1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <xen/events.h>
4 #include <xen/grant_table.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>
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>
20 #include <linux/interrupt.h>
21 #include <linux/irqreturn.h>
22 #include <linux/module.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>
39 struct start_info _xen_start_info
;
40 struct start_info
*xen_start_info
= &_xen_start_info
;
41 EXPORT_SYMBOL(xen_start_info
);
43 enum xen_domain_type xen_domain_type
= XEN_NATIVE
;
44 EXPORT_SYMBOL(xen_domain_type
);
46 struct shared_info xen_dummy_shared_info
;
47 struct shared_info
*HYPERVISOR_shared_info
= (void *)&xen_dummy_shared_info
;
49 DEFINE_PER_CPU(struct vcpu_info
*, xen_vcpu
);
50 static struct vcpu_info __percpu
*xen_vcpu_info
;
52 /* Linux <-> Xen vCPU id mapping */
53 DEFINE_PER_CPU(uint32_t, xen_vcpu_id
);
54 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id
);
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
;
60 static __read_mostly
unsigned int xen_events_irq
;
62 uint32_t xen_start_flags
;
63 EXPORT_SYMBOL(xen_start_flags
);
65 int xen_unmap_domain_gfn_range(struct vm_area_struct
*vma
,
66 int nr
, struct page
**pages
)
68 return xen_xlate_unmap_gfn_range(vma
, nr
, pages
);
70 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range
);
72 static void xen_read_wallclock(struct timespec64
*ts
)
75 struct timespec64 now
, ts_monotonic
;
76 struct shared_info
*s
= HYPERVISOR_shared_info
;
77 struct pvclock_wall_clock
*wall_clock
= &(s
->wc
);
79 /* get wallclock at system boot */
81 version
= wall_clock
->version
;
82 rmb(); /* fetch version before time */
83 now
.tv_sec
= ((uint64_t)wall_clock
->sec_hi
<< 32) | wall_clock
->sec
;
84 now
.tv_nsec
= wall_clock
->nsec
;
85 rmb(); /* fetch time before checking version */
86 } while ((wall_clock
->version
& 1) || (version
!= wall_clock
->version
));
88 /* time since system boot */
89 ktime_get_ts64(&ts_monotonic
);
90 *ts
= timespec64_add(now
, ts_monotonic
);
93 static int xen_pvclock_gtod_notify(struct notifier_block
*nb
,
94 unsigned long was_set
, void *priv
)
96 /* Protected by the calling core code serialization */
97 static struct timespec64 next_sync
;
99 struct xen_platform_op op
;
100 struct timespec64 now
, system_time
;
101 struct timekeeper
*tk
= priv
;
103 now
.tv_sec
= tk
->xtime_sec
;
104 now
.tv_nsec
= (long)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
105 system_time
= timespec64_add(now
, tk
->wall_to_monotonic
);
108 * We only take the expensive HV call when the clock was set
109 * or when the 11 minutes RTC synchronization time elapsed.
111 if (!was_set
&& timespec64_compare(&now
, &next_sync
) < 0)
114 op
.cmd
= XENPF_settime64
;
115 op
.u
.settime64
.mbz
= 0;
116 op
.u
.settime64
.secs
= now
.tv_sec
;
117 op
.u
.settime64
.nsecs
= now
.tv_nsec
;
118 op
.u
.settime64
.system_time
= timespec64_to_ns(&system_time
);
119 (void)HYPERVISOR_platform_op(&op
);
122 * Move the next drift compensation time 11 minutes
123 * ahead. That's emulating the sync_cmos_clock() update for
127 next_sync
.tv_sec
+= 11 * 60;
132 static struct notifier_block xen_pvclock_gtod_notifier
= {
133 .notifier_call
= xen_pvclock_gtod_notify
,
136 static int xen_starting_cpu(unsigned int cpu
)
138 struct vcpu_register_vcpu_info info
;
139 struct vcpu_info
*vcpup
;
143 * VCPUOP_register_vcpu_info cannot be called twice for the same
144 * vcpu, so if vcpu_info is already registered, just get out. This
145 * can happen with cpu-hotplug.
