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 #ifdef SMP 62 static void xen_hvm_cpu_resume(void); 63 #endif 64 static void xen_hvm_cpu_init(void); 65 66 /*---------------------------- Extern Declarations ---------------------------*/ 67 /* Variables used by mp_machdep to perform the bitmap IPI */ 68 extern volatile u_int cpu_ipi_pending[MAXCPU]; 69 70 /*-------------------------------- Local Types -------------------------------*/ 71 enum xen_hvm_init_type { 72 XEN_HVM_INIT_COLD, 73 XEN_HVM_INIT_CANCELLED_SUSPEND, 74 XEN_HVM_INIT_RESUME 75 }; 76 77 /*-------------------------------- Global Data -------------------------------*/ 78 enum xen_domain_type xen_domain_type = XEN_NATIVE; 79 80 #ifdef SMP 81 struct cpu_ops xen_hvm_cpu_ops = { 82 .cpu_init = xen_hvm_cpu_init, 83 .cpu_resume = xen_hvm_cpu_resume 84 }; 85 #endif 86 87 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support"); 88 89 /** 90 * If non-zero, the hypervisor has been configured to use a direct 91 * IDT event callback for interrupt injection. 92 */ 93 int xen_vector_callback_enabled; 94 95 /*------------------------------- Per-CPU Data -------------------------------*/ 96 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info); 97 DPCPU_DEFINE(struct vcpu_info *, vcpu_info); 98 99 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/ 100 /** Hypercall table accessed via HYPERVISOR_*_op() methods. */ 101 extern char *hypercall_page; 102 shared_info_t *HYPERVISOR_shared_info; 103 start_info_t *HYPERVISOR_start_info; 104 105 #ifdef SMP 106 /* XEN diverged cpu operations */ 107 static void 108 xen_hvm_cpu_resume(void) 109 { 110 u_int cpuid = PCPU_GET(cpuid); 111 112 /* 113 * Reset pending bitmap IPIs, because Xen doesn't preserve pending 114 * event channels on migration. 115 */ 116 cpu_ipi_pending[cpuid] = 0; 117 118 /* register vcpu_info area */ 119 xen_hvm_cpu_init(); 120 } 121 #endif 122 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/ 123 static uint32_t 124 xen_hvm_cpuid_base(void) 125 { 126 uint32_t base, regs[4]; 127 128 for (base = 0x40000000; base < 0x40010000; base += 0x100) { 129 do_cpuid(base, regs); 130 if (!memcmp("XenVMMXenVMM", ®s[1], 12) 131 && (regs[0] - base) >= 2) 132 return (base); 133 } 134 return (0); 135 } 136 137 /* 138 * Allocate and fill in the hypcall page. 139 */ 140 static int 141 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type) 142 { 143 uint32_t base, regs[4]; 144 int i; 145 146 if (xen_pv_domain()) { 147 /* hypercall page is already set in the PV case */ 148 return (0); 149 } 150 151 base = xen_hvm_cpuid_base(); 152 if (base == 0) 153 return (ENXIO); 154 155 if (init_type == XEN_HVM_INIT_COLD) { 156 do_cpuid(base + 1, regs); 157 printf("XEN: Hypervisor version %d.%d detected.\n", 158 regs[0] >> 16, regs[0] & 0xffff); 159 } 160 161 /* 162 * Find the hypercall pages. 163 */ 164 do_cpuid(base + 2, regs); 165 166 for (i = 0; i < regs[0]; i++) 167 wrmsr(regs[1], vtophys(&hypercall_page + i * PAGE_SIZE) + i); 168 169 return (0); 170 } 171 172 static void 173 xen_hvm_init_shared_info_page(void) 174 { 175 struct xen_add_to_physmap xatp; 176 177 if (xen_pv_domain()) { 178 /* 179 * Already setup in the PV case, shared_info is passed inside 180 * of the start_info struct at start of day. 181 */ 182 return; 183 } 184 185 if (HYPERVISOR_shared_info == NULL) { 186 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT); 187 if (HYPERVISOR_shared_info == NULL) 188 panic("Unable to allocate Xen shared info page"); 189 } 190 191 xatp.domid = DOMID_SELF; 192 xatp.idx = 0; 193 xatp.space = XENMAPSPACE_shared_info; 194 xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT; 195 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 196 panic("HYPERVISOR_memory_op failed"); 197 } 198 199 /* 200 * Tell the hypervisor how to contact us for event channel callbacks. 201 */ 202 void 203 xen_hvm_set_callback(device_t dev) 204 { 205 struct xen_hvm_param xhp; 206 int irq; 207 208 if (xen_vector_callback_enabled) 209 return; 210 211 xhp.domid = DOMID_SELF; 212 xhp.index = HVM_PARAM_CALLBACK_IRQ; 213 if (xen_feature(XENFEAT_hvm_callback_vector) != 0) { 214 int error; 215 216 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN); 217 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp); 218 if (error == 0) { 219 xen_vector_callback_enabled = 1; 220 return; 221 } 222 printf("Xen HVM callback vector registration failed (%d). " 223 "Falling back to emulated device interrupt\n", error); 224 } 225 xen_vector_callback_enabled = 0; 226 if (dev == NULL) { 227 /* 228 * Called from early boot or resume. 