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
248 .macro RESTORE_INT_REGS
258 .macro RESTORE_REGS pop=0
264 .pushsection .fixup, "ax"
278 .macro RESTORE_ALL_NMI cr3_reg:req pop=0
280 * Now switch the CR3 when PTI is enabled.
282 * We enter with kernel cr3 and switch the cr3 to the value
283 * stored on \cr3_reg, which is either a user or a kernel cr3.
285 ALTERNATIVE "jmp .Lswitched_\@", "", X86_FEATURE_PTI
287 testl $PTI_SWITCH_MASK, \cr3_reg
290 /* User cr3 in \cr3_reg - write it to hardware cr3 */
297 RESTORE_REGS pop=\pop
300 .macro CHECK_AND_APPLY_ESPFIX
301 #ifdef CONFIG_X86_ESPFIX32
302 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
304 ALTERNATIVE "jmp .Lend_\@", "", X86_BUG_ESPFIX
306 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
308 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
309 * are returning to the kernel.
310 * See comments in process.c:copy_thread() for details.
312 movb PT_OLDSS(%esp), %ah
313 movb PT_CS(%esp), %al
314 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
315 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
316 jne .Lend_\@ # returning to user-space with LDT SS
319 * Setup and switch to ESPFIX stack
321 * We're returning to userspace with a 16 bit stack. The CPU will not
322 * restore the high word of ESP for us on executing iret... This is an
323 * "official" bug of all the x86-compatible CPUs, which we can work
324 * around to make dosemu and wine happy. We do this by preloading the
325 * high word of ESP with the high word of the userspace ESP while
326 * compensating for the offset by changing to the ESPFIX segment with
327 * a base address that matches for the difference.
329 mov %esp, %edx /* load kernel esp */
330 mov PT_OLDESP(%esp), %eax /* load userspace esp */
331 mov %dx, %ax /* eax: new kernel esp */
332 sub %eax, %edx /* offset (low word is 0) */
334 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
335 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
337 pushl %eax /* new kernel esp */
339 * Disable interrupts, but do not irqtrace this section: we
340 * will soon execute iret and the tracer was already set to
341 * the irqstate after the IRET:
343 DISABLE_INTERRUPTS(CLBR_ANY)
344 lss (%esp), %esp /* switch to espfix segment */
346 #endif /* CONFIG_X86_ESPFIX32 */
350 * Called with pt_regs fully populated and kernel segments loaded,
351 * so we can access PER_CPU and use the integer registers.
353 * We need to be very careful here with the %esp switch, because an NMI
354 * can happen everywhere. If the NMI handler finds itself on the
355 * entry-stack, it will overwrite the task-stack and everything we
356 * copied there. So allocate the stack-frame on the task-stack and
357 * switch to it before we do any copying.
360 #define CS_FROM_ENTRY_STACK (1 << 31)
361 #define CS_FROM_USER_CR3 (1 << 30)
363 .macro SWITCH_TO_KERNEL_STACK
365 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
369 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
372 * %eax now contains the entry cr3 and we carry it forward in
373 * that register for the time this macro runs
377 * The high bits of the CS dword (__csh) are used for
378 * CS_FROM_ENTRY_STACK and CS_FROM_USER_CR3. Clear them in case
379 * hardware didn't do this for us.
381 andl $(0x0000ffff), PT_CS(%esp)
383 /* Are we on the entry stack? Bail out if not! */
384 movl PER_CPU_VAR(cpu_entry_area), %ecx
385 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
386 subl %esp, %ecx /* ecx = (end of entry_stack) - esp */
387 cmpl $SIZEOF_entry_stack, %ecx
390 /* Load stack pointer into %esi and %edi */
394 /* Move %edi to the top of the entry stack */
395 andl $(MASK_entry_stack), %edi
396 addl $(SIZEOF_entry_stack), %edi
398 /* Load top of task-stack into %edi */
399 movl TSS_entry2task_stack(%edi), %edi
401 /* Special case - entry from kernel mode via entry stack */
403 movl PT_EFLAGS(%esp), %ecx # mix EFLAGS and CS
404 movb PT_CS(%esp), %cl
405 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
407 movl PT_CS(%esp), %ecx
408 andl $SEGMENT_RPL_MASK, %ecx
411 jb .Lentry_from_kernel_\@
414 movl $PTREGS_SIZE, %ecx
417 testl $X86_EFLAGS_VM, PT_EFLAGS(%esi)
421 * Stack-frame contains 4 additional segment registers when
422 * coming from VM86 mode
429 /* Allocate frame on task-stack */
432 /* Switch to task-stack */
436 * We are now on the task-stack and can safely copy over the
445 .Lentry_from_kernel_\@:
448 * This handles the case when we enter the kernel from
449 * kernel-mode and %esp points to the entry-stack. When this
450 * happens we need to switch to the task-stack to run C code,
451 * but switch back to the entry-stack again when we approach
452 * iret and return to the interrupted code-path. This usually
453 * happens when we hit an exception while restoring user-space
454 * segment registers on the way back to user-space or when the
455 * sysenter handler runs with eflags.tf set.
