xref: /openbmc/linux/arch/arm/xen/enlighten.c (revision 799a545b)
1 #include <xen/xen.h>
2 #include <xen/events.h>
3 #include <xen/grant_table.h>
4 #include <xen/hvm.h>
5 #include <xen/interface/vcpu.h>
6 #include <xen/interface/xen.h>
7 #include <xen/interface/memory.h>
8 #include <xen/interface/hvm/params.h>
9 #include <xen/features.h>
10 #include <xen/platform_pci.h>
11 #include <xen/xenbus.h>
12 #include <xen/page.h>
13 #include <xen/interface/sched.h>
14 #include <xen/xen-ops.h>
15 #include <asm/xen/hypervisor.h>
16 #include <asm/xen/hypercall.h>
17 #include <asm/xen/xen-ops.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/time64.h>
33 #include <linux/timekeeping.h>
34 #include <linux/timekeeper_internal.h>
35 #include <linux/acpi.h>
36 
37 #include <linux/mm.h>
38 
39 struct start_info _xen_start_info;
40 struct start_info *xen_start_info = &_xen_start_info;
41 EXPORT_SYMBOL(xen_start_info);
42 
43 enum xen_domain_type xen_domain_type = XEN_NATIVE;
44 EXPORT_SYMBOL(xen_domain_type);
45 
46 struct shared_info xen_dummy_shared_info;
47 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
48 
49 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
50 static struct vcpu_info __percpu *xen_vcpu_info;
51 
52 /* Linux <-> Xen vCPU id mapping */
53 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
54 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
55 
56 /* These are unused until we support booting "pre-ballooned" */
57 unsigned long xen_released_pages;
58 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
59 
60 static __read_mostly unsigned int xen_events_irq;
61 
62 int xen_remap_domain_gfn_array(struct vm_area_struct *vma,
63 			       unsigned long addr,
64 			       xen_pfn_t *gfn, int nr,
65 			       int *err_ptr, pgprot_t prot,
66 			       unsigned domid,
67 			       struct page **pages)
68 {
69 	return xen_xlate_remap_gfn_array(vma, addr, gfn, nr, err_ptr,
70 					 prot, domid, pages);
71 }
72 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array);
73 
74 /* Not used by XENFEAT_auto_translated guests. */
75 int xen_remap_domain_gfn_range(struct vm_area_struct *vma,
76                               unsigned long addr,
77                               xen_pfn_t gfn, int nr,
78                               pgprot_t prot, unsigned domid,
79                               struct page **pages)
80 {
81 	return -ENOSYS;
82 }
83 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range);
84 
85 int xen_unmap_domain_gfn_range(struct vm_area_struct *vma,
86 			       int nr, struct page **pages)
87 {
88 	return xen_xlate_unmap_gfn_range(vma, nr, pages);
89 }
90 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range);
91 
92 static void xen_read_wallclock(struct timespec64 *ts)
93 {
94 	u32 version;
95 	struct timespec64 now, ts_monotonic;
96 	struct shared_info *s = HYPERVISOR_shared_info;
97 	struct pvclock_wall_clock *wall_clock = &(s->wc);
98 
99 	/* get wallclock at system boot */
100 	do {
101 		version = wall_clock->version;
102 		rmb();		/* fetch version before time */
103 		now.tv_sec  = ((uint64_t)wall_clock->sec_hi << 32) | wall_clock->sec;
104 		now.tv_nsec = wall_clock->nsec;
105 		rmb();		/* fetch time before checking version */
106 	} while ((wall_clock->version & 1) || (version != wall_clock->version));
107 
108 	/* time since system boot */
109 	ktime_get_ts64(&ts_monotonic);
110 	*ts = timespec64_add(now, ts_monotonic);
111 }
112 
113 static int xen_pvclock_gtod_notify(struct notifier_block *nb,
114 				   unsigned long was_set, void *priv)
115 {
116 	/* Protected by the calling core code serialization */
117 	static struct timespec64 next_sync;
118 
119 	struct xen_platform_op op;
120 	struct timespec64 now, system_time;
121 	struct timekeeper *tk = priv;
122 
123 	now.tv_sec = tk->xtime_sec;
124 	now.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
125 	system_time = timespec64_add(now, tk->wall_to_monotonic);
126 
127 	/*
128 	 * We only take the expensive HV call when the clock was set
129 	 * or when the 11 minutes RTC synchronization time elapsed.
130 	 */
131 	if (!was_set && timespec64_compare(&now, &next_sync) < 0)
132 		return NOTIFY_OK;
133 
134 	op.cmd = XENPF_settime64;
135 	op.u.settime64.mbz = 0;
136 	op.u.settime64.secs = now.tv_sec;
137 	op.u.settime64.nsecs = now.tv_nsec;
138 	op.u.settime64.system_time = timespec64_to_ns(&system_time);
139 	(void)HYPERVISOR_platform_op(&op);
140 
141 	/*
142 	 * Move the next drift compensation time 11 minutes
143 	 * ahead. That's emulating the sync_cmos_clock() update for
144 	 * the hardware RTC.
