xref: /linux/arch/x86/kernel/cpu/vmware.c (revision 57b7b6ac)
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/reboot.h>
30 #include <linux/static_call.h>
31 #include <asm/div64.h>
32 #include <asm/x86_init.h>
33 #include <asm/hypervisor.h>
34 #include <asm/timer.h>
35 #include <asm/apic.h>
36 #include <asm/vmware.h>
37 #include <asm/svm.h>
38 
39 #undef pr_fmt
40 #define pr_fmt(fmt)	"vmware: " fmt
41 
42 #define CPUID_VMWARE_INFO_LEAF               0x40000000
43 #define CPUID_VMWARE_FEATURES_LEAF           0x40000010
44 
45 #define GETVCPU_INFO_LEGACY_X2APIC           BIT(3)
46 #define GETVCPU_INFO_VCPU_RESERVED           BIT(31)
47 
48 #define STEALCLOCK_NOT_AVAILABLE (-1)
49 #define STEALCLOCK_DISABLED        0
50 #define STEALCLOCK_ENABLED         1
51 
52 struct vmware_steal_time {
53 	union {
54 		u64 clock;	/* stolen time counter in units of vtsc */
55 		struct {
56 			/* only for little-endian */
57 			u32 clock_low;
58 			u32 clock_high;
59 		};
60 	};
61 	u64 reserved[7];
62 };
63 
64 static unsigned long vmware_tsc_khz __ro_after_init;
65 static u8 vmware_hypercall_mode     __ro_after_init;
66 
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)67 unsigned long vmware_hypercall_slow(unsigned long cmd,
68 				    unsigned long in1, unsigned long in3,
69 				    unsigned long in4, unsigned long in5,
70 				    u32 *out1, u32 *out2, u32 *out3,
71 				    u32 *out4, u32 *out5)
72 {
73 	unsigned long out0, rbx, rcx, rdx, rsi, rdi;
74 
75 	switch (vmware_hypercall_mode) {
76 	case CPUID_VMWARE_FEATURES_ECX_VMCALL:
77 		asm_inline volatile ("vmcall"
78 				: "=a" (out0), "=b" (rbx), "=c" (rcx),
79 				"=d" (rdx), "=S" (rsi), "=D" (rdi)
80 				: "a" (VMWARE_HYPERVISOR_MAGIC),
81 				"b" (in1),
82 				"c" (cmd),
83 				"d" (in3),
84 				"S" (in4),
85 				"D" (in5)
86 				: "cc", "memory");
87 		break;
88 	case CPUID_VMWARE_FEATURES_ECX_VMMCALL:
89 		asm_inline volatile ("vmmcall"
90 				: "=a" (out0), "=b" (rbx), "=c" (rcx),
91 				"=d" (rdx), "=S" (rsi), "=D" (rdi)
92 				: "a" (VMWARE_HYPERVISOR_MAGIC),
93 				"b" (in1),
94 				"c" (cmd),
95 				"d" (in3),
96 				"S" (in4),
97 				"D" (in5)
98 				: "cc", "memory");
99 		break;
100 	default:
101 		asm_inline volatile ("movw %[port], %%dx; inl (%%dx), %%eax"
102 				: "=a" (out0), "=b" (rbx), "=c" (rcx),
103 				"=d" (rdx), "=S" (rsi), "=D" (rdi)
104 				: [port] "i" (VMWARE_HYPERVISOR_PORT),
105 				"a" (VMWARE_HYPERVISOR_MAGIC),
106 				"b" (in1),
107 				"c" (cmd),
108 				"d" (in3),
109 				"S" (in4),
110 				"D" (in5)
111 				: "cc", "memory");
112 		break;
113 	}
114 
115 	if (out1)
116 		*out1 = rbx;
117 	if (out2)
118 		*out2 = rcx;
119 	if (out3)
120 		*out3 = rdx;
121 	if (out4)
122 		*out4 = rsi;
123 	if (out5)
124 		*out5 = rdi;
125 
126 	return out0;
127 }
128 
__vmware_platform(void)129 static inline int __vmware_platform(void)
130 {
131 	u32 eax, ebx, ecx;
132 
133 	eax = vmware_hypercall3(VMWARE_CMD_GETVERSION, 0, &ebx, &ecx);
