1/*- 2 * Copyright (c) 1989, 1990 William F. Jolitz. 3 * Copyright (c) 1990 The Regents of the University of California. 4 * Copyright (c) 2007 The FreeBSD Foundation 5 * Copyright (c) 2008 The DragonFly Project. 6 * Copyright (c) 2008 Jordan Gordeev. 7 * All rights reserved. 8 * 9 * Portions of this software were developed by A. Joseph Koshy under 10 * sponsorship from the FreeBSD Foundation and Google, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37#if 0 /* JG */ 38#include "opt_atpic.h" 39#endif 40#include "opt_compat.h" 41 42#include <machine/asmacros.h> 43#include <machine/psl.h> 44#include <machine/trap.h> 45#include <machine/segments.h> 46 47#include "assym.s" 48 49 .text 50 51 .globl lwkt_switch_return 52 53/*****************************************************************************/ 54/* Trap handling */ 55/*****************************************************************************/ 56/* 57 * Trap and fault vector routines. 58 * 59 * All traps are 'interrupt gates', SDT_SYSIGT. An interrupt gate pushes 60 * state on the stack but also disables interrupts. This is important for 61 * us for the use of the swapgs instruction. We cannot be interrupted 62 * until the GS.base value is correct. For most traps, we automatically 63 * then enable interrupts if the interrupted context had them enabled. 64 * This is equivalent to the i386 port's use of SDT_SYS386TGT. 65 * 66 * The cpu will push a certain amount of state onto the kernel stack for 67 * the current process. See x86_64/include/frame.h. 68 * This includes the current RFLAGS (status register, which includes 69 * the interrupt disable state prior to the trap), the code segment register, 70 * and the return instruction pointer are pushed by the cpu. The cpu 71 * will also push an 'error' code for certain traps. We push a dummy 72 * error code for those traps where the cpu doesn't in order to maintain 73 * a consistent frame. We also push a contrived 'trap number'. 74 * 75 * The cpu does not push the general registers, we must do that, and we 76 * must restore them prior to calling 'iret'. The cpu adjusts the %cs and 77 * %ss segment registers, but does not mess with %ds, %es, or %fs. Thus we 78 * must load them with appropriate values for supervisor mode operation. 79 */ 80 81MCOUNT_LABEL(user) 82MCOUNT_LABEL(btrap) 83 84/* 85 * Interrupts are enabled for all traps, otherwise horrible livelocks 86 * can occur with the smp_invltlb and cpusync ode. 87 */ 88#if 0 89#define TRAP_NOEN(a) \ 90 subq $TF_RIP,%rsp; \ 91 movq $0,TF_XFLAGS(%rsp) ; \ 92 movq $(a),TF_TRAPNO(%rsp) ; \ 93 movq $0,TF_ADDR(%rsp) ; \ 94 movq $0,TF_ERR(%rsp) ; \ 95 jmp alltraps_noen 96#endif 97 98/* Regular traps; The cpu does not supply tf_err for these. */ 99#define TRAP(a) \ 100 subq $TF_RIP,%rsp; \ 101 movq $0,TF_XFLAGS(%rsp) ; \ 102 movq $(a),TF_TRAPNO(%rsp) ; \ 103 movq $0,TF_ADDR(%rsp) ; \ 104 movq $0,TF_ERR(%rsp) ; \ 105 jmp alltraps 106IDTVEC(dbg) 107 TRAP(T_TRCTRAP) 108IDTVEC(bpt) 109 TRAP(T_BPTFLT) 110IDTVEC(div) 