1 // SPDX-License-Identifier: GPL-2.0-only 2 #include <linux/extable.h> 3 #include <linux/uaccess.h> 4 #include <linux/sched/debug.h> 5 #include <linux/bitfield.h> 6 #include <xen/xen.h> 7 8 #include <asm/fpu/api.h> 9 #include <asm/fred.h> 10 #include <asm/sev.h> 11 #include <asm/traps.h> 12 #include <asm/kdebug.h> 13 #include <asm/insn-eval.h> 14 #include <asm/sgx.h> 15 16 static inline unsigned long *pt_regs_nr(struct pt_regs *regs, int nr) 17 { 18 int reg_offset = pt_regs_offset(regs, nr); 19 static unsigned long __dummy; 20 21 if (WARN_ON_ONCE(reg_offset < 0)) 22 return &__dummy; 23 24 return (unsigned long *)((unsigned long)regs + reg_offset); 25 } 26 27 static inline unsigned long 28 ex_fixup_addr(const struct exception_table_entry *x) 29 { 30 return (unsigned long)&x->fixup + x->fixup; 31 } 32 33 static bool ex_handler_default(const struct exception_table_entry *e, 34 struct pt_regs *regs) 35 { 36 if (e->data & EX_FLAG_CLEAR_AX) 37 regs->ax = 0; 38 if (e->data & EX_FLAG_CLEAR_DX) 39 regs->dx = 0; 40 41 regs->ip = ex_fixup_addr(e); 42 return true; 43 } 44 45 /* 46 * This is the *very* rare case where we do a "load_unaligned_zeropad()" 47 * and it's a page crosser into a non-existent page. 48 * 49 * This happens when we optimistically load a pathname a word-at-a-time 50 * and the name is less than the full word and the next page is not 51 * mapped. Typically that only happens for CONFIG_DEBUG_PAGEALLOC. 52 * 53 * NOTE! The faulting address is always a 'mov mem,reg' type instruction 54 * of size 'long', and the exception fixup must always point to right 55 * after the instruction. 56 */ 57 static bool ex_handler_zeropad(const struct exception_table_entry *e, 58 struct pt_regs *regs, 59 unsigned long fault_addr) 60 { 61 struct insn insn; 62 const unsigned long mask = sizeof(long) - 1; 63 unsigned long offset, addr, next_ip, len; 64 unsigned long *reg; 65 66 next_ip = ex_fixup_addr(e); 67 len = next_ip - regs->ip; 68 if (len > MAX_INSN_SIZE) 69 return false; 70 71 if (insn_decode(&insn, (void *) regs->ip, len, INSN_MODE_KERN)) 72 return false; 73 if (insn.length != len) 74 return false; 75 76 if (insn.opcode.bytes[0] != 0x8b) 77 return false; 78 if (insn.opnd_bytes != sizeof(long)) 79 return false; 80 81 addr = (unsigned long) insn_get_addr_ref(&insn, regs); 82 if (addr == ~0ul) 83 return false; 84 85 offset = addr & mask; 86 addr = addr & ~mask; 87 if (fault_addr != addr + sizeof(long)) 88 return false; 89 90 reg = insn_get_modrm_reg_ptr(&insn, regs); 91 if (!reg) 92 return false; 93 94 *reg = *(unsigned long *)addr >> (offset * 8); 95 return ex_handler_default(e, regs); 96 } 97 98 static bool ex_handler_fault(const struct exception_table_entry *fixup, 99 struct pt_regs *regs, int trapnr) 100 { 101 regs->ax = trapnr; 102 return ex_handler_default(fixup, regs); 103 } 104 105 static bool ex_handler_sgx(const struct exception_table_entry *fixup, 106 struct pt_regs *regs, int trapnr) 107 { 108 regs->ax = trapnr | SGX_ENCLS_FAULT_FLAG; 109 return ex_handler_default(fixup, regs); 110 } 111 112 /* 113 * Handler for when we fail to restore a task's FPU state. We should never get 114 * here because the FPU state of a task using the FPU (task->thread.fpu.state) 115 * should always be valid. However, past bugs have allowed userspace to set 116 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn(). 