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