xref: /linux/arch/x86/hyperv/hv_init.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * X86 specific Hyper-V initialization code.
3  *
4  * Copyright (C) 2016, Microsoft, Inc.
5  *
6  * Author : K. Y. Srinivasan <kys@microsoft.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms of the GNU General Public License version 2 as published
10  * by the Free Software Foundation.
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  */
19 
20 #include <linux/types.h>
21 #include <asm/hypervisor.h>
22 #include <asm/hyperv.h>
23 #include <asm/mshyperv.h>
24 #include <linux/version.h>
25 #include <linux/vmalloc.h>
26 #include <linux/mm.h>
27 #include <linux/clockchips.h>
28 
29 
30 #ifdef CONFIG_X86_64
31 
32 static struct ms_hyperv_tsc_page *tsc_pg;
33 
34 static u64 read_hv_clock_tsc(struct clocksource *arg)
35 {
36 	u64 current_tick;
37 
38 	if (tsc_pg->tsc_sequence != 0) {
39 		/*
40 		 * Use the tsc page to compute the value.
41 		 */
42 
43 		while (1) {
44 			u64 tmp;
45 			u32 sequence = tsc_pg->tsc_sequence;
46 			u64 cur_tsc;
47 			u64 scale = tsc_pg->tsc_scale;
48 			s64 offset = tsc_pg->tsc_offset;
49 
50 			rdtscll(cur_tsc);
51 			/* current_tick = ((cur_tsc *scale) >> 64) + offset */
52 			asm("mulq %3"
53 				: "=d" (current_tick), "=a" (tmp)
54 				: "a" (cur_tsc), "r" (scale));
55 
56 			current_tick += offset;
57 			if (tsc_pg->tsc_sequence == sequence)
58 				return current_tick;
59 
60 			if (tsc_pg->tsc_sequence != 0)
61 				continue;
62 			/*
63 			 * Fallback using MSR method.
64 			 */
65 			break;
66 		}
67 	}
68 	rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
69 	return current_tick;
70 }
71 
72 static struct clocksource hyperv_cs_tsc = {
73 		.name		= "hyperv_clocksource_tsc_page",
74 		.rating		= 400,
75 		.read		= read_hv_clock_tsc,
76 		.mask		= CLOCKSOURCE_MASK(64),
77 		.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
78 };
79 #endif
80 
81 static u64 read_hv_clock_msr(struct clocksource *arg)
82 {
83 	u64 current_tick;
84 	/*
85 	 * Read the partition counter to get the current tick count. This count
86 	 * is set to 0 when the partition is created and is incremented in
87 	 * 100 nanosecond units.
88 	 */
89 	rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
90 	return current_tick;
91 }
92 
93 static struct clocksource hyperv_cs_msr = {
94 	.name		= "hyperv_clocksource_msr",
95 	.rating		= 400,
96 	.read		= read_hv_clock_msr,
97 	.mask		= CLOCKSOURCE_MASK(64),
98 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
99 };
100 
101 static void *hypercall_pg;
102 struct clocksource *hyperv_cs;
103 EXPORT_SYMBOL_GPL(hyperv_cs);
104 
105 /*
106  * This function is to be invoked early in the boot sequence after the
107  * hypervisor has been detected.
108  *
109  * 1. Setup the hypercall page.
110  * 2. Register Hyper-V specific clocksource.
111  */
112 void hyperv_init(void)
113 {
114 	u64 guest_id;
115 	union hv_x64_msr_hypercall_contents hypercall_msr;
116 
117 	if (x86_hyper != &x86_hyper_ms_hyperv)
118 		return;
119 
120 	/*
121 	 * Setup the hypercall page and enable hypercalls.
122 	 * 1. Register the guest ID
123 	 * 2. Enable the hypercall and register the hypercall page
124 	 */
125 	guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
126 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
127 
128 	hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
129 	if (hypercall_pg == NULL) {
130 		wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
131 		return;
132 	}
133 
134 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
135 	hypercall_msr.enable = 1;
136 	hypercall_msr.guest_physical_address = vmalloc_to_pfn(hypercall_pg);
137 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
138 
139 	/*
140 	 * Register Hyper-V specific clocksource.
