
+
+/*
+ * 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] __cpuinitdata ;
+
+#define get_idle_for_cpu(x)     (idle_thread_array[(x)])
+#define set_idle_for_cpu(x,p)   (idle_thread_array[(x)] = (p))
+
+/*
+ * cpu_possible_map should be static, it cannot change as cpu's
+ * are onlined, or offlined. The reason is per-cpu data-structures
+ * are allocated by some modules at init time, and dont expect to
+ * do this dynamically on cpu arrival/departure.
+ * cpu_present_map on the other hand can change dynamically.
+ * In case when cpu_hotplug is not compiled, then we resort to current
+ * behaviour, which is cpu_possible == cpu_present.
+ * If cpu-hotplug is supported, then we need to preallocate for all
+ * those NR_CPUS, hence cpu_possible_map represents entire NR_CPUS range.
+ * - Ashok Raj
+ */
+#ifdef CONFIG_HOTPLUG_CPU
+#define fixup_cpu_possible_map(x)       cpu_set((x), cpu_possible_map)
+#else
+#define fixup_cpu_possible_map(x)
+#endif
+
 /*
  * Currently trivial. Write the real->protected mode
  * bootstrap into the page concerned. The caller
@@ -623,33 +655,77 @@ static int __cpuinit wakeup_secondary_via_INIT(int phys_apicid, unsigned int sta
         return (send_status | accept_status);
 }
 
+struct create_idle {
+        struct task_struct *idle;
+        struct completion done;
+        int cpu;
+};
+
+void do_fork_idle(void *_c_idle)
+{
+        struct create_idle *c_idle = _c_idle;
+
+        c_idle->idle = fork_idle(c_idle->cpu);
+        complete(&c_idle->done);
+}
+
 /*
  * Boot one CPU.
  */
 static int __cpuinit do_boot_cpu(int cpu, int apicid)
 {
-        struct task_struct *idle;
         unsigned long boot_error;
         int timeout;
         unsigned long start_rip;
+        struct create_idle c_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) {
+                c_idle.idle->thread.rsp = (unsigned long) (((struct pt_regs *)
+                        (THREAD_SIZE + (unsigned long) c_idle.idle->thread_info)) - 1);
+                init_idle(c_idle.idle, cpu);
+                goto do_rest;
+        }
+
         /*
-         * We can't use kernel_thread since we must avoid to
-         * reschedule the child.
+         * During cold boot process, keventd thread is not spun up yet.
+         * When we do cpu hot-add, we create idle threads on the fly, we should
+         * not acquire any attributes from the calling context. Hence the clean
+         * way to create kernel_threads() is to do that from keventd().
+         * We do the current_is_keventd() due to the fact that ACPI notifier
+         * was also queuing to keventd() and when the caller is already running
+         * in context of keventd(), we would end up with locking up the keventd
+         * thread.
          */
-        idle = fork_idle(cpu);
-        if (IS_ERR(idle)) {
+        if (!keventd_up() || current_is_keventd())
+                work.func(work.data);
+        else {
+                schedule_work(&work);
+                wait_for_completion(&c_idle.done);
+        }
+
+        if (IS_ERR(c_idle.idle)) {
                 printk("failed fork for CPU %d\n", cpu);
-                return PTR_ERR(idle);
+                return PTR_ERR(c_idle.idle);
         }
 
-        cpu_pda[cpu].pcurrent = idle;
+        set_idle_for_cpu(cpu, c_idle.idle);
+
+do_rest:
+
+        cpu_pda[cpu].pcurrent = c_idle.idle;
 
         start_rip = setup_trampoline();
 
-        init_rsp = idle->thread.rsp;
+        init_rsp = c_idle.idle->thread.rsp;
         per_cpu(init_tss,cpu).rsp0 = init_rsp;
         initial_code = start_secondary;
-        clear_ti_thread_flag(idle->thread_info, TIF_FORK);
+        clear_ti_thread_flag(c_idle.idle->thread_info, TIF_FORK);
 
         printk(KERN_INFO "Booting processor %d/%d rip %lx rsp %lx\n", cpu, apicid,
                start_rip, init_rsp);
@@ -925,10 +1001,9 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
                 int apicid = cpu_present_to_apicid(i);
                 if (physid_isset(apicid, phys_cpu_present_map)) {
                         cpu_set(i, cpu_present_map);
-                        /* possible map would be different if we supported real
-                           CPU hotplug. */
                         cpu_set(i, cpu_possible_map);
                 }
+                fixup_cpu_possible_map(i);
         }
 
         if (smp_sanity_check(max_cpus) < 0) {
@@ -977,9 +1052,6 @@ void __init smp_prepare_boot_cpu(void)
 
