1 /*- 2 * Copyright (c) 2004 Marcel Moolenaar 3 * Copyright (c) 2001 Doug Rabson 4 * Copyright (c) 2016 The FreeBSD Foundation 5 * All rights reserved. 6 * 7 * Portions of this software were developed by Konstantin Belousov 8 * under sponsorship from the FreeBSD Foundation. 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 #include <sys/param.h> 34 #include <sys/efi.h> 35 #include <sys/kernel.h> 36 #include <sys/linker.h> 37 #include <sys/lock.h> 38 #include <sys/module.h> 39 #include <sys/mutex.h> 40 #include <sys/clock.h> 41 #include <sys/proc.h> 42 #include <sys/rwlock.h> 43 #include <sys/sched.h> 44 #include <sys/sysctl.h> 45 #include <sys/systm.h> 46 #include <sys/vmmeter.h> 47 #include <isa/rtc.h> 48 #include <machine/fpu.h> 49 #include <machine/efi.h> 50 #include <machine/metadata.h> 51 #include <machine/md_var.h> 52 #include <machine/smp.h> 53 #include <machine/vmparam.h> 54 #include <vm/vm.h> 55 #include <vm/pmap.h> 56 #include <vm/vm_extern.h> 57 #include <vm/vm_map.h> 58 #include <vm/vm_object.h> 59 #include <vm/vm_page.h> 60 #include <vm/vm_pager.h> 61 62 static pml5_entry_t *efi_pml5; 63 static pml4_entry_t *efi_pml4; 64 static vm_object_t obj_1t1_pt; 65 static vm_page_t efi_pmltop_page; 66 static vm_pindex_t efi_1t1_idx; 67 68 void 69 efi_destroy_1t1_map(void) 70 { 71 vm_page_t m; 72 73 if (obj_1t1_pt != NULL) { 74 VM_OBJECT_RLOCK(obj_1t1_pt); 75 TAILQ_FOREACH(m, &obj_1t1_pt->memq, listq) 76 m->ref_count = VPRC_OBJREF; 77 vm_wire_sub(obj_1t1_pt->resident_page_count); 78 VM_OBJECT_RUNLOCK(obj_1t1_pt); 79 vm_object_deallocate(obj_1t1_pt); 80 } 81 82 obj_1t1_pt = NULL; 83 efi_pml4 = NULL; 84 efi_pml5 = NULL; 85 efi_pmltop_page = NULL; 86 } 87 88 /* 89 * Map a physical address from EFI runtime space into KVA space. Returns 0 to 90 * indicate a failed mapping so that the caller may handle error. 91 */ 92 vm_offset_t 93 efi_phys_to_kva(vm_paddr_t paddr) 94 { 95 96 if (paddr >= dmaplimit) 97 return (0); 98 return (PHYS_TO_DMAP(paddr)); 99 } 100 101 static vm_page_t 102 efi_1t1_page(void) 103 { 104 105 return (vm_page_grab(obj_1t1_pt, efi_1t1_idx++, VM_ALLOC_NOBUSY | 106 VM_ALLOC_WIRED | VM_ALLOC_ZERO)); 107 } 108 109 static pt_entry_t * 110 efi_1t1_pte(vm_offset_t va) 111 { 112 pml5_entry_t *pml5e; 113 pml4_entry_t *pml4e; 114 pdp_entry_t *pdpe; 115 pd_entry_t *pde; 116 pt_entry_t *pte; 117 vm_page_t m; 118 vm_pindex_t pml5_idx, pml4_idx, pdp_idx, pd_idx; 119 vm_paddr_t mphys; 120 121 pml4_idx = pmap_pml4e_index(va); 122 if (la57) { 123 pml5_idx = pmap_pml5e_index(va); 124 pml5e = &efi_pml5[pml5_idx]; 125 if (*pml5e == 0) { 126 m = efi_1t1_page(); 127 mphys = VM_PAGE_TO_PHYS(m); 128 *pml5e = mphys | X86_PG_RW | X86_PG_V; 129 } else { 130 mphys = *pml5e & PG_FRAME; 131 } 132 pml4e = (pml4_entry_t *)PHYS_TO_DMAP(mphys); 133 pml4e = &pml4e[pml4_idx]; 134 } else { 135 pml4e = &efi_pml4[pml4_idx]; 136 } 137 138 if (*pml4e == 0) { 139 m = efi_1t1_page(); 140 mphys = VM_PAGE_TO_PHYS(m); 141 *pml4e = mphys | X86_PG_RW | X86_PG_V; 142 } else { 143 mphys = *pml4e & PG_FRAME; 144 } 145 146 pdpe = (pdp_entry_t *)PHYS_TO_DMAP(mphys); 147 pdp_idx = pmap_pdpe_index(va); 148 pdpe += pdp_idx; 149 if (*pdpe == 0) { 150 m = efi_1t1_page(); 151 mphys = VM_PAGE_TO_PHYS(m); 