1// SPDX-License-Identifier: GPL-2.0-only
2#include <xen/xen.h>
3#include <xen/events.h>
4#include <xen/grant_table.h>
5#include <xen/hvm.h>
6#include <xen/interface/vcpu.h>
7#include <xen/interface/xen.h>
8#include <xen/interface/memory.h>
9#include <xen/interface/hvm/params.h>
10#include <xen/features.h>
11#include <xen/platform_pci.h>
12#include <xen/xenbus.h>
13#include <xen/page.h>
14#include <xen/interface/sched.h>
15#include <xen/xen-ops.h>
16#include <asm/xen/hypervisor.h>
17#include <asm/xen/hypercall.h>
18#include <asm/system_misc.h>
19#include <asm/efi.h>
20#include <linux/interrupt.h>
21#include <linux/irqreturn.h>
22#include <linux/module.h>
23#include <linux/of.h>
24#include <linux/of_fdt.h>
25#include <linux/of_irq.h>
26#include <linux/of_address.h>
27#include <linux/cpuidle.h>
28#include <linux/cpufreq.h>
29#include <linux/cpu.h>
30#include <linux/console.h>
31#include <linux/pvclock_gtod.h>
32#include <linux/reboot.h>
33#include <linux/time64.h>
34#include <linux/timekeeping.h>
35#include <linux/timekeeper_internal.h>
36#include <linux/acpi.h>
37#include <linux/virtio_anchor.h>
38
39#include <linux/mm.h>
40
41static struct start_info _xen_start_info;
42struct start_info *xen_start_info = &_xen_start_info;
43EXPORT_SYMBOL(xen_start_info);
44
45enum xen_domain_type xen_domain_type = XEN_NATIVE;
46EXPORT_SYMBOL(xen_domain_type);
47
48struct shared_info xen_dummy_shared_info;
49struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
50
51DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
52static struct vcpu_info __percpu *xen_vcpu_info;
53
54/* Linux <-> Xen vCPU id mapping */
55DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
56EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
57
58/* These are unused until we support booting "pre-ballooned" */
59unsigned long xen_released_pages;
60struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
61
62static __read_mostly unsigned int xen_events_irq;
63static __read_mostly phys_addr_t xen_grant_frames;
64
65#define GRANT_TABLE_INDEX   0
66#define EXT_REGION_INDEX    1
67
68uint32_t xen_start_flags;
69EXPORT_SYMBOL(xen_start_flags);
70
71int xen_unmap_domain_gfn_range(struct vm_area_struct *vma,
72			       int nr, struct page **pages)
73{
74	return xen_xlate_unmap_gfn_range(vma, nr, pages);
75}
76EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range);
77
78static void xen_read_wallclock(struct timespec64 *ts)
79{
80	u32 version;
81	struct timespec64 now, ts_monotonic;
82	struct shared_info *s = HYPERVISOR_shared_info;
83	struct pvclock_wall_clock *wall_clock = &(s->wc);
84
85	/* get wallclock at system boot */
86	do {
87		version = wall_clock->version;
88		rmb();		/* fetch version before time */
89		now.tv_sec  = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec;
90		now.tv_nsec = wall_clock->nsec;
91		rmb();		/* fetch time before checking version */
92	} while ((wall_clock->version & 1) || (version != wall_clock->version));
93
94	/* time since system boot */
95	ktime_get_ts64(&ts_monotonic);
96	*ts = timespec64_add(now, ts_monotonic);
97}
98
99static int xen_pvclock_gtod_notify(struct notifier_block *nb,
100				   unsigned long was_set, void *priv)
101{
102	/* Protected by the calling core code serialization */
103	static struct timespec64 next_sync;
104
105	struct xen_platform_op op;
106	struct timespec64 now, system_time;
107	struct timekeeper *tk = priv;
108
109	now.tv_sec = tk->xtime_sec;
110	now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
111	system_time = timespec64_add(now, tk->wall_to_monotonic);
112
113	/*
114	 * We only take the expensive HV call when the clock was set
115	 * or when the 11 minutes RTC synchronization time elapsed.
116	 */
117	if (!was_set && timespec64_compare(&now, &next_sync) < 0)
118		return NOTIFY_OK;
119
120	op.cmd = XENPF_settime64;
121	op.u.settime64.mbz = 0;
122	op.u.settime64.secs = now.tv_sec;
123	op.u.settime64.nsecs = now.tv_nsec;
124	op.u.settime64.system_time = timespec64_to_ns(&system_time);
125	(void)HYPERVISOR_platform_op(&op);
126
127	/*
128	 * Move the next drift compensation time 11 minutes
129	 * ahead. That's emulating the sync_cmos_clock() update for
130	 * the hardware RTC.
