xref: /linux/arch/x86/xen/enlighten.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
2 #include <linux/bootmem.h>
3 #endif
4 #include <linux/cpu.h>
5 #include <linux/kexec.h>
6 #include <linux/slab.h>
7 
8 #include <xen/features.h>
9 #include <xen/page.h>
10 #include <xen/interface/memory.h>
11 
12 #include <asm/xen/hypercall.h>
13 #include <asm/xen/hypervisor.h>
14 #include <asm/cpu.h>
15 #include <asm/e820/api.h>
16 
17 #include "xen-ops.h"
18 #include "smp.h"
19 #include "pmu.h"
20 
21 EXPORT_SYMBOL_GPL(hypercall_page);
22 
23 /*
24  * Pointer to the xen_vcpu_info structure or
25  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
26  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
27  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
28  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
29  * acknowledge pending events.
30  * Also more subtly it is used by the patched version of irq enable/disable
31  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
32  *
33  * The desire to be able to do those mask/unmask operations as a single
34  * instruction by using the per-cpu offset held in %gs is the real reason
35  * vcpu info is in a per-cpu pointer and the original reason for this
36  * hypercall.
37  *
38  */
39 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
40 
41 /*
42  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
43  * hypercall. This can be used both in PV and PVHVM mode. The structure
44  * overrides the default per_cpu(xen_vcpu, cpu) value.
45  */
46 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
47 
48 /* Linux <-> Xen vCPU id mapping */
49 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
50 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
51 
52 enum xen_domain_type xen_domain_type = XEN_NATIVE;
53 EXPORT_SYMBOL_GPL(xen_domain_type);
54 
55 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
56 EXPORT_SYMBOL(machine_to_phys_mapping);
57 unsigned long  machine_to_phys_nr;
58 EXPORT_SYMBOL(machine_to_phys_nr);
59 
60 struct start_info *xen_start_info;
61 EXPORT_SYMBOL_GPL(xen_start_info);
62 
63 struct shared_info xen_dummy_shared_info;
64 
65 __read_mostly int xen_have_vector_callback;
66 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
67 
68 /*
69  * NB: needs to live in .data because it's used by xen_prepare_pvh which runs
70  * before clearing the bss.
71  */
72 uint32_t xen_start_flags __attribute__((section(".data"))) = 0;
73 EXPORT_SYMBOL(xen_start_flags);
74 
75 /*
76  * Point at some empty memory to start with. We map the real shared_info
77  * page as soon as fixmap is up and running.
78  */
79 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
80 
81 /*
82  * Flag to determine whether vcpu info placement is available on all
83  * VCPUs.  We assume it is to start with, and then set it to zero on
84  * the first failure.  This is because it can succeed on some VCPUs
85  * and not others, since it can involve hypervisor memory allocation,
86  * or because the guest failed to guarantee all the appropriate
87  * constraints on all VCPUs (ie buffer can't cross a page boundary).
88  *
89  * Note that any particular CPU may be using a placed vcpu structure,
90  * but we can only optimise if the all are.
91  *
92  * 0: not available, 1: available
93  */
94 int xen_have_vcpu_info_placement = 1;
95 
96 static int xen_cpu_up_online(unsigned int cpu)
97 {
98 	xen_init_lock_cpu(cpu);
99 	return 0;
100 }
101 
102 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
103 		    int (*cpu_dead_cb)(unsigned int))
104 {
105 	int rc;
106 
107 	rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
108 				       "x86/xen/guest:prepare",
109 				       cpu_up_prepare_cb, cpu_dead_cb);
110 	if (rc >= 0) {
111 		rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
112 					       "x86/xen/guest:online",
113 					       xen_cpu_up_online, NULL);
114 		if (rc < 0)
115 			cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
116 	}
117 
118 	return rc >= 0 ? 0 : rc;
119 }
120 
121 static int xen_vcpu_setup_restore(int cpu)
122 {
123 	int rc = 0;
124 
125 	/* Any per_cpu(xen_vcpu) is stale, so reset it */
126 	xen_vcpu_info_reset(cpu);
127 
128 	/*
129 	 * For PVH and PVHVM, setup online VCPUs only. The rest will
130 	 * be handled by hotplug.
131 	 */
132 	if (xen_pv_domain() ||
133 	    (xen_hvm_domain() && cpu_online(cpu))) {
134 		rc = xen_vcpu_setup(cpu);
135 	}
136 
137 	return rc;
138 }
139 
140 /*
141  * On restore, set the vcpu placement up again.
