1 // SPDX-License-Identifier: GPL-2.0
2
3 #include <linux/console.h>
4 #include <linux/cpu.h>
5 #include <linux/instrumentation.h>
6 #include <linux/kexec.h>
7 #include <linux/memblock.h>
8 #include <linux/slab.h>
9 #include <linux/panic_notifier.h>
10
11 #include <xen/xen.h>
12 #include <xen/features.h>
13 #include <xen/interface/sched.h>
14 #include <xen/interface/version.h>
15 #include <xen/page.h>
16
17 #include <asm/xen/hypercall.h>
18 #include <asm/xen/hypervisor.h>
19 #include <asm/cpu.h>
20 #include <asm/e820/api.h>
21 #include <asm/setup.h>
22
23 #include "xen-ops.h"
24
25 DEFINE_STATIC_CALL(xen_hypercall, xen_hypercall_hvm);
26 EXPORT_STATIC_CALL_TRAMP(xen_hypercall);
27
28 /*
29 * Pointer to the xen_vcpu_info structure or
30 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
31 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
32 * but during boot it is switched to point to xen_vcpu_info.
33 * The pointer is used in xen_evtchn_do_upcall to acknowledge pending events.
34 * Make sure that xen_vcpu_info doesn't cross a page boundary by making it
35 * cache-line aligned (the struct is guaranteed to have a size of 64 bytes,
36 * which matches the cache line size of 64-bit x86 processors).
37 */
38 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
39 DEFINE_PER_CPU_ALIGNED(struct vcpu_info, xen_vcpu_info);
40
41 /* Linux <-> Xen vCPU id mapping */
42 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
43 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
44
45 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
46 EXPORT_SYMBOL(machine_to_phys_mapping);
47 unsigned long machine_to_phys_nr;
48 EXPORT_SYMBOL(machine_to_phys_nr);
49
50 struct start_info *xen_start_info;
51 EXPORT_SYMBOL_GPL(xen_start_info);
52
53 struct shared_info xen_dummy_shared_info;
54
55 __read_mostly bool xen_have_vector_callback = true;
56 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
57
58 /*
59 * NB: These need to live in .data or alike because they're used by
60 * xen_prepare_pvh() which runs before clearing the bss.
61 */
62 enum xen_domain_type __ro_after_init xen_domain_type = XEN_NATIVE;
63 EXPORT_SYMBOL_GPL(xen_domain_type);
64 uint32_t __ro_after_init xen_start_flags;
65 EXPORT_SYMBOL(xen_start_flags);
66
67 /*
68 * Point at some empty memory to start with. We map the real shared_info
69 * page as soon as fixmap is up and running.
70 */
71 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
72
73 /* Number of pages released from the initial allocation. */
74 unsigned long xen_released_pages;
75
xen_get_vendor(void)76 static __ref void xen_get_vendor(void)
77 {
78 init_cpu_devs();
79 cpu_detect(&boot_cpu_data);
80 get_cpu_vendor(&boot_cpu_data);
81 }
82
xen_hypercall_setfunc(void)83 void xen_hypercall_setfunc(void)
84 {
85 if (static_call_query(xen_hypercall) != xen_hypercall_hvm)
86 return;
87
88 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
89 boot_cpu_data.x86_vendor == X86_VENDOR_HYGON))
90 static_call_update(xen_hypercall, xen_hypercall_amd);
91 else
92 static_call_update(xen_hypercall, xen_hypercall_intel);
93 }
94
95 /*
96 * Evaluate processor vendor in order to select the correct hypercall
97 * function for HVM/PVH guests.
98 * Might be called very early in boot before vendor has been set by
99 * early_cpu_init().
100 */
__xen_hypercall_setfunc(void)101 noinstr void *__xen_hypercall_setfunc(void)
102 {
103 void (*func)(void);
104
105 /*
106 * Xen is supported only on CPUs with CPUID, so testing for
107 * X86_FEATURE_CPUID is a test for early_cpu_init() having been
108 * run.