131	 */
132	next_sync = now;
133	next_sync.tv_sec += 11 * 60;
134
135	return NOTIFY_OK;
136}
137
138static struct notifier_block xen_pvclock_gtod_notifier = {
139	.notifier_call = xen_pvclock_gtod_notify,
140};
141
142static int xen_starting_cpu(unsigned int cpu)
143{
144	struct vcpu_register_vcpu_info info;
145	struct vcpu_info *vcpup;
146	int err;
147
148	/*
149	 * VCPUOP_register_vcpu_info cannot be called twice for the same
150	 * vcpu, so if vcpu_info is already registered, just get out. This
151	 * can happen with cpu-hotplug.
152	 */
153	if (per_cpu(xen_vcpu, cpu) != NULL)
154		goto after_register_vcpu_info;
155
156	pr_info("Xen: initializing cpu%d\n", cpu);
157	vcpup = per_cpu_ptr(xen_vcpu_info, cpu);
158
159	info.mfn = percpu_to_gfn(vcpup);
160	info.offset = xen_offset_in_page(vcpup);
161
162	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
163				 &info);
164	BUG_ON(err);
165	per_cpu(xen_vcpu, cpu) = vcpup;
166
167after_register_vcpu_info:
168	enable_percpu_irq(xen_events_irq, 0);
169	return 0;
170}
171
172static int xen_dying_cpu(unsigned int cpu)
173{
174	disable_percpu_irq(xen_events_irq);
175	return 0;
176}
177
178void xen_reboot(int reason)
179{
180	struct sched_shutdown r = { .reason = reason };
181	int rc;
182
183	rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
184	BUG_ON(rc);
185}
186
187static int xen_restart(struct notifier_block *nb, unsigned long action,
188		       void *data)
189{
190	xen_reboot(SHUTDOWN_reboot);
191
192	return NOTIFY_DONE;
193}
194
195static struct notifier_block xen_restart_nb = {
196	.notifier_call = xen_restart,
197	.priority = 192,
198};
199
200static void xen_power_off(void)
201{
202	xen_reboot(SHUTDOWN_poweroff);
203}
204
205static irqreturn_t xen_arm_callback(int irq, void *arg)
206{
207	xen_evtchn_do_upcall();
208	return IRQ_HANDLED;
209}
210
211static __initdata struct {
212	const char *compat;
213	const char *prefix;
214	const char *version;
215	bool found;
216} hyper_node = {"xen,xen", "xen,xen-", NULL, false};
217
218static int __init fdt_find_hyper_node(unsigned long node, const char *uname,
219				      int depth, void *data)
220{
221	const void *s = NULL;
222	int len;
223
224	if (depth != 1 || strcmp(uname, "hypervisor") != 0)
225		return 0;
226
227	if (of_flat_dt_is_compatible(node, hyper_node.compat))
228		hyper_node.found = true;
229
230	s = of_get_flat_dt_prop(node, "compatible", &len);
231	if (strlen(hyper_node.prefix) + 3  < len &&
232	    !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix)))
233		hyper_node.version = s + strlen(hyper_node.prefix);
234
235	/*
236	 * Check if Xen supports EFI by checking whether there is the
237	 * "/hypervisor/uefi" node in DT. If so, runtime services are available
238	 * through proxy functions (e.g. in case of Xen dom0 EFI implementation
239	 * they call special hypercall which executes relevant EFI functions)
240	 * and that is why they are always enabled.
241	 */
242	if (IS_ENABLED(CONFIG_XEN_EFI)) {
243		if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) &&
244		    !efi_runtime_disabled())
245			set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
246	}
247
248	return 0;
249}
250
251/*
252 * see Documentation/devicetree/bindings/arm/xen.txt for the
253 * documentation of the Xen Device Tree format.
