xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 16f6ccde74a6f8538c62f127f17207c75f4dba7a)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
4 #include <linux/memblock.h>
5 #endif
6 #include <linux/console.h>
7 #include <linux/cpu.h>
8 #include <linux/instrumentation.h>
9 #include <linux/kexec.h>
10 #include <linux/memblock.h>
11 #include <linux/slab.h>
12 #include <linux/panic_notifier.h>
13 
14 #include <xen/xen.h>
15 #include <xen/features.h>
16 #include <xen/interface/sched.h>
17 #include <xen/interface/version.h>
18 #include <xen/page.h>
19 
20 #include <asm/xen/hypercall.h>
21 #include <asm/xen/hypervisor.h>
22 #include <asm/cpu.h>
23 #include <asm/e820/api.h>
24 #include <asm/setup.h>
25 
26 #include "xen-ops.h"
27 #include "smp.h"
28 #include "pmu.h"
29 
30 DEFINE_STATIC_CALL(xen_hypercall, xen_hypercall_hvm);
31 EXPORT_STATIC_CALL_TRAMP(xen_hypercall);
32 
33 /*
34  * Pointer to the xen_vcpu_info structure or
35  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
36  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
37  * but during boot it is switched to point to xen_vcpu_info.
38  * The pointer is used in xen_evtchn_do_upcall to acknowledge pending events.
39  * Make sure that xen_vcpu_info doesn't cross a page boundary by making it
40  * cache-line aligned (the struct is guaranteed to have a size of 64 bytes,
41  * which matches the cache line size of 64-bit x86 processors).
42  */
43 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
44 DEFINE_PER_CPU_ALIGNED(struct vcpu_info, xen_vcpu_info);
45 
46 /* Linux <-> Xen vCPU id mapping */
47 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
48 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
49 
50 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
51 EXPORT_SYMBOL(machine_to_phys_mapping);
52 unsigned long  machine_to_phys_nr;
53 EXPORT_SYMBOL(machine_to_phys_nr);
54 
55 struct start_info *xen_start_info;
56 EXPORT_SYMBOL_GPL(xen_start_info);
57 
58 struct shared_info xen_dummy_shared_info;
59 
60 __read_mostly bool xen_have_vector_callback = true;
61 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
62 
63 /*
64  * NB: These need to live in .data or alike because they're used by
65  * xen_prepare_pvh() which runs before clearing the bss.
66  */
67 enum xen_domain_type __ro_after_init xen_domain_type = XEN_NATIVE;
68 EXPORT_SYMBOL_GPL(xen_domain_type);
69 uint32_t __ro_after_init xen_start_flags;
70 EXPORT_SYMBOL(xen_start_flags);
71 
72 /*
73  * Point at some empty memory to start with. We map the real shared_info
74  * page as soon as fixmap is up and running.
75  */
76 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
77 
xen_get_vendor(void)78 static __ref void xen_get_vendor(void)
79 {
80 	init_cpu_devs();
81 	cpu_detect(&boot_cpu_data);
82 	get_cpu_vendor(&boot_cpu_data);
83 }
84 
xen_hypercall_setfunc(void)85 void xen_hypercall_setfunc(void)
86 {
87 	if (static_call_query(xen_hypercall) != xen_hypercall_hvm)
88 		return;
89 
90 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
91 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON))
92 		static_call_update(xen_hypercall, xen_hypercall_amd);
93 	else
94 		static_call_update(xen_hypercall, xen_hypercall_intel);
95 }
96 
97 /*
98  * Evaluate processor vendor in order to select the correct hypercall
99  * function for HVM/PVH guests.
100  * Might be called very early in boot before vendor has been set by
101  * early_cpu_init().
102  */
__xen_hypercall_setfunc(void)103 noinstr void *__xen_hypercall_setfunc(void)
104 {
105 	void (*func)(void);
106 
107 	/*
108 	 * Xen is supported only on CPUs with CPUID, so testing for
109 	 * X86_FEATURE_CPUID is a test for early_cpu_init() having been
110 	 * run.
