xref: /linux/arch/x86/hyperv/hv_init.c (revision 9a379e77033f02c4a071891afdf0f0a01eff8ccb)
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/apic.h>
22 #include <asm/desc.h>
23 #include <asm/hypervisor.h>
24 #include <asm/hyperv.h>
25 #include <asm/mshyperv.h>
26 #include <linux/version.h>
27 #include <linux/vmalloc.h>
28 #include <linux/mm.h>
29 #include <linux/clockchips.h>
30 #include <linux/hyperv.h>
31 #include <linux/slab.h>
32 #include <linux/cpuhotplug.h>
33 
34 #ifdef CONFIG_HYPERV_TSCPAGE
35 
36 static struct ms_hyperv_tsc_page *tsc_pg;
37 
38 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
39 {
40 	return tsc_pg;
41 }
42 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
43 
44 static u64 read_hv_clock_tsc(struct clocksource *arg)
45 {
46 	u64 current_tick = hv_read_tsc_page(tsc_pg);
47 
48 	if (current_tick == U64_MAX)
49 		rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
50 
51 	return current_tick;
52 }
53 
54 static struct clocksource hyperv_cs_tsc = {
55 		.name		= "hyperv_clocksource_tsc_page",
56 		.rating		= 400,
57 		.read		= read_hv_clock_tsc,
58 		.mask		= CLOCKSOURCE_MASK(64),
59 		.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
60 };
61 #endif
62 
63 static u64 read_hv_clock_msr(struct clocksource *arg)
64 {
65 	u64 current_tick;
66 	/*
67 	 * Read the partition counter to get the current tick count. This count
68 	 * is set to 0 when the partition is created and is incremented in
69 	 * 100 nanosecond units.
70 	 */
71 	rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
72 	return current_tick;
73 }
74 
75 static struct clocksource hyperv_cs_msr = {
76 	.name		= "hyperv_clocksource_msr",
77 	.rating		= 400,
78 	.read		= read_hv_clock_msr,
79 	.mask		= CLOCKSOURCE_MASK(64),
80 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
81 };
82 
83 void *hv_hypercall_pg;
84 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
85 struct clocksource *hyperv_cs;
86 EXPORT_SYMBOL_GPL(hyperv_cs);
87 
88 u32 *hv_vp_index;
89 EXPORT_SYMBOL_GPL(hv_vp_index);
90 
91 u32 hv_max_vp_index;
92 
93 static int hv_cpu_init(unsigned int cpu)
94 {
95 	u64 msr_vp_index;
96 
97 	hv_get_vp_index(msr_vp_index);
98 
99 	hv_vp_index[smp_processor_id()] = msr_vp_index;
100 
101 	if (msr_vp_index > hv_max_vp_index)
102 		hv_max_vp_index = msr_vp_index;
103 
104 	return 0;
105 }
106 
107 static void (*hv_reenlightenment_cb)(void);
108 
109 static void hv_reenlightenment_notify(struct work_struct *dummy)
110 {
111 	struct hv_tsc_emulation_status emu_status;
112 
113 	rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
114 
115 	/* Don't issue the callback if TSC accesses are not emulated */
116 	if (hv_reenlightenment_cb && emu_status.inprogress)
117 		hv_reenlightenment_cb();
118 }
119 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
120 
121 void hyperv_stop_tsc_emulation(void)
122 {
123 	u64 freq;
124 	struct hv_tsc_emulation_status emu_status;
125 
126 	rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
127 	emu_status.inprogress = 0;
128 	wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
129 
130 	rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
131 	tsc_khz = div64_u64(freq, 1000);
132 }
133 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
134 
135 static inline bool hv_reenlightenment_available(void)
136 {
137 	/*
138 	 * Check for required features and priviliges to make TSC frequency
139 	 * change notifications work.
140 	 */
141 	return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
142 		ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
143 		ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
144 }
145 
146 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
147 {
148 	entering_ack_irq();
149 
150 	inc_irq_stat(irq_hv_reenlightenment_count);
151 
152 	schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
153 
154 	exiting_irq();
155 }
156 
157 void set_hv_tscchange_cb(void (*cb)(void))
158 {
159 	struct hv_reenlightenment_control re_ctrl = {
160 		.vector = HYPERV_REENLIGHTENMENT_VECTOR,
161 		.enabled = 1,
162 		.target_vp = hv_vp_index[smp_processor_id()]
163 	};
164 	struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
165 
166 	if (!hv_reenlightenment_available()) {
167 		pr_warn("Hyper-V: reenlightenment support is unavailable\n");
168 		return;
169 	}
170 
171 	hv_reenlightenment_cb = cb;
172 
173 	/* Make sure callback is registered before we write to MSRs */
174 	wmb();
175 
176 	wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
177 	wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
178 }
179 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
180 
181 void clear_hv_tscchange_cb(void)
182 {
183 	struct hv_reenlightenment_control re_ctrl;
184 
185 	if (!hv_reenlightenment_available())
186 		return;
187 
188 	rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
189 	re_ctrl.enabled = 0;
190 	wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
191 
192 	hv_reenlightenment_cb = NULL;
193 }
194 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
195 
196 static int hv_cpu_die(unsigned int cpu)
197 {
198 	struct hv_reenlightenment_control re_ctrl;
199 	unsigned int new_cpu;
200 
201 	if (hv_reenlightenment_cb == NULL)
202 		return 0;
203 
204 	rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
205 	if (re_ctrl.target_vp == hv_vp_index[cpu]) {
206 		/* Reassign to some other online CPU */
207 		new_cpu = cpumask_any_but(cpu_online_mask, cpu);
208 
209 		re_ctrl.target_vp = hv_vp_index[new_cpu];
210 		wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
211 	}
212 
213 	return 0;
214 }
215 
216 /*
217  * This function is to be invoked early in the boot sequence after the
218  * hypervisor has been detected.
