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 <linux/cpu.h>
29 #include <linux/efi.h>
30 #include <linux/reboot.h>
31 #include <linux/static_call.h>
32 #include <linux/sched/cputime.h>
33 #include <asm/div64.h>
34 #include <asm/x86_init.h>
35 #include <asm/hypervisor.h>
36 #include <asm/cpuid/api.h>
37 #include <asm/timer.h>
38 #include <asm/apic.h>
39 #include <asm/vmware.h>
40 #include <asm/svm.h>
41
42 #undef pr_fmt
43 #define pr_fmt(fmt) "vmware: " fmt
44
45 #define CPUID_VMWARE_INFO_LEAF 0x40000000
46 #define CPUID_VMWARE_FEATURES_LEAF 0x40000010
47
48 #define GETVCPU_INFO_LEGACY_X2APIC BIT(3)
49 #define GETVCPU_INFO_VCPU_RESERVED BIT(31)
50
51 #define STEALCLOCK_NOT_AVAILABLE (-1)
52 #define STEALCLOCK_DISABLED 0
53 #define STEALCLOCK_ENABLED 1
54
55 struct vmware_steal_time {
56 union {
57 u64 clock; /* stolen time counter in units of vtsc */
58 struct {
59 /* only for little-endian */
60 u32 clock_low;
61 u32 clock_high;
62 };
63 };
64 u64 reserved[7];
65 };
66
67 static unsigned long vmware_tsc_khz __ro_after_init;
68 static u8 vmware_hypercall_mode __ro_after_init;
69
vmware_hypercall_slow(unsigned long cmd,unsigned long in1,unsigned long in3,unsigned long in4,unsigned long in5,u32 * out1,u32 * out2,u32 * out3,u32 * out4,u32 * out5)70 unsigned long vmware_hypercall_slow(unsigned long cmd,
71 unsigned long in1, unsigned long in3,
72 unsigned long in4, unsigned long in5,
73 u32 *out1, u32 *out2, u32 *out3,
74 u32 *out4, u32 *out5)
75 {
76 unsigned long out0, rbx, rcx, rdx, rsi, rdi;
77
78 switch (vmware_hypercall_mode) {
79 case CPUID_VMWARE_FEATURES_ECX_VMCALL:
80 asm_inline volatile ("vmcall"
81 : "=a" (out0), "=b" (rbx), "=c" (rcx),
82 "=d" (rdx), "=S" (rsi), "=D" (rdi)
83 : "a" (VMWARE_HYPERVISOR_MAGIC),
84 "b" (in1),
85 "c" (cmd),
86 "d" (in3),
87 "S" (in4),
88 "D" (in5)
89 : "cc", "memory");
90 break;
91 case CPUID_VMWARE_FEATURES_ECX_VMMCALL:
92 asm_inline volatile ("vmmcall"
93 : "=a" (out0), "=b" (rbx), "=c" (rcx),
94 "=d" (rdx), "=S" (rsi), "=D" (rdi)
95 : "a" (VMWARE_HYPERVISOR_MAGIC),
96 "b" (in1),
97 "c" (cmd),
98 "d" (in3),
99 "S" (in4),
100 "D" (in5)
101 : "cc", "memory");
102 break;
103 default:
104 asm_inline volatile ("movw %[port], %%dx; inl (%%dx), %%eax"
105 : "=a" (out0), "=b" (rbx), "=c" (rcx),
106 "=d" (rdx), "=S" (rsi), "=D" (rdi)
107 : [port] "i" (VMWARE_HYPERVISOR_PORT),
108 "a" (VMWARE_HYPERVISOR_MAGIC),
109 "b" (in1),
110 "c" (cmd),
111 "d" (in3),
112 "S" (in4),
113 "D" (in5)
114 : "cc", "memory");
115 break;
116 }
117
118 if (out1)
119 *out1 = rbx;
120 if (out2)
121 *out2 = rcx;
122 if (out3)
123 *out3 = rdx;
124 if (out4)
125 *out4 = rsi;
126 if (out5)
127 *out5 = rdi;
128
129 return out0;
130 }
131
__vmware_platform(void)132 static inline int __vmware_platform(void)
133 {
134 u32 eax, ebx, ecx;
135
136 eax = vmware_hypercall3(VMWARE_CMD_GETVERSION, 0, &ebx, &ecx);