254 */
255void __init xen_early_init(void)
256{
257	of_scan_flat_dt(fdt_find_hyper_node, NULL);
258	if (!hyper_node.found) {
259		pr_debug("No Xen support\n");
260		return;
261	}
262
263	if (hyper_node.version == NULL) {
264		pr_debug("Xen version not found\n");
265		return;
266	}
267
268	pr_info("Xen %s support found\n", hyper_node.version);
269
270	xen_domain_type = XEN_HVM_DOMAIN;
271
272	xen_setup_features();
273
274	if (xen_feature(XENFEAT_dom0))
275		xen_start_flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
276
277	if (!console_set_on_cmdline && !xen_initial_domain())
278		add_preferred_console("hvc", 0, NULL);
279}
280
281static void __init xen_acpi_guest_init(void)
282{
283#ifdef CONFIG_ACPI
284	struct xen_hvm_param a;
285	int interrupt, trigger, polarity;
286
287	a.domid = DOMID_SELF;
288	a.index = HVM_PARAM_CALLBACK_IRQ;
289
290	if (HYPERVISOR_hvm_op(HVMOP_get_param, &a)
291	    || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) {
292		xen_events_irq = 0;
293		return;
294	}
295
296	interrupt = a.value & 0xff;
297	trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
298					 : ACPI_LEVEL_SENSITIVE;
299	polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW
300					  : ACPI_ACTIVE_HIGH;
301	xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity);
302#endif
303}
304
305#ifdef CONFIG_XEN_UNPOPULATED_ALLOC
306/*
307 * A type-less specific Xen resource which contains extended regions
308 * (unused regions of guest physical address space provided by the hypervisor).
309 */
310static struct resource xen_resource = {
311	.name = "Xen unused space",
312};
313
314int __init arch_xen_unpopulated_init(struct resource **res)
315{
316	struct device_node *np;
317	struct resource *regs, *tmp_res;
318	uint64_t min_gpaddr = -1, max_gpaddr = 0;
319	unsigned int i, nr_reg = 0;
320	int rc;
321
322	if (!xen_domain())
323		return -ENODEV;
324
325	if (!acpi_disabled)
326		return -ENODEV;
327
328	np = of_find_compatible_node(NULL, NULL, "xen,xen");
329	if (WARN_ON(!np))
330		return -ENODEV;
331
332	/* Skip region 0 which is reserved for grant table space */
333	while (of_get_address(np, nr_reg + EXT_REGION_INDEX, NULL, NULL))
334		nr_reg++;
335
336	if (!nr_reg) {
337		pr_err("No extended regions are found\n");
338		of_node_put(np);
339		return -EINVAL;
340	}
341
342	regs = kcalloc(nr_reg, sizeof(*regs), GFP_KERNEL);
343	if (!regs) {
344		of_node_put(np);
345		return -ENOMEM;
346	}
347
348	/*
349	 * Create resource from extended regions provided by the hypervisor to be
350	 * used as unused address space for Xen scratch pages.
351	 */
352	for (i = 0; i < nr_reg; i++) {
353		rc = of_address_to_resource(np, i + EXT_REGION_INDEX, &regs[i]);
354		if (rc)
355			goto err;
356
357		if (max_gpaddr < regs[i].end)
358			max_gpaddr = regs[i].end;
359		if (min_gpaddr > regs[i].start)
360			min_gpaddr = regs[i].start;
361	}
362
363	xen_resource.start = min_gpaddr;
364	xen_resource.end = max_gpaddr;
365
366	/*
367	 * Mark holes between extended regions as unavailable. The rest of that
368	 * address space will be available for the allocation.
369	 */
370	for (i = 1; i < nr_reg; i++) {
371		resource_size_t start, end;
372
373		/* There is an overlap between regions */
374		if (regs[i - 1].end + 1 > regs[i].start) {
375			rc = -EINVAL;
376			goto err;
377		}
378
379		/* There is no hole between regions */
380		if (regs[i - 1].end + 1 == regs[i].start)
381			continue;
382
383		start = regs[i - 1].end + 1;
384		end = regs[i].start - 1;
385
386		tmp_res = kzalloc(sizeof(*tmp_res), GFP_KERNEL);
387		if (!tmp_res) {
388			rc = -ENOMEM;
389			goto err;
390		}
391
392		tmp_res->name = "Unavailable space";
393		tmp_res->start = start;
394		tmp_res->end = end;
395
396		rc = insert_resource(&xen_resource, tmp_res);
397		if (rc) {
398			pr_err("Cannot insert resource %pR (%d)\n", tmp_res, rc);
399			kfree(tmp_res);
400			goto err;
401		}
402	}
403
404	*res = &xen_resource;
405
406err:
407	of_node_put(np);
408	kfree(regs);
409	return rc;
410}
411#endif
412
413static void __init xen_dt_guest_init(void)
414{
415	struct device_node *xen_node;
416	struct resource res;
417
418	xen_node = of_find_compatible_node(NULL, NULL, "xen,xen");
419	if (!xen_node) {
420		pr_err("Xen support was detected before, but it has disappeared\n");
421		return;
422	}
423
424	xen_events_irq = irq_of_parse_and_map(xen_node, 0);
425
426	if (of_address_to_resource(xen_node, GRANT_TABLE_INDEX, &res)) {
427		pr_err("Xen grant table region is not found\n");
428		of_node_put(xen_node);
429		return;
430	}
431	of_node_put(xen_node);
432	xen_grant_frames = res.start;
433}
434
435static int __init xen_guest_init(void)
436{
437	struct xen_add_to_physmap xatp;
438	struct shared_info *shared_info_page = NULL;
439	int rc, cpu;
440
441	if (!xen_domain())
442		return 0;
443
444	if (IS_ENABLED(CONFIG_XEN_VIRTIO))
445		virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
446
447	if (!acpi_disabled)
448		xen_acpi_guest_init();
449	else
450		xen_dt_guest_init();
451
452	if (!xen_events_irq) {
453		pr_err("Xen event channel interrupt not found\n");
454		return -ENODEV;
455	}
456
457	/*
458	 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
459	 * parameters are found. Force enable runtime services.
