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
392 /* Are we on the entry stack? Bail out if not! */
393 movl PER_CPU_VAR(cpu_entry_area), %ecx
394 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
395 subl %esp, %ecx /* ecx = (end of entry_stack) - esp */
396 cmpl $SIZEOF_entry_stack, %ecx
399 /* Load stack pointer into %esi and %edi */
403 /* Move %edi to the top of the entry stack */
404 andl $(MASK_entry_stack), %edi
405 addl $(SIZEOF_entry_stack), %edi
407 /* Load top of task-stack into %edi */
408 movl TSS_entry2task_stack(%edi), %edi
411 * Clear unused upper bits of the dword containing the word-sized CS
412 * slot in pt_regs in case hardware didn't clear it for us.
414 andl $(0x0000ffff), PT_CS(%esp)
416 /* Special case - entry from kernel mode via entry stack */
418 movl PT_EFLAGS(%esp), %ecx # mix EFLAGS and CS
419 movb PT_CS(%esp), %cl
420 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
422 movl PT_CS(%esp), %ecx
423 andl $SEGMENT_RPL_MASK, %ecx
426 jb .Lentry_from_kernel_\@
429 movl $PTREGS_SIZE, %ecx
432 testl $X86_EFLAGS_VM, PT_EFLAGS(%esi)
436 * Stack-frame contains 4 additional segment registers when
437 * coming from VM86 mode
444 /* Allocate frame on task-stack */
447 /* Switch to task-stack */
451 * We are now on the task-stack and can safely copy over the
460 .Lentry_from_kernel_\@:
463 * This handles the case when we enter the kernel from
464 * kernel-mode and %esp points to the entry-stack. When this
465 * happens we need to switch to the task-stack to run C code,
466 * but switch back to the entry-stack again when we approach
467 * iret and return to the interrupted code-path. This usually
468 * happens when we hit an exception while restoring user-space
469 * segment registers on the way back to user-space or when the
470 * sysenter handler runs with eflags.tf set.
472 * When we switch to the task-stack here, we can't trust the
473 * contents of the entry-stack anymore, as the exception handler
474 * might be scheduled out or moved to another CPU. Therefore we
475 * copy the complete entry-stack to the task-stack and set a
476 * marker in the iret-frame (bit 31 of the CS dword) to detect
477 * what we've done on the iret path.
479 * On the iret path we copy everything back and switch to the
480 * entry-stack, so that the interrupted kernel code-path
481 * continues on the same stack it was interrupted with.
483 * Be aware that an NMI can happen anytime in this code.
485 * %esi: Entry-Stack pointer (same as %esp)
486 * %edi: Top of the task stack
487 * %eax: CR3 on kernel entry
490 /* Calculate number of bytes on the entry stack in %ecx */
493 /* %ecx to the top of entry-stack */
494 andl $(MASK_entry_stack), %ecx
495 addl $(SIZEOF_entry_stack), %ecx
497 /* Number of bytes on the entry stack to %ecx */
500 /* Mark stackframe as coming from entry stack */
501 orl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
504 * Test the cr3 used to enter the kernel and add a marker
505 * so that we can switch back to it before iret.
507 testl $PTI_SWITCH_MASK, %eax
509 orl $CS_FROM_USER_CR3, PT_CS(%esp)
512 * %esi and %edi are unchanged, %ecx contains the number of
513 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
514 * the stack-frame on task-stack and copy everything over
516 jmp .Lcopy_pt_regs_\@
522 * Switch back from the kernel stack to the entry stack.
524 * The %esp register must point to pt_regs on the task stack. It will
525 * first calculate the size of the stack-frame to copy, depending on
526 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
527 * to copy the contents of the stack over to the entry stack.
529 * We must be very careful here, as we can't trust the contents of the
530 * task-stack once we switched to the entry-stack. When an NMI happens
531 * while on the entry-stack, the NMI handler will switch back to the top
532 * of the task stack, overwriting our stack-frame we are about to copy.
533 * Therefore we switch the stack only after everything is copied over.
