1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Architecture neutral utility routines for interacting with
5 * Hyper-V. This file is specifically for code that must be
6 * built-in to the kernel image when CONFIG_HYPERV is set
7 * (vs. being in a module) because it is called from architecture
8 * specific code under arch/.
9 *
10 * Copyright (C) 2021, Microsoft, Inc.
11 *
12 * Author : Michael Kelley <mikelley@microsoft.com>
13 */
14
15 #include <linux/types.h>
16 #include <linux/acpi.h>
17 #include <linux/export.h>
18 #include <linux/bitfield.h>
19 #include <linux/cpumask.h>
20 #include <linux/panic_notifier.h>
21 #include <linux/ptrace.h>
22 #include <linux/slab.h>
23 #include <linux/dma-map-ops.h>
24 #include <asm/hyperv-tlfs.h>
25 #include <asm/mshyperv.h>
26
27 /*
28 * hv_root_partition, ms_hyperv and hv_nested are defined here with other
29 * Hyper-V specific globals so they are shared across all architectures and are
30 * built only when CONFIG_HYPERV is defined. But on x86,
31 * ms_hyperv_init_platform() is built even when CONFIG_HYPERV is not
32 * defined, and it uses these three variables. So mark them as __weak
33 * here, allowing for an overriding definition in the module containing
34 * ms_hyperv_init_platform().
35 */
36 bool __weak hv_root_partition;
37 EXPORT_SYMBOL_GPL(hv_root_partition);
38
39 bool __weak hv_nested;
40 EXPORT_SYMBOL_GPL(hv_nested);
41
42 struct ms_hyperv_info __weak ms_hyperv;
43 EXPORT_SYMBOL_GPL(ms_hyperv);
44
45 u32 *hv_vp_index;
46 EXPORT_SYMBOL_GPL(hv_vp_index);
47
48 u32 hv_max_vp_index;
49 EXPORT_SYMBOL_GPL(hv_max_vp_index);
50
51 void * __percpu *hyperv_pcpu_input_arg;
52 EXPORT_SYMBOL_GPL(hyperv_pcpu_input_arg);
53
54 void * __percpu *hyperv_pcpu_output_arg;
55 EXPORT_SYMBOL_GPL(hyperv_pcpu_output_arg);
56
57 /*
58 * Hyper-V specific initialization and shutdown code that is
59 * common across all architectures. Called from architecture
60 * specific initialization functions.
61 */
62
hv_common_free(void)63 void __init hv_common_free(void)
64 {
65 kfree(hv_vp_index);
66 hv_vp_index = NULL;
67
68 free_percpu(hyperv_pcpu_output_arg);
69 hyperv_pcpu_output_arg = NULL;
70
71 free_percpu(hyperv_pcpu_input_arg);
72 hyperv_pcpu_input_arg = NULL;
73 }
74
hv_common_init(void)75 int __init hv_common_init(void)
76 {
77 int i;
78
79 /*
80 * Hyper-V expects to get crash register data or kmsg when
81 * crash enlightment is available and system crashes. Set
82 * crash_kexec_post_notifiers to be true to make sure that
83 * calling crash enlightment interface before running kdump
84 * kernel.
85 */
86 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
87 crash_kexec_post_notifiers = true;
88 pr_info("Hyper-V: enabling crash_kexec_post_notifiers\n");
89 }
90
91 /*
92 * Allocate the per-CPU state for the hypercall input arg.
93 * If this allocation fails, we will not be able to setup
94 * (per-CPU) hypercall input page and thus this failure is
95 * fatal on Hyper-V.
96 */
97 hyperv_pcpu_input_arg = alloc_percpu(void *);
98 BUG_ON(!hyperv_pcpu_input_arg);
99
100 /* Allocate the per-CPU state for output arg for root */
101 if (hv_root_partition) {
102 hyperv_pcpu_output_arg = alloc_percpu(void *);
103 BUG_ON(!hyperv_pcpu_output_arg);
104 }
105
106 hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
107 GFP_KERNEL);
108 if (!hv_vp_index) {
109 hv_common_free();
110 return -ENOMEM;
111 }
112
113 for (i = 0; i < num_possible_cpus(); i++)
114 hv_vp_index[i] = VP_INVAL;
115
116 return 0;
117 }
118
119 /*
120 * Hyper-V specific initialization and die code for
121 * individual CPUs that is common across all architectures.
