xref: /openbmc/linux/arch/x86/xen/enlighten.c (revision 44e2971f)
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 
7 #include <xen/features.h>
8 #include <xen/page.h>
9 #include <xen/interface/memory.h>
10 
11 #include <asm/xen/hypercall.h>
12 #include <asm/xen/hypervisor.h>
13 #include <asm/cpu.h>
14 #include <asm/e820/api.h>
15 
16 #include "xen-ops.h"
17 #include "smp.h"
18 #include "pmu.h"
19 
20 EXPORT_SYMBOL_GPL(hypercall_page);
21 
22 /*
23  * Pointer to the xen_vcpu_info structure or
24  * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
25  * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
26  * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
27  * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
28  * acknowledge pending events.
29  * Also more subtly it is used by the patched version of irq enable/disable
30  * e.g. xen_irq_enable_direct and xen_iret in PV mode.
31  *
32  * The desire to be able to do those mask/unmask operations as a single
33  * instruction by using the per-cpu offset held in %gs is the real reason
34  * vcpu info is in a per-cpu pointer and the original reason for this
35  * hypercall.
36  *
37  */
38 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
39 
40 /*
41  * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
42  * hypercall. This can be used both in PV and PVHVM mode. The structure
43  * overrides the default per_cpu(xen_vcpu, cpu) value.
44  */
45 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
46 
47 /* Linux <-> Xen vCPU id mapping */
48 DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
49 EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
50 
51 enum xen_domain_type xen_domain_type = XEN_NATIVE;
52 EXPORT_SYMBOL_GPL(xen_domain_type);
53 
54 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
55 EXPORT_SYMBOL(machine_to_phys_mapping);
56 unsigned long  machine_to_phys_nr;
57 EXPORT_SYMBOL(machine_to_phys_nr);
58 
59 struct start_info *xen_start_info;
60 EXPORT_SYMBOL_GPL(xen_start_info);
61 
62 struct shared_info xen_dummy_shared_info;
63 
64 __read_mostly int xen_have_vector_callback;
65 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
66 
67 /*
68  * NB: needs to live in .data because it's used by xen_prepare_pvh which runs
69  * before clearing the bss.
70  */
71 uint32_t xen_start_flags __attribute__((section(".data"))) = 0;
72 EXPORT_SYMBOL(xen_start_flags);
73 
74 /*
75  * Point at some empty memory to start with. We map the real shared_info
76  * page as soon as fixmap is up and running.
77  */
78 struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
79 
80 /*
81  * Flag to determine whether vcpu info placement is available on all
82  * VCPUs.  We assume it is to start with, and then set it to zero on
83  * the first failure.  This is because it can succeed on some VCPUs
84  * and not others, since it can involve hypervisor memory allocation,
85  * or because the guest failed to guarantee all the appropriate
86  * constraints on all VCPUs (ie buffer can't cross a page boundary).
87  *
88  * Note that any particular CPU may be using a placed vcpu structure,
89  * but we can only optimise if the all are.
90  *
91  * 0: not available, 1: available
92  */
93 int xen_have_vcpu_info_placement = 1;
94 
95 static int xen_cpu_up_online(unsigned int cpu)
96 {
97 	xen_init_lock_cpu(cpu);
98 	return 0;
99 }
100 
101 int xen_cpuhp_setup(int (*cpu_up_prepare_cb)(unsigned int),
102 		    int (*cpu_dead_cb)(unsigned int))
103 {
104 	int rc;
105 
106 	rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
107 				       "x86/xen/guest:prepare",
108 				       cpu_up_prepare_cb, cpu_dead_cb);
109 	if (rc >= 0) {
110 		rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
111 					       "x86/xen/guest:online",
112 					       xen_cpu_up_online, NULL);
113 		if (rc < 0)
114 			cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
115 	}
116 
117 	return rc >= 0 ? 0 : rc;
118 }
119 
120 static int xen_vcpu_setup_restore(int cpu)
121 {
122 	int rc = 0;
123 
124 	/* Any per_cpu(xen_vcpu) is stale, so reset it */
125 	xen_vcpu_info_reset(cpu);
126 
127 	/*
128 	 * For PVH and PVHVM, setup online VCPUs only. The rest will
129 	 * be handled by hotplug.
130 	 */
131 	if (xen_pv_domain() ||
132 	    (xen_hvm_domain() && cpu_online(cpu))) {
133 		rc = xen_vcpu_setup(cpu);
134 	}
135 
136 	return rc;
137 }
138 
139 /*
140  * On restore, set the vcpu placement up again.
141  * If it fails, then we're in a bad state, since
142  * we can't back out from using it...