535 .macro SWITCH_TO_ENTRY_STACK
537 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
540 movl $PTREGS_SIZE, %ecx
543 testl $(X86_EFLAGS_VM), PT_EFLAGS(%esp)
546 /* Additional 4 registers to copy when returning to VM86 mode */
552 /* Initialize source and destination for movsl */
553 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
557 /* Save future stack pointer in %ebx */
560 /* Copy over the stack-frame */
566 * Switch to entry-stack - needs to happen after everything is
567 * copied because the NMI handler will overwrite the task-stack
568 * when on entry-stack
576 * This macro handles the case when we return to kernel-mode on the iret
577 * path and have to switch back to the entry stack and/or user-cr3
579 * See the comments below the .Lentry_from_kernel_\@ label in the
580 * SWITCH_TO_KERNEL_STACK macro for more details.
582 .macro PARANOID_EXIT_TO_KERNEL_MODE
585 * Test if we entered the kernel with the entry-stack. Most
586 * likely we did not, because this code only runs on the
587 * return-to-kernel path.
589 testl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
592 /* Unlikely slow-path */
594 /* Clear marker from stack-frame */
595 andl $(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
597 /* Copy the remaining task-stack contents to entry-stack */
599 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
601 /* Bytes on the task-stack to ecx */
602 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
605 /* Allocate stack-frame on entry-stack */
609 * Save future stack-pointer, we must not switch until the
610 * copy is done, otherwise the NMI handler could destroy the
611 * contents of the task-stack we are about to copy.
620 /* Safe to switch to entry-stack now */
624 * We came from entry-stack and need to check if we also need to
625 * switch back to user cr3.
627 testl $CS_FROM_USER_CR3, PT_CS(%esp)
630 /* Clear marker from stack-frame */
631 andl $(~CS_FROM_USER_CR3), PT_CS(%esp)
633 SWITCH_TO_USER_CR3 scratch_reg=%eax
641 ENTRY(__switch_to_asm)
643 * Save callee-saved registers
644 * This must match the order in struct inactive_task_frame
652 movl %esp, TASK_threadsp(%eax)
653 movl TASK_threadsp(%edx), %esp
655 #ifdef CONFIG_STACKPROTECTOR
656 movl TASK_stack_canary(%edx), %ebx
657 movl %ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
660 #ifdef CONFIG_RETPOLINE
662 * When switching from a shallower to a deeper call stack
663 * the RSB may either underflow or use entries populated
664 * with userspace addresses. On CPUs where those concerns
665 * exist, overwrite the RSB with entries which capture
666 * speculative execution to prevent attack.
668 FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
671 /* restore callee-saved registers */
681 * The unwinder expects the last frame on the stack to always be at the same
682 * offset from the end of the page, which allows it to validate the stack.
683 * Calling schedule_tail() directly would break that convention because its an
684 * asmlinkage function so its argument has to be pushed on the stack. This
685 * wrapper creates a proper "end of stack" frame header before the call.
687 ENTRY(schedule_tail_wrapper)
696 ENDPROC(schedule_tail_wrapper)
698 * A newly forked process directly context switches into this address.
700 * eax: prev task we switched from
701 * ebx: kernel thread func (NULL for user thread)
702 * edi: kernel thread arg
705 call schedule_tail_wrapper
708 jnz 1f /* kernel threads are uncommon */
711 /* When we fork, we trace the syscall return in the child, too. */
713 call syscall_return_slowpath
720 * A kernel thread is allowed to return here after successfully
721 * calling do_execve(). Exit to userspace to complete the execve()
724 movl $0, PT_EAX(%esp)
729 * Return to user mode is not as complex as all this looks,
730 * but we want the default path for a system call return to
731 * go as quickly as possible which is why some of this is
732 * less clear than it otherwise should be.
735 # userspace resumption stub bypassing syscall exit tracing
738 preempt_stop(CLBR_ANY)
741 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
742 movb PT_CS(%esp), %al
743 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
746 * We can be coming here from child spawned by kernel_thread().
748 movl PT_CS(%esp), %eax
749 andl $SEGMENT_RPL_MASK, %eax
752 jb resume_kernel # not returning to v8086 or userspace
754 ENTRY(resume_userspace)
755 DISABLE_INTERRUPTS(CLBR_ANY)
758 call prepare_exit_to_usermode
760 END(ret_from_exception)
762 #ifdef CONFIG_PREEMPT
764 DISABLE_INTERRUPTS(CLBR_ANY)
766 cmpl $0, PER_CPU_VAR(__preempt_count)
767 jnz restore_all_kernel
768 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
769 jz restore_all_kernel
770 call preempt_schedule_irq
775 GLOBAL(__begin_SYSENTER_singlestep_region)
777 * All code from here through __end_SYSENTER_singlestep_region is subject
778 * to being single-stepped if a user program sets TF and executes SYSENTER.