229 * xenpci will invoke us again later. 230 */ 231 return; 232 } 233 234 irq = pci_get_irq(dev); 235 if (irq < 16) { 236 xhp.value = HVM_CALLBACK_GSI(irq); 237 } else { 238 u_int slot; 239 u_int pin; 240 241 slot = pci_get_slot(dev); 242 pin = pci_get_intpin(dev) - 1; 243 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin); 244 } 245 246 if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0) 247 panic("Can't set evtchn callback"); 248 } 249 250 #define XEN_MAGIC_IOPORT 0x10 251 enum { 252 XMI_MAGIC = 0x49d2, 253 XMI_UNPLUG_IDE_DISKS = 0x01, 254 XMI_UNPLUG_NICS = 0x02, 255 XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04 256 }; 257 258 static void 259 xen_hvm_disable_emulated_devices(void) 260 { 261 262 if (xen_pv_domain()) { 263 /* 264 * No emulated devices in the PV case, so no need to unplug 265 * anything. 266 */ 267 return; 268 } 269 270 if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC) 271 return; 272 273 if (bootverbose) 274 printf("XEN: Disabling emulated block and network devices\n"); 275 outw(XEN_MAGIC_IOPORT, XMI_UNPLUG_IDE_DISKS|XMI_UNPLUG_NICS); 276 } 277 278 static void 279 xen_hvm_init(enum xen_hvm_init_type init_type) 280 { 281 int error; 282 int i; 283 284 if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND) 285 return; 286 287 error = xen_hvm_init_hypercall_stubs(init_type); 288 289 switch (init_type) { 290 case XEN_HVM_INIT_COLD: 291 if (error != 0) 292 return; 293 294 /* 295 * If xen_domain_type is not set at this point 296 * it means we are inside a (PV)HVM guest, because 297 * for PVH the guest type is set much earlier 298 * (see hammer_time_xen). 299 */ 300 if (!xen_domain()) { 301 xen_domain_type = XEN_HVM_DOMAIN; 302 vm_guest = VM_GUEST_XEN; 303 } 304 305 setup_xen_features(); 306 #ifdef SMP 307 cpu_ops = xen_hvm_cpu_ops; 308 #endif 309 break; 310 case XEN_HVM_INIT_RESUME: 311 if (error != 0) 312 panic("Unable to init Xen hypercall stubs on resume"); 313 314 /* Clear stale vcpu_info. */ 315 CPU_FOREACH(i) 316 DPCPU_ID_SET(i, vcpu_info, NULL); 317 break; 318 default: 319 panic("Unsupported HVM initialization type"); 320 } 321 322 xen_vector_callback_enabled = 0; 323 xen_hvm_set_callback(NULL); 324 325 /* 326 * On (PV)HVM domains we need to request the hypervisor to 327 * fill the shared info page, for PVH guest the shared_info page 328 * is passed inside the start_info struct and is already set, so this 329 * functions are no-ops. 330 */ 331 xen_hvm_init_shared_info_page(); 332 xen_hvm_disable_emulated_devices(); 333 } 334 335 void 336 xen_hvm_suspend(void) 337 { 338 } 339 340 void 341 xen_hvm_resume(bool suspend_cancelled) 342 { 343 344 xen_hvm_init(suspend_cancelled ? 345 XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME); 346 347 /* Register vcpu_info area for CPU#0. */ 348 xen_hvm_cpu_init(); 349 } 350 351 static void 352 xen_hvm_sysinit(void *arg __unused) 353 { 354 xen_hvm_init(XEN_HVM_INIT_COLD); 355 } 356 357 static void 358 xen_set_vcpu_id(void) 359 { 360 struct pcpu *pc; 361 int i; 362 363 if (!xen_hvm_domain()) 364 return; 365 366 /* Set vcpu_id to acpi_id */ 367 CPU_FOREACH(i) { 368 pc = pcpu_find(i); 369 pc->pc_vcpu_id = pc->pc_acpi_id; 370 if (bootverbose) 371 printf("XEN: CPU %u has VCPU ID %u\n", 372 i, pc->pc_vcpu_id); 373 } 374 } 375 376 static void 377 xen_hvm_cpu_init(void) 378 { 379 struct vcpu_register_vcpu_info info; 380 struct vcpu_info *vcpu_info; 381 int cpu, rc; 382 383 if (!xen_domain()) 384 return; 385 386 if (DPCPU_GET(vcpu_info) != NULL) { 387 /* 388 * vcpu_info is already set. We're resuming 389 * from a failed migration and our pre-suspend 390 * configuration is still valid. 391 */ 392 return; 393 } 394 395 vcpu_info = DPCPU_PTR(vcpu_local_info); 396 cpu = PCPU_GET(vcpu_id); 397 info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT; 398 info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info)); 399 400 rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 401 if (rc != 0) 402 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]); 403 else 404 DPCPU_SET(vcpu_info, vcpu_info); 405 } 406 407 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL); 408 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL); 409 SYSINIT(xen_set_vcpu_id, SI_SUB_CPU, SI_ORDER_ANY, xen_set_vcpu_id, NULL); 410