xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 8dda2eac)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
4 #include <linux/memblock.h>
5 #endif
6 #include <linux/cpu.h>
7 #include <linux/kexec.h>
8 #include <linux/slab.h>
9 #include <linux/panic_notifier.h>
10 
11 #include <xen/xen.h>
12 #include <xen/features.h>
13 #include <xen/page.h>
14 
15 #include <asm/xen/hypercall.h>
16 #include <asm/xen/hypervisor.h>
17 #include <asm/cpu.h>
18 #include <asm/e820/api.h>
19 
20 #include "xen-ops.h"
21 #include "smp.h"
22 #include "pmu.h"
23 
24 EXPORT_SYMBOL_GPL(hypercall_page);
25 
26 /*
27  * Pointer to the xen_vcpu_info structure or
28  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
29  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
30  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
31  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
32  * acknowledge pending events.
33  * Also more subtly it is used by the patched version of irq enable/disable
34  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
35  *
36  * The desire to be able to do those mask/unmask operations as a single
37  * instruction by using the per-cpu offset held in %gs is the real reason
38  * vcpu info is in a per-cpu pointer and the original reason for this
39  * hypercall.
40  *
41  */
42 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
43 
44 /*
45  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
46  * hypercall. This can be used both in PV and PVHVM mode. The structure
47  * overrides the default per_cpu(xen_vcpu, cpu) value.
48  */
49 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
50 
51 /* Linux <-> Xen vCPU id mapping */
52 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
53 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
54 
55 enum xen_domain_type xen_domain_type = XEN_NATIVE;
56 EXPORT_SYMBOL_GPL(xen_domain_type);
57 
58 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
59 EXPORT_SYMBOL(machine_to_phys_mapping);
60 unsigned long  machine_to_phys_nr;
61 EXPORT_SYMBOL(machine_to_phys_nr);
62 
63 struct start_info *xen_start_info;
64 EXPORT_SYMBOL_GPL(xen_start_info);
65 
66 struct shared_info xen_dummy_shared_info;
67 
68 __read_mostly int xen_have_vector_callback;
69 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
70 
71 /*
72  * NB: needs to live in .data because it's used by xen_prepare_pvh which runs
73  * before clearing the bss.
74  */
75 uint32_t xen_start_flags __section(".data") = 0;
76 EXPORT_SYMBOL(xen_start_flags);
77 
78 /*
79  * Point at some empty memory to start with. We map the real shared_info
80  * page as soon as fixmap is up and running.
81  */
82 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
83 
84 /*
85  * Flag to determine whether vcpu info placement is available on all
86  * VCPUs.  We assume it is to start with, and then set it to zero on
87  * the first failure.  This is because it can succeed on some VCPUs
88  * and not others, since it can involve hypervisor memory allocation,
89  * or because the guest failed to guarantee all the appropriate
90  * constraints on all VCPUs (ie buffer can't cross a page boundary).
91  *
92  * Note that any particular CPU may be using a placed vcpu structure,
93  * but we can only optimise if the all are.
94  *
95  * 0: not available, 1: available
96  */
97 int xen_have_vcpu_info_placement = 1;
98 
99 static int xen_cpu_up_online(unsigned int cpu)
100 {
101 	xen_init_lock_cpu(cpu);
102 	return 0;
103 }
104 
105 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
106 		    int (*cpu_dead_cb)(unsigned int))
107 {
108 	int rc;
109 
110 	rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
111 				       "x86/xen/guest:prepare",
112 				       cpu_up_prepare_cb, cpu_dead_cb);
113 	if (rc >= 0) {
114 		rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
115 					       "x86/xen/guest:online",
116 					       xen_cpu_up_online, NULL);
117 		if (rc < 0)
118 			cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
119 	}
120 
121 	return rc >= 0 ? 0 : rc;
122 }
123 
124 static int xen_vcpu_setup_restore(int cpu)
125 {
126 	int rc = 0;
127 
128 	/* Any per_cpu(xen_vcpu) is stale, so reset it */
129 	xen_vcpu_info_reset(cpu);
130 
131 	/*
132 	 * For PVH and PVHVM, setup online VCPUs only. The rest will
133 	 * be handled by hotplug.
134 	 */
135 	if (xen_pv_domain() ||
136 	    (xen_hvm_domain() && cpu_online(cpu))) {
137 		rc = xen_vcpu_setup(cpu);
138 	}
139 
140 	return rc;
141 }
142 
143 /*
144  * On restore, set the vcpu placement up again.
145  * If it fails, then we're in a bad state, since
146  * we can't back out from using it...
147  */
148 void xen_vcpu_restore(void)
149 {
150 	int cpu, rc;
151 
152 	for_each_possible_cpu(cpu) {
153 		bool other_cpu = (cpu != smp_processor_id());
154 		bool is_up;
155 
156 		if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
157 			continue;
158 
159 		/* Only Xen 4.5 and higher support this. */
160 		is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
161 					   xen_vcpu_nr(cpu), NULL) > 0;
162 
163 		if (other_cpu && is_up &&
164 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
165 			BUG();
166 
167 		if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
168 			xen_setup_runstate_info(cpu);
169 
170 		rc = xen_vcpu_setup_restore(cpu);
171 		if (rc)
172 			pr_emerg_once("vcpu restore failed for cpu=%d err=%d. "
173 					"System will hang.\n", cpu, rc);
174 		/*
175 		 * In case xen_vcpu_setup_restore() fails, do not bring up the
176 		 * VCPU. This helps us avoid the resulting OOPS when the VCPU
177 		 * accesses pvclock_vcpu_time via xen_vcpu (which is NULL.)
