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
41 static struct start_info _xen_start_info;
42 struct start_info *xen_start_info = &_xen_start_info;
43 EXPORT_SYMBOL(xen_start_info);
44
45 enum xen_domain_type xen_domain_type = XEN_NATIVE;
46 EXPORT_SYMBOL(xen_domain_type);
47
48 struct shared_info xen_dummy_shared_info;
49 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
50
51 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
52 static struct vcpu_info __percpu *xen_vcpu_info;
53
54 /* Linux <-> Xen vCPU id mapping */
55 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
56 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
57
58 /* These are unused until we support booting "pre-ballooned" */
59 unsigned long xen_released_pages;
60 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
61
62 static __read_mostly unsigned int xen_events_irq;
63 static __read_mostly phys_addr_t xen_grant_frames;
64
65 #define GRANT_TABLE_INDEX 0
66 #define EXT_REGION_INDEX 1
67
68 uint32_t xen_start_flags;
69 EXPORT_SYMBOL(xen_start_flags);
70
xen_unmap_domain_gfn_range(struct vm_area_struct * vma,int nr,struct page ** pages)71 int 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 }
76 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range);
77
xen_read_wallclock(struct timespec64 * ts)78 static 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
xen_pvclock_gtod_notify(struct notifier_block * nb,unsigned long was_set,void * priv)99 static 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
138 static struct notifier_block xen_pvclock_gtod_notifier = {
139 .notifier_call = xen_pvclock_gtod_notify,
140 };
141
xen_starting_cpu(unsigned int cpu)142 static 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
167 if (!xen_kernel_unmapped_at_usr())
168 xen_setup_runstate_info(cpu);
169
170 after_register_vcpu_info:
171 enable_percpu_irq(xen_events_irq, 0);
172 return 0;
173 }
174
xen_dying_cpu(unsigned int cpu)175 static int xen_dying_cpu(unsigned int cpu)
176 {
177 disable_percpu_irq(xen_events_irq);
178 return 0;
179 }
180
xen_reboot(int reason)181 void xen_reboot(int reason)
182 {
183 struct sched_shutdown r = { .reason = reason };
184 int rc;
185
186 rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
187 BUG_ON(rc);
188 }
189
xen_restart(struct notifier_block * nb,unsigned long action,void * data)190 static int xen_restart(struct notifier_block *nb, unsigned long action,
191 void *data)
192 {
193 xen_reboot(SHUTDOWN_reboot);
194
195 return NOTIFY_DONE;
196 }
197
198 static struct notifier_block xen_restart_nb = {
199 .notifier_call = xen_restart,
200 .priority = 192,
201 };
202
xen_power_off(void)203 static void xen_power_off(void)
204 {
205 xen_reboot(SHUTDOWN_poweroff);
206 }
207
xen_arm_callback(int irq,void * arg)208 static irqreturn_t xen_arm_callback(int irq, void *arg)
209 {
210 xen_hvm_evtchn_do_upcall();
211 return IRQ_HANDLED;
212 }
213
214 static __initdata struct {
215 const char *compat;
216 const char *prefix;
217 const char *version;
218 bool found;
219 } hyper_node = {"xen,xen", "xen,xen-", NULL, false};
220
fdt_find_hyper_node(unsigned long node,const char * uname,int depth,void * data)221 static int __init fdt_find_hyper_node(unsigned long node, const char *uname,
222 int depth, void *data)
223 {
224 const void *s = NULL;
225 int len;
226
227 if (depth != 1 || strcmp(uname, "hypervisor") != 0)
228 return 0;
229
230 if (of_flat_dt_is_compatible(node, hyper_node.compat))
231 hyper_node.found = true;
232
233 s = of_get_flat_dt_prop(node, "compatible", &len);
234 if (strlen(hyper_node.prefix) + 3 < len &&
235 !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix)))
236 hyper_node.version = s + strlen(hyper_node.prefix);
237
238 /*
239 * Check if Xen supports EFI by checking whether there is the
240 * "/hypervisor/uefi" node in DT. If so, runtime services are available
241 * through proxy functions (e.g. in case of Xen dom0 EFI implementation
242 * they call special hypercall which executes relevant EFI functions)
243 * and that is why they are always enabled.
244 */
245 if (IS_ENABLED(CONFIG_XEN_EFI)) {
246 if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) &&
247 !efi_runtime_disabled())
248 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
249 }
250
251 return 0;
252 }
253
254 /*
255 * see Documentation/devicetree/bindings/arm/xen.txt for the
256 * documentation of the Xen Device Tree format.