142  * If it fails, then we're in a bad state, since
143  * we can't back out from using it...
144  */
145 void xen_vcpu_restore(void)
146 {
147 	int cpu, rc;
148 
149 	for_each_possible_cpu(cpu) {
150 		bool other_cpu = (cpu != smp_processor_id());
151 		bool is_up;
152 
153 		if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
154 			continue;
155 
156 		/* Only Xen 4.5 and higher support this. */
157 		is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
158 					   xen_vcpu_nr(cpu), NULL) > 0;
159 
160 		if (other_cpu && is_up &&
161 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
162 			BUG();
163 
164 		if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
165 			xen_setup_runstate_info(cpu);
166 
167 		rc = xen_vcpu_setup_restore(cpu);
168 		if (rc)
169 			pr_emerg_once("vcpu restore failed for cpu=%d err=%d. "
170 					"System will hang.\n", cpu, rc);
171 		/*
172 		 * In case xen_vcpu_setup_restore() fails, do not bring up the
173 		 * VCPU. This helps us avoid the resulting OOPS when the VCPU
174 		 * accesses pvclock_vcpu_time via xen_vcpu (which is NULL.)
175 		 * Note that this does not improve the situation much -- now the
176 		 * VM hangs instead of OOPSing -- with the VCPUs that did not
177 		 * fail, spinning in stop_machine(), waiting for the failed
178 		 * VCPUs to come up.
179 		 */
180 		if (other_cpu && is_up && (rc == 0) &&
181 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
182 			BUG();
183 	}
184 }
185 
186 void xen_vcpu_info_reset(int cpu)
187 {
188 	if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
189 		per_cpu(xen_vcpu, cpu) =
190 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
191 	} else {
192 		/* Set to NULL so that if somebody accesses it we get an OOPS */
193 		per_cpu(xen_vcpu, cpu) = NULL;
194 	}
195 }
196 
197 int xen_vcpu_setup(int cpu)
198 {
199 	struct vcpu_register_vcpu_info info;
200 	int err;
201 	struct vcpu_info *vcpup;
202 
203 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
204 
205 	/*
206 	 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
207 	 * and at restore (xen_vcpu_restore). Also called for hotplugged
208 	 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
209 	 * However, the hypercall can only be done once (see below) so if a VCPU
210 	 * is offlined and comes back online then let's not redo the hypercall.
211 	 *
212 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
213 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
214 	 * use this function.
215 	 */
216 	if (xen_hvm_domain()) {
217 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
218 			return 0;
219 	}
220 
221 	if (xen_have_vcpu_info_placement) {
222 		vcpup = &per_cpu(xen_vcpu_info, cpu);
223 		info.mfn = arbitrary_virt_to_mfn(vcpup);
224 		info.offset = offset_in_page(vcpup);
225 
226 		/*
227 		 * Check to see if the hypervisor will put the vcpu_info
228 		 * structure where we want it, which allows direct access via
229 		 * a percpu-variable.
230 		 * N.B. This hypercall can _only_ be called once per CPU.
231 		 * Subsequent calls will error out with -EINVAL. This is due to
232 		 * the fact that hypervisor has no unregister variant and this
233 		 * hypercall does not allow to over-write info.mfn and
234 		 * info.offset.
235 		 */
236 		err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info,
237 					 xen_vcpu_nr(cpu), &info);
238 
239 		if (err) {
240 			pr_warn_once("register_vcpu_info failed: cpu=%d err=%d\n",
241 				     cpu, err);
242 			xen_have_vcpu_info_placement = 0;
243 		} else {
244 			/*
245 			 * This cpu is using the registered vcpu info, even if
246 			 * later ones fail to.
