/*- * Copyright (c) 1989, 1990 William F. Jolitz. * Copyright (c) 1990 The Regents of the University of California. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include "opt_atpic.h" #include "opt_compat.h" #include "opt_kdtrace.h" #include #include #include #include "assym.s" #ifdef KDTRACE_HOOKS .bss .globl dtrace_invop_jump_addr .align 8 .type dtrace_invop_jump_addr, @object .size dtrace_invop_jump_addr, 8 dtrace_invop_jump_addr: .zero 8 .globl dtrace_invop_calltrap_addr .align 8 .type dtrace_invop_calltrap_addr, @object .size dtrace_invop_calltrap_addr, 8 dtrace_invop_calltrap_addr: .zero 8 #endif .text /*****************************************************************************/ /* Trap handling */ /*****************************************************************************/ /* * Trap and fault vector routines. * * All traps are 'interrupt gates', SDT_SYSIGT. An interrupt gate pushes * state on the stack but also disables interrupts. This is important for * us for the use of the swapgs instruction. We cannot be interrupted * until the GS.base value is correct. For most traps, we automatically * then enable interrupts if the interrupted context had them enabled. * This is equivalent to the i386 port's use of SDT_SYS386TGT. * * The cpu will push a certain amount of state onto the kernel stack for * the current process. See amd64/include/frame.h. * This includes the current RFLAGS (status register, which includes * the interrupt disable state prior to the trap), the code segment register, * and the return instruction pointer are pushed by the cpu. The cpu * will also push an 'error' code for certain traps. We push a dummy * error code for those traps where the cpu doesn't in order to maintain * a consistent frame. We also push a contrived 'trap number'. * * The cpu does not push the general registers, we must do that, and we * must restore them prior to calling 'iret'. The cpu adjusts the %cs and * %ss segment registers, but does not mess with %ds, %es, or %fs. Thus we * must load them with appropriate values for supervisor mode operation. */ MCOUNT_LABEL(user) MCOUNT_LABEL(btrap) /* Traps that we leave interrupts disabled for.. */ #define TRAP_NOEN(a) \ subq $TF_RIP,%rsp; \ movq $(a),TF_TRAPNO(%rsp) ; \ movq $0,TF_ADDR(%rsp) ; \ movq $0,TF_ERR(%rsp) ; \ jmp alltraps_noen IDTVEC(dbg) TRAP_NOEN(T_TRCTRAP) IDTVEC(bpt) TRAP_NOEN(T_BPTFLT) /* Regular traps; The cpu does not supply tf_err for these. */ #define TRAP(a) \ subq $TF_RIP,%rsp; \ movq $(a),TF_TRAPNO(%rsp) ; \ movq $0,TF_ADDR(%rsp) ; \ movq $0,TF_ERR(%rsp) ; \ jmp alltraps IDTVEC(div) TRAP(T_DIVIDE) IDTVEC(ofl) TRAP(T_OFLOW) IDTVEC(bnd) TRAP(T_BOUND) IDTVEC(ill) TRAP(T_PRIVINFLT) IDTVEC(dna) TRAP(T_DNA) IDTVEC(fpusegm) TRAP(T_FPOPFLT) IDTVEC(mchk) TRAP(T_MCHK) IDTVEC(rsvd) TRAP(T_RESERVED) IDTVEC(fpu) TRAP(T_ARITHTRAP) IDTVEC(xmm) TRAP(T_XMMFLT) /* This group of traps have tf_err already pushed by the cpu */ #define TRAP_ERR(a) \ subq $TF_ERR,%rsp; \ movq $(a),TF_TRAPNO(%rsp) ; \ movq $0,TF_ADDR(%rsp) ; \ jmp alltraps IDTVEC(tss) TRAP_ERR(T_TSSFLT) IDTVEC(missing) TRAP_ERR(T_SEGNPFLT) IDTVEC(stk) TRAP_ERR(T_STKFLT) IDTVEC(align) TRAP_ERR(T_ALIGNFLT) /* * alltraps entry point. Use swapgs if this is the first time in the * kernel from userland. Reenable interrupts if they were enabled * before the trap. This approximates SDT_SYS386TGT on the i386 port. */ SUPERALIGN_TEXT .globl alltraps .type alltraps,@function alltraps: testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz alltraps_testi /* already running with kernel GS.base */ swapgs alltraps_testi: testl $PSL_I,TF_RFLAGS(%rsp) jz alltraps_pushregs sti alltraps_pushregs: movq %rdi,TF_RDI(%rsp) alltraps_pushregs_no_rdi: movq %rsi,TF_RSI(%rsp) movq %rdx,TF_RDX(%rsp) movq %rcx,TF_RCX(%rsp) movq %r8,TF_R8(%rsp) movq %r9,TF_R9(%rsp) movq %rax,TF_RAX(%rsp) movq %rbx,TF_RBX(%rsp) movq %rbp,TF_RBP(%rsp) movq %r10,TF_R10(%rsp) movq %r11,TF_R11(%rsp) movq %r12,TF_R12(%rsp) movq %r13,TF_R13(%rsp) movq %r14,TF_R14(%rsp) movq %r15,TF_R15(%rsp) FAKE_MCOUNT(TF_RIP(%rsp)) #ifdef KDTRACE_HOOKS /* * DTrace Function Boundary Trace (fbt) and Statically Defined * Trace (sdt) probes are triggered by int3 (0xcc) which causes * the #BP (T_BPTFLT) breakpoint interrupt. For all other trap * types, just handle them in the usual way. */ cmpq $T_BPTFLT,TF_TRAPNO(%rsp) jne calltrap /* Check if there is no DTrace hook registered. */ cmpq $0,dtrace_invop_jump_addr je calltrap /* * Set our jump address for the jump back in the event that * the breakpoint wasn't caused by DTrace at all. */ movq $calltrap, dtrace_invop_calltrap_addr(%rip) /* Jump to the code hooked in by DTrace. */ movq dtrace_invop_jump_addr, %rax jmpq *dtrace_invop_jump_addr #endif .globl calltrap .type calltrap,@function calltrap: movq %rsp, %rdi call trap MEXITCOUNT jmp doreti /* Handle any pending ASTs */ /* * alltraps_noen entry point. Unlike alltraps above, we want to * leave the interrupts disabled. This corresponds to * SDT_SYS386IGT on the i386 port. */ SUPERALIGN_TEXT .globl alltraps_noen .type alltraps_noen,@function alltraps_noen: testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz alltraps_pushregs /* already running with kernel GS.base */ swapgs jmp alltraps_pushregs IDTVEC(dblfault) subq $TF_ERR,%rsp movq $T_DOUBLEFLT,TF_TRAPNO(%rsp) movq $0,TF_ADDR(%rsp) movq $0,TF_ERR(%rsp) movq %rdi,TF_RDI(%rsp) movq %rsi,TF_RSI(%rsp) movq %rdx,TF_RDX(%rsp) movq %rcx,TF_RCX(%rsp) movq %r8,TF_R8(%rsp) movq %r9,TF_R9(%rsp) movq %rax,TF_RAX(%rsp) movq %rbx,TF_RBX(%rsp) movq %rbp,TF_RBP(%rsp) movq %r10,TF_R10(%rsp) movq %r11,TF_R11(%rsp) movq %r12,TF_R12(%rsp) movq %r13,TF_R13(%rsp) movq %r14,TF_R14(%rsp) movq %r15,TF_R15(%rsp) testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz 1f /* already running with kernel GS.base */ swapgs 1: movq %rsp, %rdi call dblfault_handler 2: hlt jmp 2b IDTVEC(page) subq $TF_ERR,%rsp movq $T_PAGEFLT,TF_TRAPNO(%rsp) testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz 1f /* already running with kernel GS.base */ swapgs 1: movq %rdi,TF_RDI(%rsp) /* free up a GP register */ movq %cr2,%rdi /* preserve %cr2 before .. */ movq %rdi,TF_ADDR(%rsp) /* enabling interrupts. */ testl $PSL_I,TF_RFLAGS(%rsp) jz alltraps_pushregs_no_rdi sti jmp alltraps_pushregs_no_rdi /* * We have to special-case this one. If we get a trap in doreti() at * the iretq stage, we'll reenter with the wrong gs state. We'll have * to do a special the swapgs in this case even coming from the kernel. * XXX linux has a trap handler for their equivalent of load_gs(). */ IDTVEC(prot) subq $TF_ERR,%rsp movq $T_PROTFLT,TF_TRAPNO(%rsp) movq $0,TF_ADDR(%rsp) movq %rdi,TF_RDI(%rsp) /* free up a GP register */ leaq doreti_iret(%rip),%rdi cmpq %rdi,TF_RIP(%rsp) je 2f /* kernel but with user gsbase!! */ testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz 1f /* already running with kernel GS.base */ 2: swapgs 1: testl $PSL_I,TF_RFLAGS(%rsp) jz alltraps_pushregs_no_rdi sti jmp alltraps_pushregs_no_rdi /* * Fast syscall entry point. We enter here with just our new %cs/%ss set, * and the new privilige level. We are still running on the old user stack * pointer. We have to juggle a few things around to find our stack etc. * swapgs gives us access to our PCPU space only. */ IDTVEC(fast_syscall) swapgs movq %rsp,PCPU(SCRATCH_RSP) movq PCPU(RSP0),%rsp /* Now emulate a trapframe. Make the 8 byte alignment odd for call. */ subq $TF_SIZE,%rsp /* defer TF_RSP till we have a spare register */ movq %r11,TF_RFLAGS(%rsp) movq %rcx,TF_RIP(%rsp) /* %rcx original value is in %r10 */ movq PCPU(SCRATCH_RSP),%r11 /* %r11 already saved */ movq %r11,TF_RSP(%rsp) /* user stack pointer */ sti movq $KUDSEL,TF_SS(%rsp) movq $KUCSEL,TF_CS(%rsp) movq $2,TF_ERR(%rsp) movq %rdi,TF_RDI(%rsp) /* arg 1 */ movq %rsi,TF_RSI(%rsp) /* arg 2 */ movq %rdx,TF_RDX(%rsp) /* arg 3 */ movq %r10,TF_RCX(%rsp) /* arg 4 */ movq %r8,TF_R8(%rsp) /* arg 5 */ movq %r9,TF_R9(%rsp) /* arg 6 */ movq %rax,TF_RAX(%rsp) /* syscall number */ movq %rbx,TF_RBX(%rsp) /* C preserved */ movq %rbp,TF_RBP(%rsp) /* C preserved */ movq %r12,TF_R12(%rsp) /* C preserved */ movq %r13,TF_R13(%rsp) /* C preserved */ movq %r14,TF_R14(%rsp) /* C preserved */ movq %r15,TF_R15(%rsp) /* C preserved */ FAKE_MCOUNT(TF_RIP(%rsp)) movq %rsp, %rdi call syscall movq PCPU(CURPCB),%rax testq $PCB_FULLCTX,PCB_FLAGS(%rax) jne 3f 1: /* Check for and handle AST's on return to userland */ cli movq PCPU(CURTHREAD),%rax testl $TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax) je 2f sti movq %rsp, %rdi call ast jmp 1b 2: /* restore preserved registers */ MEXITCOUNT movq TF_RDI(%rsp),%rdi /* bonus; preserve arg 1 */ movq TF_RSI(%rsp),%rsi /* bonus: preserve arg 2 */ movq TF_RDX(%rsp),%rdx /* return value 2 */ movq TF_RAX(%rsp),%rax /* return value 1 */ movq TF_RBX(%rsp),%rbx /* C preserved */ movq TF_RBP(%rsp),%rbp /* C preserved */ movq TF_R12(%rsp),%r12 /* C preserved */ movq TF_R13(%rsp),%r13 /* C preserved */ movq TF_R14(%rsp),%r14 /* C preserved */ movq TF_R15(%rsp),%r15 /* C preserved */ movq TF_RFLAGS(%rsp),%r11 /* original %rflags */ movq TF_RIP(%rsp),%rcx /* original %rip */ movq TF_RSP(%rsp),%r9 /* user stack pointer */ movq %r9,%rsp /* original %rsp */ swapgs sysretq 3: /* Requested full context restore, use doreti for that */ andq $~PCB_FULLCTX,PCB_FLAGS(%rax) MEXITCOUNT jmp doreti /* * Here for CYA insurance, in case a "syscall" instruction gets * issued from 32 bit compatability mode. MSR_CSTAR has to point * to *something* if EFER_SCE is enabled. */ IDTVEC(fast_syscall32) sysret /* * NMI handling is special. * * First, NMIs do not respect the state of the processor's RFLAGS.IF * bit and the NMI handler may be invoked at any time, including when * the processor is in a critical section with RFLAGS.IF == 0. In * particular, this means that the processor's GS.base values could be * inconsistent on entry to the handler, and so we need to read * MSR_GSBASE to determine if a 'swapgs' is needed. We use '%ebx', a * C-preserved register, to remember whether to swap GS back on the * exit path. * * Second, the processor treats NMIs specially, blocking further NMIs * until an 'iretq' instruction is executed. We therefore need to * execute the NMI handler with interrupts disabled to prevent a * nested interrupt from executing an 'iretq' instruction and * inadvertently taking the processor out of NMI mode. */ IDTVEC(nmi) subq $TF_RIP,%rsp movq $(T_NMI),TF_TRAPNO(%rsp) movq $0,TF_ADDR(%rsp) movq $0,TF_ERR(%rsp) movq %rdi,TF_RDI(%rsp) movq %rsi,TF_RSI(%rsp) movq %rdx,TF_RDX(%rsp) movq %rcx,TF_RCX(%rsp) movq %r8,TF_R8(%rsp) movq %r9,TF_R9(%rsp) movq %rax,TF_RAX(%rsp) movq %rbx,TF_RBX(%rsp) movq %rbp,TF_RBP(%rsp) movq %r10,TF_R10(%rsp) movq %r11,TF_R11(%rsp) movq %r12,TF_R12(%rsp) movq %r13,TF_R13(%rsp) movq %r14,TF_R14(%rsp) movq %r15,TF_R15(%rsp) xorl %ebx,%ebx testb $SEL_RPL_MASK,TF_CS(%rsp) jnz nmi_needswapgs /* we came from userland */ movl $MSR_GSBASE,%ecx rdmsr cmpl $VM_MAXUSER_ADDRESS >> 32,%edx jae nmi_calltrap /* GS.base holds a kernel VA */ nmi_needswapgs: incl %ebx swapgs /* Note: this label is also used by ddb and gdb: */ nmi_calltrap: FAKE_MCOUNT(TF_RIP(%rsp)) movq %rsp, %rdi call trap MEXITCOUNT testl %ebx,%ebx jz nmi_restoreregs swapgs nmi_restoreregs: movq TF_RDI(%rsp),%rdi movq TF_RSI(%rsp),%rsi movq TF_RDX(%rsp),%rdx movq TF_RCX(%rsp),%rcx movq TF_R8(%rsp),%r8 movq TF_R9(%rsp),%r9 movq TF_RAX(%rsp),%rax movq TF_RBX(%rsp),%rbx movq TF_RBP(%rsp),%rbp movq TF_R10(%rsp),%r10 movq TF_R11(%rsp),%r11 movq TF_R12(%rsp),%r12 movq TF_R13(%rsp),%r13 movq TF_R14(%rsp),%r14 movq TF_R15(%rsp),%r15 addq $TF_RIP,%rsp iretq ENTRY(fork_trampoline) movq %r12, %rdi /* function */ movq %rbx, %rsi /* arg1 */ movq %rsp, %rdx /* trapframe pointer */ call fork_exit MEXITCOUNT jmp doreti /* Handle any ASTs */ /* * To efficiently implement classification of trap and interrupt handlers * for profiling, there must be only trap handlers between the