xref: /linux/arch/x86/kernel/cpu/vmware.c (revision ca64d84e93762f4e587e040a44ad9f6089afc777)
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 
37 #undef pr_fmt
38 #define pr_fmt(fmt)	"vmware: " fmt
39 
40 #define CPUID_VMWARE_INFO_LEAF               0x40000000
41 #define CPUID_VMWARE_FEATURES_LEAF           0x40000010
42 #define CPUID_VMWARE_FEATURES_ECX_VMMCALL    BIT(0)
43 #define CPUID_VMWARE_FEATURES_ECX_VMCALL     BIT(1)
44 
45 #define VMWARE_HYPERVISOR_MAGIC	0x564D5868
46 
47 #define VMWARE_CMD_GETVERSION    10
48 #define VMWARE_CMD_GETHZ         45
49 #define VMWARE_CMD_GETVCPU_INFO  68
50 #define VMWARE_CMD_LEGACY_X2APIC  3
51 #define VMWARE_CMD_VCPU_RESERVED 31
52 #define VMWARE_CMD_STEALCLOCK    91
53 
54 #define STEALCLOCK_NOT_AVAILABLE (-1)
55 #define STEALCLOCK_DISABLED        0
56 #define STEALCLOCK_ENABLED         1
57 
58 #define VMWARE_PORT(cmd, eax, ebx, ecx, edx)				\
59 	__asm__("inl (%%dx), %%eax" :					\
60 		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :		\
61 		"a"(VMWARE_HYPERVISOR_MAGIC),				\
62 		"c"(VMWARE_CMD_##cmd),					\
63 		"d"(VMWARE_HYPERVISOR_PORT), "b"(UINT_MAX) :		\
64 		"memory")
65 
66 #define VMWARE_VMCALL(cmd, eax, ebx, ecx, edx)				\
67 	__asm__("vmcall" :						\
68 		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :		\
69 		"a"(VMWARE_HYPERVISOR_MAGIC),				\
70 		"c"(VMWARE_CMD_##cmd),					\
71 		"d"(0), "b"(UINT_MAX) :					\
72 		"memory")
73 
74 #define VMWARE_VMMCALL(cmd, eax, ebx, ecx, edx)                         \
75 	__asm__("vmmcall" :						\
76 		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :		\
77 		"a"(VMWARE_HYPERVISOR_MAGIC),				\
78 		"c"(VMWARE_CMD_##cmd),					\
79 		"d"(0), "b"(UINT_MAX) :					\
80 		"memory")
81 
82 #define VMWARE_CMD(cmd, eax, ebx, ecx, edx) do {		\
83 	switch (vmware_hypercall_mode) {			\
84 	case CPUID_VMWARE_FEATURES_ECX_VMCALL:			\
85 		VMWARE_VMCALL(cmd, eax, ebx, ecx, edx);		\
86 		break;						\
87 	case CPUID_VMWARE_FEATURES_ECX_VMMCALL:			\
88 		VMWARE_VMMCALL(cmd, eax, ebx, ecx, edx);	\
89 		break;						\
90 	default:						\
91 		VMWARE_PORT(cmd, eax, ebx, ecx, edx);		\
92 		break;						\
93 	}							\
94 	} while (0)
95 
96 struct vmware_steal_time {
97 	union {
98 		uint64_t clock;	/* stolen time counter in units of vtsc */
99 		struct {
100 			/* only for little-endian */
101 			uint32_t clock_low;
102 			uint32_t clock_high;
103 		};
104 	};
105 	uint64_t reserved[7];
106 };
107 
108 static unsigned long vmware_tsc_khz __ro_after_init;
109 static u8 vmware_hypercall_mode     __ro_after_init;
110 
111 static inline int __vmware_platform(void)
112 {
113 	uint32_t eax, ebx, ecx, edx;
114 	VMWARE_CMD(GETVERSION, eax, ebx, ecx, edx);
115 	return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
116 }
117 
118 static unsigned long vmware_get_tsc_khz(void)
119 {
120 	return vmware_tsc_khz;
121 }
122 
123 #ifdef CONFIG_PARAVIRT
124 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
125 static bool vmw_sched_clock __initdata = true;
126 static DEFINE_PER_CPU_DECRYPTED(struct vmware_steal_time, vmw_steal_time) __aligned(64);
127 static bool has_steal_clock;
128 static bool steal_acc __initdata = true; /* steal time accounting */
129 
130 static __init int setup_vmw_sched_clock(char *s)
131 {
132 	vmw_sched_clock = false;
133 	return 0;
134 }
135 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
