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