arch/x86/kernel/cpu/vmware.c

   1 /*
   2  * VMware Detection code.
   3  *
   4  * Copyright (C) 2008, VMware, Inc.
   5  * Author : Alok N Kataria <akataria@vmware.com>
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  15  * NON INFRINGEMENT.  See the GNU General Public License for more
  16  * details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  21  *
  22  */
  23
  24 #include <linux/dmi.h>
  25 #include <linux/init.h>
  26 #include <linux/export.h>
  27 #include <linux/clocksource.h>
  28 #include <asm/div64.h>
  29 #include <asm/x86_init.h>
  30 #include <asm/hypervisor.h>
  31 #include <asm/timer.h>
  32 #include <asm/apic.h>
  33 #include <asm/vmware.h>
  34
  35 #undef pr_fmt
  36 #define pr_fmt(fmt)     "vmware: " fmt
  37
  38 #define CPUID_VMWARE_INFO_LEAF               0x40000000
  39 #define CPUID_VMWARE_FEATURES_LEAF           0x40000010
  40 #define CPUID_VMWARE_FEATURES_ECX_VMMCALL    BIT(0)
  41 #define CPUID_VMWARE_FEATURES_ECX_VMCALL     BIT(1)
  42
  43 #define VMWARE_HYPERVISOR_MAGIC 0x564D5868
  44
  45 #define VMWARE_CMD_GETVERSION    10
  46 #define VMWARE_CMD_GETHZ         45
  47 #define VMWARE_CMD_GETVCPU_INFO  68
  48 #define VMWARE_CMD_LEGACY_X2APIC  3
  49 #define VMWARE_CMD_VCPU_RESERVED 31
  50
  51 #define VMWARE_PORT(cmd, eax, ebx, ecx, edx)                            \
  52         __asm__("inl (%%dx)" :                                          \
  53                 "=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :            \
  54                 "a"(VMWARE_HYPERVISOR_MAGIC),                           \
  55                 "c"(VMWARE_CMD_##cmd),                                  \
  56                 "d"(VMWARE_HYPERVISOR_PORT), "b"(UINT_MAX) :            \
  57                 "memory")
  58
  59 #define VMWARE_VMCALL(cmd, eax, ebx, ecx, edx)                          \
  60         __asm__("vmcall" :                                              \
  61                 "=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :            \
  62                 "a"(VMWARE_HYPERVISOR_MAGIC),                           \
  63                 "c"(VMWARE_CMD_##cmd),                                  \
  64                 "d"(0), "b"(UINT_MAX) :                                 \
  65                 "memory")
  66
  67 #define VMWARE_VMMCALL(cmd, eax, ebx, ecx, edx)                         \
  68         __asm__("vmmcall" :                                             \
  69                 "=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :            \
  70                 "a"(VMWARE_HYPERVISOR_MAGIC),                           \
  71                 "c"(VMWARE_CMD_##cmd),                                  \
  72                 "d"(0), "b"(UINT_MAX) :                                 \
  73                 "memory")
  74
  75 #define VMWARE_CMD(cmd, eax, ebx, ecx, edx) do {                \
  76         switch (vmware_hypercall_mode) {                        \
  77         case CPUID_VMWARE_FEATURES_ECX_VMCALL:                  \
  78                 VMWARE_VMCALL(cmd, eax, ebx, ecx, edx);         \
  79                 break;                                          \
  80         case CPUID_VMWARE_FEATURES_ECX_VMMCALL:                 \
  81                 VMWARE_VMMCALL(cmd, eax, ebx, ecx, edx);        \
  82                 break;                                          \
  83         default:                                                \
  84                 VMWARE_PORT(cmd, eax, ebx, ecx, edx);           \
  85                 break;                                          \
  86         }                                                       \
  87         } while (0)
  88
  89 static unsigned long vmware_tsc_khz __ro_after_init;
  90 static u8 vmware_hypercall_mode     __ro_after_init;
  91
  92 static inline int __vmware_platform(void)
  93 {
  94         uint32_t eax, ebx, ecx, edx;
  95         VMWARE_CMD(GETVERSION, eax, ebx, ecx, edx);
  96         return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
  97 }
  98
  99 static unsigned long vmware_get_tsc_khz(void)
 100 {
 101         return vmware_tsc_khz;
 102 }
 103
 104 #ifdef CONFIG_PARAVIRT
 105 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
 106 static int vmw_sched_clock __initdata = 1;
 107
 108 static __init int setup_vmw_sched_clock(char *s)
 109 {
 110         vmw_sched_clock = 0;
 111         return 0;
 112 }
 113 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
 114
 115 static unsigned long long notrace vmware_sched_clock(void)
 116 {
 117         unsigned long long ns;
 118
 119         ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
 120                              vmware_cyc2ns.cyc2ns_shift);
 121         ns -= vmware_cyc2ns.cyc2ns_offset;
 122         return ns;
 123 }
 124
 125 static void __init vmware_sched_clock_setup(void)
 126 {
 127         struct cyc2ns_data *d = &vmware_cyc2ns;
 128         unsigned long long tsc_now = rdtsc();
 129
 130         clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
 131                                vmware_tsc_khz, NSEC_PER_MSEC, 0);
 132         d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
 133                                            d->cyc2ns_shift);
 134
 135         pv_ops.time.sched_clock = vmware_sched_clock;
 136         pr_info("using sched offset of %llu ns\n", d->cyc2ns_offset);
 137 }
 138
 139 static void __init vmware_paravirt_ops_setup(void)
 140 {
 141         pv_info.name = "VMware hypervisor";
 142         pv_ops.cpu.io_delay = paravirt_nop;
 143
 144         if (vmware_tsc_khz && vmw_sched_clock)
 145                 vmware_sched_clock_setup();
 146 }
 147 #else
 148 #define vmware_paravirt_ops_setup() do {} while (0)
 149 #endif
 150
 151 /*
 152  * VMware hypervisor takes care of exporting a reliable TSC to the guest.
