1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2008, 2013 Citrix Systems, Inc. 5 * Copyright (c) 2012 Spectra Logic Corporation 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include <sys/param.h> 34 #include <sys/bus.h> 35 #include <sys/kernel.h> 36 #include <sys/malloc.h> 37 #include <sys/proc.h> 38 #include <sys/smp.h> 39 #include <sys/systm.h> 40 41 #include <vm/vm.h> 42 #include <vm/pmap.h> 43 #include <vm/vm_param.h> 44 45 #include <dev/pci/pcivar.h> 46 47 #include <machine/cpufunc.h> 48 #include <machine/cpu.h> 49 #include <machine/smp.h> 50 51 #include <x86/apicreg.h> 52 53 #include <xen/xen-os.h> 54 #include <xen/error.h> 55 #include <xen/features.h> 56 #include <xen/gnttab.h> 57 #include <xen/hypervisor.h> 58 #include <xen/hvm.h> 59 #include <xen/xen_intr.h> 60 61 #include <xen/interface/arch-x86/cpuid.h> 62 #include <xen/interface/hvm/params.h> 63 #include <xen/interface/vcpu.h> 64 65 /*--------------------------- Forward Declarations ---------------------------*/ 66 static void xen_hvm_cpu_init(void); 67 68 /*-------------------------------- Global Data -------------------------------*/ 69 enum xen_domain_type xen_domain_type = XEN_NATIVE; 70 71 #ifdef SMP 72 struct cpu_ops xen_hvm_cpu_ops = { 73 .cpu_init = xen_hvm_cpu_init, 74 .cpu_resume = xen_hvm_cpu_init 75 }; 76 #endif 77 78 static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support"); 79 80 /** 81 * If non-zero, the hypervisor has been configured to use a direct 82 * IDT event callback for interrupt injection. 83 */ 84 int xen_vector_callback_enabled; 85 86 /** 87 * Start info flags. ATM this only used to store the initial domain flag for 88 * PVHv2, and it's always empty for HVM guests. 89 */ 90 uint32_t hvm_start_flags; 91 92 /** 93 * Signal whether the vector injected for the event channel upcall requires to 94 * be EOI'ed on the local APIC. 95 */ 96 bool xen_evtchn_needs_ack; 97 98 /*------------------------------- Per-CPU Data -------------------------------*/ 99 DPCPU_DEFINE(struct vcpu_info, vcpu_local_info); 100 DPCPU_DEFINE(struct vcpu_info *, vcpu_info); 101 102 /*------------------ Hypervisor Access Shared Memory Regions -----------------*/ 103 shared_info_t *HYPERVISOR_shared_info; 104 105 /*------------------------------ Sysctl tunables -----------------------------*/ 106 int xen_disable_pv_disks = 0; 107 int xen_disable_pv_nics = 0; 108 TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks); 109 TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics); 110 111 /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/ 112 113 static uint32_t cpuid_base; 114 115 static uint32_t 116 xen_hvm_cpuid_base(void) 117 { 118 uint32_t base, regs[4]; 119 120 for (base = 0x40000000; base < 0x40010000; base += 0x100) { 121 do_cpuid(base, regs); 122 if (!memcmp("XenVMMXenVMM", ®s[1], 12) 123 && (regs[0] - base) >= 2) 124 return (base); 125 } 126 return (0); 127 } 128 129 static void 130 hypervisor_quirks(unsigned int major, unsigned int minor) 131 { 132 #ifdef SMP 133 if (((major < 4) || (major == 4 && minor <= 5)) && 134 msix_disable_migration == -1) { 135 /* 136 * Xen hypervisors prior to 4.6.0 do not properly 137 * handle updates to enabled MSI-X table entries, 138 * so disable MSI-X interrupt migration in that 139 * case. 140 */ 141 if (bootverbose) 142 printf( 143 "Disabling MSI-X interrupt migration due to Xen hypervisor bug.\n" 144 "Set machdep.msix_disable_migration=0 to forcefully enable it.\n"); 145 msix_disable_migration = 1; 146 } 147 #endif 148 } 149 150 static void 151 hypervisor_version(void) 152 { 153 uint32_t regs[4]; 154 int major, minor; 155 156 do_cpuid(cpuid_base + 1, regs); 157 158 major = regs[0] >> 16; 159 minor = regs[0] & 0xffff; 160 printf("XEN: Hypervisor version %d.%d detected.