152 *pdpe = mphys | X86_PG_RW | X86_PG_V; 153 } else { 154 mphys = *pdpe & PG_FRAME; 155 } 156 157 pde = (pd_entry_t *)PHYS_TO_DMAP(mphys); 158 pd_idx = pmap_pde_index(va); 159 pde += pd_idx; 160 if (*pde == 0) { 161 m = efi_1t1_page(); 162 mphys = VM_PAGE_TO_PHYS(m); 163 *pde = mphys | X86_PG_RW | X86_PG_V; 164 } else { 165 mphys = *pde & PG_FRAME; 166 } 167 168 pte = (pt_entry_t *)PHYS_TO_DMAP(mphys); 169 pte += pmap_pte_index(va); 170 KASSERT(*pte == 0, ("va %#jx *pt %#jx", va, *pte)); 171 172 return (pte); 173 } 174 175 bool 176 efi_create_1t1_map(struct efi_md *map, int ndesc, int descsz) 177 { 178 struct efi_md *p; 179 pt_entry_t *pte; 180 void *pml; 181 vm_page_t m; 182 vm_offset_t va; 183 uint64_t idx; 184 int bits, i, mode; 185 186 obj_1t1_pt = vm_pager_allocate(OBJT_PHYS, NULL, ptoa(1 + 187 NPML4EPG + NPML4EPG * NPDPEPG + NPML4EPG * NPDPEPG * NPDEPG), 188 VM_PROT_ALL, 0, NULL); 189 efi_1t1_idx = 0; 190 VM_OBJECT_WLOCK(obj_1t1_pt); 191 efi_pmltop_page = efi_1t1_page(); 192 VM_OBJECT_WUNLOCK(obj_1t1_pt); 193 pml = (void *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(efi_pmltop_page)); 194 if (la57) { 195 efi_pml5 = pml; 196 pmap_pinit_pml5(efi_pmltop_page); 197 } else { 198 efi_pml4 = pml; 199 pmap_pinit_pml4(efi_pmltop_page); 200 } 201 202 for (i = 0, p = map; i < ndesc; i++, p = efi_next_descriptor(p, 203 descsz)) { 204 if ((p->md_attr & EFI_MD_ATTR_RT) == 0) 205 continue; 206 if (p->md_virt != 0 && p->md_virt != p->md_phys) { 207 if (bootverbose) 208 printf("EFI Runtime entry %d is mapped\n", i); 209 goto fail; 210 } 211 if ((p->md_phys & EFI_PAGE_MASK) != 0) { 212 if (bootverbose) 213 printf("EFI Runtime entry %d is not aligned\n", 214 i); 215 goto fail; 216 } 217 if (p->md_phys + p->md_pages * EFI_PAGE_SIZE < p->md_phys || 218 p->md_phys + p->md_pages * EFI_PAGE_SIZE >= 219 VM_MAXUSER_ADDRESS) { 220 printf("EFI Runtime entry %d is not in mappable for RT:" 221 "base %#016jx %#jx pages\n", 222 i, (uintmax_t)p->md_phys, 223 (uintmax_t)p->md_pages); 224 goto fail; 225 } 226 if ((p->md_attr & EFI_MD_ATTR_WB) != 0) 227 mode = VM_MEMATTR_WRITE_BACK; 228 else if ((p->md_attr & EFI_MD_ATTR_WT) != 0) 229 mode = VM_MEMATTR_WRITE_THROUGH; 230 else if ((p->md_attr & EFI_MD_ATTR_WC) != 0) 231 mode = VM_MEMATTR_WRITE_COMBINING; 232 else if ((p->md_attr & EFI_MD_ATTR_WP) != 0) 233 mode = VM_MEMATTR_WRITE_PROTECTED; 234 else if ((p->md_attr & EFI_MD_ATTR_UC) != 0) 235 mode = VM_MEMATTR_UNCACHEABLE; 236 else { 237 if (bootverbose) 238 printf("EFI Runtime entry %d mapping " 239 "attributes unsupported\n", i); 240 mode = VM_MEMATTR_UNCACHEABLE; 241 } 242 bits = pmap_cache_bits(kernel_pmap, mode, FALSE) | X86_PG_RW | 243 X86_PG_V; 244 VM_OBJECT_WLOCK(obj_1t1_pt); 245 for (va = p->md_phys, idx = 0; idx < p->md_pages; idx++, 246 va += PAGE_SIZE) { 247 pte = efi_1t1_pte(va); 248 pte_store(pte, va | bits); 249 250 m = PHYS_TO_VM_PAGE(va); 251 if (m != NULL && VM_PAGE_TO_PHYS(m) == 0) { 252 vm_page_init_page(m, va, -1); 253 m->order = VM_NFREEORDER + 1; /* invalid */ 254 m->pool = VM_NFREEPOOL + 1; /* invalid */ 255 pmap_page_set_memattr_noflush(m, mode); 256 } 257 } 258 VM_OBJECT_WUNLOCK(obj_1t1_pt); 259 } 260 261 return (true); 262 263 fail: 264 efi_destroy_1t1_map(); 265 return (false); 266 } 267 268 /* 269 * Create an environment for the EFI runtime code call. The most 270 * important part is creating the required 1:1 physical->virtual 271 * mappings for the runtime segments. To do that, we manually create 272 * page table which unmap userspace but gives correct kernel mapping. 