143  */
144 void xen_vcpu_restore(void)
145 {
146 	int cpu, rc;
147 
148 	for_each_possible_cpu(cpu) {
149 		bool other_cpu = (cpu != smp_processor_id());
150 		bool is_up;
151 
152 		if (xen_vcpu_nr(cpu) == XEN_VCPU_ID_INVALID)
153 			continue;
154 
155 		/* Only Xen 4.5 and higher support this. */
156 		is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up,
157 					   xen_vcpu_nr(cpu), NULL) > 0;
158 
159 		if (other_cpu && is_up &&
160 		    HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
161 			BUG();
162 
163 		if (xen_pv_domain() || xen_feature(XENFEAT_hvm_safe_pvclock))
164 			xen_setup_runstate_info(cpu);
165 
166 		rc = xen_vcpu_setup_restore(cpu);
167 		if (rc)
168 			pr_emerg_once("vcpu restore failed for cpu=%d err=%d. "
169 					"System will hang.\n", cpu, rc);
170 		/*
171 		 * In case xen_vcpu_setup_restore() fails, do not bring up the
172 		 * VCPU. This helps us avoid the resulting OOPS when the VCPU
173 		 * accesses pvclock_vcpu_time via xen_vcpu (which is NULL.)
174 		 * Note that this does not improve the situation much -- now the
175 		 * VM hangs instead of OOPSing -- with the VCPUs that did not
176 		 * fail, spinning in stop_machine(), waiting for the failed
177 		 * VCPUs to come up.
178 		 */
179 		if (other_cpu && is_up && (rc == 0) &&
180 		    HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
181 			BUG();
182 	}
183 }
184 
185 void xen_vcpu_info_reset(int cpu)
186 {
187 	if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS) {
188 		per_cpu(xen_vcpu, cpu) =
189 			&HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
190 	} else {
191 		/* Set to NULL so that if somebody accesses it we get an OOPS */
192 		per_cpu(xen_vcpu, cpu) = NULL;
193 	}
194 }
195 
196 int xen_vcpu_setup(int cpu)
197 {
198 	struct vcpu_register_vcpu_info info;
199 	int err;
200 	struct vcpu_info *vcpup;
201 
202 	BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
203 
204 	/*
205 	 * This path is called on PVHVM at bootup (xen_hvm_smp_prepare_boot_cpu)
206 	 * and at restore (xen_vcpu_restore). Also called for hotplugged
207 	 * VCPUs (cpu_init -> xen_hvm_cpu_prepare_hvm).
208 	 * However, the hypercall can only be done once (see below) so if a VCPU
209 	 * is offlined and comes back online then let's not redo the hypercall.
210 	 *
211 	 * For PV it is called during restore (xen_vcpu_restore) and bootup
212 	 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
213 	 * use this function.
214 	 */
215 	if (xen_hvm_domain()) {
216 		if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
217 			return 0;
218 	}
219 
220 	if (xen_have_vcpu_info_placement) {
221 		vcpup = &per_cpu(xen_vcpu_info, cpu);
222 		info.mfn = arbitrary_virt_to_mfn(vcpup);
223 		info.offset = offset_in_page(vcpup);
224 
225 		/*
226 		 * Check to see if the hypervisor will put the vcpu_info
227 		 * structure where we want it, which allows direct access via
228 		 * a percpu-variable.
229 		 * N.B. This hypercall can _only_ be called once per CPU.
230 		 * Subsequent calls will error out with -EINVAL. This is due to
231 		 * the fact that hypervisor has no unregister variant and this
232 		 * hypercall does not allow to over-write info.mfn and
233 		 * info.offset.
234 		 */
235 		err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info,
236 					 xen_vcpu_nr(cpu), &info);
237 
238 		if (err) {
239 			pr_warn_once("register_vcpu_info failed: cpu=%d err=%d\n",
240 				     cpu, err);
241 			xen_have_vcpu_info_placement = 0;
242 		} else {
243 			/*
244 			 * This cpu is using the registered vcpu info, even if
245 			 * later ones fail to.