457 * When we switch to the task-stack here, we can't trust the
458 * contents of the entry-stack anymore, as the exception handler
459 * might be scheduled out or moved to another CPU. Therefore we
460 * copy the complete entry-stack to the task-stack and set a
461 * marker in the iret-frame (bit 31 of the CS dword) to detect
462 * what we've done on the iret path.
464 * On the iret path we copy everything back and switch to the
465 * entry-stack, so that the interrupted kernel code-path
466 * continues on the same stack it was interrupted with.
468 * Be aware that an NMI can happen anytime in this code.
470 * %esi: Entry-Stack pointer (same as %esp)
471 * %edi: Top of the task stack
472 * %eax: CR3 on kernel entry
475 /* Calculate number of bytes on the entry stack in %ecx */
478 /* %ecx to the top of entry-stack */
479 andl $(MASK_entry_stack), %ecx
480 addl $(SIZEOF_entry_stack), %ecx
482 /* Number of bytes on the entry stack to %ecx */
485 /* Mark stackframe as coming from entry stack */
486 orl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
489 * Test the cr3 used to enter the kernel and add a marker
490 * so that we can switch back to it before iret.
492 testl $PTI_SWITCH_MASK, %eax
494 orl $CS_FROM_USER_CR3, PT_CS(%esp)
497 * %esi and %edi are unchanged, %ecx contains the number of
498 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
499 * the stack-frame on task-stack and copy everything over
501 jmp .Lcopy_pt_regs_\@
507 * Switch back from the kernel stack to the entry stack.
509 * The %esp register must point to pt_regs on the task stack. It will
510 * first calculate the size of the stack-frame to copy, depending on
511 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
512 * to copy the contents of the stack over to the entry stack.
514 * We must be very careful here, as we can't trust the contents of the
515 * task-stack once we switched to the entry-stack. When an NMI happens
516 * while on the entry-stack, the NMI handler will switch back to the top
517 * of the task stack, overwriting our stack-frame we are about to copy.
518 * Therefore we switch the stack only after everything is copied over.
520 .macro SWITCH_TO_ENTRY_STACK
522 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
525 movl $PTREGS_SIZE, %ecx
528 testl $(X86_EFLAGS_VM), PT_EFLAGS(%esp)
531 /* Additional 4 registers to copy when returning to VM86 mode */
537 /* Initialize source and destination for movsl */
538 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
542 /* Save future stack pointer in %ebx */
545 /* Copy over the stack-frame */
551 * Switch to entry-stack - needs to happen after everything is
552 * copied because the NMI handler will overwrite the task-stack
553 * when on entry-stack
561 * This macro handles the case when we return to kernel-mode on the iret
562 * path and have to switch back to the entry stack and/or user-cr3
564 * See the comments below the .Lentry_from_kernel_\@ label in the
565 * SWITCH_TO_KERNEL_STACK macro for more details.
567 .macro PARANOID_EXIT_TO_KERNEL_MODE
570 * Test if we entered the kernel with the entry-stack. Most
571 * likely we did not, because this code only runs on the
572 * return-to-kernel path.
574 testl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
577 /* Unlikely slow-path */
579 /* Clear marker from stack-frame */
580 andl $(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
582 /* Copy the remaining task-stack contents to entry-stack */
584 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
586 /* Bytes on the task-stack to ecx */
587 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
590 /* Allocate stack-frame on entry-stack */
594 * Save future stack-pointer, we must not switch until the
595 * copy is done, otherwise the NMI handler could destroy the
596 * contents of the task-stack we are about to copy.
605 /* Safe to switch to entry-stack now */
609 * We came from entry-stack and need to check if we also need to
610 * switch back to user cr3.
612 testl $CS_FROM_USER_CR3, PT_CS(%esp)
615 /* Clear marker from stack-frame */
616 andl $(~CS_FROM_USER_CR3), PT_CS(%esp)
618 SWITCH_TO_USER_CR3 scratch_reg=%eax
626 ENTRY(__switch_to_asm)
628 * Save callee-saved registers
629 * This must match the order in struct inactive_task_frame
638 movl %esp, TASK_threadsp(%eax)
639 movl TASK_threadsp(%edx), %esp
641 #ifdef CONFIG_STACKPROTECTOR
642 movl TASK_stack_canary(%edx), %ebx
643 movl %ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
647 * When switching from a shallower to a deeper call stack
648 * the RSB may either underflow or use entries populated
649 * with userspace addresses. On CPUs where those concerns
650 * exist, overwrite the RSB with entries which capture
651 * speculative execution to prevent attack.