111 	 *
112 	 * Note that __xen_hypercall_setfunc() is noinstr only due to a nasty
113 	 * dependency chain: it is being called via the xen_hypercall static
114 	 * call when running as a PVH or HVM guest. Hypercalls need to be
115 	 * noinstr due to PV guests using hypercalls in noinstr code. So we
116 	 * can safely tag the function body as "instrumentation ok", since
117 	 * the PV guest requirement is not of interest here (xen_get_vendor()
118 	 * calls noinstr functions, and static_call_update_early() might do
119 	 * so, too).
120 	 */
121 	instrumentation_begin();
122 
123 	if (!boot_cpu_has(X86_FEATURE_CPUID))
124 		xen_get_vendor();
125 
126 	if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
127 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON))
128 		func = xen_hypercall_amd;
129 	else
130 		func = xen_hypercall_intel;
131 
132 	static_call_update_early(xen_hypercall, func);
133 
134 	instrumentation_end();
135 
136 	return func;
137 }
138 
xen_cpu_up_online(unsigned int cpu)139 static int xen_cpu_up_online(unsigned int cpu)
140 {
141 	xen_init_lock_cpu(cpu);
142 	return 0;
143 }
144 
xen_cpuhp_setup(int (* cpu_up_prepare_cb)(unsigned int),int (* cpu_dead_cb)(unsigned int))145 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
146 		    int (*cpu_dead_cb)(unsigned int))
147 {
148 	int rc;
149 
150 	rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
151 				       "x86/xen/guest:prepare",
152 				       cpu_up_prepare_cb, cpu_dead_cb);
153 	if (rc >= 0) {
154 		rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
155 					       "x86/xen/guest:online",
156 					       xen_cpu_up_online, NULL);
157 		if (rc < 0)
158 			cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
159 	}
160 
161 	return rc >= 0 ? 0 : rc;
162 }
163 
xen_vcpu_setup_restore(int cpu)164 static void xen_vcpu_setup_restore(int cpu)
165 {
166 	/* Any per_cpu(xen_vcpu) is stale, so reset it */
167 	xen_vcpu_info_reset(cpu);
168 
169 	/*
170 	 * For PVH and PVHVM, setup online VCPUs only. The rest will
171 	 * be handled by hotplug.
172 	 */
173 	if (xen_pv_domain() ||
174 	    (xen_hvm_domain() && cpu_online(cpu)))
175 		xen_vcpu_setup(cpu);
176 }
177 
178 /*
179  * On restore, set the vcpu placement up again.
180  * If it fails, then we're in a bad state, since
181  * we can't back out from using it...
182  */
xen_vcpu_restore(void)183 void xen_vcpu_restore(void)
184 {
185 	int cpu;
186 
187 	for_each_possible_cpu(cpu) {
188 		bool other_cpu = (cpu != smp_processor_id());
189 		bool is_up;
190 
191 		if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
192 			continue;
193 
194 		/* Only Xen 4.5 and higher support this. */
195 		is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
196 					   xen_vcpu_nr(cpu), NULL) > 0;
197 
198 		if (other_cpu && is_up &&
199 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
200 			BUG();
201 
202 		if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
203 			xen_setup_runstate_info(cpu);
204 
205 		xen_vcpu_setup_restore(cpu);
206 
207 		if (other_cpu && is_up &&
208 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
209 			BUG();
210 	}
211 }
212 
xen_vcpu_info_reset(int cpu)213 void xen_vcpu_info_reset(int cpu)
214 {
215 	if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
216 		per_cpu(xen_vcpu, cpu) =
217 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
218 	} else {
219 		/* Set to NULL so that if somebody accesses it we get an OOPS */
220 		per_cpu(xen_vcpu, cpu) = NULL;
221 	}
222 }
223 
xen_vcpu_setup(int cpu)224 void xen_vcpu_setup(int cpu)
225 {
226 	struct vcpu_register_vcpu_info info;
227 	int err;
228 	struct vcpu_info *vcpup;
229 
230 	BUILD_BUG_ON(sizeof(*vcpup) > SMP_CACHE_BYTES);
231 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
232 
233 	/*
234 	 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
235 	 * and at restore (xen_vcpu_restore). Also called for hotplugged
236 	 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
237 	 * However, the hypercall can only be done once (see below) so if a VCPU
238 	 * is offlined and comes back online then let's not redo the hypercall.