246 			 */
247 			per_cpu(xen_vcpu, cpu) = vcpup;
248 		}
249 	}
250 
251 	if (!xen_have_vcpu_info_placement)
252 		xen_vcpu_info_reset(cpu);
253 
254 	return ((per_cpu(xen_vcpu, cpu) == NULL) ? -ENODEV : 0);
255 }
256 
257 void xen_reboot(int reason)
258 {
259 	struct sched_shutdown r = { .reason = reason };
260 	int cpu;
261 
262 	for_each_online_cpu(cpu)
263 		xen_pmu_finish(cpu);
264 
265 	if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
266 		BUG();
267 }
268 
269 void xen_emergency_restart(void)
270 {
271 	xen_reboot(SHUTDOWN_reboot);
272 }
273 
274 static int
275 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
276 {
277 	if (!kexec_crash_loaded())
278 		xen_reboot(SHUTDOWN_crash);
279 	return NOTIFY_DONE;
280 }
281 
282 static struct notifier_block xen_panic_block = {
283 	.notifier_call = xen_panic_event,
284 	.priority = INT_MIN
285 };
286 
287 int xen_panic_handler_init(void)
288 {
289 	atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
290 	return 0;
291 }
292 
293 void xen_pin_vcpu(int cpu)
294 {
295 	static bool disable_pinning;
296 	struct sched_pin_override pin_override;
297 	int ret;
298 
299 	if (disable_pinning)
300 		return;
301 
302 	pin_override.pcpu = cpu;
303 	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
304 
305 	/* Ignore errors when removing override. */
306 	if (cpu < 0)
307 		return;
308 
309 	switch (ret) {
310 	case -ENOSYS:
311 		pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
312 			cpu);
313 		disable_pinning = true;
314 		break;
315 	case -EPERM:
316 		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
317 		disable_pinning = true;
318 		break;
319 	case -EINVAL:
320 	case -EBUSY:
321 		pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
322 			cpu);
323 		break;
324 	case 0:
325 		break;
326 	default:
327 		WARN(1, "rc %d while trying to pin vcpu\n", ret);
328 		disable_pinning = true;
329 	}
330 }
331 
332 #ifdef CONFIG_HOTPLUG_CPU
333 void xen_arch_register_cpu(int num)
334 {
335 	arch_register_cpu(num);
336 }
337 EXPORT_SYMBOL(xen_arch_register_cpu);
338 
339 void xen_arch_unregister_cpu(int num)
340 {
341 	arch_unregister_cpu(num);
342 }
343 EXPORT_SYMBOL(xen_arch_unregister_cpu);
344 #endif
345 
346 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
347 void __init arch_xen_balloon_init(struct resource *hostmem_resource)
348 {
349 	struct xen_memory_map memmap;
350 	int rc;
351 	unsigned int i, last_guest_ram;
352 	phys_addr_t max_addr = PFN_PHYS(max_pfn);
353 	struct e820_table *xen_e820_table;
354 	const struct e820_entry *entry;
355 	struct resource *res;
356 
357 	if (!xen_initial_domain())
358 		return;
359 
360 	xen_e820_table = kmalloc(sizeof(*xen_e820_table), GFP_KERNEL);
361 	if (!xen_e820_table)
362 		return;
363 
364 	memmap.nr_entries = ARRAY_SIZE(xen_e820_table->entries);
365 	set_xen_guest_handle(memmap.buffer, xen_e820_table->entries);
366 	rc = HYPERVISOR_memory_op(XENMEM_machine_memory_map, &memmap);
367 	if (rc) {
368 		pr_warn("%s: Can't read host e820 (%d)\n", __func__, rc);
369 		goto out;
370 	}
371 
372 	last_guest_ram = 0;
373 	for (i = 0; i < memmap.nr_entries; i++) {
374 		if (xen_e820_table->entries[i].addr >= max_addr)
375 			break;
376 		if (xen_e820_table->entries[i].type == E820_TYPE_RAM)
377 			last_guest_ram = i;
378 	}
379 
380 	entry = &xen_e820_table->entries[last_guest_ram];
381 	if (max_addr >= entry->addr + entry->size)
382 		goto out; /* No unallocated host RAM. */
383 
384 	hostmem_resource->start = max_addr;
385 	hostmem_resource->end = entry->addr + entry->size;
386 
387 	/*
388 	 * Mark non-RAM regions between the end of dom0 RAM and end of host RAM
389 	 * as unavailable. The rest of that region can be used for hotplug-based
390 	 * ballooning.
391 	 */
392 	for (; i < memmap.nr_entries; i++) {
393 		entry = &xen_e820_table->entries[i];
394 
395 		if (entry->type == E820_TYPE_RAM)
396 			continue;
397 
398 		if (entry->addr >= hostmem_resource->end)
399 			break;
400 
401 		res = kzalloc(sizeof(*res), GFP_KERNEL);
402 		if (!res)
403 			goto out;
404 
405 		res->name = "Unavailable host RAM";
406 		res->start = entry->addr;
407 		res->end = (entry->addr + entry->size < hostmem_resource->end) ?
408 			    entry->addr + entry->size : hostmem_resource->end;
409 		rc = insert_resource(hostmem_resource, res);
410 		if (rc) {
411 			pr_warn("%s: Can't insert [%llx - %llx) (%d)\n",
412 				__func__, res->start, res->end, rc);
413 			kfree(res);
414 			goto  out;
415 		}
416 	}
417 
418  out:
419 	kfree(xen_e820_table);
420 }
421 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
422