111 TRAP(T_DIVIDE) 112IDTVEC(ofl) 113 TRAP(T_OFLOW) 114IDTVEC(bnd) 115 TRAP(T_BOUND) 116IDTVEC(ill) 117 TRAP(T_PRIVINFLT) 118IDTVEC(dna) 119 TRAP(T_DNA) 120IDTVEC(fpusegm) 121 TRAP(T_FPOPFLT) 122IDTVEC(mchk) 123 TRAP(T_MCHK) 124IDTVEC(rsvd) 125 TRAP(T_RESERVED) 126IDTVEC(fpu) 127 TRAP(T_ARITHTRAP) 128IDTVEC(xmm) 129 TRAP(T_XMMFLT) 130 131/* This group of traps have tf_err already pushed by the cpu */ 132#define TRAP_ERR(a) \ 133 subq $TF_ERR,%rsp; \ 134 movq $(a),TF_TRAPNO(%rsp) ; \ 135 movq $0,TF_ADDR(%rsp) ; \ 136 movq $0,TF_XFLAGS(%rsp) ; \ 137 jmp alltraps 138IDTVEC(tss) 139 TRAP_ERR(T_TSSFLT) 140IDTVEC(missing) 141 TRAP_ERR(T_SEGNPFLT) 142IDTVEC(stk) 143 TRAP_ERR(T_STKFLT) 144IDTVEC(align) 145 TRAP_ERR(T_ALIGNFLT) 146 147 /* 148 * alltraps entry point. Use swapgs if this is the first time in the 149 * kernel from userland. Reenable interrupts if they were enabled 150 * before the trap. This approximates SDT_SYS386TGT on the i386 port. 151 */ 152 153 SUPERALIGN_TEXT 154 .globl alltraps 155 .type alltraps,@function 156alltraps: 157 /* Fixup %gs if coming from userland */ 158 testb $SEL_RPL_MASK,TF_CS(%rsp) 159 jz alltraps_testi 160 swapgs 161alltraps_testi: 162 testq $PSL_I,TF_RFLAGS(%rsp) 163 jz alltraps_pushregs 164 sti 165alltraps_pushregs: 166 movq %rdi,TF_RDI(%rsp) 167alltraps_pushregs_no_rdi: 168 movq %rsi,TF_RSI(%rsp) 169 movq %rdx,TF_RDX(%rsp) 170 movq %rcx,TF_RCX(%rsp) 171 movq %r8,TF_R8(%rsp) 172 movq %r9,TF_R9(%rsp) 173 movq %rax,TF_RAX(%rsp) 174 movq %rbx,TF_RBX(%rsp) 175 movq %rbp,TF_RBP(%rsp) 176 movq %r10,TF_R10(%rsp) 177 movq %r11,TF_R11(%rsp) 178 movq %r12,TF_R12(%rsp) 179 movq %r13,TF_R13(%rsp) 180 movq %r14,TF_R14(%rsp) 181 movq %r15,TF_R15(%rsp) 182 FAKE_MCOUNT(TF_RIP(%rsp)) 183 .globl calltrap 184 .type calltrap,@function 185calltrap: 186 cld 187 movq %rsp, %rdi 188 call trap 189 MEXITCOUNT 190 jmp doreti /* Handle any pending ASTs */ 191 192 /* 193 * alltraps_noen entry point. Unlike alltraps above, we want to 194 * leave the interrupts disabled. This corresponds to 195 * SDT_SYS386IGT on the i386 port. 196 */ 197 SUPERALIGN_TEXT 198 .globl alltraps_noen 199 .type alltraps_noen,@function 200alltraps_noen: 201 /* Fixup %gs if coming from userland */ 202 testb $SEL_RPL_MASK,TF_CS(%rsp) 203 jz alltraps_pushregs 204 swapgs 205 jmp alltraps_pushregs 206 207IDTVEC(dblfault) 208 subq $TF_ERR,%rsp 209 movq $T_DOUBLEFLT,TF_TRAPNO(%rsp) 210 movq $0,TF_ADDR(%rsp) 211 movq $0,TF_ERR(%rsp) 212 movq $0,TF_XFLAGS(%rsp) 213 movq %rdi,TF_RDI(%rsp) 214 movq %rsi,TF_RSI(%rsp) 215 movq %rdx,TF_RDX(%rsp) 216 movq %rcx,TF_RCX(%rsp) 217 movq %r8,TF_R8(%rsp) 218 movq %r9,TF_R9(%rsp) 219 movq %rax,TF_RAX(%rsp) 220 movq %rbx,TF_RBX(%rsp) 221 movq %rbp,TF_RBP(%rsp) 222 movq %r10,TF_R10(%rsp) 223 movq %r11,TF_R11(%rsp) 224 movq %r12,TF_R12(%rsp) 225 movq %r13,TF_R13(%rsp) 226 movq %r14,TF_R14(%rsp) 227 movq %r15,TF_R15(%rsp) 228 testb $SEL_RPL_MASK,TF_CS(%rsp) 229 jz 1f 230 swapgs 2311: movq %rsp, %rdi 232 cld 233 call dblfault_handler 2342: hlt 235 jmp 2b 