1 /*
2 * Copyright (c) 2008, 2013 Citrix Systems, Inc.
3 * Copyright (c) 2012 Spectra Logic Corporation
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30
31 #include <sys/param.h>
32 #include <sys/bus.h>
33 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/proc.h>
36 #include <sys/smp.h>
37 #include <sys/systm.h>
38
39 #include <vm/vm.h>
40 #include <vm/pmap.h>
41
42 #include <dev/pci/pcivar.h>
43
44 #include <machine/cpufunc.h>
45 #include <machine/cpu.h>
46 #include <machine/smp.h>
47
48 #include <x86/apicreg.h>
49
50 #include <xen/xen-os.h>
51 #include <xen/features.h>
52 #include <xen/gnttab.h>
53 #include <xen/hypervisor.h>
54 #include <xen/hvm.h>
55 #include <xen/xen_intr.h>
56
57 #include <xen/interface/hvm/params.h>
58 #include <xen/interface/vcpu.h>
59
60 /*--------------------------- Forward Declarations ---------------------------*/
61 static void xen_hvm_cpu_init(void);
62
63 /*-------------------------------- Local Types -------------------------------*/
64 enum xen_hvm_init_type {
65 XEN_HVM_INIT_COLD,
66 XEN_HVM_INIT_CANCELLED_SUSPEND,
67 XEN_HVM_INIT_RESUME
68 };
69
70 /*-------------------------------- Global Data -------------------------------*/
71 enum xen_domain_type xen_domain_type = XEN_NATIVE;
72
73 #ifdef SMP
74 struct cpu_ops xen_hvm_cpu_ops = {
75 .cpu_init = xen_hvm_cpu_init,
76 .cpu_resume = xen_hvm_cpu_init
77 };
78 #endif
79
80 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support");
81
82 /**
83 * If non-zero, the hypervisor has been configured to use a direct
84 * IDT event callback for interrupt injection.
85 */
86 int xen_vector_callback_enabled;
87
88 /*------------------------------- Per-CPU Data -------------------------------*/
89 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info);
90 DPCPU_DEFINE(struct vcpu_info *, vcpu_info);
91
92 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/
93 shared_info_t *HYPERVISOR_shared_info;
94 start_info_t *HYPERVISOR_start_info;
95
96
97 /*------------------------------ Sysctl tunables -----------------------------*/
98 int xen_disable_pv_disks = 0;
99 int xen_disable_pv_nics = 0;
100 TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks);
101 TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics);
102
103 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/
104 static uint32_t
xen_hvm_cpuid_base(void)105 xen_hvm_cpuid_base(void)
106 {
107 uint32_t base, regs[4];
108
109 for (base = 0x40000000; base < 0x40010000; base += 0x100) {
110 do_cpuid(base, regs);
111 if (!memcmp("XenVMMXenVMM", ®s[1], 12)
112 && (regs[0] - base) >= 2)
113 return (base);
114 }
115 return (0);
116 }
117
118 /*
119 * Allocate and fill in the hypcall page.
120 */
121 static int
xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)122 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type)
123 {
124 uint32_t base, regs[4];
125 int i;
126
127 if (xen_pv_domain()) {
128 /* hypercall page is already set in the PV case */
129 return (0);
130 }
131
132 base = xen_hvm_cpuid_base();
133 if (base == 0)
134 return (ENXIO);
135
136 if (init_type == XEN_HVM_INIT_COLD) {
137 do_cpuid(base + 1, regs);
138 printf("XEN: Hypervisor version %d.%d detected.\n",
139 regs[0] >> 16, regs[0] & 0xffff);
140 }
141
142 /*
143 * Find the hypercall pages.
144 */
145 do_cpuid(base + 2, regs);
146
147 for (i = 0; i < regs[0]; i++)
148 wrmsr(regs[1], vtophys(&hypercall_page + i * PAGE_SIZE) + i);
149
150 return (0);
151 }
152
153 static void
xen_hvm_init_shared_info_page(void)154 xen_hvm_init_shared_info_page(void)
155 {
156 struct xen_add_to_physmap xatp;
157
158 if (xen_pv_domain()) {
159 /*
160 * Already setup in the PV case, shared_info is passed inside
161 * of the start_info struct at start of day.
162 */
163 return;
164 }
165
166 if (HYPERVISOR_shared_info == NULL) {
167 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT);
168 if (HYPERVISOR_shared_info == NULL)
169 panic("Unable to allocate Xen shared info page");
170 }
171
172 xatp.domid = DOMID_SELF;
173 xatp.idx = 0;
174 xatp.space = XENMAPSPACE_shared_info;
175 xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT;
176 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
177 panic("HYPERVISOR_memory_op failed");
178 }
179
180 /*
181 * Tell the hypervisor how to contact us for event channel callbacks.
182 */
183 void
xen_hvm_set_callback(device_t dev)184 xen_hvm_set_callback(device_t dev)
185 {
186 struct xen_hvm_param xhp;
187 int irq;
188
189 if (xen_vector_callback_enabled)
190 return;
191
192 xhp.domid = DOMID_SELF;
193 xhp.index = HVM_PARAM_CALLBACK_IRQ;
194 if (xen_feature(XENFEAT_hvm_callback_vector) != 0) {
195 int error;
196
197 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN);
198 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp);
199 if (error == 0) {
200 xen_vector_callback_enabled = 1;
201 return;
202 }
203 printf("Xen HVM callback vector registration failed (%d). "
204 "Falling back to emulated device interrupt\n", error);
205 }
206 xen_vector_callback_enabled = 0;
207 if (dev == NULL) {
208 /*
209 * Called from early boot or resume.
210 * xenpci will invoke us again later.
