xref: /openbmc/linux/arch/x86/kernel/cpu/vmware.c (revision 11a163f2)
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 <asm/div64.h>
31 #include <asm/x86_init.h>
32 #include <asm/hypervisor.h>
33 #include <asm/timer.h>
34 #include <asm/apic.h>
35 #include <asm/vmware.h>
36 #include <asm/svm.h>
37 
38 #undef pr_fmt
39 #define pr_fmt(fmt)	"vmware: " fmt
40 
41 #define CPUID_VMWARE_INFO_LEAF               0x40000000
42 #define CPUID_VMWARE_FEATURES_LEAF           0x40000010
43 #define CPUID_VMWARE_FEATURES_ECX_VMMCALL    BIT(0)
44 #define CPUID_VMWARE_FEATURES_ECX_VMCALL     BIT(1)
45 
46 #define VMWARE_HYPERVISOR_MAGIC	0x564D5868
47 
48 #define VMWARE_CMD_GETVERSION    10
49 #define VMWARE_CMD_GETHZ         45
50 #define VMWARE_CMD_GETVCPU_INFO  68
51 #define VMWARE_CMD_LEGACY_X2APIC  3
52 #define VMWARE_CMD_VCPU_RESERVED 31
53 #define VMWARE_CMD_STEALCLOCK    91
54 
55 #define STEALCLOCK_NOT_AVAILABLE (-1)
56 #define STEALCLOCK_DISABLED        0
57 #define STEALCLOCK_ENABLED         1
58 
59 #define VMWARE_PORT(cmd, eax, ebx, ecx, edx)				\
60 	__asm__("inl (%%dx), %%eax" :					\
61 		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :		\
62 		"a"(VMWARE_HYPERVISOR_MAGIC),				\
63 		"c"(VMWARE_CMD_##cmd),					\
64 		"d"(VMWARE_HYPERVISOR_PORT), "b"(UINT_MAX) :		\
65 		"memory")
66 
67 #define VMWARE_VMCALL(cmd, eax, ebx, ecx, edx)				\
68 	__asm__("vmcall" :						\
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_VMMCALL(cmd, eax, ebx, ecx, edx)                         \
76 	__asm__("vmmcall" :						\
77 		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :		\
78 		"a"(VMWARE_HYPERVISOR_MAGIC),				\
79 		"c"(VMWARE_CMD_##cmd),					\
80 		"d"(0), "b"(UINT_MAX) :					\
81 		"memory")
82 
83 #define VMWARE_CMD(cmd, eax, ebx, ecx, edx) do {		\
84 	switch (vmware_hypercall_mode) {			\
85 	case CPUID_VMWARE_FEATURES_ECX_VMCALL:			\
86 		VMWARE_VMCALL(cmd, eax, ebx, ecx, edx);		\
87 		break;						\
88 	case CPUID_VMWARE_FEATURES_ECX_VMMCALL:			\
89 		VMWARE_VMMCALL(cmd, eax, ebx, ecx, edx);	\
90 		break;						\
91 	default:						\
92 		VMWARE_PORT(cmd, eax, ebx, ecx, edx);		\
93 		break;						\
94 	}							\
95 	} while (0)
96 
97 struct vmware_steal_time {
98 	union {
99 		uint64_t clock;	/* stolen time counter in units of vtsc */
100 		struct {
101 			/* only for little-endian */
102 			uint32_t clock_low;
103 			uint32_t clock_high;
104 		};
105 	};
106 	uint64_t reserved[7];
107 };
108 
109 static unsigned long vmware_tsc_khz __ro_after_init;
110 static u8 vmware_hypercall_mode     __ro_after_init;
111 
112 static inline int __vmware_platform(void)
113 {
114 	uint32_t eax, ebx, ecx, edx;
115 	VMWARE_CMD(GETVERSION, eax, ebx, ecx, edx);
116 	return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
117 }
118 
119 static unsigned long vmware_get_tsc_khz(void)
120 {
121 	return vmware_tsc_khz;
122 }
123 
124 #ifdef CONFIG_PARAVIRT
125 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
126 static bool vmw_sched_clock __initdata = true;
127 static DEFINE_PER_CPU_DECRYPTED(struct vmware_steal_time, vmw_steal_time) __aligned(64);
128 static bool has_steal_clock;
129 static bool steal_acc __initdata = true; /* steal time accounting */
130 
131 static __init int setup_vmw_sched_clock(char *s)
132 {
133 	vmw_sched_clock = false;
134 	return 0;
135 }
136 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
137 
138 static __init int parse_no_stealacc(char *arg)
139 {
140 	steal_acc = false;
141 	return 0;
142 }
143 early_param("no-steal-acc", parse_no_stealacc);
144 
145 static unsigned long long notrace vmware_sched_clock(void)
146 {
147 	unsigned long long ns;
148 
149 	ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
150 			     vmware_cyc2ns.cyc2ns_shift);
151 	ns -= vmware_cyc2ns.cyc2ns_offset;
152 	return ns;
153 }
154 
155 static void __init vmware_cyc2ns_setup(void)
156 {
157 	struct cyc2ns_data *d = &vmware_cyc2ns;
158 	unsigned long long tsc_now = rdtsc();
159 
160 	clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
161 			       vmware_tsc_khz, NSEC_PER_MSEC, 0);