779 * There is absolutely nothing that we can do to prevent this from happening
780 * (thanks Intel!). To keep our handling of this situation as simple as
781 * possible, we handle TF just like AC and NT, except that our #DB handler
782 * will ignore all of the single-step traps generated in this range.
787 * Xen doesn't set %esp to be precisely what the normal SYSENTER
788 * entry point expects, so fix it up before using the normal path.
790 ENTRY(xen_sysenter_target)
791 addl $5*4, %esp /* remove xen-provided frame */
792 jmp .Lsysenter_past_esp
796 * 32-bit SYSENTER entry.
798 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
799 * if X86_FEATURE_SEP is available. This is the preferred system call
800 * entry on 32-bit systems.
802 * The SYSENTER instruction, in principle, should *only* occur in the
803 * vDSO. In practice, a small number of Android devices were shipped
804 * with a copy of Bionic that inlined a SYSENTER instruction. This
805 * never happened in any of Google's Bionic versions -- it only happened
806 * in a narrow range of Intel-provided versions.
808 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
809 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
810 * SYSENTER does not save anything on the stack,
811 * and does not save old EIP (!!!), ESP, or EFLAGS.
813 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
814 * user and/or vm86 state), we explicitly disable the SYSENTER
815 * instruction in vm86 mode by reprogramming the MSRs.
818 * eax system call number
827 ENTRY(entry_SYSENTER_32)
829 * On entry-stack with all userspace-regs live - save and
830 * restore eflags and %eax to use it as scratch-reg for the cr3
835 BUG_IF_WRONG_CR3 no_user_check=1
836 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
840 /* Stack empty again, switch to task stack */
841 movl TSS_entry2task_stack(%esp), %esp
844 pushl $__USER_DS /* pt_regs->ss */
845 pushl %ebp /* pt_regs->sp (stashed in bp) */
846 pushfl /* pt_regs->flags (except IF = 0) */
847 orl $X86_EFLAGS_IF, (%esp) /* Fix IF */
848 pushl $__USER_CS /* pt_regs->cs */
849 pushl $0 /* pt_regs->ip = 0 (placeholder) */
850 pushl %eax /* pt_regs->orig_ax */
851 SAVE_ALL pt_regs_ax=$-ENOSYS /* save rest, stack already switched */
854 * SYSENTER doesn't filter flags, so we need to clear NT, AC
855 * and TF ourselves. To save a few cycles, we can check whether
856 * either was set instead of doing an unconditional popfq.
857 * This needs to happen before enabling interrupts so that
858 * we don't get preempted with NT set.
860 * If TF is set, we will single-step all the way to here -- do_debug
861 * will ignore all the traps. (Yes, this is slow, but so is
862 * single-stepping in general. This allows us to avoid having
863 * a more complicated code to handle the case where a user program
864 * forces us to single-step through the SYSENTER entry code.)
866 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
867 * out-of-line as an optimization: NT is unlikely to be set in the
868 * majority of the cases and instead of polluting the I$ unnecessarily,
869 * we're keeping that code behind a branch which will predict as
870 * not-taken and therefore its instructions won't be fetched.
872 testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
873 jnz .Lsysenter_fix_flags
874 .Lsysenter_flags_fixed:
877 * User mode is traced as though IRQs are on, and SYSENTER
883 call do_fast_syscall_32
884 /* XEN PV guests always use IRET path */
885 ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
886 "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
888 /* Opportunistic SYSEXIT */
889 TRACE_IRQS_ON /* User mode traces as IRQs on. */
892 * Setup entry stack - we keep the pointer in %eax and do the
893 * switch after almost all user-state is restored.
896 /* Load entry stack pointer and allocate frame for eflags/eax */
897 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
900 /* Copy eflags and eax to entry stack */
901 movl PT_EFLAGS(%esp), %edi
902 movl PT_EAX(%esp), %esi
906 /* Restore user registers and segments */
907 movl PT_EIP(%esp), %edx /* pt_regs->ip */
908 movl PT_OLDESP(%esp), %ecx /* pt_regs->sp */
909 1: mov PT_FS(%esp), %fs
912 popl %ebx /* pt_regs->bx */
913 addl $2*4, %esp /* skip pt_regs->cx and pt_regs->dx */
914 popl %esi /* pt_regs->si */
915 popl %edi /* pt_regs->di */
916 popl %ebp /* pt_regs->bp */
918 /* Switch to entry stack */
921 /* Now ready to switch the cr3 */
922 SWITCH_TO_USER_CR3 scratch_reg=%eax
925 * Restore all flags except IF. (We restore IF separately because
926 * STI gives a one-instruction window in which we won't be interrupted,
927 * whereas POPF does not.)
