1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 1991,1992 Linus Torvalds
5 * entry_32.S contains the system-call and low-level fault and trap handling routines.
7 * Stack layout while running C code:
8 * ptrace needs to have all registers on the stack.
9 * If the order here is changed, it needs to be
10 * updated in fork.c:copy_process(), signal.c:do_signal(),
11 * ptrace.c and ptrace.h
23 * 28(%esp) - %gs saved iff !CONFIG_X86_32_LAZY_GS
32 #include <linux/linkage.h>
33 #include <linux/err.h>
34 #include <asm/thread_info.h>
35 #include <asm/irqflags.h>
36 #include <asm/errno.h>
37 #include <asm/segment.h>
39 #include <asm/percpu.h>
40 #include <asm/processor-flags.h>
41 #include <asm/irq_vectors.h>
42 #include <asm/cpufeatures.h>
43 #include <asm/alternative-asm.h>
46 #include <asm/frame.h>
47 #include <asm/nospec-branch.h>
49 .section .entry.text, "ax"
52 * We use macros for low-level operations which need to be overridden
53 * for paravirtualization. The following will never clobber any registers:
54 * INTERRUPT_RETURN (aka. "iret")
55 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
56 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
58 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
59 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
60 * Allowing a register to be clobbered can shrink the paravirt replacement
61 * enough to patch inline, increasing performance.
65 # define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
67 # define preempt_stop(clobbers)
68 # define resume_kernel restore_all_kernel
71 .macro TRACE_IRQS_IRET
72 #ifdef CONFIG_TRACE_IRQFLAGS
73 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off?
80 #define PTI_SWITCH_MASK (1 << PAGE_SHIFT)
83 * User gs save/restore
85 * %gs is used for userland TLS and kernel only uses it for stack
86 * canary which is required to be at %gs:20 by gcc. Read the comment
87 * at the top of stackprotector.h for more info.
89 * Local labels 98 and 99 are used.
91 #ifdef CONFIG_X86_32_LAZY_GS
93 /* unfortunately push/pop can't be no-op */
98 addl $(4 + \pop), %esp
103 /* all the rest are no-op */
110 .macro REG_TO_PTGS reg
112 .macro SET_KERNEL_GS reg
115 #else /* CONFIG_X86_32_LAZY_GS */
128 .pushsection .fixup, "ax"
132 _ASM_EXTABLE(98b, 99b)
136 98: mov PT_GS(%esp), %gs
139 .pushsection .fixup, "ax"
140 99: movl $0, PT_GS(%esp)
143 _ASM_EXTABLE(98b, 99b)
149 .macro REG_TO_PTGS reg
150 movl \reg, PT_GS(%esp)
152 .macro SET_KERNEL_GS reg
153 movl $(__KERNEL_STACK_CANARY), \reg
157 #endif /* CONFIG_X86_32_LAZY_GS */
159 /* Unconditionally switch to user cr3 */
160 .macro SWITCH_TO_USER_CR3 scratch_reg:req
161 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
163 movl %cr3, \scratch_reg
164 orl $PTI_SWITCH_MASK, \scratch_reg
165 movl \scratch_reg, %cr3
169 .macro BUG_IF_WRONG_CR3 no_user_check=0
170 #ifdef CONFIG_DEBUG_ENTRY
171 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
172 .if \no_user_check == 0
173 /* coming from usermode? */
174 testl $SEGMENT_RPL_MASK, PT_CS(%esp)
179 testl $PTI_SWITCH_MASK, %eax
181 /* From userspace with kernel cr3 - BUG */
188 * Switch to kernel cr3 if not already loaded and return current cr3 in
191 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
192 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
193 movl %cr3, \scratch_reg
194 /* Test if we are already on kernel CR3 */
195 testl $PTI_SWITCH_MASK, \scratch_reg
197 andl $(~PTI_SWITCH_MASK), \scratch_reg
198 movl \scratch_reg, %cr3
199 /* Return original CR3 in \scratch_reg */
200 orl $PTI_SWITCH_MASK, \scratch_reg
204 .macro SAVE_ALL pt_regs_ax=%eax switch_stacks=0
217 movl $(__USER_DS), %edx
220 movl $(__KERNEL_PERCPU), %edx
224 /* Switch to kernel stack if necessary */
225 .if \switch_stacks > 0
226 SWITCH_TO_KERNEL_STACK
231 .macro SAVE_ALL_NMI cr3_reg:req
237 * Now switch the CR3 when PTI is enabled.