134 	return eax != UINT_MAX && ebx == VMWARE_HYPERVISOR_MAGIC;
135 }
136 
vmware_get_tsc_khz(void)137 static unsigned long vmware_get_tsc_khz(void)
138 {
139 	return vmware_tsc_khz;
140 }
141 
142 #ifdef CONFIG_PARAVIRT
143 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
144 static bool vmw_sched_clock __initdata = true;
145 static DEFINE_PER_CPU_DECRYPTED(struct vmware_steal_time, vmw_steal_time) __aligned(64);
146 static bool has_steal_clock;
147 static bool steal_acc __initdata = true; /* steal time accounting */
148 
setup_vmw_sched_clock(char * s)149 static __init int setup_vmw_sched_clock(char *s)
150 {
151 	vmw_sched_clock = false;
152 	return 0;
153 }
154 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
155 
parse_no_stealacc(char * arg)156 static __init int parse_no_stealacc(char *arg)
157 {
158 	steal_acc = false;
159 	return 0;
160 }
161 early_param("no-steal-acc", parse_no_stealacc);
162 
vmware_sched_clock(void)163 static noinstr u64 vmware_sched_clock(void)
164 {
165 	unsigned long long ns;
166 
167 	ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
168 			     vmware_cyc2ns.cyc2ns_shift);
169 	ns -= vmware_cyc2ns.cyc2ns_offset;
170 	return ns;
171 }
172 
vmware_cyc2ns_setup(void)173 static void __init vmware_cyc2ns_setup(void)
174 {
175 	struct cyc2ns_data *d = &vmware_cyc2ns;
176 	unsigned long long tsc_now = rdtsc();
177 
178 	clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
179 			       vmware_tsc_khz, NSEC_PER_MSEC, 0);
180 	d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
181 					   d->cyc2ns_shift);
182 
183 	pr_info("using clock offset of %llu ns\n", d->cyc2ns_offset);
184 }
185 
vmware_cmd_stealclock(u32 addr_hi,u32 addr_lo)186 static int vmware_cmd_stealclock(u32 addr_hi, u32 addr_lo)
187 {
188 	u32 info;
189 
190 	return vmware_hypercall5(VMWARE_CMD_STEALCLOCK, 0, 0, addr_hi, addr_lo,
191 				 &info);
192 }
193 
stealclock_enable(phys_addr_t pa)194 static bool stealclock_enable(phys_addr_t pa)
195 {
196 	return vmware_cmd_stealclock(upper_32_bits(pa),
197 				     lower_32_bits(pa)) == STEALCLOCK_ENABLED;
198 }
199 
__stealclock_disable(void)200 static int __stealclock_disable(void)
201 {
202 	return vmware_cmd_stealclock(0, 1);
203 }
204 
stealclock_disable(void)205 static void stealclock_disable(void)
206 {
207 	__stealclock_disable();
208 }
209 
vmware_is_stealclock_available(void)210 static bool vmware_is_stealclock_available(void)
211 {
212 	return __stealclock_disable() != STEALCLOCK_NOT_AVAILABLE;
213 }
214 
215 /**
216  * vmware_steal_clock() - read the per-cpu steal clock
217  * @cpu:            the cpu number whose steal clock we want to read
218  *
219  * The function reads the steal clock if we are on a 64-bit system, otherwise
220  * reads it in parts, checking that the high part didn't change in the
221  * meantime.
222  *
223  * Return:
224  *      The steal clock reading in ns.
