fs: Avoid premature clearing of capabilities
[cascardo/linux.git] / arch / arm / xen / enlighten.c
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(int, xen_vcpu_id) = -1;
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         /* Direct vCPU id mapping for ARM guests. */
174         per_cpu(xen_vcpu_id, cpu) = cpu;
175
176         info.mfn = virt_to_gfn(vcpup);
177         info.offset = xen_offset_in_page(vcpup);
178
179         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
180                                  &info);
181         BUG_ON(err);
182         per_cpu(xen_vcpu, cpu) = vcpup;
183
184         xen_setup_runstate_info(cpu);
185
186 after_register_vcpu_info:
187         enable_percpu_irq(xen_events_irq, 0);
188         return 0;
189 }
190
191 static int xen_dying_cpu(unsigned int cpu)
192 {
193         disable_percpu_irq(xen_events_irq);
194         return 0;
195 }
196
197 static void xen_restart(enum reboot_mode reboot_mode, const char *cmd)
198 {
199         struct sched_shutdown r = { .reason = SHUTDOWN_reboot };
200         int rc;
201         rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
202         BUG_ON(rc);
203 }
204
205 static void xen_power_off(void)
206 {
207         struct sched_shutdown r = { .reason = SHUTDOWN_poweroff };
208         int rc;
209         rc = HYPERVISOR_sched_op(SCHEDOP_shutdown, &r);
210         BUG_ON(rc);
211 }
212
213 static irqreturn_t xen_arm_callback(int irq, void *arg)
214 {
215         xen_hvm_evtchn_do_upcall();
216         return IRQ_HANDLED;
217 }
218
219 static __initdata struct {
220         const char *compat;
221         const char *prefix;
222         const char *version;
223         bool found;
224 } hyper_node = {"xen,xen", "xen,xen-", NULL, false};
225
226 static int __init fdt_find_hyper_node(unsigned long node, const char *uname,
227                                       int depth, void *data)
228 {
229         const void *s = NULL;
230         int len;
231
232         if (depth != 1 || strcmp(uname, "hypervisor") != 0)
233                 return 0;
234
235         if (of_flat_dt_is_compatible(node, hyper_node.compat))
236                 hyper_node.found = true;
237
238         s = of_get_flat_dt_prop(node, "compatible", &len);
239         if (strlen(hyper_node.prefix) + 3  < len &&
240             !strncmp(hyper_node.prefix, s, strlen(hyper_node.prefix)))
241                 hyper_node.version = s + strlen(hyper_node.prefix);
242
243         /*
244          * Check if Xen supports EFI by checking whether there is the
245          * "/hypervisor/uefi" node in DT. If so, runtime services are available
246          * through proxy functions (e.g. in case of Xen dom0 EFI implementation
247          * they call special hypercall which executes relevant EFI functions)
248          * and that is why they are always enabled.
249          */
250         if (IS_ENABLED(CONFIG_XEN_EFI)) {
251                 if ((of_get_flat_dt_subnode_by_name(node, "uefi") > 0) &&
252                     !efi_runtime_disabled())
253                         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
254         }
255
256         return 0;
257 }
258
259 /*
260  * see Documentation/devicetree/bindings/arm/xen.txt for the
261  * documentation of the Xen Device Tree format.
262  */
263 #define GRANT_TABLE_PHYSADDR 0
264 void __init xen_early_init(void)
265 {
266         of_scan_flat_dt(fdt_find_hyper_node, NULL);
267         if (!hyper_node.found) {
268                 pr_debug("No Xen support\n");
269                 return;
270         }
271
272         if (hyper_node.version == NULL) {
273                 pr_debug("Xen version not found\n");
274                 return;
275         }
276
277         pr_info("Xen %s support found\n", hyper_node.version);
278
279         xen_domain_type = XEN_HVM_DOMAIN;
280
281         xen_setup_features();
282
283         if (xen_feature(XENFEAT_dom0))
284                 xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
285         else
286                 xen_start_info->flags &= ~(SIF_INITDOMAIN|SIF_PRIVILEGED);
287
288         if (!console_set_on_cmdline && !xen_initial_domain())
289                 add_preferred_console("hvc", 0, NULL);
290 }
291
292 static void __init xen_acpi_guest_init(void)
293 {
294 #ifdef CONFIG_ACPI
295         struct xen_hvm_param a;
296         int interrupt, trigger, polarity;
297
298         a.domid = DOMID_SELF;
299         a.index = HVM_PARAM_CALLBACK_IRQ;
300
301         if (HYPERVISOR_hvm_op(HVMOP_get_param, &a)
302             || (a.value >> 56) != HVM_PARAM_CALLBACK_TYPE_PPI) {
303                 xen_events_irq = 0;
304                 return;
305         }
306
307         interrupt = a.value & 0xff;
308         trigger = ((a.value >> 8) & 0x1) ? ACPI_EDGE_SENSITIVE
309                                          : ACPI_LEVEL_SENSITIVE;
310         polarity = ((a.value >> 8) & 0x2) ? ACPI_ACTIVE_LOW
311                                           : ACPI_ACTIVE_HIGH;
312         xen_events_irq = acpi_register_gsi(NULL, interrupt, trigger, polarity);
313 #endif
314 }
315
316 static void __init xen_dt_guest_init(void)
317 {
318         struct device_node *xen_node;
319
320         xen_node = of_find_compatible_node(NULL, NULL, "xen,xen");
321         if (!xen_node) {
322                 pr_err("Xen support was detected before, but it has disappeared\n");
323                 return;
324         }
325
326         xen_events_irq = irq_of_parse_and_map(xen_node, 0);
327 }
328
329 static int __init xen_guest_init(void)
330 {
331         struct xen_add_to_physmap xatp;
332         struct shared_info *shared_info_page = NULL;
333
334         if (!xen_domain())
335                 return 0;
336
337         if (!acpi_disabled)
338                 xen_acpi_guest_init();
339         else
340                 xen_dt_guest_init();
341
342         if (!xen_events_irq) {
343                 pr_err("Xen event channel interrupt not found\n");
344                 return -ENODEV;
345         }
346
347         /*
348          * The fdt parsing codes have set EFI_RUNTIME_SERVICES if Xen EFI
349          * parameters are found. Force enable runtime services.