109 *
110 * Note that __xen_hypercall_setfunc() is noinstr only due to a nasty
111 * dependency chain: it is being called via the xen_hypercall static
112 * call when running as a PVH or HVM guest. Hypercalls need to be
113 * noinstr due to PV guests using hypercalls in noinstr code. So we
114 * the PV guest requirement is not of interest here (xen_get_vendor()
115 * calls noinstr functions, and static_call_update_early() might do
116 * so, too).
117 */
118 instrumentation_begin();
119
120 if (!boot_cpu_has(X86_FEATURE_CPUID))
121 xen_get_vendor();
122
123 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
124 boot_cpu_data.x86_vendor == X86_VENDOR_HYGON))
125 func = xen_hypercall_amd;
126 else
127 func = xen_hypercall_intel;
128
129 static_call_update_early(xen_hypercall, func);
130
131 instrumentation_end();
132
133 return func;
134 }
135
xen_cpu_up_online(unsigned int cpu)136 static int xen_cpu_up_online(unsigned int cpu)
137 {
138 xen_init_lock_cpu(cpu);
139 return 0;
140 }
141
xen_cpuhp_setup(int (* cpu_up_prepare_cb)(unsigned int),int (* cpu_dead_cb)(unsigned int))142 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
143 int (*cpu_dead_cb)(unsigned int))
144 {
145 int rc;
146
147 rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
148 "x86/xen/guest:prepare",
149 cpu_up_prepare_cb, cpu_dead_cb);
150 if (rc >= 0) {
151 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
152 "x86/xen/guest:online",
153 xen_cpu_up_online, NULL);
154 if (rc < 0)
155 cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
156 }
157
158 return rc >= 0 ? 0 : rc;
159 }
160
xen_vcpu_setup_restore(int cpu)161 static void xen_vcpu_setup_restore(int cpu)
162 {
163 /* Any per_cpu(xen_vcpu) is stale, so reset it */
164 xen_vcpu_info_reset(cpu);
165
166 /*
167 * For PVH and PVHVM, setup online VCPUs only. The rest will
168 * be handled by hotplug.
169 */
170 if (xen_pv_domain() ||
171 (xen_hvm_domain() && cpu_online(cpu)))
172 xen_vcpu_setup(cpu);
173 }
174
175 /*
176 * On restore, set the vcpu placement up again.
177 * If it fails, then we're in a bad state, since
178 * we can't back out from using it...
179 */
xen_vcpu_restore(void)180 void xen_vcpu_restore(void)
181 {
182 int cpu;
183
184 for_each_possible_cpu(cpu) {
185 bool other_cpu = (cpu != smp_processor_id());
186 bool is_up;
187
188 if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
189 continue;
190
191 /* Only Xen 4.5 and higher support this. */
192 is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
193 xen_vcpu_nr(cpu), NULL) > 0;
194
195 if (other_cpu && is_up &&
196 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
197 BUG();
198
199 if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
200 xen_setup_runstate_info(cpu);
201
202 xen_vcpu_setup_restore(cpu);
203
204 if (other_cpu && is_up &&
205 HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
206 BUG();
207 }
208 }
209
xen_vcpu_info_reset(int cpu)210 void xen_vcpu_info_reset(int cpu)
211 {
212 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
213 per_cpu(xen_vcpu, cpu) =
214 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
215 } else {
216 /* Set to NULL so that if somebody accesses it we get an OOPS */
217 per_cpu(xen_vcpu, cpu) = NULL;
218 }
219 }
220
xen_vcpu_setup(int cpu)221 void xen_vcpu_setup(int cpu)
222 {
223 struct vcpu_register_vcpu_info info;
224 int err;
225 struct vcpu_info *vcpup;
226
227 BUILD_BUG_ON(sizeof(*vcpup) > SMP_CACHE_BYTES);
228 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
229
230 /*
231 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
232 * and at restore (xen_vcpu_restore). Also called for hotplugged
233 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
234 * However, the hypercall can only be done once (see below) so if a VCPU
235 * is offlined and comes back online then let's not redo the hypercall.
