2 #include <xen/events.h>
3 #include <xen/grant_table.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>
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>
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_remap_domain_gfn_array(struct vm_area_struct
*vma
,
67 xen_pfn_t
*gfn
, int nr
,
68 int *err_ptr
, pgprot_t prot
,
72 return xen_xlate_remap_gfn_array(vma
, addr
, gfn
, nr
, err_ptr
,
75 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_array
);
77 /* Not used by XENFEAT_auto_translated guests. */
78 int xen_remap_domain_gfn_range(struct vm_area_struct
*vma
,
80 xen_pfn_t gfn
, int nr
,
81 pgprot_t prot
, unsigned domid
,
86 EXPORT_SYMBOL_GPL(xen_remap_domain_gfn_range
);
88 int xen_unmap_domain_gfn_range(struct vm_area_struct
*vma
,
89 int nr
, struct page
**pages
)
91 return xen_xlate_unmap_gfn_range(vma
, nr
, pages
);
93 EXPORT_SYMBOL_GPL(xen_unmap_domain_gfn_range
);
95 /* Not used by XENFEAT_auto_translated guests. */
96 int xen_remap_domain_mfn_array(struct vm_area_struct
*vma
,
98 xen_pfn_t
*mfn
, int nr
,
99 int *err_ptr
, pgprot_t prot
,
100 unsigned int domid
, struct page
**pages
)
104 EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_array
);
106 static void xen_read_wallclock(struct timespec64
*ts
)
109 struct timespec64 now
, ts_monotonic
;
110 struct shared_info
*s
= HYPERVISOR_shared_info
;
111 struct pvclock_wall_clock
*wall_clock
= &(s
->wc
);
113 /* get wallclock at system boot */
115 version
= wall_clock
->version
;
116 rmb(); /* fetch version before time */
117 now
.tv_sec
= ((uint64_t)wall_clock
->sec_hi
<< 32) | wall_clock
->sec
;
118 now
.tv_nsec
= wall_clock
->nsec
;
119 rmb(); /* fetch time before checking version */
120 } while ((wall_clock
->version
& 1) || (version
!= wall_clock
->version
));
122 /* time since system boot */
123 ktime_get_ts64(&ts_monotonic
);
124 *ts
= timespec64_add(now
, ts_monotonic
);
127 static int xen_pvclock_gtod_notify(struct notifier_block
*nb
,
128 unsigned long was_set
, void *priv
)
130 /* Protected by the calling core code serialization */
131 static struct timespec64 next_sync
;
133 struct xen_platform_op op
;
134 struct timespec64 now
, system_time
;
135 struct timekeeper
*tk
= priv
;
137 now
.tv_sec
= tk
->xtime_sec
;
138 now
.tv_nsec
= (long)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
139 system_time
= timespec64_add(now
, tk
->wall_to_monotonic
);
142 * We only take the expensive HV call when the clock was set
143 * or when the 11 minutes RTC synchronization time elapsed.
145 if (!was_set
&& timespec64_compare(&now
, &next_sync
) < 0)
148 op
.cmd
= XENPF_settime64
;
149 op
.u
.settime64
.mbz
= 0;
150 op
.u
.settime64
.secs
= now
.tv_sec
;
151 op
.u
.settime64
.nsecs
= now
.tv_nsec
;
152 op
.u
.settime64
.system_time
= timespec64_to_ns(&system_time
);
153 (void)HYPERVISOR_platform_op(&op
);
156 * Move the next drift compensation time 11 minutes
157 * ahead. That's emulating the sync_cmos_clock() update for
161 next_sync
.tv_sec
+= 11 * 60;
166 static struct notifier_block xen_pvclock_gtod_notifier
= {
167 .notifier_call
= xen_pvclock_gtod_notify
,
170 static int xen_starting_cpu(unsigned int cpu
)
172 struct vcpu_register_vcpu_info info
;
173 struct vcpu_info
*vcpup
;
177 * VCPUOP_register_vcpu_info cannot be called twice for the same
178 * vcpu, so if vcpu_info is already registered, just get out. This
179 * can happen with cpu-hotplug.