929 btrl $X86_EFLAGS_IF_BIT, (%esp)
930 BUG_IF_WRONG_CR3 no_user_check=1
935 * Return back to the vDSO, which will pop ecx and edx.
936 * Don't bother with DS and ES (they already contain __USER_DS).
941 .pushsection .fixup, "ax"
942 2: movl $0, PT_FS(%esp)
948 .Lsysenter_fix_flags:
949 pushl $X86_EFLAGS_FIXED
951 jmp .Lsysenter_flags_fixed
952 GLOBAL(__end_SYSENTER_singlestep_region)
953 ENDPROC(entry_SYSENTER_32)
956 * 32-bit legacy system call entry.
958 * 32-bit x86 Linux system calls traditionally used the INT $0x80
959 * instruction. INT $0x80 lands here.
961 * This entry point can be used by any 32-bit perform system calls.
962 * Instances of INT $0x80 can be found inline in various programs and
963 * libraries. It is also used by the vDSO's __kernel_vsyscall
964 * fallback for hardware that doesn't support a faster entry method.
965 * Restarted 32-bit system calls also fall back to INT $0x80
966 * regardless of what instruction was originally used to do the system
967 * call. (64-bit programs can use INT $0x80 as well, but they can
968 * only run on 64-bit kernels and therefore land in
969 * entry_INT80_compat.)
971 * This is considered a slow path. It is not used by most libc
972 * implementations on modern hardware except during process startup.
975 * eax system call number
983 ENTRY(entry_INT80_32)
985 pushl %eax /* pt_regs->orig_ax */
987 SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1 /* save rest */
990 * User mode is traced as though IRQs are on, and the interrupt gate
996 call do_int80_syscall_32
1001 SWITCH_TO_ENTRY_STACK
1002 .Lrestore_all_notrace:
1003 CHECK_AND_APPLY_ESPFIX
1005 /* Switch back to user CR3 */
1006 SWITCH_TO_USER_CR3 scratch_reg=%eax
1010 /* Restore user state */
1011 RESTORE_REGS pop=4 # skip orig_eax/error_code
1014 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1015 * when returning from IPI handler and when returning from
1016 * scheduler to user-space.
1022 PARANOID_EXIT_TO_KERNEL_MODE
1027 .section .fixup, "ax"
1029 pushl $0 # no error code
1030 pushl $do_iret_error
1032 #ifdef CONFIG_DEBUG_ENTRY
1034 * The stack-frame here is the one that iret faulted on, so its a
1035 * return-to-user frame. We are on kernel-cr3 because we come here from
1036 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1037 * as the checker expects it.
1040 SWITCH_TO_USER_CR3 scratch_reg=%eax
1044 jmp common_exception
1046 _ASM_EXTABLE(.Lirq_return, iret_exc)
1047 ENDPROC(entry_INT80_32)
1049 .macro FIXUP_ESPFIX_STACK
1051 * Switch back for ESPFIX stack to the normal zerobased stack
1053 * We can't call C functions using the ESPFIX stack. This code reads
1054 * the high word of the segment base from the GDT and swiches to the
1055 * normal stack and adjusts ESP with the matching offset.
1057 #ifdef CONFIG_X86_ESPFIX32
1058 /* fixup the stack */
1059 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1060 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1062 addl %esp, %eax /* the adjusted stack pointer */
1065 lss (%esp), %esp /* switch to the normal stack segment */
1068 .macro UNWIND_ESPFIX_STACK
1069 #ifdef CONFIG_X86_ESPFIX32
1071 /* see if on espfix stack */
1072 cmpw $__ESPFIX_SS, %ax
1074 movl $__KERNEL_DS, %eax
1077 /* switch to normal stack */
1084 * Build the entry stubs with some assembler magic.
1085 * We pack 1 stub into every 8-byte block.
