1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 3 * 4 * Copyright (c) 2004 Christian Limpach. 5 * Copyright (c) 2004-2006,2008 Kip Macy 6 * Copyright (c) 2008 The NetBSD Foundation, Inc. 7 * Copyright (c) 2013 Roger Pau Monné <roger.pau@citrix.com> 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_ddb.h" 36 #include "opt_kstack_pages.h" 37 38 #include <sys/param.h> 39 #include <sys/bus.h> 40 #include <sys/kernel.h> 41 #include <sys/reboot.h> 42 #include <sys/systm.h> 43 #include <sys/malloc.h> 44 #include <sys/linker.h> 45 #include <sys/lock.h> 46 #include <sys/rwlock.h> 47 #include <sys/boot.h> 48 #include <sys/ctype.h> 49 #include <sys/mutex.h> 50 #include <sys/smp.h> 51 52 #include <vm/vm.h> 53 #include <vm/vm_extern.h> 54 #include <vm/vm_kern.h> 55 #include <vm/vm_page.h> 56 #include <vm/vm_map.h> 57 #include <vm/vm_object.h> 58 #include <vm/vm_pager.h> 59 #include <vm/vm_param.h> 60 61 #include <machine/intr_machdep.h> 62 #include <x86/apicvar.h> 63 #include <x86/init.h> 64 #include <machine/pc/bios.h> 65 #include <machine/smp.h> 66 #include <machine/intr_machdep.h> 67 #include <machine/metadata.h> 68 69 #include <xen/xen-os.h> 70 #include <xen/hypervisor.h> 71 #include <xen/xenstore/xenstorevar.h> 72 #include <xen/xen_pv.h> 73 #include <xen/xen_msi.h> 74 75 #include <xen/interface/vcpu.h> 76 77 #include <dev/xen/timer/timer.h> 78 79 #ifdef DDB 80 #include <ddb/ddb.h> 81 #endif 82 83 /* Native initial function */ 84 extern u_int64_t hammer_time(u_int64_t, u_int64_t); 85 /* Xen initial function */ 86 uint64_t hammer_time_xen(start_info_t *, uint64_t); 87 88 #define MAX_E820_ENTRIES 128 89 90 /*--------------------------- Forward Declarations ---------------------------*/ 91 static caddr_t xen_pv_parse_preload_data(u_int64_t); 92 static void xen_pv_parse_memmap(caddr_t, vm_paddr_t *, int *); 93 94 #ifdef SMP 95 static int xen_pv_start_all_aps(void); 96 #endif 97 98 /*---------------------------- Extern Declarations ---------------------------*/ 99 #ifdef SMP 100 /* Variables used by amd64 mp_machdep to start APs */ 101 extern char *doublefault_stack; 102 extern char *mce_stack; 103 extern char *nmi_stack; 104 extern char *dbg_stack; 105 #endif 106 107 /* 108 * Placed by the linker at the end of the bss section, which is the last 109 * section loaded by Xen before loading the symtab and strtab. 110 */ 111 extern uint32_t end; 112 113 /*-------------------------------- Global Data -------------------------------*/ 114 /* Xen init_ops implementation. */ 115 struct init_ops xen_init_ops = { 116 .parse_preload_data = xen_pv_parse_preload_data, 117 .early_clock_source_init = xen_clock_init, 118 .early_delay = xen_delay, 119 .parse_memmap = xen_pv_parse_memmap, 120 #ifdef SMP 121 .start_all_aps = xen_pv_start_all_aps, 122 #endif 123 .msi_init = xen_msi_init, 124 }; 125 126 static struct bios_smap xen_smap[MAX_E820_ENTRIES]; 127 128 /*-------------------------------- Xen PV init -------------------------------*/ 129 /* 130 * First function called by the Xen PVH boot sequence. 131 * 132 * Set some Xen global variables and prepare the environment so it is 133 * as similar as possible to what native FreeBSD init function expects. 134 */ 135 uint64_t 136 hammer_time_xen(start_info_t *si, uint64_t xenstack) 137 { 138 uint64_t physfree; 139 uint64_t *PT4 = (u_int64_t *)xenstack; 140 uint64_t *PT3 = (u_int64_t *)(xenstack + PAGE_SIZE); 141 uint64_t *PT2 = (u_int64_t *)(xenstack + 2 * PAGE_SIZE); 142 int i; 143 144 xen_domain_type = XEN_PV_DOMAIN; 145 vm_guest = VM_GUEST_XEN; 146 147 if ((si == NULL) || (xenstack == 0)) { 148 xc_printf("ERROR: invalid start_info or xen stack, halting\n"); 149 HYPERVISOR_shutdown(SHUTDOWN_crash); 150 } 151 152 xc_printf("FreeBSD PVH running on %s\n", si->magic); 153 154 /* We use 3 pages of xen stack for the boot pagetables */ 155 physfree = xenstack + 3 * PAGE_SIZE - KERNBASE; 156 157 /* Setup Xen global variables */ 158 HYPERVISOR_start_info = si; 159 HYPERVISOR_shared_info = 160 (shared_info_t *)(si->shared_info + KERNBASE); 161 162 /* 163 * Setup some misc global variables for Xen devices 164 * 165 * XXX: Devices that need these specific variables should 166 * be rewritten to fetch this info by themselves from the 167 * start_info page. 168 */ 169 xen_store = (struct xenstore_domain_interface *) 170 (ptoa(si->store_mfn) + KERNBASE); 171 console_page = (char *)(ptoa(si->console.domU.mfn) + KERNBASE); 172 173 /* 174 * Use the stack Xen gives us to build the page tables 175 * as native FreeBSD expects to find them (created 176 * by the boot trampoline). 