239 * We can enter with either user or kernel cr3, the code will
240 * store the old cr3 in \cr3_reg and switches to the kernel cr3
243 SWITCH_TO_KERNEL_CR3 scratch_reg=\cr3_reg
249 * This is a sneaky trick to help the unwinder find pt_regs on the stack. The
250 * frame pointer is replaced with an encoded pointer to pt_regs. The encoding
251 * is just clearing the MSB, which makes it an invalid stack address and is also
252 * a signal to the unwinder that it's a pt_regs pointer in disguise.
254 * NOTE: This macro must be used *after* SAVE_ALL because it corrupts the
257 .macro ENCODE_FRAME_POINTER
258 #ifdef CONFIG_FRAME_POINTER
260 andl $0x7fffffff, %ebp
264 .macro RESTORE_INT_REGS
274 .macro RESTORE_REGS pop=0
280 .pushsection .fixup, "ax"
294 .macro RESTORE_ALL_NMI cr3_reg:req pop=0
296 * Now switch the CR3 when PTI is enabled.
298 * We enter with kernel cr3 and switch the cr3 to the value
299 * stored on \cr3_reg, which is either a user or a kernel cr3.
301 ALTERNATIVE "jmp .Lswitched_\@", "", X86_FEATURE_PTI
303 testl $PTI_SWITCH_MASK, \cr3_reg
306 /* User cr3 in \cr3_reg - write it to hardware cr3 */
313 RESTORE_REGS pop=\pop
316 .macro CHECK_AND_APPLY_ESPFIX
317 #ifdef CONFIG_X86_ESPFIX32
318 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
320 ALTERNATIVE "jmp .Lend_\@", "", X86_BUG_ESPFIX
322 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
324 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
325 * are returning to the kernel.
326 * See comments in process.c:copy_thread() for details.
328 movb PT_OLDSS(%esp), %ah
329 movb PT_CS(%esp), %al
330 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
331 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
332 jne .Lend_\@ # returning to user-space with LDT SS
335 * Setup and switch to ESPFIX stack
337 * We're returning to userspace with a 16 bit stack. The CPU will not
338 * restore the high word of ESP for us on executing iret... This is an
339 * "official" bug of all the x86-compatible CPUs, which we can work
340 * around to make dosemu and wine happy. We do this by preloading the
341 * high word of ESP with the high word of the userspace ESP while
342 * compensating for the offset by changing to the ESPFIX segment with
343 * a base address that matches for the difference.
345 mov %esp, %edx /* load kernel esp */
346 mov PT_OLDESP(%esp), %eax /* load userspace esp */
347 mov %dx, %ax /* eax: new kernel esp */
348 sub %eax, %edx /* offset (low word is 0) */
350 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
351 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
353 pushl %eax /* new kernel esp */
355 * Disable interrupts, but do not irqtrace this section: we
356 * will soon execute iret and the tracer was already set to
357 * the irqstate after the IRET:
359 DISABLE_INTERRUPTS(CLBR_ANY)
360 lss (%esp), %esp /* switch to espfix segment */
362 #endif /* CONFIG_X86_ESPFIX32 */
366 * Called with pt_regs fully populated and kernel segments loaded,
367 * so we can access PER_CPU and use the integer registers.
369 * We need to be very careful here with the %esp switch, because an NMI
370 * can happen everywhere. If the NMI handler finds itself on the
371 * entry-stack, it will overwrite the task-stack and everything we
372 * copied there. So allocate the stack-frame on the task-stack and
373 * switch to it before we do any copying.
376 #define CS_FROM_ENTRY_STACK (1 << 31)
377 #define CS_FROM_USER_CR3 (1 << 30)
379 .macro SWITCH_TO_KERNEL_STACK
381 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
385 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
388 * %eax now contains the entry cr3 and we carry it forward in
389 * that register for the time this macro runs
393 * The high bits of the CS dword (__csh) are used for
394 * CS_FROM_ENTRY_STACK and CS_FROM_USER_CR3. Clear them in case
395 * hardware didn't do this for us.