117 * These caused XRSTOR to fail when switching to the task, leaking the FPU 118 * registers of the task previously executing on the CPU. Mitigate this class 119 * of vulnerability by restoring from the initial state (essentially, zeroing 120 * out all the FPU registers) if we can't restore from the task's FPU state. 121 */ 122 static bool ex_handler_fprestore(const struct exception_table_entry *fixup, 123 struct pt_regs *regs) 124 { 125 regs->ip = ex_fixup_addr(fixup); 126 127 WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.", 128 (void *)instruction_pointer(regs)); 129 130 fpu_reset_from_exception_fixup(); 131 return true; 132 } 133 134 /* 135 * On x86-64, we end up being imprecise with 'access_ok()', and allow 136 * non-canonical user addresses to make the range comparisons simpler, 137 * and to not have to worry about LAM being enabled. 138 * 139 * In fact, we allow up to one page of "slop" at the sign boundary, 140 * which means that we can do access_ok() by just checking the sign 141 * of the pointer for the common case of having a small access size. 142 */ 143 static bool gp_fault_address_ok(unsigned long fault_address) 144 { 145 #ifdef CONFIG_X86_64 146 /* Is it in the "user space" part of the non-canonical space? */ 147 if (valid_user_address(fault_address)) 148 return true; 149 150 /* .. or just above it? */ 151 fault_address -= PAGE_SIZE; 152 if (valid_user_address(fault_address)) 153 return true; 154 #endif 155 return false; 156 } 157 158 static bool ex_handler_uaccess(const struct exception_table_entry *fixup, 159 struct pt_regs *regs, int trapnr, 160 unsigned long fault_address) 161 { 162 WARN_ONCE(trapnr == X86_TRAP_GP && !gp_fault_address_ok(fault_address), 163 "General protection fault in user access. Non-canonical address?"); 164 return ex_handler_default(fixup, regs); 165 } 166 167 static bool ex_handler_msr(const struct exception_table_entry *fixup, 168 struct pt_regs *regs, bool wrmsr, bool safe, int reg) 169 { 170 if (__ONCE_LITE_IF(!safe && wrmsr)) { 171 pr_warn("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pS)\n", 172 (unsigned int)regs->cx, (unsigned int)regs->dx, 173 (unsigned int)regs->ax, regs->ip, (void *)regs->ip); 174 show_stack_regs(regs); 175 } 176 177 if (__ONCE_LITE_IF(!safe && !wrmsr)) { 178 pr_warn("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pS)\n", 179 (unsigned int)regs->cx, regs->ip, (void *)regs->ip); 180 show_stack_regs(regs); 181 } 182 183 if (!wrmsr) { 184 /* Pretend that the read succeeded and returned 0. */ 185 regs->ax = 0; 186 regs->dx = 0; 187 } 188 189 if (safe) 190 *pt_regs_nr(regs, reg) = -EIO; 191 192 return ex_handler_default(fixup, regs); 193 } 194 195 static bool ex_handler_clear_fs(const struct exception_table_entry *fixup, 196 struct pt_regs *regs) 197 { 198 if (static_cpu_has(X86_BUG_NULL_SEG)) 199 asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS)); 200 asm volatile ("mov %0, %%fs" : : "rm" (0)); 201 return ex_handler_default(fixup, regs); 202 } 203 204 static bool ex_handler_imm_reg(const struct exception_table_entry *fixup, 205 struct pt_regs *regs, int reg, int imm) 206 { 207 *pt_regs_nr(regs, reg) = (long)imm; 208 return ex_handler_default(fixup, regs); 209 } 210 211 static bool ex_handler_ucopy_len(const struct exception_table_entry *fixup, 212 struct pt_regs *regs, int trapnr, 213 unsigned long fault_address, 214 int reg, int imm) 215 { 216 regs->cx = imm * regs->cx + *pt_regs_nr(regs, reg); 217 return ex_handler_uaccess(fixup, regs, trapnr, fault_address); 218 } 219 220 #ifdef CONFIG_X86_FRED 221 static bool ex_handler_eretu(const struct exception_table_entry *fixup, 222 struct pt_regs *regs, unsigned long error_code) 223 { 224 struct pt_regs *uregs = (struct pt_regs *)(regs->sp - offsetof(struct pt_regs, orig_ax)); 225 unsigned short ss = uregs->ss; 226 unsigned short cs = uregs->cs; 227 228 /* 229 * Move the NMI bit from the invalid stack frame, which caused ERETU 230 * to fault, to the fault handler's stack frame, thus to unblock NMI 231 * with the fault handler's ERETS instruction ASAP if NMI is blocked. 