1088 ENTRY(irq_entries_start)
1089 vector=FIRST_EXTERNAL_VECTOR
1090 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1091 pushl $(~vector+0x80) /* Note: always in signed byte range */
1093 jmp common_interrupt
1096 END(irq_entries_start)
1099 * the CPU automatically disables interrupts when executing an IRQ vector,
1100 * so IRQ-flags tracing has to follow that:
1102 .p2align CONFIG_X86_L1_CACHE_SHIFT
1105 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1107 SAVE_ALL switch_stacks=1
1108 ENCODE_FRAME_POINTER
1113 ENDPROC(common_interrupt)
1115 #define BUILD_INTERRUPT3(name, nr, fn) \
1119 SAVE_ALL switch_stacks=1; \
1120 ENCODE_FRAME_POINTER; \
1124 jmp ret_from_intr; \
1127 #define BUILD_INTERRUPT(name, nr) \
1128 BUILD_INTERRUPT3(name, nr, smp_##name); \
1130 /* The include is where all of the SMP etc. interrupts come from */
1131 #include <asm/entry_arch.h>
1133 ENTRY(coprocessor_error)
1136 pushl $do_coprocessor_error
1137 jmp common_exception
1138 END(coprocessor_error)
1140 ENTRY(simd_coprocessor_error)
1143 #ifdef CONFIG_X86_INVD_BUG
1144 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1145 ALTERNATIVE "pushl $do_general_protection", \
1146 "pushl $do_simd_coprocessor_error", \
1149 pushl $do_simd_coprocessor_error
1151 jmp common_exception
1152 END(simd_coprocessor_error)
1154 ENTRY(device_not_available)
1156 pushl $-1 # mark this as an int
1157 pushl $do_device_not_available
1158 jmp common_exception
1159 END(device_not_available)
1161 #ifdef CONFIG_PARAVIRT
1164 _ASM_EXTABLE(native_iret, iret_exc)
1172 jmp common_exception
1179 jmp common_exception
1185 pushl $do_invalid_op
1186 jmp common_exception
1189 ENTRY(coprocessor_segment_overrun)
1192 pushl $do_coprocessor_segment_overrun
1193 jmp common_exception
1194 END(coprocessor_segment_overrun)
1198 pushl $do_invalid_TSS
1199 jmp common_exception
1202 ENTRY(segment_not_present)
1204 pushl $do_segment_not_present
1205 jmp common_exception
1206 END(segment_not_present)
1208 ENTRY(stack_segment)
1210 pushl $do_stack_segment
1211 jmp common_exception
1214 ENTRY(alignment_check)
1216 pushl $do_alignment_check
1217 jmp common_exception
1218 END(alignment_check)
1222 pushl $0 # no error code
1223 pushl $do_divide_error
1224 jmp common_exception
1227 #ifdef CONFIG_X86_MCE
1228 ENTRY(machine_check)
1231 pushl machine_check_vector
1232 jmp common_exception
1236 ENTRY(spurious_interrupt_bug)
1239 pushl $do_spurious_interrupt_bug
1240 jmp common_exception
1241 END(spurious_interrupt_bug)
1244 ENTRY(xen_hypervisor_callback)
1245 pushl $-1 /* orig_ax = -1 => not a system call */
1247 ENCODE_FRAME_POINTER
1251 * Check to see if we got the event in the critical
1252 * region in xen_iret_direct, after we've reenabled
1253 * events and checked for pending events. This simulates
1254 * iret instruction's behaviour where it delivers a
1255 * pending interrupt when enabling interrupts:
1257 movl PT_EIP(%esp), %eax
1258 cmpl $xen_iret_start_crit, %eax
1260 cmpl $xen_iret_end_crit, %eax
1263 jmp xen_iret_crit_fixup
1265 ENTRY(xen_do_upcall)
1267 call xen_evtchn_do_upcall
1268 #ifndef CONFIG_PREEMPT
1269 call xen_maybe_preempt_hcall
1272 ENDPROC(xen_hypervisor_callback)
1275 * Hypervisor uses this for application faults while it executes.
1276 * We get here for two reasons:
1277 * 1. Fault while reloading DS, ES, FS or GS
1278 * 2. Fault while executing IRET
1279 * Category 1 we fix up by reattempting the load, and zeroing the segment
1280 * register if the load fails.
1281 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1282 * normal Linux return path in this case because if we use the IRET hypercall
1283 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1284 * We distinguish between categories by maintaining a status value in EAX.