397 andl $(0x0000ffff), PT_CS(%esp)
399 /* Are we on the entry stack? Bail out if not! */
400 movl PER_CPU_VAR(cpu_entry_area), %ecx
401 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
402 subl %esp, %ecx /* ecx = (end of entry_stack) - esp */
403 cmpl $SIZEOF_entry_stack, %ecx
406 /* Load stack pointer into %esi and %edi */
410 /* Move %edi to the top of the entry stack */
411 andl $(MASK_entry_stack), %edi
412 addl $(SIZEOF_entry_stack), %edi
414 /* Load top of task-stack into %edi */
415 movl TSS_entry2task_stack(%edi), %edi
417 /* Special case - entry from kernel mode via entry stack */
419 movl PT_EFLAGS(%esp), %ecx # mix EFLAGS and CS
420 movb PT_CS(%esp), %cl
421 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
423 movl PT_CS(%esp), %ecx
424 andl $SEGMENT_RPL_MASK, %ecx
427 jb .Lentry_from_kernel_\@
430 movl $PTREGS_SIZE, %ecx
433 testl $X86_EFLAGS_VM, PT_EFLAGS(%esi)
437 * Stack-frame contains 4 additional segment registers when
438 * coming from VM86 mode
445 /* Allocate frame on task-stack */
448 /* Switch to task-stack */
452 * We are now on the task-stack and can safely copy over the
461 .Lentry_from_kernel_\@:
464 * This handles the case when we enter the kernel from
465 * kernel-mode and %esp points to the entry-stack. When this
466 * happens we need to switch to the task-stack to run C code,
467 * but switch back to the entry-stack again when we approach
468 * iret and return to the interrupted code-path. This usually
469 * happens when we hit an exception while restoring user-space
470 * segment registers on the way back to user-space or when the
471 * sysenter handler runs with eflags.tf set.
473 * When we switch to the task-stack here, we can't trust the
474 * contents of the entry-stack anymore, as the exception handler
475 * might be scheduled out or moved to another CPU. Therefore we
476 * copy the complete entry-stack to the task-stack and set a
477 * marker in the iret-frame (bit 31 of the CS dword) to detect
478 * what we've done on the iret path.
480 * On the iret path we copy everything back and switch to the
481 * entry-stack, so that the interrupted kernel code-path
482 * continues on the same stack it was interrupted with.
484 * Be aware that an NMI can happen anytime in this code.
486 * %esi: Entry-Stack pointer (same as %esp)
487 * %edi: Top of the task stack
488 * %eax: CR3 on kernel entry
491 /* Calculate number of bytes on the entry stack in %ecx */
494 /* %ecx to the top of entry-stack */
495 andl $(MASK_entry_stack), %ecx
496 addl $(SIZEOF_entry_stack), %ecx
498 /* Number of bytes on the entry stack to %ecx */
501 /* Mark stackframe as coming from entry stack */
502 orl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
505 * Test the cr3 used to enter the kernel and add a marker
506 * so that we can switch back to it before iret.
508 testl $PTI_SWITCH_MASK, %eax
510 orl $CS_FROM_USER_CR3, PT_CS(%esp)
513 * %esi and %edi are unchanged, %ecx contains the number of
514 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
515 * the stack-frame on task-stack and copy everything over
517 jmp .Lcopy_pt_regs_\@
523 * Switch back from the kernel stack to the entry stack.
525 * The %esp register must point to pt_regs on the task stack. It will
526 * first calculate the size of the stack-frame to copy, depending on
527 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
528 * to copy the contents of the stack over to the entry stack.
530 * We must be very careful here, as we can't trust the contents of the
531 * task-stack once we switched to the entry-stack. When an NMI happens
532 * while on the entry-stack, the NMI handler will switch back to the top
533 * of the task stack, overwriting our stack-frame we are about to copy.
534 * Therefore we switch the stack only after everything is copied over.
536 .macro SWITCH_TO_ENTRY_STACK
538 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
541 movl $PTREGS_SIZE, %ecx
544 testl $(X86_EFLAGS_VM), PT_EFLAGS(%esp)
547 /* Additional 4 registers to copy when returning to VM86 mode */
553 /* Initialize source and destination for movsl */
554 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
558 /* Save future stack pointer in %ebx */
561 /* Copy over the stack-frame */
567 * Switch to entry-stack - needs to happen after everything is
568 * copied because the NMI handler will overwrite the task-stack
569 * when on entry-stack
577 * This macro handles the case when we return to kernel-mode on the iret
578 * path and have to switch back to the entry stack and/or user-cr3
580 * See the comments below the .Lentry_from_kernel_\@ label in the
581 * SWITCH_TO_KERNEL_STACK macro for more details.