236 *
237 * For PV it is called during restore (xen_vcpu_restore) and bootup
238 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
239 * use this function.
240 */
241 if (xen_hvm_domain()) {
242 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
243 return;
244 }
245
246 vcpup = &per_cpu(xen_vcpu_info, cpu);
247 info.mfn = arbitrary_virt_to_mfn(vcpup);
248 info.offset = offset_in_page(vcpup);
249
250 /*
251 * N.B. This hypercall can _only_ be called once per CPU.
252 * Subsequent calls will error out with -EINVAL. This is due to
253 * the fact that hypervisor has no unregister variant and this
254 * hypercall does not allow to over-write info.mfn and
255 * info.offset.
256 */
257 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
258 &info);
259 if (err)
260 panic("register_vcpu_info failed: cpu=%d err=%d\n", cpu, err);
261
262 per_cpu(xen_vcpu, cpu) = vcpup;
263 }
264
xen_banner(void)265 void __init xen_banner(void)
266 {
267 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
268 struct xen_extraversion extra;
269
270 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
271
272 pr_info("Booting kernel on %s\n", pv_info.name);
273 pr_info("Xen version: %u.%u%s%s\n",
274 version >> 16, version & 0xffff, extra.extraversion,
275 xen_feature(XENFEAT_mmu_pt_update_preserve_ad)
276 ? " (preserve-AD)" : "");
277 }
278
279 /* Check if running on Xen version (major, minor) or later */
xen_running_on_version_or_later(unsigned int major,unsigned int minor)280 bool xen_running_on_version_or_later(unsigned int major, unsigned int minor)
281 {
282 unsigned int version;
283
284 if (!xen_domain())
285 return false;
286
287 version = HYPERVISOR_xen_version(XENVER_version, NULL);
288 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
289 ((version >> 16) > major))
290 return true;
291 return false;
292 }
293
xen_add_preferred_consoles(void)294 void __init xen_add_preferred_consoles(void)
295 {
296 add_preferred_console("xenboot", 0, NULL);
297 if (!boot_params.screen_info.orig_video_isVGA)
298 add_preferred_console("tty", 0, NULL);
299 add_preferred_console("hvc", 0, NULL);
300 if (boot_params.screen_info.orig_video_isVGA)
301 add_preferred_console("tty", 0, NULL);
302 }
303
xen_reboot(int reason)304 void xen_reboot(int reason)
305 {
306 struct sched_shutdown r = { .reason = reason };
307 int cpu;
308
309 for_each_online_cpu(cpu)
310 xen_pmu_finish(cpu);
311
312 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
313 BUG();
314 }
315
316 static int reboot_reason = SHUTDOWN_reboot;
317 static bool xen_legacy_crash;
xen_emergency_restart(void)318 void xen_emergency_restart(void)
319 {
320 xen_reboot(reboot_reason);
321 }
322
323 static int
xen_panic_event(struct notifier_block * this,unsigned long event,void * ptr)324 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
325 {
326 if (!kexec_crash_loaded()) {
327 if (xen_legacy_crash)
328 xen_reboot(SHUTDOWN_crash);
329
330 reboot_reason = SHUTDOWN_crash;
331
332 /*
333 * If panic_timeout==0 then we are supposed to wait forever.