137 return eax != UINT_MAX && ebx == VMWARE_HYPERVISOR_MAGIC;
138 }
139
vmware_get_tsc_khz(void)140 static unsigned long vmware_get_tsc_khz(void)
141 {
142 return vmware_tsc_khz;
143 }
144
145 #ifdef CONFIG_PARAVIRT
146 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
147 static bool vmw_sched_clock __initdata = true;
148 static DEFINE_PER_CPU_DECRYPTED(struct vmware_steal_time, vmw_steal_time) __aligned(64);
149 static bool has_steal_clock;
150 static bool steal_acc __initdata = true; /* steal time accounting */
151
setup_vmw_sched_clock(char * s)152 static __init int setup_vmw_sched_clock(char *s)
153 {
154 vmw_sched_clock = false;
155 return 0;
156 }
157 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
158
parse_no_stealacc(char * arg)159 static __init int parse_no_stealacc(char *arg)
160 {
161 steal_acc = false;
162 return 0;
163 }
164 early_param("no-steal-acc", parse_no_stealacc);
165
vmware_sched_clock(void)166 static noinstr u64 vmware_sched_clock(void)
167 {
168 unsigned long long ns;
169
170 ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
171 vmware_cyc2ns.cyc2ns_shift);
172 ns -= vmware_cyc2ns.cyc2ns_offset;
173 return ns;
174 }
175
vmware_cyc2ns_setup(void)176 static void __init vmware_cyc2ns_setup(void)
177 {
178 struct cyc2ns_data *d = &vmware_cyc2ns;
179 unsigned long long tsc_now = rdtsc();
180
181 clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
182 vmware_tsc_khz, NSEC_PER_MSEC, 0);
183 d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
184 d->cyc2ns_shift);
185
186 pr_info("using clock offset of %llu ns\n", d->cyc2ns_offset);
187 }
188
vmware_cmd_stealclock(u32 addr_hi,u32 addr_lo)189 static int vmware_cmd_stealclock(u32 addr_hi, u32 addr_lo)
190 {
191 u32 info;
192
193 return vmware_hypercall5(VMWARE_CMD_STEALCLOCK, 0, 0, addr_hi, addr_lo,
194 &info);
195 }
196
stealclock_enable(phys_addr_t pa)197 static bool stealclock_enable(phys_addr_t pa)
198 {
199 return vmware_cmd_stealclock(upper_32_bits(pa),
200 lower_32_bits(pa)) == STEALCLOCK_ENABLED;
201 }
202
__stealclock_disable(void)203 static int __stealclock_disable(void)
204 {
205 return vmware_cmd_stealclock(0, 1);
206 }
207
stealclock_disable(void)208 static void stealclock_disable(void)
209 {
210 __stealclock_disable();
211 }
212
vmware_is_stealclock_available(void)213 static bool vmware_is_stealclock_available(void)
214 {
215 return __stealclock_disable() != STEALCLOCK_NOT_AVAILABLE;
216 }
217
218 /**
219 * vmware_steal_clock() - read the per-cpu steal clock
220 * @cpu: the cpu number whose steal clock we want to read
221 *
222 * The function reads the steal clock if we are on a 64-bit system, otherwise
223 * reads it in parts, checking that the high part didn't change in the
224 * meantime.
225 *
226 * Return:
227 * The steal clock reading in ns.