583 .macro PARANOID_EXIT_TO_KERNEL_MODE
586 * Test if we entered the kernel with the entry-stack. Most
587 * likely we did not, because this code only runs on the
588 * return-to-kernel path.
590 testl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
593 /* Unlikely slow-path */
595 /* Clear marker from stack-frame */
596 andl $(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
598 /* Copy the remaining task-stack contents to entry-stack */
600 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
602 /* Bytes on the task-stack to ecx */
603 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
606 /* Allocate stack-frame on entry-stack */
610 * Save future stack-pointer, we must not switch until the
611 * copy is done, otherwise the NMI handler could destroy the
612 * contents of the task-stack we are about to copy.
621 /* Safe to switch to entry-stack now */
625 * We came from entry-stack and need to check if we also need to
626 * switch back to user cr3.
628 testl $CS_FROM_USER_CR3, PT_CS(%esp)
631 /* Clear marker from stack-frame */
632 andl $(~CS_FROM_USER_CR3), PT_CS(%esp)
634 SWITCH_TO_USER_CR3 scratch_reg=%eax
642 ENTRY(__switch_to_asm)
644 * Save callee-saved registers
645 * This must match the order in struct inactive_task_frame
653 movl %esp, TASK_threadsp(%eax)
654 movl TASK_threadsp(%edx), %esp
656 #ifdef CONFIG_STACKPROTECTOR
657 movl TASK_stack_canary(%edx), %ebx
658 movl %ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
661 #ifdef CONFIG_RETPOLINE
663 * When switching from a shallower to a deeper call stack
664 * the RSB may either underflow or use entries populated
665 * with userspace addresses. On CPUs where those concerns
666 * exist, overwrite the RSB with entries which capture
667 * speculative execution to prevent attack.
669 FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
672 /* restore callee-saved registers */
682 * The unwinder expects the last frame on the stack to always be at the same
683 * offset from the end of the page, which allows it to validate the stack.
684 * Calling schedule_tail() directly would break that convention because its an
685 * asmlinkage function so its argument has to be pushed on the stack. This
686 * wrapper creates a proper "end of stack" frame header before the call.
688 ENTRY(schedule_tail_wrapper)
697 ENDPROC(schedule_tail_wrapper)
699 * A newly forked process directly context switches into this address.
701 * eax: prev task we switched from
702 * ebx: kernel thread func (NULL for user thread)
703 * edi: kernel thread arg
706 call schedule_tail_wrapper
709 jnz 1f /* kernel threads are uncommon */
712 /* When we fork, we trace the syscall return in the child, too. */
714 call syscall_return_slowpath
721 * A kernel thread is allowed to return here after successfully
722 * calling do_execve(). Exit to userspace to complete the execve()
725 movl $0, PT_EAX(%esp)
730 * Return to user mode is not as complex as all this looks,
731 * but we want the default path for a system call return to
732 * go as quickly as possible which is why some of this is
733 * less clear than it otherwise should be.
736 # userspace resumption stub bypassing syscall exit tracing
739 preempt_stop(CLBR_ANY)
742 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
743 movb PT_CS(%esp), %al
744 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
747 * We can be coming here from child spawned by kernel_thread().
749 movl PT_CS(%esp), %eax
750 andl $SEGMENT_RPL_MASK, %eax
753 jb resume_kernel # not returning to v8086 or userspace
755 ENTRY(resume_userspace)
756 DISABLE_INTERRUPTS(CLBR_ANY)
759 call prepare_exit_to_usermode
761 END(ret_from_exception)
763 #ifdef CONFIG_PREEMPT
765 DISABLE_INTERRUPTS(CLBR_ANY)
767 cmpl $0, PER_CPU_VAR(__preempt_count)
768 jnz restore_all_kernel
769 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
770 jz restore_all_kernel
771 call preempt_schedule_irq
776 GLOBAL(__begin_SYSENTER_singlestep_region)
778 * All code from here through __end_SYSENTER_singlestep_region is subject
779 * to being single-stepped if a user program sets TF and executes SYSENTER.
780 * There is absolutely nothing that we can do to prevent this from happening
781 * (thanks Intel!). To keep our handling of this situation as simple as
782 * possible, we handle TF just like AC and NT, except that our #DB handler
783 * will ignore all of the single-step traps generated in this range.