460	 */
461	if (efi_enabled(EFI_RUNTIME_SERVICES))
462		xen_efi_runtime_setup();
463
464	shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL);
465
466	if (!shared_info_page) {
467		pr_err("not enough memory\n");
468		return -ENOMEM;
469	}
470	xatp.domid = DOMID_SELF;
471	xatp.idx = 0;
472	xatp.space = XENMAPSPACE_shared_info;
473	xatp.gpfn = virt_to_gfn(shared_info_page);
474	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
475		BUG();
476
477	HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
478
479	/* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
480	 * page, we use it in the event channel upcall and in some pvclock
481	 * related functions.
482	 * The shared info contains exactly 1 CPU (the boot CPU). The guest
483	 * is required to use VCPUOP_register_vcpu_info to place vcpu info
484	 * for secondary CPUs as they are brought up.
485	 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
486	 */
487	xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info),
488				       1 << fls(sizeof(struct vcpu_info) - 1));
489	if (xen_vcpu_info == NULL)
490		return -ENOMEM;
491
492	/* Direct vCPU id mapping for ARM guests. */
493	for_each_possible_cpu(cpu)
494		per_cpu(xen_vcpu_id, cpu) = cpu;
495
496	if (!xen_grant_frames) {
497		xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames();
498		rc = xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn,
499										   &xen_auto_xlat_grant_frames.vaddr,
500										   xen_auto_xlat_grant_frames.count);
501	} else
502		rc = gnttab_setup_auto_xlat_frames(xen_grant_frames);
503	if (rc) {
504		free_percpu(xen_vcpu_info);
505		return rc;
506	}
507	gnttab_init();
508
509	/*
510	 * Making sure board specific code will not set up ops for
511	 * cpu idle and cpu freq.
512	 */
513	disable_cpuidle();
514	disable_cpufreq();
515
516	xen_init_IRQ();
517
518	if (request_percpu_irq(xen_events_irq, xen_arm_callback,
519			       "events", &xen_vcpu)) {
520		pr_err("Error request IRQ %d\n", xen_events_irq);
521		return -EINVAL;
522	}
523
524	if (xen_initial_domain())
525		pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
526
527	return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
528				 "arm/xen:starting", xen_starting_cpu,
529				 xen_dying_cpu);
530}
531early_initcall(xen_guest_init);
532
533static int xen_starting_runstate_cpu(unsigned int cpu)
534{
535	xen_setup_runstate_info(cpu);
536	return 0;
537}
538
539static int __init xen_late_init(void)
540{
541	if (!xen_domain())
542		return -ENODEV;
543
544	pm_power_off = xen_power_off;
545	register_restart_handler(&xen_restart_nb);
546	if (!xen_initial_domain()) {
547		struct timespec64 ts;
548		xen_read_wallclock(&ts);
549		do_settimeofday64(&ts);
550	}
551
552	if (xen_kernel_unmapped_at_usr())
553		return 0;
554
555	xen_time_setup_guest();
556
557	return cpuhp_setup_state(CPUHP_AP_ARM_XEN_RUNSTATE_STARTING,
558				 "arm/xen_runstate:starting",
559				 xen_starting_runstate_cpu, NULL);
560}
561late_initcall(xen_late_init);
562
563
564/* empty stubs */
565void xen_arch_pre_suspend(void) { }
566void xen_arch_post_suspend(int suspend_cancelled) { }
567void xen_timer_resume(void) { }
568void xen_arch_resume(void) { }
569void xen_arch_suspend(void) { }
570
571
572/* In the hypercall.S file. */
573EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op);
574EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op);
575EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version);
576EXPORT_SYMBOL_GPL(HYPERVISOR_console_io);
577EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op);
578EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op);
579EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op);
580EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op);
581EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op);
582EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op_raw);
583EXPORT_SYMBOL_GPL(HYPERVISOR_multicall);
584EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist);
585EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op);
586EXPORT_SYMBOL_GPL(privcmd_call);
587