232 */ 233 regs->fred_ss.nmi = uregs->fred_ss.nmi; 234 235 /* 236 * Sync event information to uregs, i.e., the ERETU return frame, but 237 * is it safe to write to the ERETU return frame which is just above 238 * current event stack frame? 239 * 240 * The RSP used by FRED to push a stack frame is not the value in %rsp, 241 * it is calculated from %rsp with the following 2 steps: 242 * 1) RSP = %rsp - (IA32_FRED_CONFIG & 0x1c0) // Reserve N*64 bytes 243 * 2) RSP = RSP & ~0x3f // Align to a 64-byte cache line 244 * when an event delivery doesn't trigger a stack level change. 245 * 246 * Here is an example with N*64 (N=1) bytes reserved: 247 * 248 * 64-byte cache line ==> ______________ 249 * |___Reserved___| 250 * |__Event_data__| 251 * |_____SS_______| 252 * |_____RSP______| 253 * |_____FLAGS____| 254 * |_____CS_______| 255 * |_____IP_______| 256 * 64-byte cache line ==> |__Error_code__| <== ERETU return frame 257 * |______________| 258 * |______________| 259 * |______________| 260 * |______________| 261 * |______________| 262 * |______________| 263 * |______________| 264 * 64-byte cache line ==> |______________| <== RSP after step 1) and 2) 265 * |___Reserved___| 266 * |__Event_data__| 267 * |_____SS_______| 268 * |_____RSP______| 269 * |_____FLAGS____| 270 * |_____CS_______| 271 * |_____IP_______| 272 * 64-byte cache line ==> |__Error_code__| <== ERETS return frame 273 * 274 * Thus a new FRED stack frame will always be pushed below a previous 275 * FRED stack frame ((N*64) bytes may be reserved between), and it is 276 * safe to write to a previous FRED stack frame as they never overlap. 277 */ 278 fred_info(uregs)->edata = fred_event_data(regs); 279 uregs->ssx = regs->ssx; 280 uregs->fred_ss.ss = ss; 281 /* The NMI bit was moved away above */ 282 uregs->fred_ss.nmi = 0; 283 uregs->csx = regs->csx; 284 uregs->fred_cs.sl = 0; 285 uregs->fred_cs.wfe = 0; 286 uregs->cs = cs; 287 uregs->orig_ax = error_code; 288 289 return ex_handler_default(fixup, regs); 290 } 291 #endif 292 293 int ex_get_fixup_type(unsigned long ip) 294 { 295 const struct exception_table_entry *e = search_exception_tables(ip); 296 297 return e ? FIELD_GET(EX_DATA_TYPE_MASK, e->data) : EX_TYPE_NONE; 298 } 299 300 int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code, 301 unsigned long fault_addr) 302 { 303 const struct exception_table_entry *e; 304 int type, reg, imm; 305 306 #ifdef CONFIG_PNPBIOS 307 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) { 308 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp; 309 extern u32 pnp_bios_is_utter_crap; 310 pnp_bios_is_utter_crap = 1; 311 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n"); 312 __asm__ volatile( 313 "movl %0, %%esp\n\t" 314 "jmp *%1\n\t" 315 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip)); 316 panic("do_trap: can't hit this"); 317 } 318 #endif 319 320 e = search_exception_tables(regs->ip); 321 if (!e) 322 return 0; 323 324 type = FIELD_GET(EX_DATA_TYPE_MASK, e->data); 325 reg = FIELD_GET(EX_DATA_REG_MASK, e->data); 326 imm = FIELD_GET(EX_DATA_IMM_MASK, e->data); 327 328 switch (type) { 329 case EX_TYPE_DEFAULT: 330 case EX_TYPE_DEFAULT_MCE_SAFE: 331 return ex_handler_default(e, regs); 332 case EX_TYPE_FAULT: 333 case EX_TYPE_FAULT_MCE_SAFE: 334 return ex_handler_fault(e, regs, trapnr); 335 case EX_TYPE_UACCESS: 336 return ex_handler_uaccess(e, regs, trapnr, fault_addr); 337 case EX_TYPE_CLEAR_FS: 338 return ex_handler_clear_fs(e, regs); 339 case EX_TYPE_FPU_RESTORE: 340 return ex_handler_fprestore(e, regs); 341 case EX_TYPE_BPF: 342 return ex_handler_bpf(e, regs); 343 case EX_TYPE_WRMSR: 344 return ex_handler_msr(e, regs, true, false, reg); 345 case EX_TYPE_RDMSR: 346 return ex_handler_msr(e, regs, false, false, reg); 347 case EX_TYPE_WRMSR_SAFE: 348 return ex_handler_msr(e, regs, true, true, reg); 349 case EX_TYPE_RDMSR_SAFE: 350 return ex_handler_msr(e, regs, false, true, reg); 351 case EX_TYPE_WRMSR_IN_MCE: 352 ex_handler_msr_mce(regs, true); 353 break; 354 case EX_TYPE_RDMSR_IN_MCE: 355 ex_handler_msr_mce(regs, false); 356 break; 357 case EX_TYPE_POP_REG: 358 regs->sp += sizeof(long); 359 fallthrough; 360 case EX_TYPE_IMM_REG: 361 return ex_handler_imm_reg(e, regs, reg, imm); 362 case EX_TYPE_FAULT_SGX: 363 return ex_handler_sgx(e, regs, trapnr); 364 case EX_TYPE_UCOPY_LEN: 365 return ex_handler_ucopy_len(e, regs, trapnr, fault_addr, reg, imm); 366 case EX_TYPE_ZEROPAD: 367 return ex_handler_zeropad(e, regs, fault_addr); 368 #ifdef CONFIG_X86_FRED 369 case EX_TYPE_ERETU: 370 return ex_handler_eretu(e, regs, error_code); 371 #endif 372 } 373 BUG(); 374 } 375 376 extern unsigned int early_recursion_flag; 377 378 /* Restricted version used during very early boot */ 379 void __init early_fixup_exception(struct pt_regs *regs, int trapnr) 380 { 381 /* Ignore early NMIs. */ 382 if (trapnr == X86_TRAP_NMI) 383 return; 384 385 if (early_recursion_flag > 2) 386 goto halt_loop; 387 388 /* 389 * Old CPUs leave the high bits of CS on the stack 390 * undefined. I'm not sure which CPUs do this, but at least 391 * the 486 DX works this way. 392 * Xen pv domains are not using the default __KERNEL_CS. 393 */ 394 if (!xen_pv_domain() && regs->cs != __KERNEL_CS) 395 goto fail; 396 397 /* 398 * The full exception fixup machinery is available as soon as 399 * the early IDT is loaded. This means that it is the 400 * responsibility of extable users to either function correctly 401 * when handlers are invoked early or to simply avoid causing 402 * exceptions before they're ready to handle them. 403 * 404 * This is better than filtering which handlers can be used, 405 * because refusing to call a handler here is guaranteed to 406 * result in a hard-to-debug panic. 407 * 408 * Keep in mind that not all vectors actually get here. Early 409 * page faults, for example, are special. 410 */ 411 if (fixup_exception(regs, trapnr, regs->orig_ax, 0)) 412 return; 413 414 if (trapnr == X86_TRAP_UD) { 415 if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN) { 416 /* Skip the ud2. */ 417 regs->ip += LEN_UD2; 418 return; 419 } 420 421 /* 422 * If this was a BUG and report_bug returns or if this 423 * was just a normal #UD, we want to continue onward and 424 * crash. 425 */ 426 } 427 428 fail: 429 early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n", 430 (unsigned)trapnr, (unsigned long)regs->cs, regs->ip, 431 regs->orig_ax, read_cr2()); 432 433 show_regs(regs); 434 435 halt_loop: 436 while (true) 437 halt(); 438 } 439