181 if (per_cpu(xen_vcpu
, cpu
) != NULL
)
182 goto after_register_vcpu_info
;
184 pr_info("Xen: initializing cpu%d\n", cpu
);
185 vcpup
= per_cpu_ptr(xen_vcpu_info
, cpu
);
187 info
.mfn
= virt_to_gfn(vcpup
);
188 info
.offset
= xen_offset_in_page(vcpup
);
190 err
= HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info
, xen_vcpu_nr(cpu
),
193 per_cpu(xen_vcpu
, cpu
) = vcpup
;
195 xen_setup_runstate_info(cpu
);
197 after_register_vcpu_info
:
198 enable_percpu_irq(xen_events_irq
, 0);
202 static int xen_dying_cpu(unsigned int cpu
)
204 disable_percpu_irq(xen_events_irq
);
208 void xen_reboot(int reason
)
210 struct sched_shutdown r
= { .reason
= reason
};
213 rc
= HYPERVISOR_sched_op(SCHEDOP_shutdown
, &r
);
217 static void xen_restart(enum reboot_mode reboot_mode
, const char *cmd
)
219 xen_reboot(SHUTDOWN_reboot
);
223 static void xen_power_off(void)
225 xen_reboot(SHUTDOWN_poweroff
);
228 static irqreturn_t
xen_arm_callback(int irq
, void *arg
)
230 xen_hvm_evtchn_do_upcall();
234 static __initdata
struct {
239 } hyper_node
= {"xen,xen", "xen,xen-", NULL
, false};
241 static int __init
fdt_find_hyper_node(unsigned long node
, const char *uname
,
242 int depth
, void *data
)
244 const void *s
= NULL
;
247 if (depth
!= 1 || strcmp(uname
, "hypervisor") != 0)
250 if (of_flat_dt_is_compatible(node
, hyper_node
.compat
))
251 hyper_node
.found
= true;
253 s
= of_get_flat_dt_prop(node
, "compatible", &len
);
254 if (strlen(hyper_node
.prefix
) + 3 < len
&&
255 !strncmp(hyper_node
.prefix
, s
, strlen(hyper_node
.prefix
)))
256 hyper_node
.version
= s
+ strlen(hyper_node
.prefix
);
259 * Check if Xen supports EFI by checking whether there is the
260 * "/hypervisor/uefi" node in DT. If so, runtime services are available
261 * through proxy functions (e.g. in case of Xen dom0 EFI implementation
262 * they call special hypercall which executes relevant EFI functions)
263 * and that is why they are always enabled.
265 if (IS_ENABLED(CONFIG_XEN_EFI
)) {
266 if ((of_get_flat_dt_subnode_by_name(node
, "uefi") > 0) &&
267 !efi_runtime_disabled())
268 set_bit(EFI_RUNTIME_SERVICES
, &efi
.flags
);
275 * see Documentation/devicetree/bindings/arm/xen.txt for the
276 * documentation of the Xen Device Tree format.
278 #define GRANT_TABLE_PHYSADDR 0
279 void __init
xen_early_init(void)
281 of_scan_flat_dt(fdt_find_hyper_node
, NULL
);
282 if (!hyper_node
.found
) {
283 pr_debug("No Xen support\n");
287 if (hyper_node
.version
== NULL
) {
288 pr_debug("Xen version not found\n");
292 pr_info("Xen %s support found\n", hyper_node
.version
);
294 xen_domain_type
= XEN_HVM_DOMAIN
;
296 xen_setup_features();
298 if (xen_feature(XENFEAT_dom0
))
299 xen_start_flags
|= SIF_INITDOMAIN
|SIF_PRIVILEGED
;
301 if (!console_set_on_cmdline
&& !xen_initial_domain())
302 add_preferred_console("hvc", 0, NULL
);
305 static void __init
xen_acpi_guest_init(void)
308 struct xen_hvm_param a
;
309 int interrupt
, trigger
, polarity
;
311 a
.domid
= DOMID_SELF
;
312 a
.index
= HVM_PARAM_CALLBACK_IRQ
;
314 if (HYPERVISOR_hvm_op(HVMOP_get_param
, &a
)
315 || (a
.value
>> 56) != HVM_PARAM_CALLBACK_TYPE_PPI
) {
320 interrupt
= a
.value
& 0xff;
321 trigger
= ((a
.value
>> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
322 : ACPI_LEVEL_SENSITIVE
;
323 polarity
= ((a
.value
>> 8) & 0x2) ? ACPI_ACTIVE_LOW
325 xen_events_irq
= acpi_register_gsi(NULL
, interrupt
, trigger
, polarity
);
329 static void __init
xen_dt_guest_init(void)
331 struct device_node
*xen_node
;
333 xen_node
= of_find_compatible_node(NULL
, NULL
, "xen,xen");
335 pr_err("Xen support was detected before, but it has disappeared\n");
339 xen_events_irq
= irq_of_parse_and_map(xen_node
, 0);
342 static int __init
xen_guest_init(void)
344 struct xen_add_to_physmap xatp
;
345 struct shared_info
*shared_info_page
= NULL
;
352 xen_acpi_guest_init();
356 if (!xen_events_irq
) {
357 pr_err("Xen event channel interrupt not found\n");
362 * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
363 * parameters are found. Force enable runtime services.