788 * Xen doesn't set %esp to be precisely what the normal SYSENTER
789 * entry point expects, so fix it up before using the normal path.
791 ENTRY(xen_sysenter_target)
792 addl $5*4, %esp /* remove xen-provided frame */
793 jmp .Lsysenter_past_esp
797 * 32-bit SYSENTER entry.
799 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
800 * if X86_FEATURE_SEP is available. This is the preferred system call
801 * entry on 32-bit systems.
803 * The SYSENTER instruction, in principle, should *only* occur in the
804 * vDSO. In practice, a small number of Android devices were shipped
805 * with a copy of Bionic that inlined a SYSENTER instruction. This
806 * never happened in any of Google's Bionic versions -- it only happened
807 * in a narrow range of Intel-provided versions.
809 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
810 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
811 * SYSENTER does not save anything on the stack,
812 * and does not save old EIP (!!!), ESP, or EFLAGS.
814 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
815 * user and/or vm86 state), we explicitly disable the SYSENTER
816 * instruction in vm86 mode by reprogramming the MSRs.
819 * eax system call number
828 ENTRY(entry_SYSENTER_32)
830 * On entry-stack with all userspace-regs live - save and
831 * restore eflags and %eax to use it as scratch-reg for the cr3
836 BUG_IF_WRONG_CR3 no_user_check=1
837 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
841 /* Stack empty again, switch to task stack */
842 movl TSS_entry2task_stack(%esp), %esp
845 pushl $__USER_DS /* pt_regs->ss */
846 pushl %ebp /* pt_regs->sp (stashed in bp) */
847 pushfl /* pt_regs->flags (except IF = 0) */
848 orl $X86_EFLAGS_IF, (%esp) /* Fix IF */
849 pushl $__USER_CS /* pt_regs->cs */
850 pushl $0 /* pt_regs->ip = 0 (placeholder) */
851 pushl %eax /* pt_regs->orig_ax */
852 SAVE_ALL pt_regs_ax=$-ENOSYS /* save rest, stack already switched */
855 * SYSENTER doesn't filter flags, so we need to clear NT, AC
856 * and TF ourselves. To save a few cycles, we can check whether
857 * either was set instead of doing an unconditional popfq.
858 * This needs to happen before enabling interrupts so that
859 * we don't get preempted with NT set.
861 * If TF is set, we will single-step all the way to here -- do_debug
862 * will ignore all the traps. (Yes, this is slow, but so is
863 * single-stepping in general. This allows us to avoid having
864 * a more complicated code to handle the case where a user program
865 * forces us to single-step through the SYSENTER entry code.)
867 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
868 * out-of-line as an optimization: NT is unlikely to be set in the
869 * majority of the cases and instead of polluting the I$ unnecessarily,
870 * we're keeping that code behind a branch which will predict as
871 * not-taken and therefore its instructions won't be fetched.
873 testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
874 jnz .Lsysenter_fix_flags
875 .Lsysenter_flags_fixed:
878 * User mode is traced as though IRQs are on, and SYSENTER
884 call do_fast_syscall_32
885 /* XEN PV guests always use IRET path */
886 ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
887 "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
889 /* Opportunistic SYSEXIT */
890 TRACE_IRQS_ON /* User mode traces as IRQs on. */
893 * Setup entry stack - we keep the pointer in %eax and do the
894 * switch after almost all user-state is restored.
897 /* Load entry stack pointer and allocate frame for eflags/eax */
898 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
901 /* Copy eflags and eax to entry stack */
902 movl PT_EFLAGS(%esp), %edi
903 movl PT_EAX(%esp), %esi
907 /* Restore user registers and segments */
908 movl PT_EIP(%esp), %edx /* pt_regs->ip */
909 movl PT_OLDESP(%esp), %ecx /* pt_regs->sp */
910 1: mov PT_FS(%esp), %fs
913 popl %ebx /* pt_regs->bx */
914 addl $2*4, %esp /* skip pt_regs->cx and pt_regs->dx */
915 popl %esi /* pt_regs->si */
916 popl %edi /* pt_regs->di */
917 popl %ebp /* pt_regs->bp */
919 /* Switch to entry stack */
922 /* Now ready to switch the cr3 */
923 SWITCH_TO_USER_CR3 scratch_reg=%eax
926 * Restore all flags except IF. (We restore IF separately because
927 * STI gives a one-instruction window in which we won't be interrupted,
928 * whereas POPF does not.)