178 		 * Note that this does not improve the situation much -- now the
179 		 * VM hangs instead of OOPSing -- with the VCPUs that did not
180 		 * fail, spinning in stop_machine(), waiting for the failed
181 		 * VCPUs to come up.
182 		 */
183 		if (other_cpu && is_up && (rc == 0) &&
184 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
185 			BUG();
186 	}
187 }
188 
189 void xen_vcpu_info_reset(int cpu)
190 {
191 	if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
192 		per_cpu(xen_vcpu, cpu) =
193 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
194 	} else {
195 		/* Set to NULL so that if somebody accesses it we get an OOPS */
196 		per_cpu(xen_vcpu, cpu) = NULL;
197 	}
198 }
199 
200 int xen_vcpu_setup(int cpu)
201 {
202 	struct vcpu_register_vcpu_info info;
203 	int err;
204 	struct vcpu_info *vcpup;
205 
206 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
207 
208 	/*
209 	 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
210 	 * and at restore (xen_vcpu_restore). Also called for hotplugged
211 	 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
212 	 * However, the hypercall can only be done once (see below) so if a VCPU
213 	 * is offlined and comes back online then let's not redo the hypercall.
214 	 *
215 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
216 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
217 	 * use this function.
218 	 */
219 	if (xen_hvm_domain()) {
220 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
221 			return 0;
222 	}
223 
224 	if (xen_have_vcpu_info_placement) {
225 		vcpup = &per_cpu(xen_vcpu_info, cpu);
226 		info.mfn = arbitrary_virt_to_mfn(vcpup);
227 		info.offset = offset_in_page(vcpup);
228 
229 		/*
230 		 * Check to see if the hypervisor will put the vcpu_info
231 		 * structure where we want it, which allows direct access via
232 		 * a percpu-variable.
233 		 * N.B. This hypercall can _only_ be called once per CPU.
234 		 * Subsequent calls will error out with -EINVAL. This is due to
235 		 * the fact that hypervisor has no unregister variant and this
236 		 * hypercall does not allow to over-write info.mfn and
237 		 * info.offset.
238 		 */
239 		err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info,
240 					 xen_vcpu_nr(cpu), &info);
241 
242 		if (err) {
243 			pr_warn_once("register_vcpu_info failed: cpu=%d err=%d\n",
244 				     cpu, err);
245 			xen_have_vcpu_info_placement = 0;
246 		} else {
247 			/*
248 			 * This cpu is using the registered vcpu info, even if
249 			 * later ones fail to.
250 			 */
251 			per_cpu(xen_vcpu, cpu) = vcpup;
252 		}
253 	}
254 
255 	if (!xen_have_vcpu_info_placement)
256 		xen_vcpu_info_reset(cpu);
257 
258 	return ((per_cpu(xen_vcpu, cpu) == NULL) ? -ENODEV : 0);
259 }
260 
261 void xen_reboot(int reason)
262 {
263 	struct sched_shutdown r = { .reason = reason };
264 	int cpu;
265 
266 	for_each_online_cpu(cpu)
267 		xen_pmu_finish(cpu);
268 
269 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
270 		BUG();
271 }
272 
273 static int reboot_reason = SHUTDOWN_reboot;
274 static bool xen_legacy_crash;
275 void xen_emergency_restart(void)
276 {
277 	xen_reboot(reboot_reason);
278 }
279 
280 static int
281 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
282 {
283 	if (!kexec_crash_loaded()) {
284 		if (xen_legacy_crash)
285 			xen_reboot(SHUTDOWN_crash);
286 
287 		reboot_reason = SHUTDOWN_crash;
288 
289 		/*
290 		 * If panic_timeout==0 then we are supposed to wait forever.
291 		 * However, to preserve original dom0 behavior we have to drop
292 		 * into hypervisor. (domU behavior is controlled by its
293 		 * config file)
294 		 */
295 		if (panic_timeout == 0)
296 			panic_timeout = -1;
297 	}
298 	return NOTIFY_DONE;
299 }
300 
301 static int __init parse_xen_legacy_crash(char *arg)
302 {
303 	xen_legacy_crash = true;
304 	return 0;
305 }
306 early_param("xen_legacy_crash", parse_xen_legacy_crash);
307 
308 static struct notifier_block xen_panic_block = {
309 	.notifier_call = xen_panic_event,
310 	.priority = INT_MIN
311 };
312 
313 int xen_panic_handler_init(void)
314 {
315 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
316 	return 0;
317 }
318 
319 void xen_pin_vcpu(int cpu)
320 {
321 	static bool disable_pinning;
322 	struct sched_pin_override pin_override;
323 	int ret;
324 
325 	if (disable_pinning)
326 		return;
327 
328 	pin_override.pcpu = cpu;
329 	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
330 
331 	/* Ignore errors when removing override. */
332 	if (cpu < 0)
333 		return;
334 
335 	switch (ret) {
336 	case -ENOSYS:
337 		pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
338 			cpu);
339 		disable_pinning = true;
340 		break;
341 	case -EPERM:
342 		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
343 		disable_pinning = true;
344 		break;
345 	case -EINVAL:
346 	case -EBUSY:
347 		pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
348 			cpu);
349 		break;
350 	case 0:
351 		break;
352 	default:
353 		WARN(1, "rc %d while trying to pin vcpu\n", ret);
354 		disable_pinning = true;
355 	}
356 }
357 
358 #ifdef CONFIG_HOTPLUG_CPU
359 void xen_arch_register_cpu(int num)
360 {
361 	arch_register_cpu(num);
362 }
363 EXPORT_SYMBOL(xen_arch_register_cpu);
364 
365 void xen_arch_unregister_cpu(int num)
366 {
367 	arch_unregister_cpu(num);
368 }
369 EXPORT_SYMBOL(xen_arch_unregister_cpu);
370 #endif
371