365 if (efi_enabled(EFI_RUNTIME_SERVICES
))
366 xen_efi_runtime_setup();
368 shared_info_page
= (struct shared_info
*)get_zeroed_page(GFP_KERNEL
);
370 if (!shared_info_page
) {
371 pr_err("not enough memory\n");
374 xatp
.domid
= DOMID_SELF
;
376 xatp
.space
= XENMAPSPACE_shared_info
;
377 xatp
.gpfn
= virt_to_gfn(shared_info_page
);
378 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap
, &xatp
))
381 HYPERVISOR_shared_info
= (struct shared_info
*)shared_info_page
;
383 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
384 * page, we use it in the event channel upcall and in some pvclock
386 * The shared info contains exactly 1 CPU (the boot CPU). The guest
387 * is required to use VCPUOP_register_vcpu_info to place vcpu info
388 * for secondary CPUs as they are brought up.
389 * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
391 xen_vcpu_info
= alloc_percpu(struct vcpu_info
);
392 if (xen_vcpu_info
== NULL
)
395 /* Direct vCPU id mapping for ARM guests. */
396 for_each_possible_cpu(cpu
)
397 per_cpu(xen_vcpu_id
, cpu
) = cpu
;
399 xen_auto_xlat_grant_frames
.count
= gnttab_max_grant_frames();
400 if (xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames
.pfn
,
401 &xen_auto_xlat_grant_frames
.vaddr
,
402 xen_auto_xlat_grant_frames
.count
)) {
403 free_percpu(xen_vcpu_info
);
407 if (!xen_initial_domain())
411 * Making sure board specific code will not set up ops for
412 * cpu idle and cpu freq.
419 if (request_percpu_irq(xen_events_irq
, xen_arm_callback
,
420 "events", &xen_vcpu
)) {
421 pr_err("Error request IRQ %d\n", xen_events_irq
);
425 xen_time_setup_guest();
427 if (xen_initial_domain())
428 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier
);
430 return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING
,
431 "arm/xen:starting", xen_starting_cpu
,
434 early_initcall(xen_guest_init
);
436 static int __init
xen_pm_init(void)
441 pm_power_off
= xen_power_off
;
442 arm_pm_restart
= xen_restart
;
443 if (!xen_initial_domain()) {
444 struct timespec64 ts
;
445 xen_read_wallclock(&ts
);
446 do_settimeofday64(&ts
);
451 late_initcall(xen_pm_init
);
455 void xen_arch_pre_suspend(void) { }
456 void xen_arch_post_suspend(int suspend_cancelled
) { }
457 void xen_timer_resume(void) { }
458 void xen_arch_resume(void) { }
459 void xen_arch_suspend(void) { }
462 /* In the hypercall.S file. */
463 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op
);
464 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op
);
465 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version
);
466 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io
);
467 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op
);
468 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op
);
469 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op
);
470 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op
);
471 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op
);
472 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op
);
473 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op
);
474 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall
);
475 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist
);
476 EXPORT_SYMBOL_GPL(HYPERVISOR_dm_op
);
477 EXPORT_SYMBOL_GPL(privcmd_call
);