228 */
vmware_steal_clock(int cpu)229 static u64 vmware_steal_clock(int cpu)
230 {
231 struct vmware_steal_time *steal = &per_cpu(vmw_steal_time, cpu);
232 u64 clock;
233
234 if (IS_ENABLED(CONFIG_64BIT))
235 clock = READ_ONCE(steal->clock);
236 else {
237 u32 initial_high, low, high;
238
239 do {
240 initial_high = READ_ONCE(steal->clock_high);
241 /* Do not reorder initial_high and high readings */
242 virt_rmb();
243 low = READ_ONCE(steal->clock_low);
244 /* Keep low reading in between */
245 virt_rmb();
246 high = READ_ONCE(steal->clock_high);
247 } while (initial_high != high);
248
249 clock = ((u64)high << 32) | low;
250 }
251
252 return mul_u64_u32_shr(clock, vmware_cyc2ns.cyc2ns_mul,
253 vmware_cyc2ns.cyc2ns_shift);
254 }
255
vmware_register_steal_time(void)256 static void vmware_register_steal_time(void)
257 {
258 int cpu = smp_processor_id();
259 struct vmware_steal_time *st = &per_cpu(vmw_steal_time, cpu);
260
261 if (!has_steal_clock)
262 return;
263
264 if (!stealclock_enable(slow_virt_to_phys(st))) {
265 has_steal_clock = false;
266 return;
267 }
268
269 pr_info("vmware-stealtime: cpu %d, pa %llx\n",
270 cpu, (unsigned long long) slow_virt_to_phys(st));
271 }
272
vmware_disable_steal_time(void)273 static void vmware_disable_steal_time(void)
274 {
275 if (!has_steal_clock)
276 return;
277
278 stealclock_disable();
279 }
280
vmware_guest_cpu_init(void)281 static void vmware_guest_cpu_init(void)
282 {
283 if (has_steal_clock)
284 vmware_register_steal_time();
285 }
286
vmware_pv_guest_cpu_reboot(void * unused)287 static void vmware_pv_guest_cpu_reboot(void *unused)
288 {
289 vmware_disable_steal_time();
290 }
291
vmware_pv_reboot_notify(struct notifier_block * nb,unsigned long code,void * unused)292 static int vmware_pv_reboot_notify(struct notifier_block *nb,
293 unsigned long code, void *unused)
294 {
295 if (code == SYS_RESTART)
296 on_each_cpu(vmware_pv_guest_cpu_reboot, NULL, 1);
297 return NOTIFY_DONE;
298 }
299
300 static struct notifier_block vmware_pv_reboot_nb = {
301 .notifier_call = vmware_pv_reboot_notify,
302 };
303
304 #ifdef CONFIG_SMP
vmware_smp_prepare_boot_cpu(void)305 static void __init vmware_smp_prepare_boot_cpu(void)
306 {
307 vmware_guest_cpu_init();
308 native_smp_prepare_boot_cpu();
309 }
310
vmware_cpu_online(unsigned int cpu)311 static int vmware_cpu_online(unsigned int cpu)
312 {
313 local_irq_disable();
314 vmware_guest_cpu_init();
315 local_irq_enable();
316 return 0;
317 }
318
vmware_cpu_down_prepare(unsigned int cpu)319 static int vmware_cpu_down_prepare(unsigned int cpu)
320 {
321 local_irq_disable();
322 vmware_disable_steal_time();
323 local_irq_enable();
324 return 0;
325 }
326 #endif
327
activate_jump_labels(void)328 static __init int activate_jump_labels(void)
329 {
330 if (has_steal_clock) {
331 static_branch_inc(¶virt_steal_enabled);
332 if (steal_acc)
333 static_branch_inc(¶virt_steal_rq_enabled);
334 }
335
336 return 0;
337 }
338 arch_initcall(activate_jump_labels);
339
vmware_paravirt_ops_setup(void)340 static void __init vmware_paravirt_ops_setup(void)
341 {
342 pv_info.name = "VMware hypervisor";
343 pv_info.io_delay = false;
344
345 if (vmware_tsc_khz == 0)
346 return;
347
348 vmware_cyc2ns_setup();
349
350 if (vmw_sched_clock)
351 paravirt_set_sched_clock(vmware_sched_clock);
352
353 if (vmware_is_stealclock_available()) {
354 has_steal_clock = true;
355 static_call_update(pv_steal_clock, vmware_steal_clock);
356
357 /* We use reboot notifier only to disable steal clock */
358 register_reboot_notifier(&vmware_pv_reboot_nb);
359
360 #ifdef CONFIG_SMP
361 smp_ops.smp_prepare_boot_cpu =
362 vmware_smp_prepare_boot_cpu;
363 if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
364 "x86/vmware:online",
365 vmware_cpu_online,
366 vmware_cpu_down_prepare) < 0)
367 pr_err("vmware_guest: Failed to install cpu hotplug callbacks\n");
368 #else
369 vmware_guest_cpu_init();
370 #endif
371 }
372 }
373 #else
374 #define vmware_paravirt_ops_setup() do {} while (0)
375 #endif
376
377 /*
378 * VMware hypervisor takes care of exporting a reliable TSC to the guest.
379 * Still, due to timing difference when running on virtual cpus, the TSC can
380 * be marked as unstable in some cases. For example, the TSC sync check at
381 * bootup can fail due to a marginal offset between vcpus' TSCs (though the
382 * TSCs do not drift from each other). Also, the ACPI PM timer clocksource
383 * is not suitable as a watchdog when running on a hypervisor because the
384 * kernel may miss a wrap of the counter if the vcpu is descheduled for a
385 * long time. To skip these checks at runtime we set these capability bits,
386 * so that the kernel could just trust the hypervisor with providing a
387 * reliable virtual TSC that is suitable for timekeeping.