136 
137 static __init int parse_no_stealacc(char *arg)
138 {
139 	steal_acc = false;
140 	return 0;
141 }
142 early_param("no-steal-acc", parse_no_stealacc);
143 
144 static unsigned long long notrace vmware_sched_clock(void)
145 {
146 	unsigned long long ns;
147 
148 	ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
149 			     vmware_cyc2ns.cyc2ns_shift);
150 	ns -= vmware_cyc2ns.cyc2ns_offset;
151 	return ns;
152 }
153 
154 static void __init vmware_cyc2ns_setup(void)
155 {
156 	struct cyc2ns_data *d = &vmware_cyc2ns;
157 	unsigned long long tsc_now = rdtsc();
158 
159 	clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
160 			       vmware_tsc_khz, NSEC_PER_MSEC, 0);
161 	d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
162 					   d->cyc2ns_shift);
163 
164 	pr_info("using clock offset of %llu ns\n", d->cyc2ns_offset);
165 }
166 
167 static int vmware_cmd_stealclock(uint32_t arg1, uint32_t arg2)
168 {
169 	uint32_t result, info;
170 
171 	asm volatile (VMWARE_HYPERCALL :
172 		"=a"(result),
173 		"=c"(info) :
174 		"a"(VMWARE_HYPERVISOR_MAGIC),
175 		"b"(0),
176 		"c"(VMWARE_CMD_STEALCLOCK),
177 		"d"(0),
178 		"S"(arg1),
179 		"D"(arg2) :
180 		"memory");
181 	return result;
182 }
183 
184 static bool stealclock_enable(phys_addr_t pa)
185 {
186 	return vmware_cmd_stealclock(upper_32_bits(pa),
187 				     lower_32_bits(pa)) == STEALCLOCK_ENABLED;
188 }
189 
190 static int __stealclock_disable(void)
191 {
192 	return vmware_cmd_stealclock(0, 1);
193 }
194 
195 static void stealclock_disable(void)
196 {
197 	__stealclock_disable();
198 }
199 
200 static bool vmware_is_stealclock_available(void)
201 {
202 	return __stealclock_disable() != STEALCLOCK_NOT_AVAILABLE;
203 }
204 
205 /**
206  * vmware_steal_clock() - read the per-cpu steal clock
207  * @cpu:            the cpu number whose steal clock we want to read
208  *
209  * The function reads the steal clock if we are on a 64-bit system, otherwise
210  * reads it in parts, checking that the high part didn't change in the
211  * meantime.
212  *
213  * Return:
214  *      The steal clock reading in ns.
215  */
216 static uint64_t vmware_steal_clock(int cpu)
217 {
218 	struct vmware_steal_time *steal = &per_cpu(vmw_steal_time, cpu);
219 	uint64_t clock;
220 
221 	if (IS_ENABLED(CONFIG_64BIT))
222 		clock = READ_ONCE(steal->clock);
223 	else {
224 		uint32_t initial_high, low, high;
225 
226 		do {
227 			initial_high = READ_ONCE(steal->clock_high);
228 			/* Do not reorder initial_high and high readings */
229 			virt_rmb();
230 			low = READ_ONCE(steal->clock_low);
231 			/* Keep low reading in between */
232 			virt_rmb();
233 			high = READ_ONCE(steal->clock_high);
234 		} while (initial_high != high);
235 
236 		clock = ((uint64_t)high << 32) | low;
237 	}
238 
239 	return mul_u64_u32_shr(clock, vmware_cyc2ns.cyc2ns_mul,
240 			     vmware_cyc2ns.cyc2ns_shift);
241 }
242 
243 static void vmware_register_steal_time(void)
244 {
245 	int cpu = smp_processor_id();
246 	struct vmware_steal_time *st = &per_cpu(vmw_steal_time, cpu);
247 
248 	if (!has_steal_clock)
249 		return;
250 
251 	if (!stealclock_enable(slow_virt_to_phys(st))) {
252 		has_steal_clock = false;
253 		return;
254 	}
255 
256 	pr_info("vmware-stealtime: cpu %d, pa %llx\n",
257 		cpu, (unsigned long long) slow_virt_to_phys(st));
258 }
259 