930 btrl $X86_EFLAGS_IF_BIT, (%esp)
931 BUG_IF_WRONG_CR3 no_user_check=1
936 * Return back to the vDSO, which will pop ecx and edx.
937 * Don't bother with DS and ES (they already contain __USER_DS).
942 .pushsection .fixup, "ax"
943 2: movl $0, PT_FS(%esp)
949 .Lsysenter_fix_flags:
950 pushl $X86_EFLAGS_FIXED
952 jmp .Lsysenter_flags_fixed
953 GLOBAL(__end_SYSENTER_singlestep_region)
954 ENDPROC(entry_SYSENTER_32)
957 * 32-bit legacy system call entry.
959 * 32-bit x86 Linux system calls traditionally used the INT $0x80
960 * instruction. INT $0x80 lands here.
962 * This entry point can be used by any 32-bit perform system calls.
963 * Instances of INT $0x80 can be found inline in various programs and
964 * libraries. It is also used by the vDSO's __kernel_vsyscall
965 * fallback for hardware that doesn't support a faster entry method.
966 * Restarted 32-bit system calls also fall back to INT $0x80
967 * regardless of what instruction was originally used to do the system
968 * call. (64-bit programs can use INT $0x80 as well, but they can
969 * only run on 64-bit kernels and therefore land in
970 * entry_INT80_compat.)
972 * This is considered a slow path. It is not used by most libc
973 * implementations on modern hardware except during process startup.
976 * eax system call number
984 ENTRY(entry_INT80_32)
986 pushl %eax /* pt_regs->orig_ax */
988 SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1 /* save rest */
991 * User mode is traced as though IRQs are on, and the interrupt gate
997 call do_int80_syscall_32
1002 SWITCH_TO_ENTRY_STACK
1003 .Lrestore_all_notrace:
1004 CHECK_AND_APPLY_ESPFIX
1006 /* Switch back to user CR3 */
1007 SWITCH_TO_USER_CR3 scratch_reg=%eax
1011 /* Restore user state */
1012 RESTORE_REGS pop=4 # skip orig_eax/error_code
1015 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1016 * when returning from IPI handler and when returning from
1017 * scheduler to user-space.
1023 PARANOID_EXIT_TO_KERNEL_MODE
1028 .section .fixup, "ax"
1030 pushl $0 # no error code
1031 pushl $do_iret_error
1033 #ifdef CONFIG_DEBUG_ENTRY
1035 * The stack-frame here is the one that iret faulted on, so its a
1036 * return-to-user frame. We are on kernel-cr3 because we come here from
1037 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1038 * as the checker expects it.
1041 SWITCH_TO_USER_CR3 scratch_reg=%eax
1045 jmp common_exception
1047 _ASM_EXTABLE(.Lirq_return, iret_exc)
1048 ENDPROC(entry_INT80_32)
1050 .macro FIXUP_ESPFIX_STACK
1052 * Switch back for ESPFIX stack to the normal zerobased stack
1054 * We can't call C functions using the ESPFIX stack. This code reads
1055 * the high word of the segment base from the GDT and swiches to the
1056 * normal stack and adjusts ESP with the matching offset.
1058 #ifdef CONFIG_X86_ESPFIX32
1059 /* fixup the stack */
1060 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1061 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1063 addl %esp, %eax /* the adjusted stack pointer */
1066 lss (%esp), %esp /* switch to the normal stack segment */
1069 .macro UNWIND_ESPFIX_STACK
1070 #ifdef CONFIG_X86_ESPFIX32
1072 /* see if on espfix stack */
1073 cmpw $__ESPFIX_SS, %ax
1075 movl $__KERNEL_DS, %eax
1078 /* switch to normal stack */
1085 * Build the entry stubs with some assembler magic.
1086 * We pack 1 stub into every 8-byte block.