239 	 *
240 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
241 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
242 	 * use this function.
243 	 */
244 	if (xen_hvm_domain()) {
245 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
246 			return;
247 	}
248 
249 	vcpup = &per_cpu(xen_vcpu_info, cpu);
250 	info.mfn = arbitrary_virt_to_mfn(vcpup);
251 	info.offset = offset_in_page(vcpup);
252 
253 	/*
254 	 * N.B. This hypercall can _only_ be called once per CPU.
255 	 * Subsequent calls will error out with -EINVAL. This is due to
256 	 * the fact that hypervisor has no unregister variant and this
257 	 * hypercall does not allow to over-write info.mfn and
258 	 * info.offset.
259 	 */
260 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
261 				 &info);
262 	if (err)
263 		panic("register_vcpu_info failed: cpu=%d err=%d\n", cpu, err);
264 
265 	per_cpu(xen_vcpu, cpu) = vcpup;
266 }
267 
xen_banner(void)268 void __init xen_banner(void)
269 {
270 	unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
271 	struct xen_extraversion extra;
272 
273 	HYPERVISOR_xen_version(XENVER_extraversion, &extra);
274 
275 	pr_info("Booting kernel on %s\n", pv_info.name);
276 	pr_info("Xen version: %u.%u%s%s\n",
277 		version >> 16, version & 0xffff, extra.extraversion,
278 		xen_feature(XENFEAT_mmu_pt_update_preserve_ad)
279 		? " (preserve-AD)" : "");
280 }
281 
282 /* Check if running on Xen version (major, minor) or later */
xen_running_on_version_or_later(unsigned int major,unsigned int minor)283 bool xen_running_on_version_or_later(unsigned int major, unsigned int minor)
284 {
285 	unsigned int version;
286 
287 	if (!xen_domain())
288 		return false;
289 
290 	version = HYPERVISOR_xen_version(XENVER_version, NULL);
291 	if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
292 		((version >> 16) > major))
293 		return true;
294 	return false;
295 }
296 
xen_add_preferred_consoles(void)297 void __init xen_add_preferred_consoles(void)
298 {
299 	add_preferred_console("xenboot", 0, NULL);
300 	if (!boot_params.screen_info.orig_video_isVGA)
301 		add_preferred_console("tty", 0, NULL);
302 	add_preferred_console("hvc", 0, NULL);
303 	if (boot_params.screen_info.orig_video_isVGA)
304 		add_preferred_console("tty", 0, NULL);
305 }
306 
xen_reboot(int reason)307 void xen_reboot(int reason)
308 {
309 	struct sched_shutdown r = { .reason = reason };
310 	int cpu;
311 
312 	for_each_online_cpu(cpu)
313 		xen_pmu_finish(cpu);
314 
315 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
316 		BUG();
317 }
318 
319 static int reboot_reason = SHUTDOWN_reboot;
320 static bool xen_legacy_crash;
xen_emergency_restart(void)321 void xen_emergency_restart(void)
322 {
323 	xen_reboot(reboot_reason);
324 }
325 
326 static int
xen_panic_event(struct notifier_block * this,unsigned long event,void * ptr)327 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
328 {
329 	if (!kexec_crash_loaded()) {
330 		if (xen_legacy_crash)
331 			xen_reboot(SHUTDOWN_crash);
332 
333 		reboot_reason = SHUTDOWN_crash;
334 
335 		/*
336 		 * If panic_timeout==0 then we are supposed to wait forever.