225  */
vmware_steal_clock(int cpu)226 static u64 vmware_steal_clock(int cpu)
227 {
228 	struct vmware_steal_time *steal = &per_cpu(vmw_steal_time, cpu);
229 	u64 clock;
230 
231 	if (IS_ENABLED(CONFIG_64BIT))
232 		clock = READ_ONCE(steal->clock);
233 	else {
234 		u32 initial_high, low, high;
235 
236 		do {
237 			initial_high = READ_ONCE(steal->clock_high);
238 			/* Do not reorder initial_high and high readings */
239 			virt_rmb();
240 			low = READ_ONCE(steal->clock_low);
241 			/* Keep low reading in between */
242 			virt_rmb();
243 			high = READ_ONCE(steal->clock_high);
244 		} while (initial_high != high);
245 
246 		clock = ((u64)high << 32) | low;
247 	}
248 
249 	return mul_u64_u32_shr(clock, vmware_cyc2ns.cyc2ns_mul,
250 			     vmware_cyc2ns.cyc2ns_shift);
251 }
252 
vmware_register_steal_time(void)253 static void vmware_register_steal_time(void)
254 {
255 	int cpu = smp_processor_id();
256 	struct vmware_steal_time *st = &per_cpu(vmw_steal_time, cpu);
257 
258 	if (!has_steal_clock)
259 		return;
260 
261 	if (!stealclock_enable(slow_virt_to_phys(st))) {
262 		has_steal_clock = false;
263 		return;
264 	}
265 
266 	pr_info("vmware-stealtime: cpu %d, pa %llx\n",
267 		cpu, (unsigned long long) slow_virt_to_phys(st));
268 }
269 
vmware_disable_steal_time(void)270 static void vmware_disable_steal_time(void)
271 {
272 	if (!has_steal_clock)
273 		return;
274 
275 	stealclock_disable();
276 }
277 
vmware_guest_cpu_init(void)278 static void vmware_guest_cpu_init(void)
279 {
280 	if (has_steal_clock)
281 		vmware_register_steal_time();
282 }
283 
vmware_pv_guest_cpu_reboot(void * unused)284 static void vmware_pv_guest_cpu_reboot(void *unused)
285 {
286 	vmware_disable_steal_time();
287 }
288 
vmware_pv_reboot_notify(struct notifier_block * nb,unsigned long code,void * unused)289 static int vmware_pv_reboot_notify(struct notifier_block *nb,
290 				unsigned long code, void *unused)
291 {
292 	if (code == SYS_RESTART)
293 		on_each_cpu(vmware_pv_guest_cpu_reboot, NULL, 1);
294 	return NOTIFY_DONE;
295 }
296 
297 static struct notifier_block vmware_pv_reboot_nb = {
298 	.notifier_call = vmware_pv_reboot_notify,
299 };
300 
301 #ifdef CONFIG_SMP
vmware_smp_prepare_boot_cpu(void)302 static void __init vmware_smp_prepare_boot_cpu(void)
303 {
304 	vmware_guest_cpu_init();
305 	native_smp_prepare_boot_cpu();
306 }
307 
vmware_cpu_online(unsigned int cpu)308 static int vmware_cpu_online(unsigned int cpu)
309 {
310 	local_irq_disable();
311 	vmware_guest_cpu_init();
312 	local_irq_enable();
313 	return 0;
314 }
315 
vmware_cpu_down_prepare(unsigned int cpu)316 static int vmware_cpu_down_prepare(unsigned int cpu)
317 {
318 	local_irq_disable();
319 	vmware_disable_steal_time();
320 	local_irq_enable();
321 	return 0;
322 }
323 #endif
324 
activate_jump_labels(void)325 static __init int activate_jump_labels(void)
326 {
327 	if (has_steal_clock) {
328 		static_key_slow_inc(&paravirt_steal_enabled);
329 		if (steal_acc)
330 			static_key_slow_inc(&paravirt_steal_rq_enabled);
331 	}
332 
333 	return 0;
334 }
335 arch_initcall(activate_jump_labels);
336 
vmware_paravirt_ops_setup(void)337 static void __init vmware_paravirt_ops_setup(void)
338 {
339 	pv_info.name = "VMware hypervisor";
340 	pv_ops.cpu.io_delay = paravirt_nop;
341 
342 	if (vmware_tsc_khz == 0)
343 		return;
344 
345 	vmware_cyc2ns_setup();
346 
347 	if (vmw_sched_clock)
348 		paravirt_set_sched_clock(vmware_sched_clock);
349 
350 	if (vmware_is_stealclock_available()) {
351 		has_steal_clock = true;
352 		static_call_update(pv_steal_clock, vmware_steal_clock);
353 
354 		/* We use reboot notifier only to disable steal clock */
355 		register_reboot_notifier(&vmware_pv_reboot_nb);
356 
357 #ifdef CONFIG_SMP
358 		smp_ops.smp_prepare_boot_cpu =
359 			vmware_smp_prepare_boot_cpu;
360 		if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
361 					      "x86/vmware:online",
362 					      vmware_cpu_online,
363 					      vmware_cpu_down_prepare) < 0)
364 			pr_err("vmware_guest: Failed to install cpu hotplug callbacks\n");
365 #else
366 		vmware_guest_cpu_init();
367 #endif
368 	}
369 }
370 #else
371 #define vmware_paravirt_ops_setup() do {} while (0)
372 #endif
373 
374 /*
375  * VMware hypervisor takes care of exporting a reliable TSC to the guest.