1089 ENTRY(irq_entries_start)
1090 vector=FIRST_EXTERNAL_VECTOR
1091 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1092 pushl $(~vector+0x80) /* Note: always in signed byte range */
1094 jmp common_interrupt
1097 END(irq_entries_start)
1100 * the CPU automatically disables interrupts when executing an IRQ vector,
1101 * so IRQ-flags tracing has to follow that:
1103 .p2align CONFIG_X86_L1_CACHE_SHIFT
1106 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1108 SAVE_ALL switch_stacks=1
1109 ENCODE_FRAME_POINTER
1114 ENDPROC(common_interrupt)
1116 #define BUILD_INTERRUPT3(name, nr, fn) \
1120 SAVE_ALL switch_stacks=1; \
1121 ENCODE_FRAME_POINTER; \
1125 jmp ret_from_intr; \
1128 #define BUILD_INTERRUPT(name, nr) \
1129 BUILD_INTERRUPT3(name, nr, smp_##name); \
1131 /* The include is where all of the SMP etc. interrupts come from */
1132 #include <asm/entry_arch.h>
1134 ENTRY(coprocessor_error)
1137 pushl $do_coprocessor_error
1138 jmp common_exception
1139 END(coprocessor_error)
1141 ENTRY(simd_coprocessor_error)
1144 #ifdef CONFIG_X86_INVD_BUG
1145 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1146 ALTERNATIVE "pushl $do_general_protection", \
1147 "pushl $do_simd_coprocessor_error", \
1150 pushl $do_simd_coprocessor_error
1152 jmp common_exception
1153 END(simd_coprocessor_error)
1155 ENTRY(device_not_available)
1157 pushl $-1 # mark this as an int
1158 pushl $do_device_not_available
1159 jmp common_exception
1160 END(device_not_available)
1162 #ifdef CONFIG_PARAVIRT
1165 _ASM_EXTABLE(native_iret, iret_exc)
1173 jmp common_exception
1180 jmp common_exception
1186 pushl $do_invalid_op
1187 jmp common_exception
1190 ENTRY(coprocessor_segment_overrun)
1193 pushl $do_coprocessor_segment_overrun
1194 jmp common_exception
1195 END(coprocessor_segment_overrun)
1199 pushl $do_invalid_TSS
1200 jmp common_exception
1203 ENTRY(segment_not_present)
1205 pushl $do_segment_not_present
1206 jmp common_exception
1207 END(segment_not_present)
1209 ENTRY(stack_segment)
1211 pushl $do_stack_segment
1212 jmp common_exception
1215 ENTRY(alignment_check)
1217 pushl $do_alignment_check
1218 jmp common_exception
1219 END(alignment_check)
1223 pushl $0 # no error code
1224 pushl $do_divide_error
1225 jmp common_exception
1228 #ifdef CONFIG_X86_MCE
1229 ENTRY(machine_check)
1232 pushl machine_check_vector
1233 jmp common_exception
1237 ENTRY(spurious_interrupt_bug)
1240 pushl $do_spurious_interrupt_bug
1241 jmp common_exception
1242 END(spurious_interrupt_bug)
1244 #ifdef CONFIG_XEN_PV
1245 ENTRY(xen_hypervisor_callback)
1246 pushl $-1 /* orig_ax = -1 => not a system call */
1248 ENCODE_FRAME_POINTER
1252 * Check to see if we got the event in the critical
1253 * region in xen_iret_direct, after we've reenabled
1254 * events and checked for pending events. This simulates
1255 * iret instruction's behaviour where it delivers a
1256 * pending interrupt when enabling interrupts:
1258 movl PT_EIP(%esp), %eax
1259 cmpl $xen_iret_start_crit, %eax
1261 cmpl $xen_iret_end_crit, %eax
1264 jmp xen_iret_crit_fixup
1266 ENTRY(xen_do_upcall)
1268 call xen_evtchn_do_upcall
1269 #ifndef CONFIG_PREEMPT
1270 call xen_maybe_preempt_hcall
1273 ENDPROC(xen_hypervisor_callback)
1276 * Hypervisor uses this for application faults while it executes.
1277 * We get here for two reasons:
1278 * 1. Fault while reloading DS, ES, FS or GS
1279 * 2. Fault while executing IRET
1280 * Category 1 we fix up by reattempting the load, and zeroing the segment
1281 * register if the load fails.
1282 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1283 * normal Linux return path in this case because if we use the IRET hypercall
1284 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1285 * We distinguish between categories by maintaining a status value in EAX.