\n", major, minor); 161 162 hypervisor_quirks(major, minor); 163 } 164 165 /* 166 * Allocate and fill in the hypcall page. 167 */ 168 int 169 xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type) 170 { 171 uint32_t regs[4]; 172 173 /* Legacy PVH will get here without the cpuid leaf being set. */ 174 if (cpuid_base == 0) 175 cpuid_base = xen_hvm_cpuid_base(); 176 if (cpuid_base == 0) 177 return (ENXIO); 178 179 if (xen_domain() && init_type == XEN_HVM_INIT_LATE) { 180 /* 181 * If the domain type is already set we can assume that the 182 * hypercall page has been populated too, so just print the 183 * version (and apply any quirks) and exit. 184 */ 185 hypervisor_version(); 186 return 0; 187 } 188 189 if (init_type == XEN_HVM_INIT_LATE) 190 hypervisor_version(); 191 192 /* 193 * Find the hypercall pages. 194 */ 195 do_cpuid(cpuid_base + 2, regs); 196 if (regs[0] != 1) 197 return (EINVAL); 198 199 wrmsr(regs[1], (init_type == XEN_HVM_INIT_EARLY) 200 ? ((vm_paddr_t)&hypercall_page - KERNBASE) 201 : vtophys(&hypercall_page)); 202 203 return (0); 204 } 205 206 static void 207 xen_hvm_init_shared_info_page(void) 208 { 209 struct xen_add_to_physmap xatp; 210 211 if (xen_pv_domain()) { 212 /* 213 * Already setup in the PV case, shared_info is passed inside 214 * of the start_info struct at start of day. 215 */ 216 return; 217 } 218 219 if (HYPERVISOR_shared_info == NULL) { 220 HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT); 221 if (HYPERVISOR_shared_info == NULL) 222 panic("Unable to allocate Xen shared info page"); 223 } 224 225 xatp.domid = DOMID_SELF; 226 xatp.idx = 0; 227 xatp.space = XENMAPSPACE_shared_info; 228 xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT; 229 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) 230 panic("HYPERVISOR_memory_op failed"); 231 } 232 233 static int 234 set_percpu_callback(unsigned int vcpu) 235 { 236 struct xen_hvm_evtchn_upcall_vector vec; 237 int error; 238 239 vec.vcpu = vcpu; 240 vec.vector = IDT_EVTCHN; 241 error = HYPERVISOR_hvm_op(HVMOP_set_evtchn_upcall_vector, &vec); 242 243 return (error != 0 ? xen_translate_error(error) : 0); 244 } 245 246 /* 247 * Tell the hypervisor how to contact us for event channel callbacks. 248 */ 249 void 250 xen_hvm_set_callback(device_t dev) 251 { 252 struct xen_hvm_param xhp; 253 int irq; 254 255 if (xen_vector_callback_enabled) 256 return; 257 258 xhp.domid = DOMID_SELF; 259 xhp.index = HVM_PARAM_CALLBACK_IRQ; 260 if (xen_feature(XENFEAT_hvm_callback_vector) != 0) { 261 int error; 262 263 error = set_percpu_callback(0); 264 if (error == 0) { 265 xen_evtchn_needs_ack = true; 266 /* Trick toolstack to think we are enlightened */ 267 xhp.value = 1; 268 } else 269 xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN); 270 error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp); 271 if (error == 0) { 272 xen_vector_callback_enabled = 1; 273 return; 274 } else if (xen_evtchn_needs_ack) 275 panic("Unable to setup fake HVM param: %d", error); 276 277 printf("Xen HVM callback vector registration failed (%d). " 278 "Falling back to emulated device interrupt\n", error); 279 } 280 xen_vector_callback_enabled = 0; 281 if (dev == NULL) { 282 /* 283 * Called from early boot or resume. 284 * xenpci will invoke us again later. 285 */ 286 return; 287 } 288 289 irq = pci_get_irq(dev); 290 if (irq < 16) { 291 xhp.value = HVM_CALLBACK_GSI(irq); 292 } else { 293 u_int slot; 294 u_int pin; 295 296 slot = pci_get_slot(dev); 297 pin = pci_get_intpin(dev) - 1; 298 xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin); 299 } 300 301 if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0) 302 panic("Can't set evtchn callback"); 303 } 304 305 #define XEN_MAGIC_IOPORT 0x10 306 enum { 307 XMI_MAGIC = 0x49d2, 308 XMI_UNPLUG_IDE_DISKS = 0x01, 309 XMI_UNPLUG_NICS = 0x02, 310 XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04 311 }; 312 313 static void 314 xen_hvm_disable_emulated_devices(void) 315 { 316 u_short disable_devs = 0; 317 318 if (xen_pv_domain()) { 319 /* 320 * No emulated devices in the PV case, so no need to unplug 321 * anything. 