337 		 * However, to preserve original dom0 behavior we have to drop
338 		 * into hypervisor. (domU behavior is controlled by its
339 		 * config file)
340 		 */
341 		if (panic_timeout == 0)
342 			panic_timeout = -1;
343 	}
344 	return NOTIFY_DONE;
345 }
346 
parse_xen_legacy_crash(char * arg)347 static int __init parse_xen_legacy_crash(char *arg)
348 {
349 	xen_legacy_crash = true;
350 	return 0;
351 }
352 early_param("xen_legacy_crash", parse_xen_legacy_crash);
353 
354 static struct notifier_block xen_panic_block = {
355 	.notifier_call = xen_panic_event,
356 	.priority = INT_MIN
357 };
358 
xen_panic_handler_init(void)359 int xen_panic_handler_init(void)
360 {
361 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
362 	return 0;
363 }
364 
xen_pin_vcpu(int cpu)365 void xen_pin_vcpu(int cpu)
366 {
367 	static bool disable_pinning;
368 	struct sched_pin_override pin_override;
369 	int ret;
370 
371 	if (disable_pinning)
372 		return;
373 
374 	pin_override.pcpu = cpu;
375 	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
376 
377 	/* Ignore errors when removing override. */
378 	if (cpu < 0)
379 		return;
380 
381 	switch (ret) {
382 	case -ENOSYS:
383 		pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
384 			cpu);
385 		disable_pinning = true;
386 		break;
387 	case -EPERM:
388 		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
389 		disable_pinning = true;
390 		break;
391 	case -EINVAL:
392 	case -EBUSY:
393 		pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
394 			cpu);
395 		break;
396 	case 0:
397 		break;
398 	default:
399 		WARN(1, "rc %d while trying to pin vcpu\n", ret);
400 		disable_pinning = true;
401 	}
402 }
403 
404 #ifdef CONFIG_HOTPLUG_CPU
xen_arch_register_cpu(int num)405 void xen_arch_register_cpu(int num)
406 {
407 	arch_register_cpu(num);
408 }
409 EXPORT_SYMBOL(xen_arch_register_cpu);
410 
xen_arch_unregister_cpu(int num)411 void xen_arch_unregister_cpu(int num)
412 {
413 	arch_unregister_cpu(num);
414 }
415 EXPORT_SYMBOL(xen_arch_unregister_cpu);
416 #endif
417 
418 /* Amount of extra memory space we add to the e820 ranges */
419 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
420 
xen_add_extra_mem(unsigned long start_pfn,unsigned long n_pfns)421 void __init xen_add_extra_mem(unsigned long start_pfn, unsigned long n_pfns)
422 {
423 	unsigned int i;
424 
425 	/*
426 	 * No need to check for zero size, should happen rarely and will only
427 	 * write a new entry regarded to be unused due to zero size.
428 	 */
429 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
430 		/* Add new region. */
431 		if (xen_extra_mem[i].n_pfns == 0) {
432 			xen_extra_mem[i].start_pfn = start_pfn;
433 			xen_extra_mem[i].n_pfns = n_pfns;
434 			break;
435 		}
436 		/* Append to existing region. */
437 		if (xen_extra_mem[i].start_pfn + xen_extra_mem[i].n_pfns ==
438 		    start_pfn) {
439 			xen_extra_mem[i].n_pfns += n_pfns;
440 			break;
441 		}
442 	}
443 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
444 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
445 
446 	memblock_reserve(PFN_PHYS(start_pfn), PFN_PHYS(n_pfns));
447 }
448 
449 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC
arch_xen_unpopulated_init(struct resource ** res)450 int __init arch_xen_unpopulated_init(struct resource **res)
451 {
452 	unsigned int i;
453 
454 	if (!xen_domain())
455 		return -ENODEV;
456 
457 	/* Must be set strictly before calling xen_free_unpopulated_pages(). */
458 	*res = &iomem_resource;
459 
460 	/*
461 	 * Initialize with pages from the extra memory regions (see
462 	 * arch/x86/xen/setup.c).
463 	 */
464 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
465 		unsigned int j;
466 
467 		for (j = 0; j < xen_extra_mem[i].n_pfns; j++) {
468 			struct page *pg =
469 				pfn_to_page(xen_extra_mem[i].start_pfn + j);
470 
471 			xen_free_unpopulated_pages(1, &pg);
472 		}
473 
474 		/* Zero so region is not also added to the balloon driver. */
475 		xen_extra_mem[i].n_pfns = 0;
476 	}
477 
478 	return 0;
479 }
480 #endif
481