388 */
vmware_set_capabilities(void)389 static void __init vmware_set_capabilities(void)
390 {
391 setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
392 setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
393 if (vmware_tsc_khz)
394 setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
395 if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMCALL)
396 setup_force_cpu_cap(X86_FEATURE_VMCALL);
397 else if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMMCALL)
398 setup_force_cpu_cap(X86_FEATURE_VMW_VMMCALL);
399 }
400
vmware_platform_setup(void)401 static void __init vmware_platform_setup(void)
402 {
403 u32 eax, ebx, ecx;
404 u64 lpj, tsc_khz;
405
406 eax = vmware_hypercall3(VMWARE_CMD_GETHZ, UINT_MAX, &ebx, &ecx);
407
408 if (ebx != UINT_MAX) {
409 lpj = tsc_khz = eax | (((u64)ebx) << 32);
410 do_div(tsc_khz, 1000);
411 WARN_ON(tsc_khz >> 32);
412 pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
413 (unsigned long) tsc_khz / 1000,
414 (unsigned long) tsc_khz % 1000);
415
416 if (!preset_lpj) {
417 do_div(lpj, HZ);
418 preset_lpj = lpj;
419 }
420
421 vmware_tsc_khz = tsc_khz;
422 x86_platform.calibrate_tsc = vmware_get_tsc_khz;
423 x86_platform.calibrate_cpu = vmware_get_tsc_khz;
424
425 #ifdef CONFIG_X86_LOCAL_APIC
426 /* Skip lapic calibration since we know the bus frequency. */
427 lapic_timer_period = ecx / HZ;
428 pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
429 ecx);
430 #endif
431 } else {
432 pr_warn("Failed to get TSC freq from the hypervisor\n");
433 }
434
435 if (cc_platform_has(CC_ATTR_GUEST_SEV_SNP) && !efi_enabled(EFI_BOOT))
436 x86_init.mpparse.find_mptable = mpparse_find_mptable;
437
438 vmware_paravirt_ops_setup();
439
440 #ifdef CONFIG_X86_IO_APIC
441 no_timer_check = 1;
442 #endif
443
444 vmware_set_capabilities();
445 }
446
vmware_select_hypercall(void)447 static u8 __init vmware_select_hypercall(void)
448 {
449 int eax, ebx, ecx, edx;
450
451 cpuid(CPUID_VMWARE_FEATURES_LEAF, &eax, &ebx, &ecx, &edx);
452 return (ecx & (CPUID_VMWARE_FEATURES_ECX_VMMCALL |
453 CPUID_VMWARE_FEATURES_ECX_VMCALL));
454 }
455
456 /*
457 * While checking the dmi string information, just checking the product
458 * serial key should be enough, as this will always have a VMware
459 * specific string when running under VMware hypervisor.
460 * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
461 * intentionally defaults to 0.
462 */
vmware_platform(void)463 static u32 __init vmware_platform(void)
464 {
465 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
466 unsigned int eax;
467 unsigned int hyper_vendor_id[3];
468
469 cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
470 &hyper_vendor_id[1], &hyper_vendor_id[2]);
471 if (!memcmp(hyper_vendor_id, "VMwareVMware", 12)) {
472 if (eax >= CPUID_VMWARE_FEATURES_LEAF)
473 vmware_hypercall_mode =
474 vmware_select_hypercall();
475
476 pr_info("hypercall mode: 0x%02x\n",
477 (unsigned int) vmware_hypercall_mode);
478
479 return CPUID_VMWARE_INFO_LEAF;
480 }
481 } else if (dmi_available && dmi_name_in_serial("VMware") &&
482 __vmware_platform())
483 return 1;
484
485 return 0;
486 }
487
488 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
vmware_legacy_x2apic_available(void)489 static bool __init vmware_legacy_x2apic_available(void)
490 {
491 u32 eax;
492
493 eax = vmware_hypercall1(VMWARE_CMD_GETVCPU_INFO, 0);
494 return !(eax & GETVCPU_INFO_VCPU_RESERVED) &&
495 (eax & GETVCPU_INFO_LEGACY_X2APIC);
496 }
497
498 #ifdef CONFIG_INTEL_TDX_GUEST
499 /*
500 * TDCALL[TDG.VP.VMCALL] uses %rax (arg0) and %rcx (arg2). Therefore,
501 * we remap those registers to %r12 and %r13, respectively.
