1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * X86 specific Hyper-V initialization code.
4 *
5 * Copyright (C) 2016, Microsoft, Inc.
6 *
7 * Author : K. Y. Srinivasan <kys@microsoft.com>
8 */
9
10 #include <linux/efi.h>
11 #include <linux/types.h>
12 #include <asm/apic.h>
13 #include <asm/desc.h>
14 #include <asm/hypervisor.h>
15 #include <asm/hyperv-tlfs.h>
16 #include <asm/mshyperv.h>
17 #include <linux/version.h>
18 #include <linux/vmalloc.h>
19 #include <linux/mm.h>
20 #include <linux/hyperv.h>
21 #include <linux/slab.h>
22 #include <linux/cpuhotplug.h>
23 #include <clocksource/hyperv_timer.h>
24
25 void *hv_hypercall_pg;
26 EXPORT_SYMBOL_GPL(hv_hypercall_pg);
27
28 u32 *hv_vp_index;
29 EXPORT_SYMBOL_GPL(hv_vp_index);
30
31 struct hv_vp_assist_page **hv_vp_assist_page;
32 EXPORT_SYMBOL_GPL(hv_vp_assist_page);
33
34 void __percpu **hyperv_pcpu_input_arg;
35 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
36
37 u32 hv_max_vp_index;
38 EXPORT_SYMBOL_GPL(hv_max_vp_index);
39
hv_alloc_hyperv_page(void)40 void *hv_alloc_hyperv_page(void)
41 {
42 BUILD_BUG_ON(PAGE_SIZE != HV_HYP_PAGE_SIZE);
43
44 return (void *)__get_free_page(GFP_KERNEL);
45 }
46 EXPORT_SYMBOL_GPL(hv_alloc_hyperv_page);
47
hv_free_hyperv_page(unsigned long addr)48 void hv_free_hyperv_page(unsigned long addr)
49 {
50 free_page(addr);
51 }
52 EXPORT_SYMBOL_GPL(hv_free_hyperv_page);
53
hv_cpu_init(unsigned int cpu)54 static int hv_cpu_init(unsigned int cpu)
55 {
56 u64 msr_vp_index;
57 struct hv_vp_assist_page **hvp = &hv_vp_assist_page[smp_processor_id()];
58 void **input_arg;
59 struct page *pg;
60
61 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
62 pg = alloc_page(GFP_KERNEL);
63 if (unlikely(!pg))
64 return -ENOMEM;
65 *input_arg = page_address(pg);
66
67 hv_get_vp_index(msr_vp_index);
68
69 hv_vp_index[smp_processor_id()] = msr_vp_index;
70
71 if (msr_vp_index > hv_max_vp_index)
72 hv_max_vp_index = msr_vp_index;
73
74 if (!hv_vp_assist_page)
75 return 0;
76
77 /*
78 * The VP ASSIST PAGE is an "overlay" page (see Hyper-V TLFS's Section
79 * 5.2.1 "GPA Overlay Pages"). Here it must be zeroed out to make sure
80 * we always write the EOI MSR in hv_apic_eoi_write() *after* the
81 * EOI optimization is disabled in hv_cpu_die(), otherwise a CPU may
82 * not be stopped in the case of CPU offlining and the VM will hang.
83 */
84 if (!*hvp) {
85 *hvp = __vmalloc(PAGE_SIZE, GFP_KERNEL | __GFP_ZERO,
86 PAGE_KERNEL);
87 }
88
89 if (*hvp) {
90 u64 val;
91
92 val = vmalloc_to_pfn(*hvp);
93 val = (val << HV_X64_MSR_VP_ASSIST_PAGE_ADDRESS_SHIFT) |
94 HV_X64_MSR_VP_ASSIST_PAGE_ENABLE;
95
96 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, val);
97 }
98
99 return 0;
100 }
101
102 static void (*hv_reenlightenment_cb)(void);
103
hv_reenlightenment_notify(struct work_struct * dummy)104 static void hv_reenlightenment_notify(struct work_struct *dummy)
105 {
106 struct hv_tsc_emulation_status emu_status;
107
108 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
109
110 /* Don't issue the callback if TSC accesses are not emulated */
111 if (hv_reenlightenment_cb && emu_status.inprogress)
112 hv_reenlightenment_cb();
113 }
114 static DECLARE_DELAYED_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
115
hyperv_stop_tsc_emulation(void)116 void hyperv_stop_tsc_emulation(void)
117 {
118 u64 freq;
119 struct hv_tsc_emulation_status emu_status;
120
121 rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
122 emu_status.inprogress = 0;
123 wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
124
125 rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
126 tsc_khz = div64_u64(freq, 1000);
127 }
128 EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
129
hv_reenlightenment_available(void)130 static inline bool hv_reenlightenment_available(void)
131 {
132 /*
133 * Check for required features and priviliges to make TSC frequency
134 * change notifications work.