236 237IDTVEC(page) 238 subq $TF_ERR,%rsp 239 movq $T_PAGEFLT,TF_TRAPNO(%rsp) 240 /* Fixup %gs if coming from userland */ 241 testb $SEL_RPL_MASK,TF_CS(%rsp) 242 jz 1f 243 swapgs 2441: 245 movq %rdi,TF_RDI(%rsp) /* free up a GP register */ 246 movq %cr2,%rdi /* preserve %cr2 before .. */ 247 movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */ 248 movq $0,TF_XFLAGS(%rsp) 249 testq $PSL_I,TF_RFLAGS(%rsp) 250 jz alltraps_pushregs_no_rdi 251 sti 252 jmp alltraps_pushregs_no_rdi 253 254 /* 255 * We have to special-case this one. If we get a trap in doreti() at 256 * the iretq stage, we'll reenter with the wrong gs state. We'll have 257 * to do a special the swapgs in this case even coming from the kernel. 258 * XXX linux has a trap handler for their equivalent of load_gs(). 259 */ 260IDTVEC(prot) 261 subq $TF_ERR,%rsp 262 movq $T_PROTFLT,TF_TRAPNO(%rsp) 263 movq $0,TF_ADDR(%rsp) 264 movq $0,TF_XFLAGS(%rsp) 265 movq %rdi,TF_RDI(%rsp) /* free up a GP register */ 266 267 /* 268 * Fixup %gs if coming from userland. Handle the special case where 269 * %fs faults in doreti at the iretq instruction itself. 270 */ 271 leaq doreti_iret(%rip),%rdi 272 cmpq %rdi,TF_RIP(%rsp) /* special iretq fault case */ 273 je 2f 274 testb $SEL_RPL_MASK,TF_CS(%rsp) /* check if from userland */ 275 jz 1f 2762: 277 swapgs 2781: 279 testq $PSL_I,TF_RFLAGS(%rsp) 280 jz alltraps_pushregs_no_rdi 281 sti 282 jmp alltraps_pushregs_no_rdi 283 284/* 285 * Fast syscall entry point. We enter here with just our new %cs/%ss set, 286 * and the new privilige level. We are still running on the old user stack 287 * pointer. We have to juggle a few things around to find our stack etc. 288 * swapgs gives us access to our PCPU space only. 289 */ 290IDTVEC(fast_syscall) 291 swapgs 292 movq %rsp,PCPU(scratch_rsp) 293 movq PCPU(common_tss) + TSS_RSP0, %rsp 294 /* Now emulate a trapframe. Make the 8 byte alignment odd for call. */ 295 subq $TF_SIZE,%rsp 296 /* defer TF_RSP till we have a spare register */ 297 movq %r11,TF_RFLAGS(%rsp) 298 movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */ 299 movq PCPU(scratch_rsp),%r11 /* %r11 already saved */ 300 movq %r11,TF_RSP(%rsp) /* user stack pointer */ 301 orl $RQF_QUICKRET,PCPU(reqflags) 302 sti 303 movq $KUDSEL,TF_SS(%rsp) 304 movq $KUCSEL,TF_CS(%rsp) 305 movq $2,TF_ERR(%rsp) 306 movq $T_FAST_SYSCALL,TF_TRAPNO(%rsp) /* for the vkernel */ 307 movq $0,TF_XFLAGS(%rsp) /* note: used in signal frame */ 308 movq %rdi,TF_RDI(%rsp) /* arg 1 */ 309 movq %rsi,TF_RSI(%rsp) /* arg 2 */ 310 movq %rdx,TF_RDX(%rsp) /* arg 3 */ 311 movq %r10,TF_RCX(%rsp) /* arg 4 */ 312 movq %r8,TF_R8(%rsp) /* arg 5 */ 313 movq %r9,TF_R9(%rsp) /* arg 6 */ 314 movq %rax,TF_RAX(%rsp) /* syscall number */ 315 movq %rbx,TF_RBX(%rsp) /* C preserved */ 316 movq %rbp,TF_RBP(%rsp) /* C preserved */ 317 movq %r12,TF_R12(%rsp) /* C preserved */ 318 movq %r13,TF_R13(%rsp) /* C preserved */ 319 movq %r14,TF_R14(%rsp) /* C preserved */ 320 movq %r15,TF_R15(%rsp) /* C preserved */ 321 