260 static void vmware_disable_steal_time(void)
261 {
262 	if (!has_steal_clock)
263 		return;
264 
265 	stealclock_disable();
266 }
267 
268 static void vmware_guest_cpu_init(void)
269 {
270 	if (has_steal_clock)
271 		vmware_register_steal_time();
272 }
273 
274 static void vmware_pv_guest_cpu_reboot(void *unused)
275 {
276 	vmware_disable_steal_time();
277 }
278 
279 static int vmware_pv_reboot_notify(struct notifier_block *nb,
280 				unsigned long code, void *unused)
281 {
282 	if (code == SYS_RESTART)
283 		on_each_cpu(vmware_pv_guest_cpu_reboot, NULL, 1);
284 	return NOTIFY_DONE;
285 }
286 
287 static struct notifier_block vmware_pv_reboot_nb = {
288 	.notifier_call = vmware_pv_reboot_notify,
289 };
290 
291 #ifdef CONFIG_SMP
292 static void __init vmware_smp_prepare_boot_cpu(void)
293 {
294 	vmware_guest_cpu_init();
295 	native_smp_prepare_boot_cpu();
296 }
297 
298 static int vmware_cpu_online(unsigned int cpu)
299 {
300 	local_irq_disable();
301 	vmware_guest_cpu_init();
302 	local_irq_enable();
303 	return 0;
304 }
305 
306 static int vmware_cpu_down_prepare(unsigned int cpu)
307 {
308 	local_irq_disable();
309 	vmware_disable_steal_time();
310 	local_irq_enable();
311 	return 0;
312 }
313 #endif
314 
315 static __init int activate_jump_labels(void)
316 {
317 	if (has_steal_clock) {
318 		static_key_slow_inc(&paravirt_steal_enabled);
319 		if (steal_acc)
320 			static_key_slow_inc(&paravirt_steal_rq_enabled);
321 	}
322 
323 	return 0;
324 }
325 arch_initcall(activate_jump_labels);
326 
327 static void __init vmware_paravirt_ops_setup(void)
328 {
329 	pv_info.name = "VMware hypervisor";
330 	pv_ops.cpu.io_delay = paravirt_nop;
331 
332 	if (vmware_tsc_khz == 0)
333 		return;
334 
335 	vmware_cyc2ns_setup();
336 
337 	if (vmw_sched_clock)
338 		pv_ops.time.sched_clock = vmware_sched_clock;
339 
340 	if (vmware_is_stealclock_available()) {
341 		has_steal_clock = true;
342 		pv_ops.time.steal_clock = vmware_steal_clock;
343 
344 		/* We use reboot notifier only to disable steal clock */
345 		register_reboot_notifier(&vmware_pv_reboot_nb);
346 
347 #ifdef CONFIG_SMP
348 		smp_ops.smp_prepare_boot_cpu =
349 			vmware_smp_prepare_boot_cpu;
350 		if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
351 					      "x86/vmware:online",
352 					      vmware_cpu_online,
353 					      vmware_cpu_down_prepare) < 0)
354 			pr_err("vmware_guest: Failed to install cpu hotplug callbacks\n");
355 #else
356 		vmware_guest_cpu_init();
357 #endif
358 	}
359 }
360 #else
361 #define vmware_paravirt_ops_setup() do {} while (0)
362 #endif
363 
364 /*
365  * VMware hypervisor takes care of exporting a reliable TSC to the guest.
366  * Still, due to timing difference when running on virtual cpus, the TSC can
367  * be marked as unstable in some cases. For example, the TSC sync check at
368  * bootup can fail due to a marginal offset between vcpus' TSCs (though the
369  * TSCs do not drift from each other).  Also, the ACPI PM timer clocksource
370  * is not suitable as a watchdog when running on a hypervisor because the
371  * kernel may miss a wrap of the counter if the vcpu is descheduled for a
372  * long time. To skip these checks at runtime we set these capability bits,
373  * so that the kernel could just trust the hypervisor with providing a
374  * reliable virtual TSC that is suitable for timekeeping.