334 * However, to preserve original dom0 behavior we have to drop
335 * into hypervisor. (domU behavior is controlled by its
336 * config file)
337 */
338 if (panic_timeout == 0)
339 panic_timeout = -1;
340 }
341 return NOTIFY_DONE;
342 }
343
parse_xen_legacy_crash(char * arg)344 static int __init parse_xen_legacy_crash(char *arg)
345 {
346 xen_legacy_crash = true;
347 return 0;
348 }
349 early_param("xen_legacy_crash", parse_xen_legacy_crash);
350
351 static struct notifier_block xen_panic_block = {
352 .notifier_call = xen_panic_event,
353 .priority = INT_MIN
354 };
355
xen_panic_handler_init(void)356 int xen_panic_handler_init(void)
357 {
358 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
359 return 0;
360 }
361
xen_pin_vcpu(int cpu)362 void xen_pin_vcpu(int cpu)
363 {
364 static bool disable_pinning;
365 struct sched_pin_override pin_override;
366 int ret;
367
368 if (disable_pinning)
369 return;
370
371 pin_override.pcpu = cpu;
372 ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
373
374 /* Ignore errors when removing override. */
375 if (cpu < 0)
376 return;
377
378 switch (ret) {
379 case -ENOSYS:
380 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
381 cpu);
382 disable_pinning = true;
383 break;
384 case -EPERM:
385 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
386 disable_pinning = true;
387 break;
388 case -EINVAL:
389 case -EBUSY:
390 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
391 cpu);
392 break;
393 case 0:
394 break;
395 default:
396 WARN(1, "rc %d while trying to pin vcpu\n", ret);
397 disable_pinning = true;
398 }
399 }
400
401 #ifdef CONFIG_HOTPLUG_CPU
xen_arch_register_cpu(int num)402 void xen_arch_register_cpu(int num)
403 {
404 arch_register_cpu(num);
405 }
406 EXPORT_SYMBOL(xen_arch_register_cpu);
407
xen_arch_unregister_cpu(int num)408 void xen_arch_unregister_cpu(int num)
409 {
410 arch_unregister_cpu(num);
411 }
412 EXPORT_SYMBOL(xen_arch_unregister_cpu);
413 #endif
414
415 /* Amount of extra memory space we add to the e820 ranges */
416 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
417
xen_add_extra_mem(unsigned long start_pfn,unsigned long n_pfns)418 void __init xen_add_extra_mem(unsigned long start_pfn, unsigned long n_pfns)
419 {
420 unsigned int i;
421
422 /*
423 * No need to check for zero size, should happen rarely and will only
424 * write a new entry regarded to be unused due to zero size.
425 */
426 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
427 /* Add new region. */
428 if (xen_extra_mem[i].n_pfns == 0) {
429 xen_extra_mem[i].start_pfn = start_pfn;
430 xen_extra_mem[i].n_pfns = n_pfns;
431 break;
432 }
433 /* Append to existing region. */
434 if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
435 start_pfn) {
436 xen_extra_mem[i].n_pfns += n_pfns;
437 break;
438 }
439 }
440 if (i == XEN_EXTRA_MEM_MAX_REGIONS)
441 printk(KERN_WARNING "Warning: not enough extra memory regions\n");
442
443 memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
444 }
445
446 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC
arch_xen_unpopulated_init(struct resource ** res)447 int __init arch_xen_unpopulated_init(struct resource **res)
448 {
449 unsigned int i;
450
451 if (!xen_domain())
452 return -ENODEV;
453
454 /* Must be set strictly before calling xen_free_unpopulated_pages(). */
455 *res = &iomem_resource;
456
457 /*
458 * Initialize with pages from the extra memory regions (see
459 * arch/x86/xen/setup.c).
460 */
461 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
462 unsigned int j;
463
464 for (j = 0; j < xen_extra_mem[i].n_pfns; j++) {
465 struct page *pg =
466 pfn_to_page(xen_extra_mem[i].start_pfn + j);
467
468 xen_free_unpopulated_pages(1, &pg);
469 }
470
471 /*
472 * Account for the region being in the physmap but unpopulated.
473 * The value in xen_released_pages is used by the balloon
474 * driver to know how much of the physmap is unpopulated and
475 * set an accurate initial memory target.
476 */
477 xen_released_pages += xen_extra_mem[i].n_pfns;
478 /* Zero so region is not also added to the balloon driver. */
479 xen_extra_mem[i].n_pfns = 0;
480 }
481
482 return 0;
483 }
484 #endif
485