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-tlfs.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
hv_get_tsc_page(void)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
read_hv_clock_tsc(struct clocksource * arg)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
read_hv_clock_msr(struct clocksource * arg)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 struct hv_vp_assist_page **hv_vp_assist_page;
92 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
93
94 void __percpu **hyperv_pcpu_input_arg;
95 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
96
97 u32 hv_max_vp_index;
98
hv_cpu_init(unsigned int cpu)99 static int hv_cpu_init(unsigned int cpu)
100 {
101 u64 msr_vp_index;
102 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
103 void **input_arg;
104
105 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
106 *input_arg = page_address(alloc_page(GFP_KERNEL));
107
108 hv_get_vp_index(msr_vp_index);
109
110 hv_vp_index[smp_processor_id()] = msr_vp_index;
111
112 if (msr_vp_index > hv_max_vp_index)
113 hv_max_vp_index = msr_vp_index;
114
115 if (!hv_vp_assist_page)
116 return 0;
117
118 if (!*hvp)
119 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
120
121 if (*hvp) {
122 u64 val;
123
124 val = vmalloc_to_pfn(*hvp);
125 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
126 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
127
128 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
129 }
130
131 return 0;
132 }
133
134 static void (*hv_reenlightenment_cb)(void);
135
hv_reenlightenment_notify(struct work_struct * dummy)136 static void hv_reenlightenment_notify(struct work_struct *dummy)
137 {
138 struct hv_tsc_emulation_status emu_status;
139
140 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
141
142 /* Don't issue the callback if TSC accesses are not emulated */
143 if (hv_reenlightenment_cb && emu_status.inprogress)
144 hv_reenlightenment_cb();
145 }
146 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
147
hyperv_stop_tsc_emulation(void)148 void hyperv_stop_tsc_emulation(void)
149 {
150 u64 freq;
151 struct hv_tsc_emulation_status emu_status;
152
153 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
154 emu_status.inprogress = 0;
155 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
156
157 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
158 tsc_khz = div64_u64(freq, 1000);
159 }
160 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
161
hv_reenlightenment_available(void)162 static inline bool hv_reenlightenment_available(void)
163 {
164 /*
165 * Check for required features and priviliges to make TSC frequency
166 * change notifications work.
167 */
168 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
169 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
170 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
171 }
172
hyperv_reenlightenment_intr(struct pt_regs * regs)173 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
174 {
175 entering_ack_irq();
176
177 inc_irq_stat(irq_hv_reenlightenment_count);
178
179 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
180
181 exiting_irq();
182 }
183
set_hv_tscchange_cb(void (* cb)(void))184 void set_hv_tscchange_cb(void (*cb)(void))
185 {
186 struct hv_reenlightenment_control re_ctrl = {
187 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
188 .enabled = 1,
189 .target_vp = hv_vp_index[smp_processor_id()]
190 };
191 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
192
193 if (!hv_reenlightenment_available()) {
194 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
195 return;
196 }
197
198 hv_reenlightenment_cb = cb;
199
200 /* Make sure callback is registered before we write to MSRs */
201 wmb();
202
203 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
204 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
205 }
206 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
207
clear_hv_tscchange_cb(void)208 void clear_hv_tscchange_cb(void)
209 {
210 struct hv_reenlightenment_control re_ctrl;
211
212 if (!hv_reenlightenment_available())
213 return;
214
215 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
216 re_ctrl.enabled = 0;
217 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
218
219 hv_reenlightenment_cb = NULL;
220 }
221 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
222
hv_cpu_die(unsigned int cpu)223 static int hv_cpu_die(unsigned int cpu)
224 {
225 struct hv_reenlightenment_control re_ctrl;
226 unsigned int new_cpu;
227 unsigned long flags;
228 void **input_arg;
229 void *input_pg = NULL;
230
231 local_irq_save(flags);
232 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
233 input_pg = *input_arg;
234 *input_arg = NULL;
235 local_irq_restore(flags);
236 free_page((unsigned long)input_pg);
237
238 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
239 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
240
241 if (hv_reenlightenment_cb == NULL)
242 return 0;
243
244 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
245 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
246 /* Reassign to some other online CPU */
247 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
248
249 re_ctrl.target_vp = hv_vp_index[new_cpu];
250 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
251 }
252
253 return 0;
254 }
255
256 /*
257 * This function is to be invoked early in the boot sequence after the
258 * hypervisor has been detected.
259 *
260 * 1. Setup the hypercall page.
261 * 2. Register Hyper-V specific clocksource.
262 * 3. Setup Hyper-V specific APIC entry points.
263 */
hyperv_init(void)264 void __init hyperv_init(void)
265 {
266 u64 guest_id, required_msrs;
267 union hv_x64_msr_hypercall_contents hypercall_msr;
268 int cpuhp, i;
269
270 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
271 return;
272
273 /* Absolutely required MSRs */
274 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
275 HV_X64_MSR_VP_INDEX_AVAILABLE;
276
277 if ((ms_hyperv.features & required_msrs) != required_msrs)
278 return;
279
280 /*
281 * Allocate the per-CPU state for the hypercall input arg.
282 * If this allocation fails, we will not be able to setup
283 * (per-CPU) hypercall input page and thus this failure is
284 * fatal on Hyper-V.