135 */
136 return ms_hyperv.features & HV_X64_ACCESS_FREQUENCY_MSRS &&
137 ms_hyperv.misc_features & HV_FEATURE_FREQUENCY_MSRS_AVAILABLE &&
138 ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT;
139 }
140
hyperv_reenlightenment_intr(struct pt_regs * regs)141 __visible void __irq_entry hyperv_reenlightenment_intr(struct pt_regs *regs)
142 {
143 entering_ack_irq();
144
145 inc_irq_stat(irq_hv_reenlightenment_count);
146
147 schedule_delayed_work(&hv_reenlightenment_work, HZ/10);
148
149 exiting_irq();
150 }
151
set_hv_tscchange_cb(void (* cb)(void))152 void set_hv_tscchange_cb(void (*cb)(void))
153 {
154 struct hv_reenlightenment_control re_ctrl = {
155 .vector = HYPERV_REENLIGHTENMENT_VECTOR,
156 .enabled = 1,
157 .target_vp = hv_vp_index[smp_processor_id()]
158 };
159 struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
160
161 if (!hv_reenlightenment_available()) {
162 pr_warn("Hyper-V: reenlightenment support is unavailable\n");
163 return;
164 }
165
166 hv_reenlightenment_cb = cb;
167
168 /* Make sure callback is registered before we write to MSRs */
169 wmb();
170
171 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
172 wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
173 }
174 EXPORT_SYMBOL_GPL(set_hv_tscchange_cb);
175
clear_hv_tscchange_cb(void)176 void clear_hv_tscchange_cb(void)
177 {
178 struct hv_reenlightenment_control re_ctrl;
179
180 if (!hv_reenlightenment_available())
181 return;
182
183 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
184 re_ctrl.enabled = 0;
185 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
186
187 hv_reenlightenment_cb = NULL;
188 }
189 EXPORT_SYMBOL_GPL(clear_hv_tscchange_cb);
190
hv_cpu_die(unsigned int cpu)191 static int hv_cpu_die(unsigned int cpu)
192 {
193 struct hv_reenlightenment_control re_ctrl;
194 unsigned int new_cpu;
195 unsigned long flags;
196 void **input_arg;
197 void *input_pg = NULL;
198
199 local_irq_save(flags);
200 input_arg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
201 input_pg = *input_arg;
202 *input_arg = NULL;
203 local_irq_restore(flags);
204 free_page((unsigned long)input_pg);
205
206 if (hv_vp_assist_page && hv_vp_assist_page[cpu])
207 wrmsrl(HV_X64_MSR_VP_ASSIST_PAGE, 0);
208
209 if (hv_reenlightenment_cb == NULL)
210 return 0;
211
212 rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
213 if (re_ctrl.target_vp == hv_vp_index[cpu]) {
214 /* Reassign to some other online CPU */
215 new_cpu = cpumask_any_but(cpu_online_mask, cpu);
216
217 re_ctrl.target_vp = hv_vp_index[new_cpu];
218 wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
219 }
220
221 return 0;
222 }
223
hv_pci_init(void)224 static int __init hv_pci_init(void)
225 {
226 int gen2vm = efi_enabled(EFI_BOOT);
227
228 /*
229 * For Generation-2 VM, we exit from pci_arch_init() by returning 0.
230 * The purpose is to suppress the harmless warning:
231 * "PCI: Fatal: No config space access function found"
232 */
233 if (gen2vm)
234 return 0;
235
236 /* For Generation-1 VM, we'll proceed in pci_arch_init(). */
237 return 1;
238 }
239
240 /*
241 * This function is to be invoked early in the boot sequence after the
242 * hypervisor has been detected.
243 *
244 * 1. Setup the hypercall page.
245 * 2. Register Hyper-V specific clocksource.
246 * 3. Setup Hyper-V specific APIC entry points.
247 */
hyperv_init(void)248 void __init hyperv_init(void)
249 {
250 u64 guest_id, required_msrs;
251 union hv_x64_msr_hypercall_contents hypercall_msr;
252 int cpuhp, i;
253
254 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
255 return;
256
257 /* Absolutely required MSRs */
258 required_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
259 HV_X64_MSR_VP_INDEX_AVAILABLE;
260
261 if ((ms_hyperv.features & required_msrs) != required_msrs)
262 return;
263
264 /*
265 * Allocate the per-CPU state for the hypercall input arg.
266 * If this allocation fails, we will not be able to setup
267 * (per-CPU) hypercall input page and thus this failure is
268 * fatal on Hyper-V.
269 */
270 hyperv_pcpu_input_arg = alloc_percpu(void *);
271
272 BUG_ON(hyperv_pcpu_input_arg == NULL);
273
274 /* Allocate percpu VP index */
275 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
276 GFP_KERNEL);
277 if (!hv_vp_index)
278 return;
279
280 for (i = 0; i < num_possible_cpus(); i++)
281 hv_vp_index[i] = VP_INVAL;
282
283 hv_vp_assist_page = kcalloc(num_possible_cpus(),
284 sizeof(*hv_vp_assist_page), GFP_KERNEL);
285 if (!hv_vp_assist_page) {
286 ms_hyperv.hints &= ~HV_X64_ENLIGHTENED_VMCS_RECOMMENDED;
287 goto free_vp_index;
288 }
289
290 cpuhp = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
291 hv_cpu_init, hv_cpu_die);
292 if (cpuhp < 0)
293 goto free_vp_assist_page;
294
295 /*
296 * Setup the hypercall page and enable hypercalls.