147 if (per_cpu(xen_vcpu
, cpu
) != NULL
)
148 goto after_register_vcpu_info
;
150 pr_info("Xen: initializing cpu%d\n", cpu
);
151 vcpup
= per_cpu_ptr(xen_vcpu_info
, cpu
);
153 info
.mfn
= virt_to_gfn(vcpup
);
154 info
.offset
= xen_offset_in_page(vcpup
);
156 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, xen_vcpu_nr(cpu
),
159 per_cpu(xen_vcpu
, cpu
) = vcpup
;
161 xen_setup_runstate_info(cpu
);
163 after_register_vcpu_info
:
164 enable_percpu_irq(xen_events_irq
, 0);
168 static int xen_dying_cpu(unsigned int cpu
)
170 disable_percpu_irq(xen_events_irq
);
174 void xen_reboot(int reason
)
176 struct sched_shutdown r
= { .reason
= reason
};
179 rc
= HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
);
183 static void xen_restart(enum reboot_mode reboot_mode
, const char *cmd
)
185 xen_reboot(SHUTDOWN_reboot
);
189 static void xen_power_off(void)
191 xen_reboot(SHUTDOWN_poweroff
);
194 static irqreturn_t
xen_arm_callback(int irq
, void *arg
)
196 xen_hvm_evtchn_do_upcall();
200 static __initdata
struct {
205 } hyper_node
= {"xen,xen", "xen,xen-", NULL
, false};
207 static int __init
fdt_find_hyper_node(unsigned long node
, const char *uname
,
208 int depth
, void *data
)
210 const void *s
= NULL
;
213 if (depth
!= 1 || strcmp(uname
, "hypervisor") != 0)
216 if (of_flat_dt_is_compatible(node
, hyper_node
.compat
))
217 hyper_node
.found
= true;
219 s
= of_get_flat_dt_prop(node
, "compatible", &len
);
220 if (strlen(hyper_node
.prefix
) + 3 < len
&&
221 !strncmp(hyper_node
.prefix
, s
, strlen(hyper_node
.prefix
)))
222 hyper_node
.version
= s
+ strlen(hyper_node
.prefix
);
225 * Check if Xen supports EFI by checking whether there is the
226 * "/hypervisor/uefi" node in DT. If so, runtime services are available
227 * through proxy functions (e.g. in case of Xen dom0 EFI implementation
228 * they call special hypercall which executes relevant EFI functions)
229 * and that is why they are always enabled.
231 if (IS_ENABLED(CONFIG_XEN_EFI
)) {
232 if ((of_get_flat_dt_subnode_by_name(node
, "uefi") > 0) &&
233 !efi_runtime_disabled())
234 set_bit(EFI_RUNTIME_SERVICES
, &efi
.flags
);
241 * see Documentation/devicetree/bindings/arm/xen.txt for the
242 * documentation of the Xen Device Tree format.
244 #define GRANT_TABLE_PHYSADDR 0
245 void __init
xen_early_init(void)
247 of_scan_flat_dt(fdt_find_hyper_node
, NULL
);
248 if (!hyper_node
.found
) {
249 pr_debug("No Xen support\n");
253 if (hyper_node
.version
== NULL
) {
254 pr_debug("Xen version not found\n");
258 pr_info("Xen %s support found\n", hyper_node
.version
);
260 xen_domain_type
= XEN_HVM_DOMAIN
;
262 xen_setup_features();
264 if (xen_feature(XENFEAT_dom0
))
265 xen_start_flags
|= SIF_INITDOMAIN
|SIF_PRIVILEGED
;
267 if (!console_set_on_cmdline
&& !xen_initial_domain())
268 add_preferred_console("hvc", 0, NULL
);
271 static void __init
xen_acpi_guest_init(void)
274 struct xen_hvm_param a
;
275 int interrupt
, trigger
, polarity
;
277 a
.domid
= DOMID_SELF
;
278 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
280 if (HYPERVISOR_hvm_op(HVMOP_get_param
, &a
)
281 || (a
.value
>> 56) != HVM_PARAM_CALLBACK_TYPE_PPI
) {
286 interrupt
= a
.value
& 0xff;
287 trigger
= ((a
.value
>> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
288 : ACPI_LEVEL_SENSITIVE
;
289 polarity
= ((a
.value
>> 8) & 0x2) ? ACPI_ACTIVE_LOW
291 xen_events_irq
= acpi_register_gsi(NULL
, interrupt
, trigger
, polarity
);
295 static void __init
xen_dt_guest_init(void)
297 struct device_node
*xen_node
;
299 xen_node
= of_find_compatible_node(NULL
, NULL
, "xen,xen");
301 pr_err("Xen support was detected before, but it has disappeared\n");
305 xen_events_irq
= irq_of_parse_and_map(xen_node
, 0);
308 static int __init
xen_guest_init(void)
310 struct xen_add_to_physmap xatp
;
311 struct shared_info
*shared_info_page
= NULL
;
318 xen_acpi_guest_init();
322 if (!xen_events_irq
) {
323 pr_err("Xen event channel interrupt not found\n");
328 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
329 * parameters are found. Force enable runtime services.