162 	d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
163 					   d->cyc2ns_shift);
164 
165 	pr_info("using clock offset of %llu ns\n", d->cyc2ns_offset);
166 }
167 
168 static int vmware_cmd_stealclock(uint32_t arg1, uint32_t arg2)
169 {
170 	uint32_t result, info;
171 
172 	asm volatile (VMWARE_HYPERCALL :
173 		"=a"(result),
174 		"=c"(info) :
175 		"a"(VMWARE_HYPERVISOR_MAGIC),
176 		"b"(0),
177 		"c"(VMWARE_CMD_STEALCLOCK),
178 		"d"(0),
179 		"S"(arg1),
180 		"D"(arg2) :
181 		"memory");
182 	return result;
183 }
184 
185 static bool stealclock_enable(phys_addr_t pa)
186 {
187 	return vmware_cmd_stealclock(upper_32_bits(pa),
188 				     lower_32_bits(pa)) == STEALCLOCK_ENABLED;
189 }
190 
191 static int __stealclock_disable(void)
192 {
193 	return vmware_cmd_stealclock(0, 1);
194 }
195 
196 static void stealclock_disable(void)
197 {
198 	__stealclock_disable();
199 }
200 
201 static bool vmware_is_stealclock_available(void)
202 {
203 	return __stealclock_disable() != STEALCLOCK_NOT_AVAILABLE;
204 }
205 
206 /**
207  * vmware_steal_clock() - read the per-cpu steal clock
208  * @cpu:            the cpu number whose steal clock we want to read
209  *
210  * The function reads the steal clock if we are on a 64-bit system, otherwise
211  * reads it in parts, checking that the high part didn't change in the
212  * meantime.
213  *
214  * Return:
215  *      The steal clock reading in ns.
216  */
217 static uint64_t vmware_steal_clock(int cpu)
218 {
219 	struct vmware_steal_time *steal = &per_cpu(vmw_steal_time, cpu);
220 	uint64_t clock;
221 
222 	if (IS_ENABLED(CONFIG_64BIT))
223 		clock = READ_ONCE(steal->clock);
224 	else {
225 		uint32_t initial_high, low, high;
226 
227 		do {
228 			initial_high = READ_ONCE(steal->clock_high);
229 			/* Do not reorder initial_high and high readings */
230 			virt_rmb();
231 			low = READ_ONCE(steal->clock_low);
232 			/* Keep low reading in between */
233 			virt_rmb();
234 			high = READ_ONCE(steal->clock_high);
235 		} while (initial_high != high);
236 
237 		clock = ((uint64_t)high << 32) | low;
238 	}
239 
240 	return mul_u64_u32_shr(clock, vmware_cyc2ns.cyc2ns_mul,
241 			     vmware_cyc2ns.cyc2ns_shift);
242 }
243 
244 static void vmware_register_steal_time(void)
245 {
246 	int cpu = smp_processor_id();
247 	struct vmware_steal_time *st = &per_cpu(vmw_steal_time, cpu);
248 
249 	if (!has_steal_clock)
250 		return;
251 
252 	if (!stealclock_enable(slow_virt_to_phys(st))) {
253 		has_steal_clock = false;
254 		return;
255 	}
256 
257 	pr_info("vmware-stealtime: cpu %d, pa %llx\n",
258 		cpu, (unsigned long long) slow_virt_to_phys(st));
259 }
260 
261 static void vmware_disable_steal_time(void)
262 {
263 	if (!has_steal_clock)
264 		return;
265 
266 	stealclock_disable();
267 }
268 
269 static void vmware_guest_cpu_init(void)
270 {
271 	if (has_steal_clock)
272 		vmware_register_steal_time();
273 }
274 
275 static void vmware_pv_guest_cpu_reboot(void *unused)
276 {
277 	vmware_disable_steal_time();
278 }
279 
280 static int vmware_pv_reboot_notify(struct notifier_block *nb,
281 				unsigned long code, void *unused)
282 {
283 	if (code == SYS_RESTART)
284 		on_each_cpu(vmware_pv_guest_cpu_reboot, NULL, 1);
285 	return NOTIFY_DONE;
286 }
287 
288 static struct notifier_block vmware_pv_reboot_nb = {
289 	.notifier_call = vmware_pv_reboot_notify,
290 };
291 
292 #ifdef CONFIG_SMP
293 static void __init vmware_smp_prepare_boot_cpu(void)
294 {
295 	vmware_guest_cpu_init();
296 	native_smp_prepare_boot_cpu();
297 }
298 
299 static int vmware_cpu_online(unsigned int cpu)
300 {
301 	local_irq_disable();
302 	vmware_guest_cpu_init();
303 	local_irq_enable();
304 	return 0;
305 }
306 
307 static int vmware_cpu_down_prepare(unsigned int cpu)
308 {
309 	local_irq_disable();
310 	vmware_disable_steal_time();
311 	local_irq_enable();
312 	return 0;
313 }
314 #endif
315 
316 static __init int activate_jump_labels(void)
317 {
318 	if (has_steal_clock) {
319 		static_key_slow_inc(&paravirt_steal_enabled);
320 		if (steal_acc)