273 * The 1:1 mappings for runtime segments usually occupy low 4G of the 274 * physical address map. 275 * 276 * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT 277 * service, because there are some BIOSes which fail to correctly 278 * relocate itself on the call, requiring both 1:1 and virtual 279 * mapping. As result, we must provide 1:1 mapping anyway, so no 280 * reason to bother with the virtual map, and no need to add a 281 * complexity into loader. 282 * 283 * There is no need to disable interrupts around the change of %cr3, 284 * the kernel mappings are correct, while we only grabbed the 285 * userspace portion of VA. Interrupts handlers must not access 286 * userspace. Having interrupts enabled fixes the issue with 287 * firmware/SMM long operation, which would negatively affect IPIs, 288 * esp. TLB shootdown requests. 289 */ 290 int 291 efi_arch_enter(void) 292 { 293 pmap_t curpmap; 294 uint64_t cr3; 295 296 curpmap = PCPU_GET(curpmap); 297 PMAP_LOCK_ASSERT(curpmap, MA_OWNED); 298 curthread->td_md.md_efirt_dis_pf = vm_fault_disable_pagefaults(); 299 300 /* 301 * IPI TLB shootdown handler invltlb_pcid_handler() reloads 302 * %cr3 from the curpmap->pm_cr3, which would disable runtime 303 * segments mappings. Block the handler's action by setting 304 * curpmap to impossible value. See also comment in 305 * pmap.c:pmap_activate_sw(). 306 */ 307 if (pmap_pcid_enabled && !invpcid_works) 308 PCPU_SET(curpmap, NULL); 309 310 cr3 = VM_PAGE_TO_PHYS(efi_pmltop_page); 311 if (pmap_pcid_enabled) 312 cr3 |= pmap_get_pcid(curpmap); 313 load_cr3(cr3); 314 /* 315 * If PCID is enabled, the clear CR3_PCID_SAVE bit in the loaded %cr3 316 * causes TLB invalidation. 317 */ 318 if (!pmap_pcid_enabled) 319 invltlb(); 320 return (0); 321 } 322 323 void 324 efi_arch_leave(void) 325 { 326 pmap_t curpmap; 327 uint64_t cr3; 328 329 curpmap = &curproc->p_vmspace->vm_pmap; 330 cr3 = curpmap->pm_cr3; 331 if (pmap_pcid_enabled) { 332 cr3 |= pmap_get_pcid(curpmap); 333 if (!invpcid_works) 334 PCPU_SET(curpmap, curpmap); 335 } 336 load_cr3(cr3); 337 if (!pmap_pcid_enabled) 338 invltlb(); 339 vm_fault_enable_pagefaults(curthread->td_md.md_efirt_dis_pf); 340 } 341 342 /* XXX debug stuff */ 343 static int 344 efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS) 345 { 346 struct efi_tm tm; 347 int error, val; 348 349 val = 0; 350 error = sysctl_handle_int(oidp, &val, 0, req); 351 if (error != 0 || req->newptr == NULL) 352 return (error); 353 error = efi_get_time(&tm); 354 if (error == 0) { 355 uprintf("EFI reports: Year %d Month %d Day %d Hour %d Min %d " 356 "Sec %d\n", tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour, 357 tm.tm_min, tm.tm_sec); 358 } 359 return (error); 360 } 361 362 SYSCTL_PROC(_debug, OID_AUTO, efi_time, 363 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, 364 efi_time_sysctl_handler, "I", 365 ""); 366