375  */
376 static void __init vmware_set_capabilities(void)
377 {
378 	setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
379 	setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
380 	if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMCALL)
381 		setup_force_cpu_cap(X86_FEATURE_VMCALL);
382 	else if (vmware_hypercall_mode == CPUID_VMWARE_FEATURES_ECX_VMMCALL)
383 		setup_force_cpu_cap(X86_FEATURE_VMW_VMMCALL);
384 }
385 
386 static void __init vmware_platform_setup(void)
387 {
388 	uint32_t eax, ebx, ecx, edx;
389 	uint64_t lpj, tsc_khz;
390 
391 	VMWARE_CMD(GETHZ, eax, ebx, ecx, edx);
392 
393 	if (ebx != UINT_MAX) {
394 		lpj = tsc_khz = eax | (((uint64_t)ebx) << 32);
395 		do_div(tsc_khz, 1000);
396 		WARN_ON(tsc_khz >> 32);
397 		pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
398 			(unsigned long) tsc_khz / 1000,
399 			(unsigned long) tsc_khz % 1000);
400 
401 		if (!preset_lpj) {
402 			do_div(lpj, HZ);
403 			preset_lpj = lpj;
404 		}
405 
406 		vmware_tsc_khz = tsc_khz;
407 		x86_platform.calibrate_tsc = vmware_get_tsc_khz;
408 		x86_platform.calibrate_cpu = vmware_get_tsc_khz;
409 
410 #ifdef CONFIG_X86_LOCAL_APIC
411 		/* Skip lapic calibration since we know the bus frequency. */
412 		lapic_timer_period = ecx / HZ;
413 		pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
414 			ecx);
415 #endif
416 	} else {
417 		pr_warn("Failed to get TSC freq from the hypervisor\n");
418 	}
419 
420 	vmware_paravirt_ops_setup();
421 
422 #ifdef CONFIG_X86_IO_APIC
423 	no_timer_check = 1;
424 #endif
425 
426 	vmware_set_capabilities();
427 }
428 
429 static u8 __init vmware_select_hypercall(void)
430 {
431 	int eax, ebx, ecx, edx;
432 
433 	cpuid(CPUID_VMWARE_FEATURES_LEAF, &eax, &ebx, &ecx, &edx);
434 	return (ecx & (CPUID_VMWARE_FEATURES_ECX_VMMCALL |
435 		       CPUID_VMWARE_FEATURES_ECX_VMCALL));
436 }
437 
438 /*
439  * While checking the dmi string information, just checking the product
440  * serial key should be enough, as this will always have a VMware
441  * specific string when running under VMware hypervisor.
442  * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
443  * intentionally defaults to 0.
444  */
445 static uint32_t __init vmware_platform(void)
446 {
447 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
448 		unsigned int eax;
449 		unsigned int hyper_vendor_id[3];
450 
451 		cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
452 		      &hyper_vendor_id[1], &hyper_vendor_id[2]);
453 		if (!memcmp(hyper_vendor_id, "VMwareVMware", 12)) {
454 			if (eax >= CPUID_VMWARE_FEATURES_LEAF)
455 				vmware_hypercall_mode =
456 					vmware_select_hypercall();
457 
458 			pr_info("hypercall mode: 0x%02x\n",
459 				(unsigned int) vmware_hypercall_mode);
460 
461 			return CPUID_VMWARE_INFO_LEAF;
462 		}
463 	} else if (dmi_available && dmi_name_in_serial("VMware") &&
464 		   __vmware_platform())
465 		return 1;
466 
467 	return 0;
468 }
469 
470 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
471 static bool __init vmware_legacy_x2apic_available(void)
472 {
473 	uint32_t eax, ebx, ecx, edx;
474 	VMWARE_CMD(GETVCPU_INFO, eax, ebx, ecx, edx);
475 	return (eax & (1 << VMWARE_CMD_VCPU_RESERVED)) == 0 &&
476 	       (eax & (1 << VMWARE_CMD_LEGACY_X2APIC)) != 0;
477 }
478 
479 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
480 	.name			= "VMware",
481 	.detect			= vmware_platform,
482 	.type			= X86_HYPER_VMWARE,
483 	.init.init_platform	= vmware_platform_setup,
484 	.init.x2apic_available	= vmware_legacy_x2apic_available,
485 };
486