257 */
xen_early_init(void)258 void __init xen_early_init(void)
259 {
260 of_scan_flat_dt(fdt_find_hyper_node, NULL);
261 if (!hyper_node.found) {
262 pr_debug("No Xen support\n");
263 return;
264 }
265
266 if (hyper_node.version == NULL) {
267 pr_debug("Xen version not found\n");
268 return;
269 }
270
271 pr_info("Xen %s support found\n", hyper_node.version);
272
273 xen_domain_type = XEN_HVM_DOMAIN;
274
275 xen_setup_features();
276
277 if (xen_feature(XENFEAT_dom0))
278 xen_start_flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
279
280 if (!console_set_on_cmdline && !xen_initial_domain())
281 add_preferred_console("hvc", 0, NULL);
282 }
283
xen_acpi_guest_init(void)284 static void __init xen_acpi_guest_init(void)
285 {
286 #ifdef CONFIG_ACPI
287 struct xen_hvm_param a;
288 int interrupt, trigger, polarity;
289
290 a.domid = DOMID_SELF;
291 a.index = HVM_PARAM_CALLBACK_IRQ;
292
293 if (HYPERVISOR_hvm_op(HVMOP_get_param, &a)
294 || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) {
295 xen_events_irq = 0;
296 return;
297 }
298
299 interrupt = a.value & 0xff;
300 trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
301 : ACPI_LEVEL_SENSITIVE;
302 polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW
303 : ACPI_ACTIVE_HIGH;
304 xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity);
305 #endif
306 }
307
308 #ifdef CONFIG_XEN_UNPOPULATED_ALLOC
309 /*
310 * A type-less specific Xen resource which contains extended regions
311 * (unused regions of guest physical address space provided by the hypervisor).
312 */
313 static struct resource xen_resource = {
314 .name = "Xen unused space",
315 };
316
arch_xen_unpopulated_init(struct resource ** res)317 int __init arch_xen_unpopulated_init(struct resource **res)
318 {
319 struct device_node *np;
320 struct resource *regs, *tmp_res;
321 uint64_t min_gpaddr = -1, max_gpaddr = 0;
322 unsigned int i, nr_reg = 0;
323 int rc;
324
325 if (!xen_domain())
326 return -ENODEV;
327
328 if (!acpi_disabled)
329 return -ENODEV;
330
331 np = of_find_compatible_node(NULL, NULL, "xen,xen");
332 if (WARN_ON(!np))
333 return -ENODEV;
334
335 /* Skip region 0 which is reserved for grant table space */
336 while (of_get_address(np, nr_reg + EXT_REGION_INDEX, NULL, NULL))
337 nr_reg++;
338
339 if (!nr_reg) {
340 pr_err("No extended regions are found\n");
341 of_node_put(np);
342 return -EINVAL;
343 }
344
345 regs = kcalloc(nr_reg, sizeof(*regs), GFP_KERNEL);
346 if (!regs) {
347 of_node_put(np);
348 return -ENOMEM;
349 }
350
351 /*
352 * Create resource from extended regions provided by the hypervisor to be
353 * used as unused address space for Xen scratch pages.
354 */
355 for (i = 0; i < nr_reg; i++) {
356 rc = of_address_to_resource(np, i + EXT_REGION_INDEX, ®s[i]);
357 if (rc)
358 goto err;
359
360 if (max_gpaddr < regs[i].end)
361 max_gpaddr = regs[i].end;
362 if (min_gpaddr > regs[i].start)
363 min_gpaddr = regs[i].start;
364 }
365
366 xen_resource.start = min_gpaddr;
367 xen_resource.end = max_gpaddr;
368
369 /*
370 * Mark holes between extended regions as unavailable. The rest of that
371 * address space will be available for the allocation.
372 */
373 for (i = 1; i < nr_reg; i++) {
374 resource_size_t start, end;
375
376 /* There is an overlap between regions */
377 if (regs[i - 1].end + 1 > regs[i].start) {
378 rc = -EINVAL;
379 goto err;
380 }
381
382 /* There is no hole between regions */
383 if (regs[i - 1].end + 1 == regs[i].start)
384 continue;
385
386 start = regs[i - 1].end + 1;
387 end = regs[i].start - 1;
388
389 tmp_res = kzalloc(sizeof(*tmp_res), GFP_KERNEL);
390 if (!tmp_res) {
391 rc = -ENOMEM;
392 goto err;
393 }
394
395 tmp_res->name = "Unavailable space";
396 tmp_res->start = start;
397 tmp_res->end = end;
398
399 rc = insert_resource(&xen_resource, tmp_res);
400 if (rc) {
401 pr_err("Cannot insert resource %pR (%d)\n", tmp_res, rc);
402 kfree(tmp_res);
403 goto err;
404 }
405 }
406
407 *res = &xen_resource;
408
409 err:
410 of_node_put(np);
411 kfree(regs);
412 return rc;
413 }
414 #endif
415
xen_dt_guest_init(void)416 static void __init xen_dt_guest_init(void)
417 {
418 struct device_node *xen_node;
419 struct resource res;
420
421 xen_node = of_find_compatible_node(NULL, NULL, "xen,xen");
422 if (!xen_node) {
423 pr_err("Xen support was detected before, but it has disappeared\n");
424 return;
425 }
426
427 xen_events_irq = irq_of_parse_and_map(xen_node, 0);
428
429 if (of_address_to_resource(xen_node, GRANT_TABLE_INDEX, &res)) {
430 pr_err("Xen grant table region is not found\n");
431 of_node_put(xen_node);
432 return;
433 }
434 of_node_put(xen_node);
435 xen_grant_frames = res.start;
436 }
437
xen_guest_init(void)438 static int __init xen_guest_init(void)
439 {
440 struct xen_add_to_physmap xatp;
441 struct shared_info *shared_info_page = NULL;
442 int rc, cpu;
443
444 if (!xen_domain())
445 return 0;
446
447 if (IS_ENABLED(CONFIG_XEN_VIRTIO))
448 virtio_set_mem_acc_cb(xen_virtio_mem_acc);
449
450 if (!acpi_disabled)
451 xen_acpi_guest_init();
452 else
453 xen_dt_guest_init();
454
455 if (!xen_events_irq) {
456 pr_err("Xen event channel interrupt not found\n");
457 return -ENODEV;
458 }
459
460 /*
461 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
462 * parameters are found. Force enable runtime services.