247 			 */
248 			per_cpu(xen_vcpu, cpu) = vcpup;
249 		}
250 	}
251 
252 	if (!xen_have_vcpu_info_placement)
253 		xen_vcpu_info_reset(cpu);
254 
255 	return ((per_cpu(xen_vcpu, cpu) == NULL) ? -ENODEV : 0);
256 }
257 
258 void xen_reboot(int reason)
259 {
260 	struct sched_shutdown r = { .reason = reason };
261 	int cpu;
262 
263 	for_each_online_cpu(cpu)
264 		xen_pmu_finish(cpu);
265 
266 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
267 		BUG();
268 }
269 
270 void xen_emergency_restart(void)
271 {
272 	xen_reboot(SHUTDOWN_reboot);
273 }
274 
275 static int
276 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
277 {
278 	if (!kexec_crash_loaded())
279 		xen_reboot(SHUTDOWN_crash);
280 	return NOTIFY_DONE;
281 }
282 
283 static struct notifier_block xen_panic_block = {
284 	.notifier_call = xen_panic_event,
285 	.priority = INT_MIN
286 };
287 
288 int xen_panic_handler_init(void)
289 {
290 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
291 	return 0;
292 }
293 
294 void xen_pin_vcpu(int cpu)
295 {
296 	static bool disable_pinning;
297 	struct sched_pin_override pin_override;
298 	int ret;
299 
300 	if (disable_pinning)
301 		return;
302 
303 	pin_override.pcpu = cpu;
304 	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
305 
306 	/* Ignore errors when removing override. */
307 	if (cpu < 0)
308 		return;
309 
310 	switch (ret) {
311 	case -ENOSYS:
312 		pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
313 			cpu);
314 		disable_pinning = true;
315 		break;
316 	case -EPERM:
317 		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
318 		disable_pinning = true;
319 		break;
320 	case -EINVAL:
321 	case -EBUSY:
322 		pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
323 			cpu);
324 		break;
325 	case 0:
326 		break;
327 	default:
328 		WARN(1, "rc %d while trying to pin vcpu\n", ret);
329 		disable_pinning = true;
330 	}
331 }
332 
333 #ifdef CONFIG_HOTPLUG_CPU
334 void xen_arch_register_cpu(int num)
335 {
336 	arch_register_cpu(num);
337 }
338 EXPORT_SYMBOL(xen_arch_register_cpu);
339 
340 void xen_arch_unregister_cpu(int num)
341 {
342 	arch_unregister_cpu(num);
343 }
344 EXPORT_SYMBOL(xen_arch_unregister_cpu);
345 #endif
346 
347 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
348 void __init arch_xen_balloon_init(struct resource *hostmem_resource)
349 {
350 	struct xen_memory_map memmap;
351 	int rc;
352 	unsigned int i, last_guest_ram;
353 	phys_addr_t max_addr = PFN_PHYS(max_pfn);
354 	struct e820_table *xen_e820_table;
355 	const struct e820_entry *entry;
356 	struct resource *res;
357 
358 	if (!xen_initial_domain())
359 		return;
360 
361 	xen_e820_table = kmalloc(sizeof(*xen_e820_table), GFP_KERNEL);
362 	if (!xen_e820_table)
363 		return;
364 
365 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table->entries);
366 	set_xen_guest_handle(memmap.buffer, xen_e820_table->entries);
367 	rc = HYPERVISOR_memory_op(XENMEM_machine_memory_map, &memmap);
368 	if (rc) {
369 		pr_warn("%s: Can't read host e820 (%d)\n", __func__, rc);
370 		goto out;
371 	}
372 
373 	last_guest_ram = 0;
374 	for (i = 0; i < memmap.nr_entries; i++) {
375 		if (xen_e820_table->entries[i].addr >= max_addr)
376 			break;
377 		if (xen_e820_table->entries[i].type == E820_TYPE_RAM)
378 			last_guest_ram = i;
379 	}
380 
381 	entry = &xen_e820_table->entries[last_guest_ram];
382 	if (max_addr >= entry->addr + entry->size)
383 		goto out; /* No unallocated host RAM. */
384 
385 	hostmem_resource->start = max_addr;
386 	hostmem_resource->end = entry->addr + entry->size;
387 
388 	/*
389 	 * Mark non-RAM regions between the end of dom0 RAM and end of host RAM
390 	 * as unavailable. The rest of that region can be used for hotplug-based
391 	 * ballooning.
392 	 */
393 	for (; i < memmap.nr_entries; i++) {
394 		entry = &xen_e820_table->entries[i];
395 
396 		if (entry->type == E820_TYPE_RAM)
397 			continue;
398 
399 		if (entry->addr >= hostmem_resource->end)
400 			break;
401 
402 		res = kzalloc(sizeof(*res), GFP_KERNEL);
403 		if (!res)
404 			goto out;
405 
406 		res->name = "Unavailable host RAM";
407 		res->start = entry->addr;
408 		res->end = (entry->addr + entry->size < hostmem_resource->end) ?
409 			    entry->addr + entry->size : hostmem_resource->end;
410 		rc = insert_resource(hostmem_resource, res);
411 		if (rc) {
412 			pr_warn("%s: Can't insert [%llx - %llx) (%d)\n",
413 				__func__, res->start, res->end, rc);
414 			kfree(res);
415 			goto  out;
416 		}
417 	}
418 
419  out:
420 	kfree(xen_e820_table);
421 }
422 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
423