376  * Still, due to timing difference when running on virtual cpus, the TSC can
377  * be marked as unstable in some cases. For example, the TSC sync check at
378  * bootup can fail due to a marginal offset between vcpus' TSCs (though the
379  * TSCs do not drift from each other).  Also, the ACPI PM timer clocksource
380  * is not suitable as a watchdog when running on a hypervisor because the
381  * kernel may miss a wrap of the counter if the vcpu is descheduled for a
382  * long time. To skip these checks at runtime we set these capability bits,
383  * so that the kernel could just trust the hypervisor with providing a
384  * reliable virtual TSC that is suitable for timekeeping.
385  */
vmware_set_capabilities(void)386 static void __init vmware_set_capabilities(void)
387 {
388 	setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
389 	setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
390 	if (vmware_tsc_khz)
391 		setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
392 	if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMCALL)
393 		setup_force_cpu_cap(X86_FEATURE_VMCALL);
394 	else if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMMCALL)
395 		setup_force_cpu_cap(X86_FEATURE_VMW_VMMCALL);
396 }
397 
vmware_platform_setup(void)398 static void __init vmware_platform_setup(void)
399 {
400 	u32 eax, ebx, ecx;
401 	u64 lpj, tsc_khz;
402 
403 	eax = vmware_hypercall3(VMWARE_CMD_GETHZ, UINT_MAX, &ebx, &ecx);
404 
405 	if (ebx != UINT_MAX) {
406 		lpj = tsc_khz = eax | (((u64)ebx) << 32);
407 		do_div(tsc_khz, 1000);
408 		WARN_ON(tsc_khz >> 32);
409 		pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
410 			(unsigned long) tsc_khz / 1000,
411 			(unsigned long) tsc_khz % 1000);
412 
413 		if (!preset_lpj) {
414 			do_div(lpj, HZ);
415 			preset_lpj = lpj;
416 		}
417 
418 		vmware_tsc_khz = tsc_khz;
419 		x86_platform.calibrate_tsc = vmware_get_tsc_khz;
420 		x86_platform.calibrate_cpu = vmware_get_tsc_khz;
421 
422 #ifdef CONFIG_X86_LOCAL_APIC
423 		/* Skip lapic calibration since we know the bus frequency. */
424 		lapic_timer_period = ecx / HZ;
425 		pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
426 			ecx);
427 #endif
428 	} else {
429 		pr_warn("Failed to get TSC freq from the hypervisor\n");
430 	}
431 
432 	vmware_paravirt_ops_setup();
433 
434 #ifdef CONFIG_X86_IO_APIC
435 	no_timer_check = 1;
436 #endif
437 
438 	vmware_set_capabilities();
439 }
440 
vmware_select_hypercall(void)441 static u8 __init vmware_select_hypercall(void)
442 {
443 	int eax, ebx, ecx, edx;
444 
445 	cpuid(CPUID_VMWARE_FEATURES_LEAF, &eax, &ebx, &ecx, &edx);
446 	return (ecx & (CPUID_VMWARE_FEATURES_ECX_VMMCALL |
447 		       CPUID_VMWARE_FEATURES_ECX_VMCALL));
448 }
449 
450 /*
451  * While checking the dmi string information, just checking the product
452  * serial key should be enough, as this will always have a VMware
453  * specific string when running under VMware hypervisor.
454  * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
455  * intentionally defaults to 0.
456  */
vmware_platform(void)457 static u32 __init vmware_platform(void)
458 {
459 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
460 		unsigned int eax;
461 		unsigned int hyper_vendor_id[3];
462 
463 		cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
464 		      &hyper_vendor_id[1], &hyper_vendor_id[2]);
465 		if (!memcmp(hyper_vendor_id, "VMwareVMware", 12)) {
466 			if (eax >= CPUID_VMWARE_FEATURES_LEAF)
467 				vmware_hypercall_mode =
468 					vmware_select_hypercall();
469 
470 			pr_info("hypercall mode: 0x%02x\n",
471 				(unsigned int) vmware_hypercall_mode);
472 
473 			return CPUID_VMWARE_INFO_LEAF;
474 		}
475 	} else if (dmi_available && dmi_name_in_serial("VMware") &&
476 		   __vmware_platform())
477 		return 1;
478 
479 	return 0;
480 }
481 
482 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
vmware_legacy_x2apic_available(void)483 static bool __init vmware_legacy_x2apic_available(void)
484 {
485 	u32 eax;
486 
487 	eax = vmware_hypercall1(VMWARE_CMD_GETVCPU_INFO, 0);
488 	return !(eax & GETVCPU_INFO_VCPU_RESERVED) &&
489 		(eax & GETVCPU_INFO_LEGACY_X2APIC);
490 }
491 
492 #ifdef CONFIG_INTEL_TDX_GUEST
493 /*
494  * TDCALL[TDG.VP.VMCALL] uses %rax (arg0) and %rcx (arg2). Therefore,
495  * we remap those registers to %r12 and %r13, respectively.