297 * 1. Register the guest ID
298 * 2. Enable the hypercall and register the hypercall page
299 */
300 guest_id = generate_guest_id(0, LINUX_VERSION_CODE, 0);
301 wrmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
302
303 hv_hypercall_pg = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
304 if (hv_hypercall_pg == NULL) {
305 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
306 goto remove_cpuhp_state;
307 }
308
309 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
310 hypercall_msr.enable = 1;
311 hypercall_msr.guest_physical_address = vmalloc_to_pfn(hv_hypercall_pg);
312 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
313
314 hv_apic_init();
315
316 x86_init.pci.arch_init = hv_pci_init;
317
318 return;
319
320 remove_cpuhp_state:
321 cpuhp_remove_state(cpuhp);
322 free_vp_assist_page:
323 kfree(hv_vp_assist_page);
324 hv_vp_assist_page = NULL;
325 free_vp_index:
326 kfree(hv_vp_index);
327 hv_vp_index = NULL;
328 }
329
330 /*
331 * This routine is called before kexec/kdump, it does the required cleanup.
332 */
hyperv_cleanup(void)333 void hyperv_cleanup(void)
334 {
335 union hv_x64_msr_hypercall_contents hypercall_msr;
336
337 /* Reset our OS id */
338 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
339
340 /*
341 * Reset hypercall page reference before reset the page,
342 * let hypercall operations fail safely rather than
343 * panic the kernel for using invalid hypercall page
344 */
345 hv_hypercall_pg = NULL;
346
347 /* Reset the hypercall page */
348 hypercall_msr.as_uint64 = 0;
349 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
350
351 /* Reset the TSC page */
352 hypercall_msr.as_uint64 = 0;
353 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
354 }
355 EXPORT_SYMBOL_GPL(hyperv_cleanup);
356
hyperv_report_panic(struct pt_regs * regs,long err)357 void hyperv_report_panic(struct pt_regs *regs, long err)
358 {
359 static bool panic_reported;
360 u64 guest_id;
361
362 /*
363 * We prefer to report panic on 'die' chain as we have proper
364 * registers to report, but if we miss it (e.g. on BUG()) we need
365 * to report it on 'panic'.
366 */
367 if (panic_reported)
368 return;
369 panic_reported = true;
370
371 rdmsrl(HV_X64_MSR_GUEST_OS_ID, guest_id);
372
373 wrmsrl(HV_X64_MSR_CRASH_P0, err);
374 wrmsrl(HV_X64_MSR_CRASH_P1, guest_id);
375 wrmsrl(HV_X64_MSR_CRASH_P2, regs->ip);
376 wrmsrl(HV_X64_MSR_CRASH_P3, regs->ax);
377 wrmsrl(HV_X64_MSR_CRASH_P4, regs->sp);
378
379 /*
380 * Let Hyper-V know there is crash data available
381 */
382 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
383 }
384 EXPORT_SYMBOL_GPL(hyperv_report_panic);
385
386 /**
387 * hyperv_report_panic_msg - report panic message to Hyper-V
388 * @pa: physical address of the panic page containing the message
389 * @size: size of the message in the page
390 */
hyperv_report_panic_msg(phys_addr_t pa,size_t size)391 void hyperv_report_panic_msg(phys_addr_t pa, size_t size)
392 {
393 /*
394 * P3 to contain the physical address of the panic page & P4 to
395 * contain the size of the panic data in that page. Rest of the
396 * registers are no-op when the NOTIFY_MSG flag is set.
397 */
398 wrmsrl(HV_X64_MSR_CRASH_P0, 0);
399 wrmsrl(HV_X64_MSR_CRASH_P1, 0);
400 wrmsrl(HV_X64_MSR_CRASH_P2, 0);
401 wrmsrl(HV_X64_MSR_CRASH_P3, pa);
402 wrmsrl(HV_X64_MSR_CRASH_P4, size);
403
404 /*
405 * Let Hyper-V know there is crash data available along with
406 * the panic message.
407 */
408 wrmsrl(HV_X64_MSR_CRASH_CTL,
409 (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
410 }
411 EXPORT_SYMBOL_GPL(hyperv_report_panic_msg);
412
hv_is_hyperv_initialized(void)413 bool hv_is_hyperv_initialized(void)
414 {
415 union hv_x64_msr_hypercall_contents hypercall_msr;
416
417 /*
418 * Ensure that we're really on Hyper-V, and not a KVM or Xen
419 * emulation of Hyper-V
420 */
421 if (x86_hyper_type != X86_HYPER_MS_HYPERV)
422 return false;
423
424 /*
425 * Verify that earlier initialization succeeded by checking
426 * that the hypercall page is setup
427 */
428 hypercall_msr.as_uint64 = 0;
429 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
430
431 return hypercall_msr.enable;
432 }
433 EXPORT_SYMBOL_GPL(hv_is_hyperv_initialized);
434