177 */ 178 for (i = 0; i < (PAGE_SIZE / sizeof(uint64_t)); i++) { 179 /* 180 * Each slot of the level 4 pages points 181 * to the same level 3 page 182 */ 183 PT4[i] = ((uint64_t)&PT3[0]) - KERNBASE; 184 PT4[i] |= PG_V | PG_RW | PG_U; 185 186 /* 187 * Each slot of the level 3 pages points 188 * to the same level 2 page 189 */ 190 PT3[i] = ((uint64_t)&PT2[0]) - KERNBASE; 191 PT3[i] |= PG_V | PG_RW | PG_U; 192 193 /* 194 * The level 2 page slots are mapped with 195 * 2MB pages for 1GB. 196 */ 197 PT2[i] = i * (2 * 1024 * 1024); 198 PT2[i] |= PG_V | PG_RW | PG_PS | PG_U; 199 } 200 load_cr3(((uint64_t)&PT4[0]) - KERNBASE); 201 202 /* Set the hooks for early functions that diverge from bare metal */ 203 init_ops = xen_init_ops; 204 apic_ops = xen_apic_ops; 205 206 /* Now we can jump into the native init function */ 207 return (hammer_time(0, physfree)); 208 } 209 210 /*-------------------------------- PV specific -------------------------------*/ 211 #ifdef SMP 212 static bool 213 start_xen_ap(int cpu) 214 { 215 struct vcpu_guest_context *ctxt; 216 int ms, cpus = mp_naps; 217 const size_t stacksize = kstack_pages * PAGE_SIZE; 218 219 /* allocate and set up an idle stack data page */ 220 bootstacks[cpu] = 221 (void *)kmem_malloc(kernel_arena, stacksize, M_WAITOK | M_ZERO); 222 doublefault_stack = 223 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO); 224 mce_stack = 225 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO); 226 nmi_stack = 227 (char *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO); 228 dbg_stack = 229 (void *)kmem_malloc(kernel_arena, PAGE_SIZE, M_WAITOK | M_ZERO); 230 dpcpu = 231 (void *)kmem_malloc(kernel_arena, DPCPU_SIZE, M_WAITOK | M_ZERO); 232 233 bootSTK = (char *)bootstacks[cpu] + kstack_pages * PAGE_SIZE - 8; 234 bootAP = cpu; 235 236 ctxt = malloc(sizeof(*ctxt), M_TEMP, M_WAITOK | M_ZERO); 237 238 ctxt->flags = VGCF_IN_KERNEL; 239 ctxt->user_regs.rip = (unsigned long) init_secondary; 240 ctxt->user_regs.rsp = (unsigned long) bootSTK; 241 242 /* Set the AP to use the same page tables */ 243 ctxt->ctrlreg[3] = KPML4phys; 244 245 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) 246 panic("unable to initialize AP#%d", cpu); 247 248 free(ctxt, M_TEMP); 249 250 /* Launch the vCPU */ 251 if (HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) 252 panic("unable to start AP#%d", cpu); 253 254 /* Wait up to 5 seconds for it to start. */ 255 for (ms = 0; ms < 5000; ms++) { 256 if (mp_naps > cpus) 257 return (true); 258 DELAY(1000); 259 } 260 261 return (false); 262 } 263 264 static int 265 xen_pv_start_all_aps(void) 266 { 267 int cpu; 268 269 mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); 270 271 for (cpu = 1; cpu < mp_ncpus; cpu++) { 272 273 /* attempt to start the Application Processor */ 274 if (!start_xen_ap(cpu)) 275 panic("AP #%d failed to start!", cpu); 276 277 CPU_SET(cpu, &all_cpus); /* record AP in CPU map */ 278 } 279 280 return (mp_naps); 281 } 282 #endif /* SMP */ 283 284 /* 285 * Functions to convert the "extra" parameters passed by Xen 286 * into FreeBSD boot options. 