releases/4.14.34/x86-microcode-synchronize-late-microcode-loading.patch

   1 From a5321aec6412b20b5ad15db2d6b916c05349dbff Mon Sep 17 00:00:00 2001
   2 From: Ashok Raj <ashok.raj@intel.com>
   3 Date: Wed, 28 Feb 2018 11:28:46 +0100
   4 Subject: x86/microcode: Synchronize late microcode loading
   5
   6 From: Ashok Raj <ashok.raj@intel.com>
   7
   8 commit a5321aec6412b20b5ad15db2d6b916c05349dbff upstream.
   9
  10 Original idea by Ashok, completely rewritten by Borislav.
  11
  12 Before you read any further: the early loading method is still the
  13 preferred one and you should always do that. The following patch is
  14 improving the late loading mechanism for long running jobs and cloud use
  15 cases.
  16
  17 Gather all cores and serialize the microcode update on them by doing it
  18 one-by-one to make the late update process as reliable as possible and
  19 avoid potential issues caused by the microcode update.
  20
  21 [ Borislav: Rewrite completely. ]
  22
  23 Co-developed-by: Borislav Petkov <bp@suse.de>
  24 Signed-off-by: Ashok Raj <ashok.raj@intel.com>
  25 Signed-off-by: Borislav Petkov <bp@suse.de>
  26 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
  27 Tested-by: Tom Lendacky <thomas.lendacky@amd.com>
  28 Tested-by: Ashok Raj <ashok.raj@intel.com>
  29 Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com>
  30 Cc: Arjan Van De Ven <arjan.van.de.ven@intel.com>
  31 Link: https://lkml.kernel.org/r/20180228102846.13447-8-bp@alien8.de
  32 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  33
  34 ---
  35  arch/x86/kernel/cpu/microcode/core.c |  118 +++++++++++++++++++++++++++--------
  36  1 file changed, 92 insertions(+), 26 deletions(-)
  37
  38 --- a/arch/x86/kernel/cpu/microcode/core.c
  39 +++ b/arch/x86/kernel/cpu/microcode/core.c
  40 @@ -22,13 +22,16 @@
  41  #define pr_fmt(fmt) "microcode: " fmt
  42
  43  #include <linux/platform_device.h>
  44 +#include <linux/stop_machine.h>
  45  #include <linux/syscore_ops.h>
  46  #include <linux/miscdevice.h>
  47  #include <linux/capability.h>
  48  #include <linux/firmware.h>
  49  #include <linux/kernel.h>
  50 +#include <linux/delay.h>
  51  #include <linux/mutex.h>
  52  #include <linux/cpu.h>
  53 +#include <linux/nmi.h>
  54  #include <linux/fs.h>
  55  #include <linux/mm.h>
  56
  57 @@ -64,6 +67,11 @@ LIST_HEAD(microcode_cache);
  58   */
  59  static DEFINE_MUTEX(microcode_mutex);
  60
  61 +/*
  62 + * Serialize late loading so that CPUs get updated one-by-one.
  63 + */
  64 +static DEFINE_SPINLOCK(update_lock);
  65 +
  66  struct ucode_cpu_info          ucode_cpu_info[NR_CPUS];
  67
  68  struct cpu_info_ctx {
  69 @@ -486,6 +494,19 @@ static void __exit microcode_dev_exit(vo
  70  /* fake device for request_firmware */
  71  static struct platform_device  *microcode_pdev;
  72
  73 +/*
  74 + * Late loading dance. Why the heavy-handed stomp_machine effort?
  75 + *
  76 + * - HT siblings must be idle and not execute other code while the other sibling
  77 + *   is loading microcode in order to avoid any negative interactions caused by
  78 + *   the loading.
  79 + *
  80 + * - In addition, microcode update on the cores must be serialized until this
  81 + *   requirement can be relaxed in the future. Right now, this is conservative
  82 + *   and good.