141 	 */
142 #ifdef CONFIG_X86_64
143 	if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
144 		union hv_x64_msr_hypercall_contents tsc_msr;
145 
146 		tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
147 		if (!tsc_pg)
148 			goto register_msr_cs;
149 
150 		hyperv_cs = &hyperv_cs_tsc;
151 
152 		rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
153 
154 		tsc_msr.enable = 1;
155 		tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
156 
157 		wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
158 		clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
159 		return;
160 	}
161 register_msr_cs:
162 #endif
163 	/*
164 	 * For 32 bit guests just use the MSR based mechanism for reading
165 	 * the partition counter.
166 	 */
167 
168 	hyperv_cs = &hyperv_cs_msr;
169 	if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE)
170 		clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
171 }
172 
173 /*
174  * This routine is called before kexec/kdump, it does the required cleanup.
175  */
176 void hyperv_cleanup(void)
177 {
178 	union hv_x64_msr_hypercall_contents hypercall_msr;
179 
180 	/* Reset our OS id */
181 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
182 
183 	/* Reset the hypercall page */
184 	hypercall_msr.as_uint64 = 0;
185 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
186 
187 	/* Reset the TSC page */
188 	hypercall_msr.as_uint64 = 0;
189 	wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
190 }
191 EXPORT_SYMBOL_GPL(hyperv_cleanup);
192 
193 /*
194  * hv_do_hypercall- Invoke the specified hypercall
195  */
196 u64 hv_do_hypercall(u64 control, void *input, void *output)
197 {
198 	u64 input_address = (input) ? virt_to_phys(input) : 0;
199 	u64 output_address = (output) ? virt_to_phys(output) : 0;
200 #ifdef CONFIG_X86_64
201 	u64 hv_status = 0;
202 
203 	if (!hypercall_pg)
204 		return (u64)ULLONG_MAX;
205 
206 	__asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
207 	__asm__ __volatile__("call *%3" : "=a" (hv_status) :
208 			     "c" (control), "d" (input_address),
209 			     "m" (hypercall_pg));
210 
211 	return hv_status;
212 
213 #else
214 
215 	u32 control_hi = control >> 32;
216 	u32 control_lo = control & 0xFFFFFFFF;
217 	u32 hv_status_hi = 1;
218 	u32 hv_status_lo = 1;
219 	u32 input_address_hi = input_address >> 32;
220 	u32 input_address_lo = input_address & 0xFFFFFFFF;
221 	u32 output_address_hi = output_address >> 32;
222 	u32 output_address_lo = output_address & 0xFFFFFFFF;
223 
224 	if (!hypercall_pg)
225 		return (u64)ULLONG_MAX;
226 
227 	__asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
228 			      "=a"(hv_status_lo) : "d" (control_hi),
229 			      "a" (control_lo), "b" (input_address_hi),
230 			      "c" (input_address_lo), "D"(output_address_hi),
231 			      "S"(output_address_lo), "m" (hypercall_pg));
232 
233 	return hv_status_lo | ((u64)hv_status_hi << 32);
234 #endif /* !x86_64 */
235 }
236 EXPORT_SYMBOL_GPL(hv_do_hypercall);
237 
238 void hyperv_report_panic(struct pt_regs *regs)
239 {
240 	static bool panic_reported;
241 
242 	/*
243 	 * We prefer to report panic on 'die' chain as we have proper
244 	 * registers to report, but if we miss it (e.g. on BUG()) we need
245 	 * to report it on 'panic'.
246 	 */
247 	if (panic_reported)
248 		return;
249 	panic_reported = true;
250 
251 	wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip);
252 	wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax);
253 	wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx);
254 	wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx);
255 	wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx);
256 
257 	/*
258 	 * Let Hyper-V know there is crash data available
259 	 */
260 	wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
261 }
262 EXPORT_SYMBOL_GPL(hyperv_report_panic);
263 
264 bool hv_is_hypercall_page_setup(void)
265 {
266 	union hv_x64_msr_hypercall_contents hypercall_msr;
267 
268 	/* Check if the hypercall page is setup */
269 	hypercall_msr.as_uint64 = 0;
270 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
271 
272 	if (!hypercall_msr.enable)
273 		return false;
274 
275 	return true;
276 }
277 EXPORT_SYMBOL_GPL(hv_is_hypercall_page_setup);
278