321 			static_key_slow_inc(&paravirt_steal_rq_enabled);
322 	}
323 
324 	return 0;
325 }
326 arch_initcall(activate_jump_labels);
327 
328 static void __init vmware_paravirt_ops_setup(void)
329 {
330 	pv_info.name = "VMware hypervisor";
331 	pv_ops.cpu.io_delay = paravirt_nop;
332 
333 	if (vmware_tsc_khz == 0)
334 		return;
335 
336 	vmware_cyc2ns_setup();
337 
338 	if (vmw_sched_clock)
339 		pv_ops.time.sched_clock = vmware_sched_clock;
340 
341 	if (vmware_is_stealclock_available()) {
342 		has_steal_clock = true;
343 		pv_ops.time.steal_clock = vmware_steal_clock;
344 
345 		/* We use reboot notifier only to disable steal clock */
346 		register_reboot_notifier(&vmware_pv_reboot_nb);
347 
348 #ifdef CONFIG_SMP
349 		smp_ops.smp_prepare_boot_cpu =
350 			vmware_smp_prepare_boot_cpu;
351 		if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
352 					      "x86/vmware:online",
353 					      vmware_cpu_online,
354 					      vmware_cpu_down_prepare) < 0)
355 			pr_err("vmware_guest: Failed to install cpu hotplug callbacks\n");
356 #else
357 		vmware_guest_cpu_init();
358 #endif
359 	}
360 }
361 #else
362 #define vmware_paravirt_ops_setup() do {} while (0)
363 #endif
364 
365 /*
366  * VMware hypervisor takes care of exporting a reliable TSC to the guest.
367  * Still, due to timing difference when running on virtual cpus, the TSC can
368  * be marked as unstable in some cases. For example, the TSC sync check at
369  * bootup can fail due to a marginal offset between vcpus' TSCs (though the
370  * TSCs do not drift from each other).  Also, the ACPI PM timer clocksource
371  * is not suitable as a watchdog when running on a hypervisor because the
372  * kernel may miss a wrap of the counter if the vcpu is descheduled for a
373  * long time. To skip these checks at runtime we set these capability bits,
374  * so that the kernel could just trust the hypervisor with providing a
375  * reliable virtual TSC that is suitable for timekeeping.
376  */
377 static void __init vmware_set_capabilities(void)
378 {
379 	setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
380 	setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
381 	if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMCALL)
382 		setup_force_cpu_cap(X86_FEATURE_VMCALL);
383 	else if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMMCALL)
384 		setup_force_cpu_cap(X86_FEATURE_VMW_VMMCALL);
385 }
386 
387 static void __init vmware_platform_setup(void)
388 {
389 	uint32_t eax, ebx, ecx, edx;
390 	uint64_t lpj, tsc_khz;
391 
392 	VMWARE_CMD(GETHZ, eax, ebx, ecx, edx);
393 
394 	if (ebx != UINT_MAX) {
395 		lpj = tsc_khz = eax | (((uint64_t)ebx) << 32);
396 		do_div(tsc_khz, 1000);
397 		WARN_ON(tsc_khz >> 32);
398 		pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
399 			(unsigned long) tsc_khz / 1000,
400 			(unsigned long) tsc_khz % 1000);
401 
402 		if (!preset_lpj) {
403 			do_div(lpj, HZ);
404 			preset_lpj = lpj;
405 		}
406 
407 		vmware_tsc_khz = tsc_khz;
408 		x86_platform.calibrate_tsc = vmware_get_tsc_khz;
409 		x86_platform.calibrate_cpu = vmware_get_tsc_khz;
410 
411 #ifdef CONFIG_X86_LOCAL_APIC
412 		/* Skip lapic calibration since we know the bus frequency. */
413 		lapic_timer_period = ecx / HZ;
414 		pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
415 			ecx);
416 #endif
417 	} else {
418 		pr_warn("Failed to get TSC freq from the hypervisor\n");
419 	}
420 
421 	vmware_paravirt_ops_setup();
422 
423 #ifdef CONFIG_X86_IO_APIC
424 	no_timer_check = 1;
425 #endif
426 
427 	vmware_set_capabilities();
428 }
429 
430 static u8 __init vmware_select_hypercall(void)
431 {
432 	int eax, ebx, ecx, edx;
433 
434 	cpuid(CPUID_VMWARE_FEATURES_LEAF, &eax, &ebx, &ecx, &edx);
435 	return (ecx & (CPUID_VMWARE_FEATURES_ECX_VMMCALL |
436 		       CPUID_VMWARE_FEATURES_ECX_VMCALL));
437 }
438 
439 /*
440  * While checking the dmi string information, just checking the product
441  * serial key should be enough, as this will always have a VMware
442  * specific string when running under VMware hypervisor.