1286 ENTRY(xen_failsafe_callback)
1291 3: mov 12(%esp), %fs
1292 4: mov 16(%esp), %gs
1293 /* EAX == 0 => Category 1 (Bad segment)
1294 EAX != 0 => Category 2 (Bad IRET) */
1300 5: pushl $-1 /* orig_ax = -1 => not a system call */
1302 ENCODE_FRAME_POINTER
1303 jmp ret_from_exception
1305 .section .fixup, "ax"
1319 _ASM_EXTABLE(1b, 6b)
1320 _ASM_EXTABLE(2b, 7b)
1321 _ASM_EXTABLE(3b, 8b)
1322 _ASM_EXTABLE(4b, 9b)
1323 ENDPROC(xen_failsafe_callback)
1325 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1326 xen_evtchn_do_upcall)
1328 #endif /* CONFIG_XEN */
1330 #if IS_ENABLED(CONFIG_HYPERV)
1332 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1333 hyperv_vector_handler)
1335 BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1336 hyperv_reenlightenment_intr)
1338 BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1339 hv_stimer0_vector_handler)
1341 #endif /* CONFIG_HYPERV */
1345 pushl $do_page_fault
1347 jmp common_exception
1351 /* the function address is in %gs's slot on the stack */
1356 movl $(__USER_DS), %eax
1359 movl $(__KERNEL_PERCPU), %eax
1367 SWITCH_TO_KERNEL_STACK
1368 ENCODE_FRAME_POINTER
1372 movl PT_GS(%esp), %edi # get the function address
1373 movl PT_ORIG_EAX(%esp), %edx # get the error code
1374 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1378 movl %esp, %eax # pt_regs pointer
1380 jmp ret_from_exception
1381 END(common_exception)
1385 * Entry from sysenter is now handled in common_exception
1388 pushl $-1 # mark this as an int
1390 jmp common_exception
1394 * NMI is doubly nasty. It can happen on the first instruction of
1395 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1396 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1397 * switched stacks. We handle both conditions by simply checking whether we
1398 * interrupted kernel code running on the SYSENTER stack.
1403 #ifdef CONFIG_X86_ESPFIX32
1406 cmpw $__ESPFIX_SS, %ax
1408 je .Lnmi_espfix_stack
1411 pushl %eax # pt_regs->orig_ax
1412 SAVE_ALL_NMI cr3_reg=%edi
1413 ENCODE_FRAME_POINTER
1414 xorl %edx, %edx # zero error code
1415 movl %esp, %eax # pt_regs pointer
1417 /* Are we currently on the SYSENTER stack? */
1418 movl PER_CPU_VAR(cpu_entry_area), %ecx
1419 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1420 subl %eax, %ecx /* ecx = (end of entry_stack) - esp */
1421 cmpl $SIZEOF_entry_stack, %ecx
1422 jb .Lnmi_from_sysenter_stack
1424 /* Not on SYSENTER stack. */
1428 .Lnmi_from_sysenter_stack:
1430 * We're on the SYSENTER stack. Switch off. No one (not even debug)
1431 * is using the thread stack right now, so it's safe for us to use it.
1434 movl PER_CPU_VAR(cpu_current_top_of_stack), %esp
1439 CHECK_AND_APPLY_ESPFIX
1440 RESTORE_ALL_NMI cr3_reg=%edi pop=4
1443 #ifdef CONFIG_X86_ESPFIX32
1446 * create the pointer to lss back
1451 /* copy the iret frame of 12 bytes */
1456 SAVE_ALL_NMI cr3_reg=%edi
1457 ENCODE_FRAME_POINTER
1458 FIXUP_ESPFIX_STACK # %eax == %esp
1459 xorl %edx, %edx # zero error code
1461 RESTORE_ALL_NMI cr3_reg=%edi
1462 lss 12+4(%esp), %esp # back to espfix stack
1469 pushl $-1 # mark this as an int
1471 SAVE_ALL switch_stacks=1
1472 ENCODE_FRAME_POINTER
1474 xorl %edx, %edx # zero error code
1475 movl %esp, %eax # pt_regs pointer
1477 jmp ret_from_exception
1480 ENTRY(general_protection)
1481 pushl $do_general_protection
1482 jmp common_exception
1483 END(general_protection)
1485 #ifdef CONFIG_KVM_GUEST
1486 ENTRY(async_page_fault)
1488 pushl $do_async_page_fault
1489 jmp common_exception
1490 END(async_page_fault)
1493 ENTRY(rewind_stack_do_exit)
1494 /* Prevent any naive code from trying to unwind to our caller. */
1497 movl PER_CPU_VAR(cpu_current_top_of_stack), %esi
1498 leal -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1502 END(rewind_stack_do_exit)