502 */
vmware_tdx_hypercall(unsigned long cmd,unsigned long in1,unsigned long in3,unsigned long in4,unsigned long in5,u32 * out1,u32 * out2,u32 * out3,u32 * out4,u32 * out5)503 unsigned long vmware_tdx_hypercall(unsigned long cmd,
504 unsigned long in1, unsigned long in3,
505 unsigned long in4, unsigned long in5,
506 u32 *out1, u32 *out2, u32 *out3,
507 u32 *out4, u32 *out5)
508 {
509 struct tdx_module_args args = {};
510
511 if (!hypervisor_is_type(X86_HYPER_VMWARE)) {
512 pr_warn_once("Incorrect usage\n");
513 return ULONG_MAX;
514 }
515
516 if (cmd & ~VMWARE_CMD_MASK) {
517 pr_warn_once("Out of range command %lx\n", cmd);
518 return ULONG_MAX;
519 }
520
521 args.rbx = in1;
522 args.rdx = in3;
523 args.rsi = in4;
524 args.rdi = in5;
525 args.r10 = VMWARE_TDX_VENDOR_LEAF;
526 args.r11 = VMWARE_TDX_HCALL_FUNC;
527 args.r12 = VMWARE_HYPERVISOR_MAGIC;
528 args.r13 = cmd;
529 /* CPL */
530 args.r15 = 0;
531
532 __tdx_hypercall(&args);
533
534 if (out1)
535 *out1 = args.rbx;
536 if (out2)
537 *out2 = args.r13;
538 if (out3)
539 *out3 = args.rdx;
540 if (out4)
541 *out4 = args.rsi;
542 if (out5)
543 *out5 = args.rdi;
544
545 return args.r12;
546 }
547 EXPORT_SYMBOL_GPL(vmware_tdx_hypercall);
548 #endif
549
550 #ifdef CONFIG_AMD_MEM_ENCRYPT
vmware_sev_es_hcall_prepare(struct ghcb * ghcb,struct pt_regs * regs)551 static void vmware_sev_es_hcall_prepare(struct ghcb *ghcb,
552 struct pt_regs *regs)
553 {
554 /* Copy VMWARE specific Hypercall parameters to the GHCB */
555 ghcb_set_rip(ghcb, regs->ip);
556 ghcb_set_rbx(ghcb, regs->bx);
557 ghcb_set_rcx(ghcb, regs->cx);
558 ghcb_set_rdx(ghcb, regs->dx);
559 ghcb_set_rsi(ghcb, regs->si);
560 ghcb_set_rdi(ghcb, regs->di);
561 ghcb_set_rbp(ghcb, regs->bp);
562 }
563
vmware_sev_es_hcall_finish(struct ghcb * ghcb,struct pt_regs * regs)564 static bool vmware_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
565 {
566 if (!(ghcb_rbx_is_valid(ghcb) &&
567 ghcb_rcx_is_valid(ghcb) &&
568 ghcb_rdx_is_valid(ghcb) &&
569 ghcb_rsi_is_valid(ghcb) &&
570 ghcb_rdi_is_valid(ghcb) &&
571 ghcb_rbp_is_valid(ghcb)))
572 return false;
573
574 regs->bx = ghcb_get_rbx(ghcb);
575 regs->cx = ghcb_get_rcx(ghcb);
576 regs->dx = ghcb_get_rdx(ghcb);
577 regs->si = ghcb_get_rsi(ghcb);
578 regs->di = ghcb_get_rdi(ghcb);
579 regs->bp = ghcb_get_rbp(ghcb);
580
581 return true;
582 }
583 #endif
584
585 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
586 .name = "VMware",
587 .detect = vmware_platform,
588 .type = X86_HYPER_VMWARE,
589 .init.init_platform = vmware_platform_setup,
590 .init.x2apic_available = vmware_legacy_x2apic_available,
591 #ifdef CONFIG_AMD_MEM_ENCRYPT
592 .runtime.sev_es_hcall_prepare = vmware_sev_es_hcall_prepare,
593 .runtime.sev_es_hcall_finish = vmware_sev_es_hcall_finish,
594 #endif
595 };
596