287 */ 288 static void 289 xen_pv_set_env(void) 290 { 291 char *cmd_line_next, *cmd_line; 292 size_t env_size; 293 294 cmd_line = HYPERVISOR_start_info->cmd_line; 295 env_size = sizeof(HYPERVISOR_start_info->cmd_line); 296 297 /* Skip leading spaces */ 298 for (; isspace(*cmd_line) && (env_size != 0); cmd_line++) 299 env_size--; 300 301 /* Replace ',' with '\0' */ 302 for (cmd_line_next = cmd_line; strsep(&cmd_line_next, ",") != NULL;) 303 ; 304 305 init_static_kenv(cmd_line, 0); 306 } 307 308 static void 309 xen_pv_set_boothowto(void) 310 { 311 int i; 312 char *env; 313 314 /* get equivalents from the environment */ 315 for (i = 0; howto_names[i].ev != NULL; i++) { 316 if ((env = kern_getenv(howto_names[i].ev)) != NULL) { 317 boothowto |= howto_names[i].mask; 318 freeenv(env); 319 } 320 } 321 } 322 323 #ifdef DDB 324 /* 325 * The way Xen loads the symtab is different from the native boot loader, 326 * because it's tailored for NetBSD. So we have to adapt and use the same 327 * method as NetBSD. Portions of the code below have been picked from NetBSD: 328 * sys/kern/kern_ksyms.c CVS Revision 1.71. 329 */ 330 static void 331 xen_pv_parse_symtab(void) 332 { 333 Elf_Ehdr *ehdr; 334 Elf_Shdr *shdr; 335 vm_offset_t sym_end; 336 uint32_t size; 337 int i, j; 338 339 size = end; 340 sym_end = HYPERVISOR_start_info->mod_start != 0 ? 341 HYPERVISOR_start_info->mod_start : 342 HYPERVISOR_start_info->mfn_list; 343 344 /* 345 * Make sure the size is right headed, sym_end is just a 346 * high boundary, but at least allows us to fail earlier. 347 */ 348 if ((vm_offset_t)&end + size > sym_end) { 349 xc_printf("Unable to load ELF symtab: size mismatch\n"); 350 return; 351 } 352 353 ehdr = (Elf_Ehdr *)(&end + 1); 354 if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) || 355 ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || 356 ehdr->e_version > 1) { 357 xc_printf("Unable to load ELF symtab: invalid symbol table\n"); 358 return; 359 } 360 361 shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff); 362 /* Find the symbol table and the corresponding string table. */ 363 for (i = 1; i < ehdr->e_shnum; i++) { 364 if (shdr[i].sh_type != SHT_SYMTAB) 365 continue; 366 if (shdr[i].sh_offset == 0) 367 continue; 368 ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset); 369 ksymtab_size = shdr[i].sh_size; 370 j = shdr[i].sh_link; 371 if (shdr[j].sh_offset == 0) 372 continue; /* Can this happen? */ 373 kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset); 374 break; 375 } 376 377 if (ksymtab == 0 || kstrtab == 0) { 378 xc_printf( 379 "Unable to load ELF symtab: could not find symtab or strtab\n"); 380 return; 381 } 382 } 383 #endif 384 385 static caddr_t 386 xen_pv_parse_preload_data(u_int64_t modulep) 387 { 388 caddr_t kmdp; 389 vm_ooffset_t off; 390 vm_paddr_t metadata; 391 char *envp; 392 393 if (HYPERVISOR_start_info->mod_start != 0) { 394 preload_metadata = (caddr_t)(HYPERVISOR_start_info->mod_start); 395 396 kmdp = preload_search_by_type("elf kernel"); 397 if (kmdp == NULL) 398 kmdp = preload_search_by_type("elf64 kernel"); 399 KASSERT(kmdp != NULL, ("unable to find kernel")); 400 401 /* 402 * Xen has relocated the metadata and the modules, 403 * so we need to recalculate it's position. This is 404 * done by saving the original modulep address and 405 * then calculating the offset with mod_start, 406 * which contains the relocated modulep address. 407 */ 408 metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t); 409 off = HYPERVISOR_start_info->mod_start - metadata; 410 411 preload_bootstrap_relocate(off); 412 413 boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int); 414 envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *); 415 if (envp != NULL) 416 envp += off; 417 init_static_kenv(envp, 0); 418 } else { 419 /* Parse the extra boot information given by Xen */ 420 xen_pv_set_env(); 421 xen_pv_set_boothowto(); 422 kmdp = NULL; 423 } 424 425 #ifdef DDB 426 xen_pv_parse_symtab(); 427 #endif 428 return (kmdp); 429 } 430 431 static void 432 xen_pv_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx) 433 { 434 struct xen_memory_map memmap; 435 u_int32_t size; 436 int rc; 437 438 /* Fetch the E820 map from Xen */ 439 memmap.nr_entries = MAX_E820_ENTRIES; 440 set_xen_guest_handle(memmap.buffer, xen_smap); 441 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); 442 if (rc) 443 panic("unable to fetch Xen E820 memory map"); 444 size = memmap.nr_entries * sizeof(xen_smap[0]); 445 446 bios_add_smap_entries(xen_smap, size, physmap, physmap_idx); 447 } 448