285 */
286 hyperv_pcpu_input_arg = alloc_percpu(void *);
287
288 BUG_ON(hyperv_pcpu_input_arg == NULL);
289
290 /* Allocate percpu VP index */
291 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
292 GFP_KERNEL);
293 if (!hv_vp_index)
294 return;
295
296 for (i = 0; i < num_possible_cpus(); i++)
297 hv_vp_index[i] = VP_INVAL;
298
299 hv_vp_assist_page = kcalloc(num_possible_cpus(),
300 sizeof(*hv_vp_assist_page), GFP_KERNEL);
301 if (!hv_vp_assist_page) {
302 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
303 goto free_vp_index;
304 }
305
306 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
307 hv_cpu_init, hv_cpu_die);
308 if (cpuhp < 0)
309 goto free_vp_assist_page;
310
311 /*
312 * Setup the hypercall page and enable hypercalls.
313 * 1. Register the guest ID
314 * 2. Enable the hypercall and register the hypercall page
315 */
316 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
317 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
318
319 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
320 if (hv_hypercall_pg == NULL) {
321 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
322 goto remove_cpuhp_state;
323 }
324
325 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
326 hypercall_msr.enable = 1;
327 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
328 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
329
330 hv_apic_init();
331
332 /*
333 * Register Hyper-V specific clocksource.
334 */
335 #ifdef CONFIG_HYPERV_TSCPAGE
336 if (ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE) {
337 union hv_x64_msr_hypercall_contents tsc_msr;
338
339 tsc_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
340 if (!tsc_pg)
341 goto register_msr_cs;
342
343 hyperv_cs = &hyperv_cs_tsc;
344
345 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
346
347 tsc_msr.enable = 1;
348 tsc_msr.guest_physical_address = vmalloc_to_pfn(tsc_pg);
349
350 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
351
352 hyperv_cs_tsc.archdata.vclock_mode = VCLOCK_HVCLOCK;
353
354 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
355 return;
356 }
357 register_msr_cs:
358 #endif
359 /*
360 * For 32 bit guests just use the MSR based mechanism for reading
361 * the partition counter.
362 */
363
364 hyperv_cs = &hyperv_cs_msr;
365 if (ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE)
366 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
367
368 return;
369
370 remove_cpuhp_state:
371 cpuhp_remove_state(cpuhp);
372 free_vp_assist_page:
373 kfree(hv_vp_assist_page);
374 hv_vp_assist_page = NULL;
375 free_vp_index:
376 kfree(hv_vp_index);
377 hv_vp_index = NULL;
378 }
379
380 /*
381 * This routine is called before kexec/kdump, it does the required cleanup.
382 */
hyperv_cleanup(void)383 void hyperv_cleanup(void)
384 {
385 union hv_x64_msr_hypercall_contents hypercall_msr;
386
387 /* Reset our OS id */
388 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
389
390 /* Reset the hypercall page */
391 hypercall_msr.as_uint64 = 0;
392 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
393
394 /* Reset the TSC page */
395 hypercall_msr.as_uint64 = 0;
396 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
397 }
398 EXPORT_SYMBOL_GPL(hyperv_cleanup);
399
hyperv_report_panic(struct pt_regs * regs,long err)400 void hyperv_report_panic(struct pt_regs *regs, long err)
401 {
402 static bool panic_reported;
403 u64 guest_id;
404
405 /*
406 * We prefer to report panic on 'die' chain as we have proper
407 * registers to report, but if we miss it (e.g. on BUG()) we need
408 * to report it on 'panic'.
409 */
410 if (panic_reported)
411 return;
412 panic_reported = true;
413
414 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
415
416 wrmsrl(HV_X64_MSR_CRASH_P0, err);
417 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
418 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
419 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
420 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
421
422 /*
423 * Let Hyper-V know there is crash data available
424 */
425 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
426 }
427 EXPORT_SYMBOL_GPL(hyperv_report_panic);
428
429 /**
430 * hyperv_report_panic_msg - report panic message to Hyper-V
431 * @pa: physical address of the panic page containing the message
432 * @size: size of the message in the page
433 */
hyperv_report_panic_msg(phys_addr_t pa,size_t size)434 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
435 {
436 /*
437 * P3 to contain the physical address of the panic page & P4 to
438 * contain the size of the panic data in that page. Rest of the
439 * registers are no-op when the NOTIFY_MSG flag is set.
440 */
441 wrmsrl(HV_X64_MSR_CRASH_P0, 0);
442 wrmsrl(HV_X64_MSR_CRASH_P1, 0);
443 wrmsrl(HV_X64_MSR_CRASH_P2, 0);
444 wrmsrl(HV_X64_MSR_CRASH_P3, pa);
445 wrmsrl(HV_X64_MSR_CRASH_P4, size);
446
447 /*
448 * Let Hyper-V know there is crash data available along with
449 * the panic message.
450 */
451 wrmsrl(HV_X64_MSR_CRASH_CTL,
452 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
453 }
454 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
455
hv_is_hyperv_initialized(void)456 bool hv_is_hyperv_initialized(void)
457 {
458 union hv_x64_msr_hypercall_contents hypercall_msr;
459
460 /*
461 * Ensure that we're really on Hyper-V, and not a KVM or Xen
462 * emulation of Hyper-V
463 */
464 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
465 return false;
466
467 /*
468 * Verify that earlier initialization succeeded by checking
469 * that the hypercall page is setup
470 */
471 hypercall_msr.as_uint64 = 0;
472 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
473
474 return hypercall_msr.enable;
475 }
476 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
477