496  */
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)497 unsigned long vmware_tdx_hypercall(unsigned long cmd,
498 				   unsigned long in1, unsigned long in3,
499 				   unsigned long in4, unsigned long in5,
500 				   u32 *out1, u32 *out2, u32 *out3,
501 				   u32 *out4, u32 *out5)
502 {
503 	struct tdx_module_args args = {};
504 
505 	if (!hypervisor_is_type(X86_HYPER_VMWARE)) {
506 		pr_warn_once("Incorrect usage\n");
507 		return ULONG_MAX;
508 	}
509 
510 	if (cmd & ~VMWARE_CMD_MASK) {
511 		pr_warn_once("Out of range command %lx\n", cmd);
512 		return ULONG_MAX;
513 	}
514 
515 	args.rbx = in1;
516 	args.rdx = in3;
517 	args.rsi = in4;
518 	args.rdi = in5;
519 	args.r10 = VMWARE_TDX_VENDOR_LEAF;
520 	args.r11 = VMWARE_TDX_HCALL_FUNC;
521 	args.r12 = VMWARE_HYPERVISOR_MAGIC;
522 	args.r13 = cmd;
523 	/* CPL */
524 	args.r15 = 0;
525 
526 	__tdx_hypercall(&args);
527 
528 	if (out1)
529 		*out1 = args.rbx;
530 	if (out2)
531 		*out2 = args.r13;
532 	if (out3)
533 		*out3 = args.rdx;
534 	if (out4)
535 		*out4 = args.rsi;
536 	if (out5)
537 		*out5 = args.rdi;
538 
539 	return args.r12;
540 }
541 EXPORT_SYMBOL_GPL(vmware_tdx_hypercall);
542 #endif
543 
544 #ifdef CONFIG_AMD_MEM_ENCRYPT
vmware_sev_es_hcall_prepare(struct ghcb * ghcb,struct pt_regs * regs)545 static void vmware_sev_es_hcall_prepare(struct ghcb *ghcb,
546 					struct pt_regs *regs)
547 {
548 	/* Copy VMWARE specific Hypercall parameters to the GHCB */
549 	ghcb_set_rip(ghcb, regs->ip);
550 	ghcb_set_rbx(ghcb, regs->bx);
551 	ghcb_set_rcx(ghcb, regs->cx);
552 	ghcb_set_rdx(ghcb, regs->dx);
553 	ghcb_set_rsi(ghcb, regs->si);
554 	ghcb_set_rdi(ghcb, regs->di);
555 	ghcb_set_rbp(ghcb, regs->bp);
556 }
557 
vmware_sev_es_hcall_finish(struct ghcb * ghcb,struct pt_regs * regs)558 static bool vmware_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
559 {
560 	if (!(ghcb_rbx_is_valid(ghcb) &&
561 	      ghcb_rcx_is_valid(ghcb) &&
562 	      ghcb_rdx_is_valid(ghcb) &&
563 	      ghcb_rsi_is_valid(ghcb) &&
564 	      ghcb_rdi_is_valid(ghcb) &&
565 	      ghcb_rbp_is_valid(ghcb)))
566 		return false;
567 
568 	regs->bx = ghcb_get_rbx(ghcb);
569 	regs->cx = ghcb_get_rcx(ghcb);
570 	regs->dx = ghcb_get_rdx(ghcb);
571 	regs->si = ghcb_get_rsi(ghcb);
572 	regs->di = ghcb_get_rdi(ghcb);
573 	regs->bp = ghcb_get_rbp(ghcb);
574 
575 	return true;
576 }
577 #endif
578 
579 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
580 	.name				= "VMware",
581 	.detect				= vmware_platform,
582 	.type				= X86_HYPER_VMWARE,
583 	.init.init_platform		= vmware_platform_setup,
584 	.init.x2apic_available		= vmware_legacy_x2apic_available,
585 #ifdef CONFIG_AMD_MEM_ENCRYPT
586 	.runtime.sev_es_hcall_prepare	= vmware_sev_es_hcall_prepare,
587 	.runtime.sev_es_hcall_finish	= vmware_sev_es_hcall_finish,
588 #endif
589 };
590