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