 153  * Still, due to timing difference when running on virtual cpus, the TSC can
 154  * be marked as unstable in some cases. For example, the TSC sync check at
 155  * bootup can fail due to a marginal offset between vcpus' TSCs (though the
 156  * TSCs do not drift from each other).  Also, the ACPI PM timer clocksource
 157  * is not suitable as a watchdog when running on a hypervisor because the
 158  * kernel may miss a wrap of the counter if the vcpu is descheduled for a
 159  * long time. To skip these checks at runtime we set these capability bits,
 160  * so that the kernel could just trust the hypervisor with providing a
 161  * reliable virtual TSC that is suitable for timekeeping.
 162  */
 163 static void __init vmware_set_capabilities(void)
 164 {
 165         setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
 166         setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
 167         if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMCALL)
 168                 setup_force_cpu_cap(X86_FEATURE_VMCALL);
 169         else if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMMCALL)
 170                 setup_force_cpu_cap(X86_FEATURE_VMW_VMMCALL);
 171 }
 172
 173 static void __init vmware_platform_setup(void)
 174 {
 175         uint32_t eax, ebx, ecx, edx;
 176         uint64_t lpj, tsc_khz;
 177
 178         VMWARE_CMD(GETHZ, eax, ebx, ecx, edx);
 179
 180         if (ebx != UINT_MAX) {
 181                 lpj = tsc_khz = eax | (((uint64_t)ebx) << 32);
 182                 do_div(tsc_khz, 1000);
 183                 WARN_ON(tsc_khz >> 32);
 184                 pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
 185                         (unsigned long) tsc_khz / 1000,
 186                         (unsigned long) tsc_khz % 1000);
 187
 188                 if (!preset_lpj) {
 189                         do_div(lpj, HZ);
 190                         preset_lpj = lpj;
 191                 }
 192
 193                 vmware_tsc_khz = tsc_khz;
 194                 x86_platform.calibrate_tsc = vmware_get_tsc_khz;
 195                 x86_platform.calibrate_cpu = vmware_get_tsc_khz;
 196
 197 #ifdef CONFIG_X86_LOCAL_APIC
 198                 /* Skip lapic calibration since we know the bus frequency. */
 199                 lapic_timer_period = ecx / HZ;
 200                 pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
 201                         ecx);
 202 #endif
 203         } else {
 204                 pr_warn("Failed to get TSC freq from the hypervisor\n");
 205         }
 206
 207         vmware_paravirt_ops_setup();
 208
 209 #ifdef CONFIG_X86_IO_APIC
 210         no_timer_check = 1;
 211 #endif
 212
 213         vmware_set_capabilities();
 214 }
 215
 216 static u8 vmware_select_hypercall(void)
 217 {
 218         int eax, ebx, ecx, edx;
 219
 220         cpuid(CPUID_VMWARE_FEATURES_LEAF, &eax, &ebx, &ecx, &edx);
 221         return (ecx & (CPUID_VMWARE_FEATURES_ECX_VMMCALL |
 222                        CPUID_VMWARE_FEATURES_ECX_VMCALL));
 223 }
 224
 225 /*
 226  * While checking the dmi string information, just checking the product
 227  * serial key should be enough, as this will always have a VMware
 228  * specific string when running under VMware hypervisor.
 229  * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
 230  * intentionally defaults to 0.
 231  */
 232 static uint32_t __init vmware_platform(void)
 233 {
 234         if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
 235                 unsigned int eax;
 236                 unsigned int hyper_vendor_id[3];
 237
 238                 cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
 239                       &hyper_vendor_id[1], &hyper_vendor_id[2]);
 240                 if (!memcmp(hyper_vendor_id, "VMwareVMware", 12)) {
 241                         if (eax >= CPUID_VMWARE_FEATURES_LEAF)
 242                                 vmware_hypercall_mode =
 243                                         vmware_select_hypercall();
 244
 245                         pr_info("hypercall mode: 0x%02x\n",
 246                                 (unsigned int) vmware_hypercall_mode);
 247
 248                         return CPUID_VMWARE_INFO_LEAF;
 249                 }
 250         } else if (dmi_available && dmi_name_in_serial("VMware") &&
 251                    __vmware_platform())
 252                 return 1;
 253
 254         return 0;
 255 }
 256
 257 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
 258 static bool __init vmware_legacy_x2apic_available(void)
 259 {
 260         uint32_t eax, ebx, ecx, edx;
 261         VMWARE_CMD(GETVCPU_INFO, eax, ebx, ecx, edx);
 262         return (eax & (1 << VMWARE_CMD_VCPU_RESERVED)) == 0 &&
 263                (eax & (1 << VMWARE_CMD_LEGACY_X2APIC)) != 0;
 264 }
 265
 266 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
 267         .name                   = "VMware",
 268         .detect                 = vmware_platform,
 269         .type                   = X86_HYPER_VMWARE,
 270         .init.init_platform     = vmware_platform_setup,
 271         .init.x2apic_available  = vmware_legacy_x2apic_available,
 272 };