FAKE_MCOUNT(TF_RIP(%rsp)) 322 movq %rsp, %rdi 323 call syscall2 324 325 /* 326 * Fast return from system call 327 */ 328 cli 329 testl $RQF_IPIQ|RQF_TIMER|RQF_INTPEND|RQF_AST_MASK,PCPU(reqflags) 330 jnz 1f 331 testl $RQF_QUICKRET,PCPU(reqflags) 332 jz 1f 333 MEXITCOUNT 334 movq TF_RDI(%rsp),%rdi 335 movq TF_RSI(%rsp),%rsi 336 movq TF_RDX(%rsp),%rdx 337 movq TF_RAX(%rsp),%rax 338 movq TF_RFLAGS(%rsp),%r11 339 movq TF_RIP(%rsp),%rcx 340 movq TF_RSP(%rsp),%rsp 341 swapgs 342 sysretq 343 /* 344 * Normal slow / full iret 345 */ 3461: 347 MEXITCOUNT 348 jmp doreti 349 350/* 351 * Here for CYA insurance, in case a "syscall" instruction gets 352 * issued from 32 bit compatability mode. MSR_CSTAR has to point 353 * to *something* if EFER_SCE is enabled. 354 */ 355IDTVEC(fast_syscall32) 356 sysret 357 358/* 359 * NMI handling is special. 360 * 361 * First, NMIs do not respect the state of the processor's RFLAGS.IF 362 * bit and the NMI handler may be invoked at any time, including when 363 * the processor is in a critical section with RFLAGS.IF == 0. In 364 * particular, this means that the processor's GS.base values could be 365 * inconsistent on entry to the handler, and so we need to read 366 * MSR_GSBASE to determine if a 'swapgs' is needed. We use '%ebx', a 367 * C-preserved register, to remember whether to swap GS back on the 368 * exit path. 369 * 370 * Second, the processor treats NMIs specially, blocking further NMIs 371 * until an 'iretq' instruction is executed. We therefore need to 372 * execute the NMI handler with interrupts disabled to prevent a 373 * nested interrupt from executing an 'iretq' instruction and 374 * inadvertently taking the processor out of NMI mode. 375 * 376 * Third, the NMI handler runs on its own stack (tss_ist1), shared 377 * with the double fault handler. 378 */ 379 380IDTVEC(nmi) 381 subq $TF_RIP,%rsp 382 movq $(T_NMI),TF_TRAPNO(%rsp) 383 movq $0,TF_ADDR(%rsp) 384 movq $0,TF_ERR(%rsp) 385 movq $0,TF_XFLAGS(%rsp) 386 movq %rdi,TF_RDI(%rsp) 387 movq %rsi,TF_RSI(%rsp) 388 movq %rdx,TF_RDX(%rsp) 389 movq %rcx,TF_RCX(%rsp) 390 movq %r8,TF_R8(%rsp) 391 movq %r9,TF_R9(%rsp) 392 movq %rax,TF_RAX(%rsp) 393 movq %rbx,TF_RBX(%rsp) 394 movq %rbp,TF_RBP(%rsp) 395 movq %r10,TF_R10(%rsp) 396 movq %r11,TF_R11(%rsp) 397 movq %r12,TF_R12(%rsp) 398 movq %r13,TF_R13(%rsp) 399 movq %r14,TF_R14(%rsp) 400 movq %r15,TF_R15(%rsp) 401 xorl %ebx,%ebx 402 testb $SEL_RPL_MASK,TF_CS(%rsp) 403 jnz nmi_needswapgs /* we came from userland */ 404 movl $MSR_GSBASE,%ecx 405 rdmsr 406 cmpl $VM_MAX_USER_ADDRESS >> 32,%edx 407 jae nmi_calltrap /* GS.base holds a kernel VA */ 408nmi_needswapgs: 409 incl %ebx 410 swapgs 411/* Note: this label is also used by ddb and gdb: */ 412nmi_calltrap: 413 FAKE_MCOUNT(TF_RIP(%rsp)) 414 cld 415 movq %rsp, %rdi 416 call trap 417 MEXITCOUNT 418 testl %ebx,%ebx 419 jz nmi_restoreregs 420 swapgs 421nmi_restoreregs: 422 movq TF_RDI(%rsp),%rdi 423 movq TF_RSI(%rsp),%rsi 424 movq TF_RDX(%rsp),%rdx 425 movq TF_RCX(%rsp),%rcx 426 movq TF_R8(%rsp),%r8 