 /*
  * Entry point to boot a CPU.
- *
- * This is all __cpuinit, not __devinit for now because we don't support
- * CPU hotplug (yet).
  */
 int __cpuinit __cpu_up(unsigned int cpu)
 {
@@ -996,6 +1068,14 @@ int __cpuinit __cpu_up(unsigned int cpu)
                 return -EINVAL;
         }
 
+        /*
+         * Already booted CPU?
+         */
+        if (cpu_isset(cpu, cpu_callin_map)) {
+                Dprintk("do_boot_cpu %d Already started\n", cpu);
+                return -ENOSYS;
+        }
+
         /* Boot it! */
         err = do_boot_cpu(cpu, apicid);
         if (err < 0) {
@@ -1008,7 +1088,9 @@ int __cpuinit __cpu_up(unsigned int cpu)
 
         while (!cpu_isset(cpu, cpu_online_map))
                 cpu_relax();
-        return 0;
+        err = 0;
+
+        return err;
 }
 
 /*
@@ -1016,7 +1098,9 @@ int __cpuinit __cpu_up(unsigned int cpu)
  */
 void __init smp_cpus_done(unsigned int max_cpus)
 {
+#ifndef CONFIG_HOTPLUG_CPU
         zap_low_mappings();
+#endif
         smp_cleanup_boot();
 
 #ifdef CONFIG_X86_IO_APIC
@@ -1028,3 +1112,94 @@ void __init smp_cpus_done(unsigned int max_cpus)
 
         check_nmi_watchdog();
 }
+
+#ifdef CONFIG_HOTPLUG_CPU
+
+static void
+remove_siblinginfo(int cpu)
+{
+        int sibling;
+
+        for_each_cpu_mask(sibling, cpu_sibling_map[cpu])
+                cpu_clear(cpu, cpu_sibling_map[sibling]);
+        for_each_cpu_mask(sibling, cpu_core_map[cpu])
+                cpu_clear(cpu, cpu_core_map[sibling]);
+        cpus_clear(cpu_sibling_map[cpu]);
+        cpus_clear(cpu_core_map[cpu]);
+        phys_proc_id[cpu] = BAD_APICID;
+        cpu_core_id[cpu] = BAD_APICID;
+}
+
+void remove_cpu_from_maps(void)
+{
+        int cpu = smp_processor_id();
+
+        cpu_clear(cpu, cpu_callout_map);
+        cpu_clear(cpu, cpu_callin_map);
+        clear_bit(cpu, &cpu_initialized); /* was set by cpu_init() */
+}
+
+int __cpu_disable(void)
+{
+        int cpu = smp_processor_id();
+
+        /*
+         * Perhaps use cpufreq to drop frequency, but that could go
+         * into generic code.
+         *
+         * We won't take down the boot processor on i386 due to some
+         * interrupts only being able to be serviced by the BSP.
+         * Especially so if we're not using an IOAPIC   -zwane
+         */
+        if (cpu == 0)
+                return -EBUSY;
+
+        disable_APIC_timer();
+
+        /*
+         * HACK:
+         * Allow any queued timer interrupts to get serviced
+         * This is only a temporary solution until we cleanup
+         * fixup_irqs as we do for IA64.
+         */
+        local_irq_enable();
+        mdelay(1);
+
+        local_irq_disable();
+        remove_siblinginfo(cpu);
+
+        /* It's now safe to remove this processor from the online map */
+        cpu_clear(cpu, cpu_online_map);
+        remove_cpu_from_maps();
+        fixup_irqs(cpu_online_map);
+        return 0;
+}
+
+void __cpu_die(unsigned int cpu)
+{
+        /* We don't do anything here: idle task is faking death itself. */
+        unsigned int i;
+
+        for (i = 0; i < 10; i++) {
+                /* They ack this in play_dead by setting CPU_DEAD */
+                if (per_cpu(cpu_state, cpu) == CPU_DEAD)
+                        return;
+                current->state = TASK_UNINTERRUPTIBLE;
+                schedule_timeout(HZ/10);
+        }
+        printk(KERN_ERR "CPU %u didn't die...\n", cpu);
+}
+
+#else /* ... !CONFIG_HOTPLUG_CPU */
+
+int __cpu_disable(void)
+{
+        return -ENOSYS;
+}
+
+void __cpu_die(unsigned int cpu)
+{
+        /* We said "no" in __cpu_disable */
+        BUG();
+}
+#endif /* CONFIG_HOTPLUG_CPU */