219  *
220  * 1. Setup the hypercall page.
221  * 2. Register Hyper-V specific clocksource.
222  */
223 void hyperv_init(void)
224 {
225 	u64 guest_id, required_msrs;
226 	union hv_x64_msr_hypercall_contents hypercall_msr;
227 
228 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
229 		return;
230 
231 	/* Absolutely required MSRs */
232 	required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
233 		HV_X64_MSR_VP_INDEX_AVAILABLE;
234 
235 	if ((ms_hyperv.features & required_msrs) != required_msrs)
236 		return;
237 
238 	/* Allocate percpu VP index */
239 	hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
240 				    GFP_KERNEL);
241 	if (!hv_vp_index)
242 		return;
243 
244 	if (cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
245 			      hv_cpu_init, hv_cpu_die) < 0)
246 		goto free_vp_index;
247 
248 	/*
249 	 * Setup the hypercall page and enable hypercalls.
250 	 * 1. Register the guest ID
251 	 * 2. Enable the hypercall and register the hypercall page
252 	 */
253 	guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
254 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
255 
256 	hv_hypercall_pg  = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
257 	if (hv_hypercall_pg == NULL) {
258 		wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
259 		goto free_vp_index;
260 	}
261 
262 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
263 	hypercall_msr.enable = 1;
264 	hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
265 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
266 
267 	hyper_alloc_mmu();
268 
269 	/*
270 	 * Register Hyper-V specific clocksource.
271 	 */
272 #ifdef CONFIG_HYPERV_TSCPAGE
273 	if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
274 		union hv_x64_msr_hypercall_contents tsc_msr;
275 
276 		tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
277 		if (!tsc_pg)
278 			goto register_msr_cs;
279 
280 		hyperv_cs = &hyperv_cs_tsc;
281 
282 		rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
283 
284 		tsc_msr.enable = 1;
285 		tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
286 
287 		wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
288 
289 		hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
290 
291 		clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
292 		return;
293 	}
294 register_msr_cs:
295 #endif
296 	/*
297 	 * For 32 bit guests just use the MSR based mechanism for reading
298 	 * the partition counter.
299 	 */
300 
301 	hyperv_cs = &hyperv_cs_msr;
302 	if (ms_hyperv.features & HV_X64_MSR_TIME_REF_COUNT_AVAILABLE)
303 		clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
304 
305 	return;
306 
307 free_vp_index:
308 	kfree(hv_vp_index);
309 	hv_vp_index = NULL;
310 }
311 
312 /*
313  * This routine is called before kexec/kdump, it does the required cleanup.
314  */
315 void hyperv_cleanup(void)
316 {
317 	union hv_x64_msr_hypercall_contents hypercall_msr;
318 
319 	/* Reset our OS id */
320 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
321 
322 	/* Reset the hypercall page */
323 	hypercall_msr.as_uint64 = 0;
324 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
325 
326 	/* Reset the TSC page */
327 	hypercall_msr.as_uint64 = 0;
328 	wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
329 }
330 EXPORT_SYMBOL_GPL(hyperv_cleanup);
331 
332 void hyperv_report_panic(struct pt_regs *regs, long err)
333 {
334 	static bool panic_reported;
335 	u64 guest_id;
336 
337 	/*
338 	 * We prefer to report panic on 'die' chain as we have proper
339 	 * registers to report, but if we miss it (e.g. on BUG()) we need
340 	 * to report it on 'panic'.
341 	 */
342 	if (panic_reported)
343 		return;
344 	panic_reported = true;
345 
346 	rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
347 
348 	wrmsrl(HV_X64_MSR_CRASH_P0, err);
349 	wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
350 	wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
351 	wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
352 	wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
353 
354 	/*
355 	 * Let Hyper-V know there is crash data available
356 	 */
357 	wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
358 }
359 EXPORT_SYMBOL_GPL(hyperv_report_panic);
360 
361 bool hv_is_hyperv_initialized(void)
362 {
363 	union hv_x64_msr_hypercall_contents hypercall_msr;
364 
365 	/*
366 	 * Ensure that we're really on Hyper-V, and not a KVM or Xen
367 	 * emulation of Hyper-V
368 	 */
369 	if (x86_hyper_type != X86_HYPER_MS_HYPERV)
370 		return false;
371 
372 	/*
373 	 * Verify that earlier initialization succeeded by checking
374 	 * that the hypercall page is setup
375 	 */
376 	hypercall_msr.as_uint64 = 0;
377 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
378 
379 	return hypercall_msr.enable;
380 }
381 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
382