427 movq TF_R9(%rsp),%r9 428 movq TF_RAX(%rsp),%rax 429 movq TF_RBX(%rsp),%rbx 430 movq TF_RBP(%rsp),%rbp 431 movq TF_R10(%rsp),%r10 432 movq TF_R11(%rsp),%r11 433 movq TF_R12(%rsp),%r12 434 movq TF_R13(%rsp),%r13 435 movq TF_R14(%rsp),%r14 436 movq TF_R15(%rsp),%r15 437 addq $TF_RIP,%rsp 438 iretq 439 440/* 441 * This function is what cpu_heavy_restore jumps to after a new process 442 * is created. The LWKT subsystem switches while holding a critical 443 * section and we maintain that abstraction here (e.g. because 444 * cpu_heavy_restore needs it due to PCB_*() manipulation), then get out of 445 * it before calling the initial function (typically fork_return()) and/or 446 * returning to user mode. 447 * 448 * The MP lock is not held at any point but the critcount is bumped 449 * on entry to prevent interruption of the trampoline at a bad point. 450 * 451 * This is effectively what td->td_switch() returns to. It 'returns' the 452 * old thread in %rax and since this is not returning to a td->td_switch() 453 * call from lwkt_switch() we must handle the cleanup for the old thread 454 * by calling lwkt_switch_return(). 455 * 456 * fork_trampoline(%rax:otd, %rbx:func, %r12:arg) 457 */ 458ENTRY(fork_trampoline) 459 movq %rax,%rdi 460 call lwkt_switch_return 461 movq PCPU(curthread),%rax 462 decl TD_CRITCOUNT(%rax) 463 464 /* 465 * cpu_set_fork_handler intercepts this function call to 466 * have this call a non-return function to stay in kernel mode. 467 * 468 * initproc has its own fork handler, start_init(), which DOES 469 * return. 470 * 471 * %rbx - chaining function (typically fork_return) 472 * %r12 -> %rdi (argument) 473 * frame-> %rsi (trap frame) 474 * 475 * void (func:rbx)(arg:rdi, trapframe:rsi) 476 */ 477 movq %rsp, %rsi /* pass trapframe by reference */ 478 movq %r12, %rdi /* arg1 */ 479 call *%rbx /* function */ 480 481 /* cut from syscall */ 482 483 sti 484 call splz 485 486 /* 487 * Return via doreti to handle ASTs. 488 * 489 * trapframe is at the top of the stack. 490 */ 491 MEXITCOUNT 492 jmp doreti 493 494/* 495 * To efficiently implement classification of trap and interrupt handlers 496 * for profiling, there must be only trap handlers between the labels btrap 497 * and bintr, and only interrupt handlers between the labels bintr and 498 * eintr. This is implemented (partly) by including files that contain 499 * some of the handlers. Before including the files, set up a normal asm 500 * environment so that the included files doen't need to know that they are 501 * included. 502 */ 503 504#ifdef COMPAT_IA32 505 .data 506 .p2align 4 507 .text 508 SUPERALIGN_TEXT 509 510#include <x86_64/ia32/ia32_exception.S> 511#endif 512 513 .data 514 .p2align 4 515 .text 516 SUPERALIGN_TEXT 517MCOUNT_LABEL(bintr) 518 519#if 0 /* JG */ 520#include <x86_64/x86_64/apic_vector.S> 521#endif 522 523#ifdef DEV_ATPIC 524 .data 525 .p2align 4 526 .text 527 SUPERALIGN_TEXT 528 529#include <x86_64/isa/atpic_vector.S> 530#endif 531 532 .text 533MCOUNT_LABEL(eintr) 534 535