labels btrap * and bintr, and only interrupt handlers between the labels bintr and * eintr. This is implemented (partly) by including files that contain * some of the handlers. Before including the files, set up a normal asm * environment so that the included files doen't need to know that they are * included. */ #ifdef COMPAT_IA32 .data .p2align 4 .text SUPERALIGN_TEXT #include #endif .data .p2align 4 .text SUPERALIGN_TEXT MCOUNT_LABEL(bintr) #include #ifdef DEV_ATPIC .data .p2align 4 .text SUPERALIGN_TEXT #include #endif .text MCOUNT_LABEL(eintr) /* * void doreti(struct trapframe) * * Handle return from interrupts, traps and syscalls. */ .text SUPERALIGN_TEXT .type doreti,@function doreti: FAKE_MCOUNT($bintr) /* init "from" bintr -> doreti */ /* * Check if ASTs can be handled now. */ testb $SEL_RPL_MASK,TF_CS(%rsp) /* are we returning to user mode? */ jz doreti_exit /* can't handle ASTs now if not */ doreti_ast: /* * Check for ASTs atomically with returning. Disabling CPU * interrupts provides sufficient locking eve in the SMP case, * since we will be informed of any new ASTs by an IPI. */ cli movq PCPU(CURTHREAD),%rax testl $TDF_ASTPENDING | TDF_NEEDRESCHED,TD_FLAGS(%rax) je doreti_exit sti movq %rsp, %rdi /* pass a pointer to the trapframe */ call ast jmp doreti_ast /* * doreti_exit: pop registers, iret. * * The segment register pop is a special case, since it may * fault if (for example) a sigreturn specifies bad segment * registers. The fault is handled in trap.c. */ doreti_exit: MEXITCOUNT movq TF_RDI(%rsp),%rdi movq TF_RSI(%rsp),%rsi movq TF_RDX(%rsp),%rdx movq TF_RCX(%rsp),%rcx movq TF_R8(%rsp),%r8 movq TF_R9(%rsp),%r9 movq TF_RAX(%rsp),%rax movq TF_RBX(%rsp),%rbx movq TF_RBP(%rsp),%rbp movq TF_R10(%rsp),%r10 movq TF_R11(%rsp),%r11 movq TF_R12(%rsp),%r12 movq TF_R13(%rsp),%r13 movq TF_R14(%rsp),%r14 movq TF_R15(%rsp),%r15 testb $SEL_RPL_MASK,TF_CS(%rsp) /* Did we come from kernel? */ jz 1f /* keep running with kernel GS.base */ cli swapgs 1: addq $TF_RIP,%rsp /* skip over tf_err, tf_trapno */ .globl doreti_iret doreti_iret: iretq /* * doreti_iret_fault. Alternative return code for * the case where we get a fault in the doreti_exit code * above. trap() (amd64/amd64/trap.c) catches this specific * case, sends the process a signal and continues in the * corresponding place in the code below. */ ALIGN_TEXT .globl doreti_iret_fault doreti_iret_fault: subq $TF_RIP,%rsp /* space including tf_err, tf_trapno */ testl $PSL_I,TF_RFLAGS(%rsp) jz 1f sti 1: movq %rdi,TF_RDI(%rsp) movq %rsi,TF_RSI(%rsp) movq %rdx,TF_RDX(%rsp) movq %rcx,TF_RCX(%rsp) movq %r8,TF_R8(%rsp) movq %r9,TF_R9(%rsp) movq %rax,TF_RAX(%rsp) movq %rbx,TF_RBX(%rsp) movq %rbp,TF_RBP(%rsp) movq %r10,TF_R10(%rsp) movq %r11,TF_R11(%rsp) movq %r12,TF_R12(%rsp) movq %r13,TF_R13(%rsp) movq %r14,TF_R14(%rsp) movq %r15,TF_R15(%rsp) movq $T_PROTFLT,TF_TRAPNO(%rsp) movq $0,TF_ERR(%rsp) /* XXX should be the error code */ movq $0,TF_ADDR(%rsp) FAKE_MCOUNT(TF_RIP(%rsp)) jmp calltrap