322 */ 323 if (xen_disable_pv_disks != 0 || xen_disable_pv_nics != 0) 324 printf("PV devices cannot be disabled in PV guests\n"); 325 return; 326 } 327 328 if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC) 329 return; 330 331 if (xen_disable_pv_disks == 0) { 332 if (bootverbose) 333 printf("XEN: disabling emulated disks\n"); 334 disable_devs |= XMI_UNPLUG_IDE_DISKS; 335 } 336 if (xen_disable_pv_nics == 0) { 337 if (bootverbose) 338 printf("XEN: disabling emulated nics\n"); 339 disable_devs |= XMI_UNPLUG_NICS; 340 } 341 342 if (disable_devs != 0) 343 outw(XEN_MAGIC_IOPORT, disable_devs); 344 } 345 346 static void 347 xen_hvm_init(enum xen_hvm_init_type init_type) 348 { 349 int error; 350 int i; 351 352 if (init_type == XEN_HVM_INIT_CANCELLED_SUSPEND) 353 return; 354 355 error = xen_hvm_init_hypercall_stubs(init_type); 356 357 switch (init_type) { 358 case XEN_HVM_INIT_LATE: 359 if (error != 0) 360 return; 361 362 /* 363 * If xen_domain_type is not set at this point 364 * it means we are inside a (PV)HVM guest, because 365 * for PVH the guest type is set much earlier 366 * (see hammer_time_xen). 367 */ 368 if (!xen_domain()) { 369 xen_domain_type = XEN_HVM_DOMAIN; 370 vm_guest = VM_GUEST_XEN; 371 } 372 373 setup_xen_features(); 374 #ifdef SMP 375 cpu_ops = xen_hvm_cpu_ops; 376 #endif 377 break; 378 case XEN_HVM_INIT_RESUME: 379 if (error != 0) 380 panic("Unable to init Xen hypercall stubs on resume"); 381 382 /* Clear stale vcpu_info. */ 383 CPU_FOREACH(i) 384 DPCPU_ID_SET(i, vcpu_info, NULL); 385 break; 386 default: 387 panic("Unsupported HVM initialization type"); 388 } 389 390 xen_vector_callback_enabled = 0; 391 xen_evtchn_needs_ack = false; 392 xen_hvm_set_callback(NULL); 393 394 /* 395 * On (PV)HVM domains we need to request the hypervisor to 396 * fill the shared info page, for PVH guest the shared_info page 397 * is passed inside the start_info struct and is already set, so this 398 * functions are no-ops. 399 */ 400 xen_hvm_init_shared_info_page(); 401 xen_hvm_disable_emulated_devices(); 402 } 403 404 void 405 xen_hvm_suspend(void) 406 { 407 } 408 409 void 410 xen_hvm_resume(bool suspend_cancelled) 411 { 412 413 xen_hvm_init(suspend_cancelled ? 414 XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME); 415 416 /* Register vcpu_info area for CPU#0. */ 417 xen_hvm_cpu_init(); 418 } 419 420 static void 421 xen_hvm_sysinit(void *arg __unused) 422 { 423 xen_hvm_init(XEN_HVM_INIT_LATE); 424 } 425 SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL); 426 427 static void 428 xen_hvm_cpu_init(void) 429 { 430 struct vcpu_register_vcpu_info info; 431 struct vcpu_info *vcpu_info; 432 uint32_t regs[4]; 433 int cpu, rc; 434 435 if (!xen_domain()) 436 return; 437 438 if (DPCPU_GET(vcpu_info) != NULL) { 439 /* 440 * vcpu_info is already set. We're resuming 441 * from a failed migration and our pre-suspend 442 * configuration is still valid. 443 */ 444 return; 445 } 446 447 /* 448 * Set vCPU ID. If available fetch the ID from CPUID, if not just use 449 * the ACPI ID. 450 */ 451 KASSERT(cpuid_base != 0, ("Invalid base Xen CPUID leaf")); 452 cpuid_count(cpuid_base + 4, 0, regs); 453 KASSERT((regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) || 454 !xen_pv_domain(), 455 ("Xen PV domain without vcpu_id in cpuid")); 456 PCPU_SET(vcpu_id, (regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) ? 457 regs[1] : PCPU_GET(acpi_id)); 458 459 if (xen_evtchn_needs_ack && !IS_BSP()) { 460 /* 461 * Setup the per-vpcu event channel upcall vector. This is only 462 * required when using the new HVMOP_set_evtchn_upcall_vector 463 * hypercall, which allows using a different vector for each 464 * vCPU. Note that FreeBSD uses the same vector for all vCPUs 465 * because it's not dynamically allocated. 466 */ 467 rc = set_percpu_callback(PCPU_GET(vcpu_id)); 468 if (rc != 0) 469 panic("Event channel upcall vector setup failed: %d", 470 rc); 471 } 472 473 /* 474 * Set the vCPU info. 475 * 476 * NB: the vCPU info for vCPUs < 32 can be fetched from the shared info 477 * page, but in order to make sure the mapping code is correct always 478 * attempt to map the vCPU info at a custom place. 479 */ 480 vcpu_info = DPCPU_PTR(vcpu_local_info); 481 cpu = PCPU_GET(vcpu_id); 482 info.mfn = vtophys(vcpu_info) >> PAGE_SHIFT; 483 info.offset = vtophys(vcpu_info) - trunc_page(vtophys(vcpu_info)); 484 485 rc = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info); 486 if (rc != 0) 487 DPCPU_SET(vcpu_info, &HYPERVISOR_shared_info->vcpu_info[cpu]); 488 else 489 DPCPU_SET(vcpu_info, vcpu_info); 490 } 491 SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL); 492