122 * Called by the CPU hotplug mechanism.
123 */
124
hv_common_cpu_init(unsigned int cpu)125 int hv_common_cpu_init(unsigned int cpu)
126 {
127 void **inputarg, **outputarg;
128 u64 msr_vp_index;
129 gfp_t flags;
130 int pgcount = hv_root_partition ? 2 : 1;
131
132 /* hv_cpu_init() can be called with IRQs disabled from hv_resume() */
133 flags = irqs_disabled() ? GFP_ATOMIC : GFP_KERNEL;
134
135 inputarg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
136 *inputarg = kmalloc(pgcount * HV_HYP_PAGE_SIZE, flags);
137 if (!(*inputarg))
138 return -ENOMEM;
139
140 if (hv_root_partition) {
141 outputarg = (void **)this_cpu_ptr(hyperv_pcpu_output_arg);
142 *outputarg = (char *)(*inputarg) + HV_HYP_PAGE_SIZE;
143 }
144
145 msr_vp_index = hv_get_register(HV_REGISTER_VP_INDEX);
146
147 hv_vp_index[cpu] = msr_vp_index;
148
149 if (msr_vp_index > hv_max_vp_index)
150 hv_max_vp_index = msr_vp_index;
151
152 return 0;
153 }
154
hv_common_cpu_die(unsigned int cpu)155 int hv_common_cpu_die(unsigned int cpu)
156 {
157 unsigned long flags;
158 void **inputarg, **outputarg;
159 void *mem;
160
161 local_irq_save(flags);
162
163 inputarg = (void **)this_cpu_ptr(hyperv_pcpu_input_arg);
164 mem = *inputarg;
165 *inputarg = NULL;
166
167 if (hv_root_partition) {
168 outputarg = (void **)this_cpu_ptr(hyperv_pcpu_output_arg);
169 *outputarg = NULL;
170 }
171
172 local_irq_restore(flags);
173
174 kfree(mem);
175
176 return 0;
177 }
178
179 /* Bit mask of the extended capability to query: see HV_EXT_CAPABILITY_xxx */
hv_query_ext_cap(u64 cap_query)180 bool hv_query_ext_cap(u64 cap_query)
181 {
182 /*
183 * The address of the 'hv_extended_cap' variable will be used as an
184 * output parameter to the hypercall below and so it should be
185 * compatible with 'virt_to_phys'. Which means, it's address should be
186 * directly mapped. Use 'static' to keep it compatible; stack variables
187 * can be virtually mapped, making them incompatible with
188 * 'virt_to_phys'.
189 * Hypercall input/output addresses should also be 8-byte aligned.
190 */
191 static u64 hv_extended_cap __aligned(8);
192 static bool hv_extended_cap_queried;
193 u64 status;
194
195 /*
196 * Querying extended capabilities is an extended hypercall. Check if the
197 * partition supports extended hypercall, first.
198 */
199 if (!(ms_hyperv.priv_high & HV_ENABLE_EXTENDED_HYPERCALLS))
200 return false;
201
202 /* Extended capabilities do not change at runtime. */
203 if (hv_extended_cap_queried)
204 return hv_extended_cap & cap_query;
205
206 status = hv_do_hypercall(HV_EXT_CALL_QUERY_CAPABILITIES, NULL,
207 &hv_extended_cap);
208
209 /*
210 * The query extended capabilities hypercall should not fail under
211 * any normal circumstances. Avoid repeatedly making the hypercall, on
212 * error.