145 	 */
146 	next_sync = now;
147 	next_sync.tv_sec += 11 * 60;
148 
149 	return NOTIFY_OK;
150 }
151 
152 static struct notifier_block xen_pvclock_gtod_notifier = {
153 	.notifier_call = xen_pvclock_gtod_notify,
154 };
155 
156 static int xen_starting_cpu(unsigned int cpu)
157 {
158 	struct vcpu_register_vcpu_info info;
159 	struct vcpu_info *vcpup;
160 	int err;
161 
162 	/*
163 	 * VCPUOP_register_vcpu_info cannot be called twice for the same
164 	 * vcpu, so if vcpu_info is already registered, just get out. This
165 	 * can happen with cpu-hotplug.
166 	 */
167 	if (per_cpu(xen_vcpu, cpu) != NULL)
168 		goto after_register_vcpu_info;
169 
170 	pr_info("Xen: initializing cpu%d\n", cpu);
171 	vcpup = per_cpu_ptr(xen_vcpu_info, cpu);
172 
173 	info.mfn = virt_to_gfn(vcpup);
174 	info.offset = xen_offset_in_page(vcpup);
175 
176 	err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
177 				 &info);
178 	BUG_ON(err);
179 	per_cpu(xen_vcpu, cpu) = vcpup;
180 
181 	xen_setup_runstate_info(cpu);
182 
183 after_register_vcpu_info:
184 	enable_percpu_irq(xen_events_irq, 0);
185 	return 0;
186 }
187 
188 static int xen_dying_cpu(unsigned int cpu)
189 {
190 	disable_percpu_irq(xen_events_irq);
191 	return 0;
192 }
193 
194 static void xen_restart(enum reboot_mode reboot_mode, const char *cmd)
195 {
196 	struct sched_shutdown r = { .reason = SHUTDOWN_reboot };
197 	int rc;
198 	rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
199 	BUG_ON(rc);
200 }
201 
202 static void xen_power_off(void)
203 {
204 	struct sched_shutdown r = { .reason = SHUTDOWN_poweroff };
205 	int rc;
206 	rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
207 	BUG_ON(rc);
208 }
209 
210 static irqreturn_t xen_arm_callback(int irq, void *arg)
211 {
212 	xen_hvm_evtchn_do_upcall();
213 	return IRQ_HANDLED;
214 }
215 
216 static __initdata struct {
217 	const char *compat;
218 	const char *prefix;
219 	const char *version;
220 	bool found;
221 } hyper_node = {"xen,xen", "xen,xen-", NULL, false};
222 
223 static int __init fdt_find_hyper_node(unsigned long node, const char *uname,
224 				      int depth, void *data)
225 {
226 	const void *s = NULL;
227 	int len;
228 
229 	if (depth != 1 || strcmp(uname, "hypervisor") != 0)
230 		return 0;
231 
232 	if (of_flat_dt_is_compatible(node, hyper_node.compat))
233 		hyper_node.found = true;
234 
235 	s = of_get_flat_dt_prop(node, "compatible", &len);
236 	if (strlen(hyper_node.prefix) + 3  < len &&
237 	    !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix)))
238 		hyper_node.version = s + strlen(hyper_node.prefix);
239 
240 	/*
241 	 * Check if Xen supports EFI by checking whether there is the
242 	 * "/hypervisor/uefi" node in DT. If so, runtime services are available
243 	 * through proxy functions (e.g. in case of Xen dom0 EFI implementation
244 	 * they call special hypercall which executes relevant EFI functions)
245 	 * and that is why they are always enabled.
246 	 */
247 	if (IS_ENABLED(CONFIG_XEN_EFI)) {
248 		if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) &&
249 		    !efi_runtime_disabled())
250 			set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
251 	}
252 
253 	return 0;
254 }
255 
256 /*
257  * see Documentation/devicetree/bindings/arm/xen.txt for the
258  * documentation of the Xen Device Tree format.