331 if (efi_enabled(EFI_RUNTIME_SERVICES
))
332 xen_efi_runtime_setup();
334 shared_info_page
= (struct shared_info
*)get_zeroed_page(GFP_KERNEL
);
336 if (!shared_info_page
) {
337 pr_err("not enough memory\n");
340 xatp
.domid
= DOMID_SELF
;
342 xatp
.space
= XENMAPSPACE_shared_info
;
343 xatp
.gpfn
= virt_to_gfn(shared_info_page
);
344 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
347 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
349 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
350 * page, we use it in the event channel upcall and in some pvclock
352 * The shared info contains exactly 1 CPU (the boot CPU). The guest
353 * is required to use VCPUOP_register_vcpu_info to place vcpu info
354 * for secondary CPUs as they are brought up.
355 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
357 xen_vcpu_info
= alloc_percpu(struct vcpu_info
);
358 if (xen_vcpu_info
== NULL
)
361 /* Direct vCPU id mapping for ARM guests. */
362 for_each_possible_cpu(cpu
)
363 per_cpu(xen_vcpu_id
, cpu
) = cpu
;
365 xen_auto_xlat_grant_frames
.count
= gnttab_max_grant_frames();
366 if (xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames
.pfn
,
367 &xen_auto_xlat_grant_frames
.vaddr
,
368 xen_auto_xlat_grant_frames
.count
)) {
369 free_percpu(xen_vcpu_info
);
373 if (!xen_initial_domain())
377 * Making sure board specific code will not set up ops for
378 * cpu idle and cpu freq.
385 if (request_percpu_irq(xen_events_irq
, xen_arm_callback
,
386 "events", &xen_vcpu
)) {
387 pr_err("Error request IRQ %d\n", xen_events_irq
);
391 xen_time_setup_guest();
393 if (xen_initial_domain())
394 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier
);
396 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING
,
397 "arm/xen:starting", xen_starting_cpu
,
400 early_initcall(xen_guest_init
);
402 static int __init
xen_pm_init(void)
407 pm_power_off
= xen_power_off
;
408 arm_pm_restart
= xen_restart
;
409 if (!xen_initial_domain()) {
410 struct timespec64 ts
;
411 xen_read_wallclock(&ts
);
412 do_settimeofday64(&ts
);
417 late_initcall(xen_pm_init
);
421 void xen_arch_pre_suspend(void) { }
422 void xen_arch_post_suspend(int suspend_cancelled
) { }
423 void xen_timer_resume(void) { }
424 void xen_arch_resume(void) { }
425 void xen_arch_suspend(void) { }
428 /* In the hypercall.S file. */
429 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op
);
430 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op
);
431 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version
);
432 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io
);
433 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op
);
434 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op
);
435 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op
);
436 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op
);
437 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op
);
438 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op
);
439 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op_raw
);
440 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall
);
441 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist
);
442 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op
);
443 EXPORT_SYMBOL_GPL(privcmd_call
);