213 */
214 hv_extended_cap_queried = true;
215 if (!hv_result_success(status)) {
216 pr_err("Hyper-V: Extended query capabilities hypercall failed 0x%llx\n",
217 status);
218 return false;
219 }
220
221 return hv_extended_cap & cap_query;
222 }
223 EXPORT_SYMBOL_GPL(hv_query_ext_cap);
224
hv_setup_dma_ops(struct device * dev,bool coherent)225 void hv_setup_dma_ops(struct device *dev, bool coherent)
226 {
227 /*
228 * Hyper-V does not offer a vIOMMU in the guest
229 * VM, so pass 0/NULL for the IOMMU settings
230 */
231 arch_setup_dma_ops(dev, 0, 0, NULL, coherent);
232 }
233 EXPORT_SYMBOL_GPL(hv_setup_dma_ops);
234
hv_is_hibernation_supported(void)235 bool hv_is_hibernation_supported(void)
236 {
237 return !hv_root_partition && acpi_sleep_state_supported(ACPI_STATE_S4);
238 }
239 EXPORT_SYMBOL_GPL(hv_is_hibernation_supported);
240
241 /*
242 * Default function to read the Hyper-V reference counter, independent
243 * of whether Hyper-V enlightened clocks/timers are being used. But on
244 * architectures where it is used, Hyper-V enlightenment code in
245 * hyperv_timer.c may override this function.
246 */
__hv_read_ref_counter(void)247 static u64 __hv_read_ref_counter(void)
248 {
249 return hv_get_register(HV_REGISTER_TIME_REF_COUNT);
250 }
251
252 u64 (*hv_read_reference_counter)(void) = __hv_read_ref_counter;
253 EXPORT_SYMBOL_GPL(hv_read_reference_counter);
254
255 /* These __weak functions provide default "no-op" behavior and
256 * may be overridden by architecture specific versions. Architectures
257 * for which the default "no-op" behavior is sufficient can leave
258 * them unimplemented and not be cluttered with a bunch of stub
259 * functions in arch-specific code.
260 */
261
hv_is_isolation_supported(void)262 bool __weak hv_is_isolation_supported(void)
263 {
264 return false;
265 }
266 EXPORT_SYMBOL_GPL(hv_is_isolation_supported);
267
hv_isolation_type_snp(void)268 bool __weak hv_isolation_type_snp(void)
269 {
270 return false;
271 }
272 EXPORT_SYMBOL_GPL(hv_isolation_type_snp);
273
hv_setup_vmbus_handler(void (* handler)(void))274 void __weak hv_setup_vmbus_handler(void (*handler)(void))
275 {
276 }
277 EXPORT_SYMBOL_GPL(hv_setup_vmbus_handler);
278
hv_remove_vmbus_handler(void)279 void __weak hv_remove_vmbus_handler(void)
280 {
281 }
282 EXPORT_SYMBOL_GPL(hv_remove_vmbus_handler);
283
hv_setup_kexec_handler(void (* handler)(void))284 void __weak hv_setup_kexec_handler(void (*handler)(void))
285 {
286 }
287 EXPORT_SYMBOL_GPL(hv_setup_kexec_handler);
288
hv_remove_kexec_handler(void)289 void __weak hv_remove_kexec_handler(void)
290 {
291 }
292 EXPORT_SYMBOL_GPL(hv_remove_kexec_handler);
293
hv_setup_crash_handler(void (* handler)(struct pt_regs * regs))294 void __weak hv_setup_crash_handler(void (*handler)(struct pt_regs *regs))
295 {
296 }
297 EXPORT_SYMBOL_GPL(hv_setup_crash_handler);
298
hv_remove_crash_handler(void)299 void __weak hv_remove_crash_handler(void)
300 {
301 }
302 EXPORT_SYMBOL_GPL(hv_remove_crash_handler);
303
hyperv_cleanup(void)304 void __weak hyperv_cleanup(void)
305 {
306 }
307 EXPORT_SYMBOL_GPL(hyperv_cleanup);
308
hv_ghcb_hypercall(u64 control,void * input,void * output,u32 input_size)309 u64 __weak hv_ghcb_hypercall(u64 control, void *input, void *output, u32 input_size)
310 {
311 return HV_STATUS_INVALID_PARAMETER;
312 }
313 EXPORT_SYMBOL_GPL(hv_ghcb_hypercall);
314
hv_map_memory(void * addr,unsigned long size)315 void __weak *hv_map_memory(void *addr, unsigned long size)
316 {
317 return NULL;
318 }
319 EXPORT_SYMBOL_GPL(hv_map_memory);
320
hv_unmap_memory(void * addr)321 void __weak hv_unmap_memory(void *addr)
322 {
323 }
324 EXPORT_SYMBOL_GPL(hv_unmap_memory);
325