1#include <linux/cpu.h>
2#include <linux/kexec.h>
3
4#include <xen/features.h>
5#include <xen/page.h>
6
7#include <asm/xen/hypercall.h>
8#include <asm/xen/hypervisor.h>
9#include <asm/cpu.h>
10#include <asm/e820/api.h>
11
12#include "xen-ops.h"
13#include "smp.h"
14#include "pmu.h"
15
16EXPORT_SYMBOL_GPL(hypercall_page);
17
18/*
19 * Pointer to the xen_vcpu_info structure or
20 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
21 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
22 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
23 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
24 * acknowledge pending events.
25 * Also more subtly it is used by the patched version of irq enable/disable
26 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
27 *
28 * The desire to be able to do those mask/unmask operations as a single
29 * instruction by using the per-cpu offset held in %gs is the real reason
30 * vcpu info is in a per-cpu pointer and the original reason for this
31 * hypercall.
32 *
33 */
34DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
35
36/*
37 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
38 * hypercall. This can be used both in PV and PVHVM mode. The structure
39 * overrides the default per_cpu(xen_vcpu, cpu) value.
40 */
41DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
42
43/* Linux <-> Xen vCPU id mapping */
44DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
45EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
46
47enum xen_domain_type xen_domain_type = XEN_NATIVE;
48EXPORT_SYMBOL_GPL(xen_domain_type);
49
50unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
51EXPORT_SYMBOL(machine_to_phys_mapping);
52unsigned long machine_to_phys_nr;
53EXPORT_SYMBOL(machine_to_phys_nr);
54
55struct start_info *xen_start_info;
56EXPORT_SYMBOL_GPL(xen_start_info);
57
58struct shared_info xen_dummy_shared_info;
59
60__read_mostly int xen_have_vector_callback;
61EXPORT_SYMBOL_GPL(xen_have_vector_callback);
62
63/*
64 * Point at some empty memory to start with. We map the real shared_info
65 * page as soon as fixmap is up and running.
66 */
67struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
68
69/*
70 * Flag to determine whether vcpu info placement is available on all
71 * VCPUs. We assume it is to start with, and then set it to zero on
72 * the first failure. This is because it can succeed on some VCPUs
73 * and not others, since it can involve hypervisor memory allocation,
74 * or because the guest failed to guarantee all the appropriate
75 * constraints on all VCPUs (ie buffer can't cross a page boundary).
76 *
77 * Note that any particular CPU may be using a placed vcpu structure,
78 * but we can only optimise if the all are.
79 *
80 * 0: not available, 1: available
81 */
82int xen_have_vcpu_info_placement = 1;
83
84static int xen_cpu_up_online(unsigned int cpu)
85{
86 xen_init_lock_cpu(cpu);
87 return 0;
88}
89
90int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
91 int (*cpu_dead_cb)(unsigned int))
92{
93 int rc;
94
95 rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
96 "x86/xen/hvm_guest:prepare",
97 cpu_up_prepare_cb, cpu_dead_cb);
98 if (rc >= 0) {
99 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
100 "x86/xen/hvm_guest:online",
101 xen_cpu_up_online, NULL);
102 if (rc < 0)
103 cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
104 }
105
106 return rc >= 0 ? 0 : rc;
107}
108
109static void clamp_max_cpus(void)
110{
111#ifdef CONFIG_SMP
112 if (setup_max_cpus > MAX_VIRT_CPUS)
113 setup_max_cpus = MAX_VIRT_CPUS;
114#endif
115}
116
117void xen_vcpu_setup(int cpu)
118{
119 struct vcpu_register_vcpu_info info;
120 int err;
121 struct vcpu_info *vcpup;
122
123 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
124
125 /*
126 * This path is called twice on PVHVM - first during bootup via
127 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
128 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
129 * As we can only do the VCPUOP_register_vcpu_info once lets
130 * not over-write its result.
131 *
132 * For PV it is called during restore (xen_vcpu_restore) and bootup
133 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
134 * use this function.
135 */
136 if (xen_hvm_domain()) {
137 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
138 return;
139 }
140 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
141 per_cpu(xen_vcpu, cpu) =
142 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
143
144 if (!xen_have_vcpu_info_placement) {
145 if (cpu >= MAX_VIRT_CPUS)
146 clamp_max_cpus();
147 return;
148 }
149
150 vcpup = &per_cpu(xen_vcpu_info, cpu);
151 info.mfn = arbitrary_virt_to_mfn(vcpup);
152 info.offset = offset_in_page(vcpup);
153
154 /* Check to see if the hypervisor will put the vcpu_info
155 structure where we want it, which allows direct access via
156 a percpu-variable.
157 N.B. This hypercall can _only_ be called once per CPU. Subsequent
158 calls will error out with -EINVAL. This is due to the fact that
159 hypervisor has no unregister variant and this hypercall does not
160 allow to over-write info.mfn and info.offset.
161 */
162 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
163 &info);
164
165 if (err) {
166 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
167 xen_have_vcpu_info_placement = 0;
168 clamp_max_cpus();
169 } else {
170 /* This cpu is using the registered vcpu info, even if
171 later ones fail to. */
172 per_cpu(xen_vcpu, cpu) = vcpup;
173 }
174}
175
176void xen_reboot(int reason)
177{
178 struct sched_shutdown r = { .reason = reason };
179 int cpu;
180
181 for_each_online_cpu(cpu)
182 xen_pmu_finish(cpu);
183
184 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
185 BUG();
186}
187
188void xen_emergency_restart(void)
189{
190 xen_reboot(SHUTDOWN_reboot);
191}
192
193static int
194xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
195{
196 if (!kexec_crash_loaded())
197 xen_reboot(SHUTDOWN_crash);
198 return NOTIFY_DONE;
199}
200
201static struct notifier_block xen_panic_block = {
202 .notifier_call = xen_panic_event,
203 .priority = INT_MIN
204};
205
206int xen_panic_handler_init(void)
207{
208 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
209 return 0;
210}
211
212void xen_pin_vcpu(int cpu)
213{
214 static bool disable_pinning;
215 struct sched_pin_override pin_override;
216 int ret;
217
218 if (disable_pinning)
219 return;
220
221 pin_override.pcpu = cpu;
222 ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
223
224 /* Ignore errors when removing override. */
225 if (cpu < 0)
226 return;
227
228 switch (ret) {
229 case -ENOSYS:
230 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
231 cpu);
232 disable_pinning = true;
233 break;
234 case -EPERM:
235 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
236 disable_pinning = true;
237 break;
238 case -EINVAL:
239 case -EBUSY:
240 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
241 cpu);
242 break;
243 case 0:
244 break;
245 default:
246 WARN(1, "rc %d while trying to pin vcpu\n", ret);
247 disable_pinning = true;
248 }
249}
250
251#ifdef CONFIG_HOTPLUG_CPU
252void xen_arch_register_cpu(int num)
253{
254 arch_register_cpu(num);
255}
256EXPORT_SYMBOL(xen_arch_register_cpu);
257
258void xen_arch_unregister_cpu(int num)
259{
260 arch_unregister_cpu(num);
261}
262EXPORT_SYMBOL(xen_arch_unregister_cpu);
263#endif
264