211 */
212 return;
213 }
214
215 irq = pci_get_irq(dev);
216 if (irq < 16) {
217 xhp.value = HVM_CALLBACK_GSI(irq);
218 } else {
219 u_int slot;
220 u_int pin;
221
222 slot = pci_get_slot(dev);
223 pin = pci_get_intpin(dev) - 1;
224 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin);
225 }
226
227 if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0)
228 panic("Can't set evtchn callback");
229 }
230
231 #define XEN_MAGIC_IOPORT 0x10
232 enum {
233 XMI_MAGIC = 0x49d2,
234 XMI_UNPLUG_IDE_DISKS = 0x01,
235 XMI_UNPLUG_NICS = 0x02,
236 XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04
237 };
238
239 static void
xen_hvm_disable_emulated_devices(void)240 xen_hvm_disable_emulated_devices(void)
241 {
242 u_short disable_devs = 0;
243
244 if (xen_pv_domain()) {
245 /*
246 * No emulated devices in the PV case, so no need to unplug
247 * anything.
248 */
249 if (xen_disable_pv_disks != 0 || xen_disable_pv_nics != 0)
250 printf("PV devices cannot be disabled in PV guests\n");
251 return;
252 }
253
254 if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC)
255 return;
256
257 if (xen_disable_pv_disks == 0) {
258 if (bootverbose)
259 printf("XEN: disabling emulated disks\n");
260 disable_devs |= XMI_UNPLUG_IDE_DISKS;
261 }
262 if (xen_disable_pv_nics == 0) {
263 if (bootverbose)
264 printf("XEN: disabling emulated nics\n");
265 disable_devs |= XMI_UNPLUG_NICS;
266 }
267
268 if (disable_devs != 0)
269 outw(XEN_MAGIC_IOPORT, disable_devs);
270 }
271
272 static void
xen_hvm_init(enum xen_hvm_init_type init_type)273 xen_hvm_init(enum xen_hvm_init_type init_type)
274 {
275 int error;
276 int i;
277
278 if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND)
279 return;
280
281 error = xen_hvm_init_hypercall_stubs(init_type);
282
283 switch (init_type) {
284 case XEN_HVM_INIT_COLD:
285 if (error != 0)
286 return;
287
288 /*
289 * If xen_domain_type is not set at this point
290 * it means we are inside a (PV)HVM guest, because
291 * for PVH the guest type is set much earlier
292 * (see hammer_time_xen).
293 */
294 if (!xen_domain()) {
295 xen_domain_type = XEN_HVM_DOMAIN;
296 vm_guest = VM_GUEST_XEN;
297 }
298
299 setup_xen_features();
300 #ifdef SMP
301 cpu_ops = xen_hvm_cpu_ops;
302 #endif
303 break;
304 case XEN_HVM_INIT_RESUME:
305 if (error != 0)
306 panic("Unable to init Xen hypercall stubs on resume");
307
308 /* Clear stale vcpu_info. */
309 CPU_FOREACH(i)
310 DPCPU_ID_SET(i, vcpu_info, NULL);
311 break;
312 default:
313 panic("Unsupported HVM initialization type");
314 }
315
316 xen_vector_callback_enabled = 0;
317 xen_hvm_set_callback(NULL);
318
319 /*
320 * On (PV)HVM domains we need to request the hypervisor to
321 * fill the shared info page, for PVH guest the shared_info page
322 * is passed inside the start_info struct and is already set, so this
323 * functions are no-ops.
324 */
325 xen_hvm_init_shared_info_page();
326 xen_hvm_disable_emulated_devices();
327 }
328
329 void
xen_hvm_suspend(void)330 xen_hvm_suspend(void)
331 {
332 }
333
334 void
xen_hvm_resume(bool suspend_cancelled)335 xen_hvm_resume(bool suspend_cancelled)
336 {
337
338 xen_hvm_init(suspend_cancelled ?
339 XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME);
340
341 /* Register vcpu_info area for CPU#0. */
342 xen_hvm_cpu_init();
343 }
344
345 static void
xen_hvm_sysinit(void * arg __unused)346 xen_hvm_sysinit(void *arg __unused)
347 {
348 xen_hvm_init(XEN_HVM_INIT_COLD);
349 }
350
351 static void
xen_set_vcpu_id(void)352 xen_set_vcpu_id(void)
353 {
354 struct pcpu *pc;
355 int i;
356
357 if (!xen_hvm_domain())
358 return;
359
360 /* Set vcpu_id to acpi_id */
361 CPU_FOREACH(i) {
362 pc = pcpu_find(i);
363 pc->pc_vcpu_id = pc->pc_acpi_id;
364 if (bootverbose)
365 printf("XEN: CPU %u has VCPU ID %u\n",
366 i, pc->pc_vcpu_id);
367 }
368 }
369
370 static void
xen_hvm_cpu_init(void)371 xen_hvm_cpu_init(void)
372 {
373 struct vcpu_register_vcpu_info info;
374 struct vcpu_info *vcpu_info;
375 int cpu, rc;
376
377 if (!xen_domain())
378 return;
379
380 if (DPCPU_GET(vcpu_info) != NULL) {
381 /*
382 * vcpu_info is already set. We're resuming
383 * from a failed migration and our pre-suspend
384 * configuration is still valid.
385 */
386 return;
387 }
388
389 vcpu_info = DPCPU_PTR(vcpu_local_info);
390 cpu = PCPU_GET(vcpu_id);
391 info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT;
392 info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info));
393
394 rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
395 if (rc != 0)
396 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]);
397 else
398 DPCPU_SET(vcpu_info, vcpu_info);
399 }
400
401 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL);
402 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL);
403 SYSINIT(xen_set_vcpu_id, SI_SUB_CPU, SI_ORDER_ANY, xen_set_vcpu_id, NULL);
404