forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-05-10 37f49e37ab4cb5d0bc4c60eb5c6d4dd57db767bb 
 kernel/arch/x86/kernel/cpu/vmware.c
....@@ -25,39 +25,94 @@
2525 #include <linux/init.h>
2626 #include <linux/export.h>
2727 #include <linux/clocksource.h>
28
+#include <linux/cpu.h>
29
+#include <linux/reboot.h>
2830 #include <asm/div64.h>
2931 #include <asm/x86_init.h>
3032 #include <asm/hypervisor.h>
3133 #include <asm/timer.h>
3234 #include <asm/apic.h>
35
+#include <asm/vmware.h>
36
+#include <asm/svm.h>
3337 
3438 #undef pr_fmt
3539 #define pr_fmt(fmt)	"vmware: " fmt
3640 
37
-#define CPUID_VMWARE_INFO_LEAF	0x40000000
38
-#define VMWARE_HYPERVISOR_MAGIC	0x564D5868
39
-#define VMWARE_HYPERVISOR_PORT	0x5658
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)
4045 
41
-#define VMWARE_PORT_CMD_GETVERSION	10
42
-#define VMWARE_PORT_CMD_GETHZ		45
43
-#define VMWARE_PORT_CMD_GETVCPU_INFO	68
44
-#define VMWARE_PORT_CMD_LEGACY_X2APIC	3
45
-#define VMWARE_PORT_CMD_VCPU_RESERVED	31
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
4658 
4759 #define VMWARE_PORT(cmd, eax, ebx, ecx, edx)				\
4860 	__asm__("inl (%%dx), %%eax" :					\
49
-			"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :	\
50
-			"0"(VMWARE_HYPERVISOR_MAGIC),			\
51
-			"1"(VMWARE_PORT_CMD_##cmd),			\
52
-			"2"(VMWARE_HYPERVISOR_PORT), "3"(UINT_MAX) :	\
53
-			"memory");
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
+};
54108 
55109 static unsigned long vmware_tsc_khz __ro_after_init;
110
+static u8 vmware_hypercall_mode     __ro_after_init;
56111 
57112 static inline int __vmware_platform(void)
58113 {
59114 	uint32_t eax, ebx, ecx, edx;
60
-	VMWARE_PORT(GETVERSION, eax, ebx, ecx, edx);
115
+	VMWARE_CMD(GETVERSION, eax, ebx, ecx, edx);
61116 	return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
62117 }
63118 
....@@ -68,14 +123,24 @@
68123 
69124 #ifdef CONFIG_PARAVIRT
70125 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
71
-static int vmw_sched_clock __initdata = 1;
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 */
72130 
73131 static __init int setup_vmw_sched_clock(char *s)
74132 {
75
-	vmw_sched_clock = 0;
133
+	vmw_sched_clock = false;
76134 	return 0;
77135 }
78136 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);
79144 
80145 static unsigned long long notrace vmware_sched_clock(void)
81146 {
....@@ -87,7 +152,7 @@
87152 	return ns;
88153 }
89154 
90
-static void __init vmware_sched_clock_setup(void)
155
+static void __init vmware_cyc2ns_setup(void)
91156 {
92157 	struct cyc2ns_data *d = &vmware_cyc2ns;
93158 	unsigned long long tsc_now = rdtsc();
....@@ -97,17 +162,201 @@
97162 	d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
98163 					   d->cyc2ns_shift);
99164 
100
-	pv_time_ops.sched_clock = vmware_sched_clock;
101
-	pr_info("using sched offset of %llu ns\n", d->cyc2ns_offset);
165
+	pr_info("using clock offset of %llu ns\n", d->cyc2ns_offset);
102166 }
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);
103327 
104328 static void __init vmware_paravirt_ops_setup(void)
105329 {
106330 	pv_info.name = "VMware hypervisor";
107
-	pv_cpu_ops.io_delay = paravirt_nop;
331
+	pv_ops.cpu.io_delay = paravirt_nop;
108332 
109
-	if (vmware_tsc_khz && vmw_sched_clock)
110
-		vmware_sched_clock_setup();
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
+	}
111360 }
112361 #else
113362 #define vmware_paravirt_ops_setup() do {} while (0)
....@@ -129,6 +378,10 @@
129378 {
130379 	setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
131380 	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);
132385 }
133386 
134387 static void __init vmware_platform_setup(void)
....@@ -136,7 +389,7 @@
136389 	uint32_t eax, ebx, ecx, edx;
137390 	uint64_t lpj, tsc_khz;
138391 
139
-	VMWARE_PORT(GETHZ, eax, ebx, ecx, edx);
392
+	VMWARE_CMD(GETHZ, eax, ebx, ecx, edx);
140393 
141394 	if (ebx != UINT_MAX) {
142395 		lpj = tsc_khz = eax | (((uint64_t)ebx) << 32);
....@@ -157,7 +410,7 @@
157410 
158411 #ifdef CONFIG_X86_LOCAL_APIC
159412 		/* Skip lapic calibration since we know the bus frequency. */
160
-		lapic_timer_frequency = ecx / HZ;
413
+		lapic_timer_period = ecx / HZ;
161414 		pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
162415 			ecx);
163416 #endif
....@@ -174,10 +427,21 @@
174427 	vmware_set_capabilities();
175428 }
176429 
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
+
177439 /*
178440  * While checking the dmi string information, just checking the product
179441  * serial key should be enough, as this will always have a VMware
180442  * specific string when running under VMware hypervisor.
443
+ * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
444
+ * intentionally defaults to 0.
181445  */
182446 static uint32_t __init vmware_platform(void)
183447 {
....@@ -187,8 +451,16 @@
187451 
188452 		cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
189453 		      &hyper_vendor_id[1], &hyper_vendor_id[2]);
190
-		if (!memcmp(hyper_vendor_id, "VMwareVMware", 12))
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
+
191462 			return CPUID_VMWARE_INFO_LEAF;
463
+		}
192464 	} else if (dmi_available && dmi_name_in_serial("VMware") &&
193465 		   __vmware_platform())
194466 		return 1;
....@@ -200,15 +472,54 @@
200472 static bool __init vmware_legacy_x2apic_available(void)
201473 {
202474 	uint32_t eax, ebx, ecx, edx;
203
-	VMWARE_PORT(GETVCPU_INFO, eax, ebx, ecx, edx);
204
-	return (eax & (1 << VMWARE_PORT_CMD_VCPU_RESERVED)) == 0 &&
205
-	       (eax & (1 << VMWARE_PORT_CMD_LEGACY_X2APIC)) != 0;
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;
206478 }
207479 
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
+
208515 const __initconst struct hypervisor_x86 x86_hyper_vmware = {
209
-	.name			= "VMware",
210
-	.detect			= vmware_platform,
211
-	.type			= X86_HYPER_VMWARE,
212
-	.init.init_platform	= vmware_platform_setup,
213
-	.init.x2apic_available	= vmware_legacy_x2apic_available,
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
214525 };