1287 ENTRY(xen_failsafe_callback)
1292 3: mov 12(%esp), %fs
1293 4: mov 16(%esp), %gs
1294 /* EAX == 0 => Category 1 (Bad segment)
1295 EAX != 0 => Category 2 (Bad IRET) */
1301 5: pushl $-1 /* orig_ax = -1 => not a system call */
1303 ENCODE_FRAME_POINTER
1304 jmp ret_from_exception
1306 .section .fixup, "ax"
1320 _ASM_EXTABLE(1b, 6b)
1321 _ASM_EXTABLE(2b, 7b)
1322 _ASM_EXTABLE(3b, 8b)
1323 _ASM_EXTABLE(4b, 9b)
1324 ENDPROC(xen_failsafe_callback)
1325 #endif /* CONFIG_XEN_PV */
1327 #ifdef CONFIG_XEN_PVHVM
1328 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1329 xen_evtchn_do_upcall)
1333 #if IS_ENABLED(CONFIG_HYPERV)
1335 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1336 hyperv_vector_handler)
1338 BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1339 hyperv_reenlightenment_intr)
1341 BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1342 hv_stimer0_vector_handler)
1344 #endif /* CONFIG_HYPERV */
1348 pushl $do_page_fault
1350 jmp common_exception
1354 /* the function address is in %gs's slot on the stack */
1359 movl $(__USER_DS), %eax
1362 movl $(__KERNEL_PERCPU), %eax
1370 SWITCH_TO_KERNEL_STACK
1371 ENCODE_FRAME_POINTER
1375 movl PT_GS(%esp), %edi # get the function address
1376 movl PT_ORIG_EAX(%esp), %edx # get the error code
1377 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1381 movl %esp, %eax # pt_regs pointer
1383 jmp ret_from_exception
1384 END(common_exception)
1388 * Entry from sysenter is now handled in common_exception
1391 pushl $-1 # mark this as an int
1393 jmp common_exception
1397 * NMI is doubly nasty. It can happen on the first instruction of
1398 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1399 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1400 * switched stacks. We handle both conditions by simply checking whether we
1401 * interrupted kernel code running on the SYSENTER stack.
1406 #ifdef CONFIG_X86_ESPFIX32
1409 cmpw $__ESPFIX_SS, %ax
1411 je .Lnmi_espfix_stack
1414 pushl %eax # pt_regs->orig_ax
1415 SAVE_ALL_NMI cr3_reg=%edi
1416 ENCODE_FRAME_POINTER
1417 xorl %edx, %edx # zero error code
1418 movl %esp, %eax # pt_regs pointer
1420 /* Are we currently on the SYSENTER stack? */
1421 movl PER_CPU_VAR(cpu_entry_area), %ecx
1422 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1423 subl %eax, %ecx /* ecx = (end of entry_stack) - esp */
1424 cmpl $SIZEOF_entry_stack, %ecx
1425 jb .Lnmi_from_sysenter_stack
1427 /* Not on SYSENTER stack. */
1431 .Lnmi_from_sysenter_stack:
1433 * We're on the SYSENTER stack. Switch off. No one (not even debug)
1434 * is using the thread stack right now, so it's safe for us to use it.
1437 movl PER_CPU_VAR(cpu_current_top_of_stack), %esp
1442 CHECK_AND_APPLY_ESPFIX
1443 RESTORE_ALL_NMI cr3_reg=%edi pop=4
1446 #ifdef CONFIG_X86_ESPFIX32
1449 * create the pointer to lss back
1454 /* copy the iret frame of 12 bytes */
1459 SAVE_ALL_NMI cr3_reg=%edi
1460 ENCODE_FRAME_POINTER
1461 FIXUP_ESPFIX_STACK # %eax == %esp
1462 xorl %edx, %edx # zero error code
1464 RESTORE_ALL_NMI cr3_reg=%edi
1465 lss 12+4(%esp), %esp # back to espfix stack
1472 pushl $-1 # mark this as an int
1474 SAVE_ALL switch_stacks=1
1475 ENCODE_FRAME_POINTER
1477 xorl %edx, %edx # zero error code
1478 movl %esp, %eax # pt_regs pointer
1480 jmp ret_from_exception
1483 ENTRY(general_protection)
1484 pushl $do_general_protection
1485 jmp common_exception
1486 END(general_protection)
1488 #ifdef CONFIG_KVM_GUEST
1489 ENTRY(async_page_fault)
1491 pushl $do_async_page_fault
1492 jmp common_exception
1493 END(async_page_fault)
1496 ENTRY(rewind_stack_do_exit)
1497 /* Prevent any naive code from trying to unwind to our caller. */
1500 movl PER_CPU_VAR(cpu_current_top_of_stack), %esi
1501 leal -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1505 END(rewind_stack_do_exit)