arch/powerpc/platforms/powernv/smp.c

   1 /*
   2  * SMP support for PowerNV machines.
   3  *
   4  * Copyright 2011 IBM Corp.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/sched.h>
  15 #include <linux/smp.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/delay.h>
  18 #include <linux/init.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/cpu.h>
  21
  22 #include <asm/irq.h>
  23 #include <asm/smp.h>
  24 #include <asm/paca.h>
  25 #include <asm/machdep.h>
  26 #include <asm/cputable.h>
  27 #include <asm/firmware.h>
  28 #include <asm/rtas.h>
  29 #include <asm/vdso_datapage.h>
  30 #include <asm/cputhreads.h>
  31 #include <asm/xics.h>
  32 #include <asm/opal.h>
  33
  34 #include "powernv.h"
  35
  36 #ifdef DEBUG
  37 #include <asm/udbg.h>
  38 #define DBG(fmt...) udbg_printf(fmt)
  39 #else
  40 #define DBG(fmt...)
  41 #endif
  42
  43 static void pnv_smp_setup_cpu(int cpu)
  44 {
  45         if (cpu != boot_cpuid)
  46                 xics_setup_cpu();
  47 }
  48
  49 int pnv_smp_kick_cpu(int nr)
  50 {
  51         unsigned int pcpu = get_hard_smp_processor_id(nr);
  52         unsigned long start_here = __pa(*((unsigned long *)
  53                                           generic_secondary_smp_init));
  54         long rc;
  55
  56         BUG_ON(nr < 0 || nr >= NR_CPUS);
  57
  58         /*
  59          * If we already started or OPALv2 is not supported, we just
  60          * kick the CPU via the PACA
  61          */
  62         if (paca[nr].cpu_start || !firmware_has_feature(FW_FEATURE_OPALv2))
  63                 goto kick;
  64
  65         /*
  66          * At this point, the CPU can either be spinning on the way in
  67          * from kexec or be inside OPAL waiting to be started for the
  68          * first time. OPAL v3 allows us to query OPAL to know if it
  69          * has the CPUs, so we do that
  70          */
  71         if (firmware_has_feature(FW_FEATURE_OPALv3)) {
  72                 uint8_t status;
  73
  74                 rc = opal_query_cpu_status(pcpu, &status);
  75                 if (rc != OPAL_SUCCESS) {
  76                         pr_warn("OPAL Error %ld querying CPU %d state\n",
  77                                 rc, nr);
  78                         return -ENODEV;
  79                 }
  80
  81                 /*
  82                  * Already started, just kick it, probably coming from
  83                  * kexec and spinning
  84                  */
  85                 if (status == OPAL_THREAD_STARTED)
  86                         goto kick;
  87
  88                 /*
  89                  * Available/inactive, let's kick it
  90                  */
  91                 if (status == OPAL_THREAD_INACTIVE) {
  92                         pr_devel("OPAL: Starting CPU %d (HW 0x%x)...\n",
  93                                  nr, pcpu);
  94                         rc = opal_start_cpu(pcpu, start_here);
  95                         if (rc != OPAL_SUCCESS) {
  96                                 pr_warn("OPAL Error %ld starting CPU %d\n",
  97                                         rc, nr);
  98                                 return -ENODEV;
  99                         }
 100                 } else {
 101                         /*
 102                          * An unavailable CPU (or any other unknown status)
 103                          * shouldn't be started. It should also
 104                          * not be in the possible map but currently it can
 105                          * happen
 106                          */
 107                         pr_devel("OPAL: CPU %d (HW 0x%x) is unavailable"
 108                                  " (status %d)...\n", nr, pcpu, status);
 109                         return -ENODEV;
 110                 }
 111         } else {
 112                 /*
 113                  * On OPAL v2, we just kick it and hope for the best,
 114                  * we must not test the error from opal_start_cpu() or
 115                  * we would fail to get CPUs from kexec.