  83 + */
  84 +#define SPINUNIT 100 /* 100 nsec */
  85 +
  86  static int check_online_cpus(void)
  87  {
  88         if (num_online_cpus() == num_present_cpus())
  89 @@ -496,23 +517,85 @@ static int check_online_cpus(void)
  90         return -EINVAL;
  91  }
  92
  93 -static enum ucode_state reload_for_cpu(int cpu)
  94 +static atomic_t late_cpus;
  95 +
  96 +/*
  97 + * Returns:
  98 + * < 0 - on error
  99 + *   0 - no update done
 100 + *   1 - microcode was updated
 101 + */
 102 +static int __reload_late(void *info)
 103  {
 104 -       struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
 105 +       unsigned int timeout = NSEC_PER_SEC;
 106 +       int all_cpus = num_online_cpus();
 107 +       int cpu = smp_processor_id();
 108 +       enum ucode_state err;
 109 +       int ret = 0;
 110 +
 111 +       atomic_dec(&late_cpus);
 112 +
 113 +       /*
 114 +        * Wait for all CPUs to arrive. A load will not be attempted unless all
 115 +        * CPUs show up.
 116 +        * */
 117 +       while (atomic_read(&late_cpus)) {
 118 +               if (timeout < SPINUNIT) {
 119 +                       pr_err("Timeout while waiting for CPUs rendezvous, remaining: %d\n",
 120 +                               atomic_read(&late_cpus));
 121 +                       return -1;
 122 +               }
 123 +
 124 +               ndelay(SPINUNIT);
 125 +               timeout -= SPINUNIT;
 126 +
 127 +               touch_nmi_watchdog();
 128 +       }
 129 +
 130 +       spin_lock(&update_lock);
 131 +       apply_microcode_local(&err);
 132 +       spin_unlock(&update_lock);
 133 +
 134 +       if (err > UCODE_NFOUND) {
 135 +               pr_warn("Error reloading microcode on CPU %d\n", cpu);
 136 +               ret = -1;
 137 +       } else if (err == UCODE_UPDATED) {
 138 +               ret = 1;
 139 +       }
 140
 141 -       if (!uci->valid)
 142 -               return UCODE_OK;
 143 +       atomic_inc(&late_cpus);
 144
 145 -       return apply_microcode_on_target(cpu);
 146 +       while (atomic_read(&late_cpus) != all_cpus)
 147 +               cpu_relax();
 148 +
 149 +       return ret;
 150 +}
 151 +
 152 +/*
 153 + * Reload microcode late on all CPUs. Wait for a sec until they
 154 + * all gather together.
 155 + */
 156 +static int microcode_reload_late(void)
 157 +{
 158 +       int ret;
 159 +
 160 +       atomic_set(&late_cpus, num_online_cpus());
 161 +
 162 +       ret = stop_machine_cpuslocked(__reload_late, NULL, cpu_online_mask);
 163 +       if (ret < 0)
 164 +               return ret;
 165 +       else if (ret > 0)
 166 +               microcode_check();
 167 +
 168 +       return ret;
 169  }
 170
 171  static ssize_t reload_store(struct device *dev,
 172                             struct device_attribute *attr,
 173                             const char *buf, size_t size)
 174  {
 175 -       int cpu, bsp = boot_cpu_data.cpu_index;
 176         enum ucode_state tmp_ret = UCODE_OK;
 177 -       bool do_callback = false;
 178 +       int bsp = boot_cpu_data.cpu_index;
 179         unsigned long val;
 180         ssize_t ret = 0;
 181
 182 @@ -534,30 +617,13 @@ static ssize_t reload_store(struct devic
 183                 goto put;
 184
 185         mutex_lock(&microcode_mutex);
 186 -
 187 -       for_each_online_cpu(cpu) {
 188 -               tmp_ret = reload_for_cpu(cpu);
 189 -               if (tmp_ret > UCODE_NFOUND) {
 190 -                       pr_warn("Error reloading microcode on CPU %d\n", cpu);
 191 -
 192 -                       /* set retval for the first encountered reload error */
 193 -                       if (!ret)
 194 -                               ret = -EINVAL;
 195 -               }
 196 -
 197 -               if (tmp_ret == UCODE_UPDATED)
 198 -                       do_callback = true;
 199 -       }
 200 -
 201 -       if (!ret && do_callback)
 202 -               microcode_check();
 203 -
 204 +       ret = microcode_reload_late();
 205         mutex_unlock(&microcode_mutex);
 206
 207  put:
 208         put_online_cpus();
 209
 210 -       if (!ret)
 211 +       if (ret >= 0)
 212                 ret = size;
 213
 214         return ret;