350          */
351         if (efi_enabled(EFI_RUNTIME_SERVICES))
352                 xen_efi_runtime_setup();
353
354         shared_info_page = (struct shared_info *)get_zeroed_page(GFP_KERNEL);
355
356         if (!shared_info_page) {
357                 pr_err("not enough memory\n");
358                 return -ENOMEM;
359         }
360         xatp.domid = DOMID_SELF;
361         xatp.idx = 0;
362         xatp.space = XENMAPSPACE_shared_info;
363         xatp.gpfn = virt_to_gfn(shared_info_page);
364         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
365                 BUG();
366
367         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
368
369         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
370          * page, we use it in the event channel upcall and in some pvclock
371          * related functions. 
372          * The shared info contains exactly 1 CPU (the boot CPU). The guest
373          * is required to use VCPUOP_register_vcpu_info to place vcpu info
374          * for secondary CPUs as they are brought up.
375          * For uniformity we use VCPUOP_register_vcpu_info even on cpu0.
376          */
377         xen_vcpu_info = __alloc_percpu(sizeof(struct vcpu_info),
378                                                sizeof(struct vcpu_info));
379         if (xen_vcpu_info == NULL)
380                 return -ENOMEM;
381
382         /* Direct vCPU id mapping for ARM guests. */
383         per_cpu(xen_vcpu_id, 0) = 0;
384
385         xen_auto_xlat_grant_frames.count = gnttab_max_grant_frames();
386         if (xen_xlate_map_ballooned_pages(&xen_auto_xlat_grant_frames.pfn,
387                                           &xen_auto_xlat_grant_frames.vaddr,
388                                           xen_auto_xlat_grant_frames.count)) {
389                 free_percpu(xen_vcpu_info);
390                 return -ENOMEM;
391         }
392         gnttab_init();
393         if (!xen_initial_domain())
394                 xenbus_probe(NULL);
395
396         /*
397          * Making sure board specific code will not set up ops for
398          * cpu idle and cpu freq.
399          */
400         disable_cpuidle();
401         disable_cpufreq();
402
403         xen_init_IRQ();
404
405         if (request_percpu_irq(xen_events_irq, xen_arm_callback,
406                                "events", &xen_vcpu)) {
407                 pr_err("Error request IRQ %d\n", xen_events_irq);
408                 return -EINVAL;
409         }
410
411         xen_time_setup_guest();
412
413         if (xen_initial_domain())
414                 pvclock_gtod_register_notifier(&xen_pvclock_gtod_notifier);
415
416         return cpuhp_setup_state(CPUHP_AP_ARM_XEN_STARTING,
417                                  "AP_ARM_XEN_STARTING", xen_starting_cpu,
418                                  xen_dying_cpu);
419 }
420 early_initcall(xen_guest_init);
421
422 static int __init xen_pm_init(void)
423 {
424         if (!xen_domain())
425                 return -ENODEV;
426
427         pm_power_off = xen_power_off;
428         arm_pm_restart = xen_restart;
429         if (!xen_initial_domain()) {
430                 struct timespec64 ts;
431                 xen_read_wallclock(&ts);
432                 do_settimeofday64(&ts);
433         }
434
435         return 0;
436 }
437 late_initcall(xen_pm_init);
438
439
440 /* empty stubs */
441 void xen_arch_pre_suspend(void) { }
442 void xen_arch_post_suspend(int suspend_cancelled) { }
443 void xen_timer_resume(void) { }
444 void xen_arch_resume(void) { }
445 void xen_arch_suspend(void) { }
446
447
448 /* In the hypercall.S file. */
449 EXPORT_SYMBOL_GPL(HYPERVISOR_event_channel_op);
450 EXPORT_SYMBOL_GPL(HYPERVISOR_grant_table_op);
451 EXPORT_SYMBOL_GPL(HYPERVISOR_xen_version);
452 EXPORT_SYMBOL_GPL(HYPERVISOR_console_io);
453 EXPORT_SYMBOL_GPL(HYPERVISOR_sched_op);
454 EXPORT_SYMBOL_GPL(HYPERVISOR_hvm_op);
455 EXPORT_SYMBOL_GPL(HYPERVISOR_memory_op);
456 EXPORT_SYMBOL_GPL(HYPERVISOR_physdev_op);
457 EXPORT_SYMBOL_GPL(HYPERVISOR_vcpu_op);
458 EXPORT_SYMBOL_GPL(HYPERVISOR_tmem_op);
459 EXPORT_SYMBOL_GPL(HYPERVISOR_platform_op);
460 EXPORT_SYMBOL_GPL(HYPERVISOR_multicall);
461 EXPORT_SYMBOL_GPL(HYPERVISOR_vm_assist);
462 EXPORT_SYMBOL_GPL(privcmd_call);