xref: /linux/arch/x86/kernel/cpu/vmware.c (revision 44f57d78)
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 <asm/div64.h>
29 #include <asm/x86_init.h>
30 #include <asm/hypervisor.h>
31 #include <asm/timer.h>
32 #include <asm/apic.h>
33 
34 #undef pr_fmt
35 #define pr_fmt(fmt)	"vmware: " fmt
36 
37 #define CPUID_VMWARE_INFO_LEAF	0x40000000
38 #define VMWARE_HYPERVISOR_MAGIC	0x564D5868
39 #define VMWARE_HYPERVISOR_PORT	0x5658
40 
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 
47 #define VMWARE_PORT(cmd, eax, ebx, ecx, edx)				\
48 	__asm__("inl (%%dx)" :						\
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");
54 
55 static unsigned long vmware_tsc_khz __ro_after_init;
56 
57 static inline int __vmware_platform(void)
58 {
59 	uint32_t eax, ebx, ecx, edx;
60 	VMWARE_PORT(GETVERSION, eax, ebx, ecx, edx);
61 	return eax != (uint32_t)-1 && ebx == VMWARE_HYPERVISOR_MAGIC;
62 }
63 
64 static unsigned long vmware_get_tsc_khz(void)
65 {
66 	return vmware_tsc_khz;
67 }
68 
69 #ifdef CONFIG_PARAVIRT
70 static struct cyc2ns_data vmware_cyc2ns __ro_after_init;
71 static int vmw_sched_clock __initdata = 1;
72 
73 static __init int setup_vmw_sched_clock(char *s)
74 {
75 	vmw_sched_clock = 0;
76 	return 0;
77 }
78 early_param("no-vmw-sched-clock", setup_vmw_sched_clock);
79 
80 static unsigned long long notrace vmware_sched_clock(void)
81 {
82 	unsigned long long ns;
83 
84 	ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul,
85 			     vmware_cyc2ns.cyc2ns_shift);
86 	ns -= vmware_cyc2ns.cyc2ns_offset;
87 	return ns;
88 }
89 
90 static void __init vmware_sched_clock_setup(void)
91 {
92 	struct cyc2ns_data *d = &vmware_cyc2ns;
93 	unsigned long long tsc_now = rdtsc();
94 
95 	clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift,
96 			       vmware_tsc_khz, NSEC_PER_MSEC, 0);
97 	d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul,
98 					   d->cyc2ns_shift);
99 
100 	pv_ops.time.sched_clock = vmware_sched_clock;
101 	pr_info("using sched offset of %llu ns\n", d->cyc2ns_offset);
102 }
103 
104 static void __init vmware_paravirt_ops_setup(void)
105 {
106 	pv_info.name = "VMware hypervisor";
107 	pv_ops.cpu.io_delay = paravirt_nop;
108 
109 	if (vmware_tsc_khz && vmw_sched_clock)
110 		vmware_sched_clock_setup();
111 }
112 #else
113 #define vmware_paravirt_ops_setup() do {} while (0)
114 #endif
115 
116 /*
117  * VMware hypervisor takes care of exporting a reliable TSC to the guest.
118  * Still, due to timing difference when running on virtual cpus, the TSC can
119  * be marked as unstable in some cases. For example, the TSC sync check at
120  * bootup can fail due to a marginal offset between vcpus' TSCs (though the
121  * TSCs do not drift from each other).  Also, the ACPI PM timer clocksource
122  * is not suitable as a watchdog when running on a hypervisor because the
123  * kernel may miss a wrap of the counter if the vcpu is descheduled for a
124  * long time. To skip these checks at runtime we set these capability bits,
125  * so that the kernel could just trust the hypervisor with providing a
126  * reliable virtual TSC that is suitable for timekeeping.
127  */
128 static void __init vmware_set_capabilities(void)
129 {
130 	setup_force_cpu_cap(X86_FEATURE_CONSTANT_TSC);
131 	setup_force_cpu_cap(X86_FEATURE_TSC_RELIABLE);
132 }
133 
134 static void __init vmware_platform_setup(void)
135 {
136 	uint32_t eax, ebx, ecx, edx;
137 	uint64_t lpj, tsc_khz;
138 
139 	VMWARE_PORT(GETHZ, eax, ebx, ecx, edx);
140 
141 	if (ebx != UINT_MAX) {
142 		lpj = tsc_khz = eax | (((uint64_t)ebx) << 32);
143 		do_div(tsc_khz, 1000);
144 		WARN_ON(tsc_khz >> 32);
145 		pr_info("TSC freq read from hypervisor : %lu.%03lu MHz\n",
146 			(unsigned long) tsc_khz / 1000,
147 			(unsigned long) tsc_khz % 1000);
148 
149 		if (!preset_lpj) {
150 			do_div(lpj, HZ);
151 			preset_lpj = lpj;
152 		}
153 
154 		vmware_tsc_khz = tsc_khz;
155 		x86_platform.calibrate_tsc = vmware_get_tsc_khz;
156 		x86_platform.calibrate_cpu = vmware_get_tsc_khz;
157 
158 #ifdef CONFIG_X86_LOCAL_APIC
159 		/* Skip lapic calibration since we know the bus frequency. */
160 		lapic_timer_frequency = ecx / HZ;
161 		pr_info("Host bus clock speed read from hypervisor : %u Hz\n",
162 			ecx);
163 #endif
164 	} else {
165 		pr_warn("Failed to get TSC freq from the hypervisor\n");
166 	}
167 
168 	vmware_paravirt_ops_setup();
169 
170 #ifdef CONFIG_X86_IO_APIC
171 	no_timer_check = 1;
172 #endif
173 
174 	vmware_set_capabilities();
175 }
176 
177 /*
178  * While checking the dmi string information, just checking the product
179  * serial key should be enough, as this will always have a VMware
180  * specific string when running under VMware hypervisor.
181  */
182 static uint32_t __init vmware_platform(void)
183 {
184 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
185 		unsigned int eax;
186 		unsigned int hyper_vendor_id[3];
187 
188 		cpuid(CPUID_VMWARE_INFO_LEAF, &eax, &hyper_vendor_id[0],
189 		      &hyper_vendor_id[1], &hyper_vendor_id[2]);
190 		if (!memcmp(hyper_vendor_id, "VMwareVMware", 12))
191 			return CPUID_VMWARE_INFO_LEAF;
192 	} else if (dmi_available && dmi_name_in_serial("VMware") &&
193 		   __vmware_platform())
194 		return 1;
195 
196 	return 0;
197 }
198 
199 /* Checks if hypervisor supports x2apic without VT-D interrupt remapping. */
200 static bool __init vmware_legacy_x2apic_available(void)
201 {
202 	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;
206 }
207 
208 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,
214 };
215