 116                  */
 117                 opal_start_cpu(pcpu, start_here);
 118         }
 119  kick:
 120         return smp_generic_kick_cpu(nr);
 121 }
 122
 123 #ifdef CONFIG_HOTPLUG_CPU
 124
 125 static int pnv_smp_cpu_disable(void)
 126 {
 127         int cpu = smp_processor_id();
 128
 129         /* This is identical to pSeries... might consolidate by
 130          * moving migrate_irqs_away to a ppc_md with default to
 131          * the generic fixup_irqs. --BenH.
 132          */
 133         set_cpu_online(cpu, false);
 134         vdso_data->processorCount--;
 135         if (cpu == boot_cpuid)
 136                 boot_cpuid = cpumask_any(cpu_online_mask);
 137         xics_migrate_irqs_away();
 138         return 0;
 139 }
 140
 141 static void pnv_smp_cpu_kill_self(void)
 142 {
 143         unsigned int cpu;
 144
 145         /* Standard hot unplug procedure */
 146         local_irq_disable();
 147         idle_task_exit();
 148         current->active_mm = NULL; /* for sanity */
 149         cpu = smp_processor_id();
 150         DBG("CPU%d offline\n", cpu);
 151         generic_set_cpu_dead(cpu);
 152         smp_wmb();
 153
 154         /* We don't want to take decrementer interrupts while we are offline,
 155          * so clear LPCR:PECE1. We keep PECE2 enabled.
 156          */
 157         mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) & ~(u64)LPCR_PECE1);
 158         while (!generic_check_cpu_restart(cpu)) {
 159                 power7_nap();
 160                 if (!generic_check_cpu_restart(cpu)) {
 161                         DBG("CPU%d Unexpected exit while offline !\n", cpu);
 162                         /* We may be getting an IPI, so we re-enable
 163                          * interrupts to process it, it will be ignored
 164                          * since we aren't online (hopefully)
 165                          */
 166                         local_irq_enable();
 167                         local_irq_disable();
 168                 }
 169         }
 170         mtspr(SPRN_LPCR, mfspr(SPRN_LPCR) | LPCR_PECE1);
 171         DBG("CPU%d coming online...\n", cpu);
 172 }
 173
 174 #endif /* CONFIG_HOTPLUG_CPU */
 175
 176 static struct smp_ops_t pnv_smp_ops = {
 177         .message_pass   = smp_muxed_ipi_message_pass,
 178         .cause_ipi      = NULL, /* Filled at runtime by xics_smp_probe() */
 179         .probe          = xics_smp_probe,
 180         .kick_cpu       = pnv_smp_kick_cpu,
 181         .setup_cpu      = pnv_smp_setup_cpu,
 182         .cpu_bootable   = smp_generic_cpu_bootable,
 183 #ifdef CONFIG_HOTPLUG_CPU
 184         .cpu_disable    = pnv_smp_cpu_disable,
 185         .cpu_die        = generic_cpu_die,
 186 #endif /* CONFIG_HOTPLUG_CPU */
 187 };
 188
 189 /* This is called very early during platform setup_arch */
 190 void __init pnv_smp_init(void)
 191 {
 192         smp_ops = &pnv_smp_ops;
 193
 194         /* XXX We don't yet have a proper entry point from HAL, for
 195          * now we rely on kexec-style entry from BML
 196          */
 197
 198 #ifdef CONFIG_PPC_RTAS
 199         /* Non-lpar has additional take/give timebase */
 200         if (rtas_token("freeze-time-base") != RTAS_UNKNOWN_SERVICE) {
 201                 smp_ops->give_timebase = rtas_give_timebase;
 202                 smp_ops->take_timebase = rtas_take_timebase;
 203         }
 204 #endif /* CONFIG_PPC_RTAS */
 205
 206 #ifdef CONFIG_HOTPLUG_CPU
 207         ppc_md.cpu_die  = pnv_smp_cpu_kill_self;
 208 #endif
 209 }