463 */
464 if (efi_enabled(EFI_RUNTIME_SERVICES))
465 xen_efi_runtime_setup();
466
467 shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL);
468
469 if (!shared_info_page) {
470 pr_err("not enough memory\n");
471 return -ENOMEM;
472 }
473 xatp.domid = DOMID_SELF;
474 xatp.idx = 0;
475 xatp.space = XENMAPSPACE_shared_info;
476 xatp.gpfn = virt_to_gfn(shared_info_page);
477 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
478 BUG();
479
480 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
481
482 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
483 * page, we use it in the event channel upcall and in some pvclock
484 * related functions.
485 * The shared info contains exactly 1 CPU (the boot CPU). The guest
486 * is required to use VCPUOP_register_vcpu_info to place vcpu info
487 * for secondary CPUs as they are brought up.
488 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
489 */
490 xen_vcpu_info = alloc_percpu(struct vcpu_info);
491 if (xen_vcpu_info == NULL)
492 return -ENOMEM;
493
494 /* Direct vCPU id mapping for ARM guests. */
495 for_each_possible_cpu(cpu)
496 per_cpu(xen_vcpu_id, cpu) = cpu;
497
498 if (!xen_grant_frames) {
499 xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames();
500 rc = xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn,
501 &xen_auto_xlat_grant_frames.vaddr,
502 xen_auto_xlat_grant_frames.count);
503 } else
504 rc = gnttab_setup_auto_xlat_frames(xen_grant_frames);
505 if (rc) {
506 free_percpu(xen_vcpu_info);
507 return rc;
508 }
509 gnttab_init();
510
511 /*
512 * Making sure board specific code will not set up ops for
513 * cpu idle and cpu freq.
514 */
515 disable_cpuidle();
516 disable_cpufreq();
517
518 xen_init_IRQ();
519
520 if (request_percpu_irq(xen_events_irq, xen_arm_callback,
521 "events", &xen_vcpu)) {
522 pr_err("Error request IRQ %d\n", xen_events_irq);
523 return -EINVAL;
524 }
525
526 if (!xen_kernel_unmapped_at_usr())
527 xen_time_setup_guest();
528
529 if (xen_initial_domain())
530 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
531
532 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
533 "arm/xen:starting", xen_starting_cpu,
534 xen_dying_cpu);
535 }
536 early_initcall(xen_guest_init);
537
xen_pm_init(void)538 static int __init xen_pm_init(void)
539 {
540 if (!xen_domain())
541 return -ENODEV;
542
543 pm_power_off = xen_power_off;
544 register_restart_handler(&xen_restart_nb);
545 if (!xen_initial_domain()) {
546 struct timespec64 ts;
547 xen_read_wallclock(&ts);
548 do_settimeofday64(&ts);
549 }
550
551 return 0;
552 }
553 late_initcall(xen_pm_init);
554
555
556 /* empty stubs */
xen_arch_pre_suspend(void)557 void xen_arch_pre_suspend(void) { }
xen_arch_post_suspend(int suspend_cancelled)558 void xen_arch_post_suspend(int suspend_cancelled) { }
xen_timer_resume(void)559 void xen_timer_resume(void) { }
xen_arch_resume(void)560 void xen_arch_resume(void) { }
xen_arch_suspend(void)561 void xen_arch_suspend(void) { }
562
563
564 /* In the hypercall.S file. */
565 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op);
566 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op);
567 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version);
568 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io);
569 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op);
570 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op);
571 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op);
572 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op);
573 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op);
574 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op_raw);
575 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall);
576 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist);
577 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op);
578 EXPORT_SYMBOL_GPL(privcmd_call);
579