443  * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
444  * intentionally defaults to 0.
445  */
446 static uint32_t __init vmware_platform(void)
447 {
448 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
449 		unsigned int eax;
450 		unsigned int hyper_vendor_id[3];
451 
452 		cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
453 		      &hyper_vendor_id[1], &hyper_vendor_id[2]);
454 		if (!memcmp(hyper_vendor_id, "VMwareVMware", 12)) {
455 			if (eax >= CPUID_VMWARE_FEATURES_LEAF)
456 				vmware_hypercall_mode =
457 					vmware_select_hypercall();
458 
459 			pr_info("hypercall mode: 0x%02x\n",
460 				(unsigned int) vmware_hypercall_mode);
461 
462 			return CPUID_VMWARE_INFO_LEAF;
463 		}
464 	} else if (dmi_available && dmi_name_in_serial("VMware") &&
465 		   __vmware_platform())
466 		return 1;
467 
468 	return 0;
469 }
470 
471 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
472 static bool __init vmware_legacy_x2apic_available(void)
473 {
474 	uint32_t eax, ebx, ecx, edx;
475 	VMWARE_CMD(GETVCPU_INFO, eax, ebx, ecx, edx);
476 	return (eax & (1 << VMWARE_CMD_VCPU_RESERVED)) == 0 &&
477 	       (eax & (1 << VMWARE_CMD_LEGACY_X2APIC)) != 0;
478 }
479 
480 #ifdef CONFIG_AMD_MEM_ENCRYPT
481 static void vmware_sev_es_hcall_prepare(struct ghcb *ghcb,
482 					struct pt_regs *regs)
483 {
484 	/* Copy VMWARE specific Hypercall parameters to the GHCB */
485 	ghcb_set_rip(ghcb, regs->ip);
486 	ghcb_set_rbx(ghcb, regs->bx);
487 	ghcb_set_rcx(ghcb, regs->cx);
488 	ghcb_set_rdx(ghcb, regs->dx);
489 	ghcb_set_rsi(ghcb, regs->si);
490 	ghcb_set_rdi(ghcb, regs->di);
491 	ghcb_set_rbp(ghcb, regs->bp);
492 }
493 
494 static bool vmware_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
495 {
496 	if (!(ghcb_rbx_is_valid(ghcb) &&
497 	      ghcb_rcx_is_valid(ghcb) &&
498 	      ghcb_rdx_is_valid(ghcb) &&
499 	      ghcb_rsi_is_valid(ghcb) &&
500 	      ghcb_rdi_is_valid(ghcb) &&
501 	      ghcb_rbp_is_valid(ghcb)))
502 		return false;
503 
504 	regs->bx = ghcb->save.rbx;
505 	regs->cx = ghcb->save.rcx;
506 	regs->dx = ghcb->save.rdx;
507 	regs->si = ghcb->save.rsi;
508 	regs->di = ghcb->save.rdi;
509 	regs->bp = ghcb->save.rbp;
510 
511 	return true;
512 }
513 #endif
514 
515 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
516 	.name				= "VMware",
517 	.detect				= vmware_platform,
518 	.type				= X86_HYPER_VMWARE,
519 	.init.init_platform		= vmware_platform_setup,
520 	.init.x2apic_available		= vmware_legacy_x2apic_available,
521 #ifdef CONFIG_AMD_MEM_ENCRYPT
522 	.runtime.sev_es_hcall_prepare	= vmware_sev_es_hcall_prepare,
523 	.runtime.sev_es_hcall_finish	= vmware_sev_es_hcall_finish,
524 #endif
525 };
526