259  */
260 #define GRANT_TABLE_PHYSADDR 0
261 void __init xen_early_init(void)
262 {
263 	of_scan_flat_dt(fdt_find_hyper_node, NULL);
264 	if (!hyper_node.found) {
265 		pr_debug("No Xen support\n");
266 		return;
267 	}
268 
269 	if (hyper_node.version == NULL) {
270 		pr_debug("Xen version not found\n");
271 		return;
272 	}
273 
274 	pr_info("Xen %s support found\n", hyper_node.version);
275 
276 	xen_domain_type = XEN_HVM_DOMAIN;
277 
278 	xen_setup_features();
279 
280 	if (xen_feature(XENFEAT_dom0))
281 		xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
282 	else
283 		xen_start_info->flags &= ~(SIF_INITDOMAIN|SIF_PRIVILEGED);
284 
285 	if (!console_set_on_cmdline && !xen_initial_domain())
286 		add_preferred_console("hvc", 0, NULL);
287 }
288 
289 static void __init xen_acpi_guest_init(void)
290 {
291 #ifdef CONFIG_ACPI
292 	struct xen_hvm_param a;
293 	int interrupt, trigger, polarity;
294 
295 	a.domid = DOMID_SELF;
296 	a.index = HVM_PARAM_CALLBACK_IRQ;
297 
298 	if (HYPERVISOR_hvm_op(HVMOP_get_param, &a)
299 	    || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) {
300 		xen_events_irq = 0;
301 		return;
302 	}
303 
304 	interrupt = a.value & 0xff;
305 	trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
306 					 : ACPI_LEVEL_SENSITIVE;
307 	polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW
308 					  : ACPI_ACTIVE_HIGH;
309 	xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity);
310 #endif
311 }
312 
313 static void __init xen_dt_guest_init(void)
314 {
315 	struct device_node *xen_node;
316 
317 	xen_node = of_find_compatible_node(NULL, NULL, "xen,xen");
318 	if (!xen_node) {
319 		pr_err("Xen support was detected before, but it has disappeared\n");
320 		return;
321 	}
322 
323 	xen_events_irq = irq_of_parse_and_map(xen_node, 0);
324 }
325 
326 static int __init xen_guest_init(void)
327 {
328 	struct xen_add_to_physmap xatp;
329 	struct shared_info *shared_info_page = NULL;
330 	int cpu;
331 
332 	if (!xen_domain())
333 		return 0;
334 
335 	if (!acpi_disabled)
336 		xen_acpi_guest_init();
337 	else
338 		xen_dt_guest_init();
339 
340 	if (!xen_events_irq) {
341 		pr_err("Xen event channel interrupt not found\n");
342 		return -ENODEV;
343 	}
344 
345 	/*
346 	 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
347 	 * parameters are found. Force enable runtime services.
348 	 */
349 	if (efi_enabled(EFI_RUNTIME_SERVICES))
350 		xen_efi_runtime_setup();
351 
352 	shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL);
353 
354 	if (!shared_info_page) {
355 		pr_err("not enough memory\n");
356 		return -ENOMEM;
357 	}
358 	xatp.domid = DOMID_SELF;
359 	xatp.idx = 0;
360 	xatp.space = XENMAPSPACE_shared_info;
361 	xatp.gpfn = virt_to_gfn(shared_info_page);
362 	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
363 		BUG();
364 
365 	HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
366 
367 	/* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
368 	 * page, we use it in the event channel upcall and in some pvclock
369 	 * related functions.
370 	 * The shared info contains exactly 1 CPU (the boot CPU). The guest
371 	 * is required to use VCPUOP_register_vcpu_info to place vcpu info
372 	 * for secondary CPUs as they are brought up.
373 	 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
374 	 */
375 	xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info),
376 			                       sizeof(struct vcpu_info));
377 	if (xen_vcpu_info == NULL)
378 		return -ENOMEM;
379 
380 	/* Direct vCPU id mapping for ARM guests. */
381 	for_each_possible_cpu(cpu)
382 		per_cpu(xen_vcpu_id, cpu) = cpu;
383 
384 	xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames();
385 	if (xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn,
386 					  &xen_auto_xlat_grant_frames.vaddr,
387 					  xen_auto_xlat_grant_frames.count)) {
388 		free_percpu(xen_vcpu_info);
389 		return -ENOMEM;
390 	}
391 	gnttab_init();
392 	if (!xen_initial_domain())
393 		xenbus_probe(NULL);
394 
395 	/*
396 	 * Making sure board specific code will not set up ops for
397 	 * cpu idle and cpu freq.
398 	 */
399 	disable_cpuidle();
400 	disable_cpufreq();
401 
402 	xen_init_IRQ();
403 
404 	if (request_percpu_irq(xen_events_irq, xen_arm_callback,
405 			       "events", &xen_vcpu)) {
406 		pr_err("Error request IRQ %d\n", xen_events_irq);
407 		return -EINVAL;
408 	}
409 
410 	xen_time_setup_guest();
411 
412 	if (xen_initial_domain())
413 		pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
414 
415 	return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
416 				 "AP_ARM_XEN_STARTING", xen_starting_cpu,
417 				 xen_dying_cpu);
418 }
419 early_initcall(xen_guest_init);
420 
421 static int __init xen_pm_init(void)
422 {
423 	if (!xen_domain())
424 		return -ENODEV;
425 
426 	pm_power_off = xen_power_off;
427 	arm_pm_restart = xen_restart;
428 	if (!xen_initial_domain()) {
429 		struct timespec64 ts;
430 		xen_read_wallclock(&ts);
431 		do_settimeofday64(&ts);
432 	}
433 
434 	return 0;
435 }
436 late_initcall(xen_pm_init);
437 
438 
439 /* empty stubs */
440 void xen_arch_pre_suspend(void) { }
441 void xen_arch_post_suspend(int suspend_cancelled) { }
442 void xen_timer_resume(void) { }
443 void xen_arch_resume(void) { }
444 void xen_arch_suspend(void) { }
445 
446 
447 /* In the hypercall.S file. */
448 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op);
449 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op);
450 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version);
451 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io);
452 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op);
453 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op);
454 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op);
455 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op);
456 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op);
457 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op);
458 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op);
459 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall);
460 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist);
461 EXPORT_SYMBOL_GPL(privcmd_call);
462