653 FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
655 /* restore callee-saved registers */
666 * The unwinder expects the last frame on the stack to always be at the same
667 * offset from the end of the page, which allows it to validate the stack.
668 * Calling schedule_tail() directly would break that convention because its an
669 * asmlinkage function so its argument has to be pushed on the stack. This
670 * wrapper creates a proper "end of stack" frame header before the call.
672 ENTRY(schedule_tail_wrapper)
681 ENDPROC(schedule_tail_wrapper)
683 * A newly forked process directly context switches into this address.
685 * eax: prev task we switched from
686 * ebx: kernel thread func (NULL for user thread)
687 * edi: kernel thread arg
690 call schedule_tail_wrapper
693 jnz 1f /* kernel threads are uncommon */
696 /* When we fork, we trace the syscall return in the child, too. */
698 call syscall_return_slowpath
705 * A kernel thread is allowed to return here after successfully
706 * calling do_execve(). Exit to userspace to complete the execve()
709 movl $0, PT_EAX(%esp)
714 * Return to user mode is not as complex as all this looks,
715 * but we want the default path for a system call return to
716 * go as quickly as possible which is why some of this is
717 * less clear than it otherwise should be.
720 # userspace resumption stub bypassing syscall exit tracing
723 preempt_stop(CLBR_ANY)
726 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
727 movb PT_CS(%esp), %al
728 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
731 * We can be coming here from child spawned by kernel_thread().
733 movl PT_CS(%esp), %eax
734 andl $SEGMENT_RPL_MASK, %eax
737 jb resume_kernel # not returning to v8086 or userspace
739 ENTRY(resume_userspace)
740 DISABLE_INTERRUPTS(CLBR_ANY)
743 call prepare_exit_to_usermode
745 END(ret_from_exception)
747 #ifdef CONFIG_PREEMPT
749 DISABLE_INTERRUPTS(CLBR_ANY)
751 cmpl $0, PER_CPU_VAR(__preempt_count)
752 jnz restore_all_kernel
753 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
754 jz restore_all_kernel
755 call preempt_schedule_irq
760 GLOBAL(__begin_SYSENTER_singlestep_region)
762 * All code from here through __end_SYSENTER_singlestep_region is subject
763 * to being single-stepped if a user program sets TF and executes SYSENTER.
764 * There is absolutely nothing that we can do to prevent this from happening
765 * (thanks Intel!). To keep our handling of this situation as simple as
766 * possible, we handle TF just like AC and NT, except that our #DB handler
767 * will ignore all of the single-step traps generated in this range.
772 * Xen doesn't set %esp to be precisely what the normal SYSENTER
773 * entry point expects, so fix it up before using the normal path.
775 ENTRY(xen_sysenter_target)
776 addl $5*4, %esp /* remove xen-provided frame */
777 jmp .Lsysenter_past_esp
781 * 32-bit SYSENTER entry.
783 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
784 * if X86_FEATURE_SEP is available. This is the preferred system call
785 * entry on 32-bit systems.
787 * The SYSENTER instruction, in principle, should *only* occur in the
788 * vDSO. In practice, a small number of Android devices were shipped
789 * with a copy of Bionic that inlined a SYSENTER instruction. This
790 * never happened in any of Google's Bionic versions -- it only happened
791 * in a narrow range of Intel-provided versions.
793 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
794 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
795 * SYSENTER does not save anything on the stack,
796 * and does not save old EIP (!!!), ESP, or EFLAGS.
798 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
799 * user and/or vm86 state), we explicitly disable the SYSENTER
800 * instruction in vm86 mode by reprogramming the MSRs.
803 * eax system call number
812 ENTRY(entry_SYSENTER_32)
814 * On entry-stack with all userspace-regs live - save and
815 * restore eflags and %eax to use it as scratch-reg for the cr3
820 BUG_IF_WRONG_CR3 no_user_check=1
821 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
825 /* Stack empty again, switch to task stack */
826 movl TSS_entry2task_stack(%esp), %esp
829 pushl $__USER_DS /* pt_regs->ss */
830 pushl %ebp /* pt_regs->sp (stashed in bp) */
831 pushfl /* pt_regs->flags (except IF = 0) */
832 orl $X86_EFLAGS_IF, (%esp) /* Fix IF */
833 pushl $__USER_CS /* pt_regs->cs */
834 pushl $0 /* pt_regs->ip = 0 (placeholder) */
835 pushl %eax /* pt_regs->orig_ax */
836 SAVE_ALL pt_regs_ax=$-ENOSYS /* save rest, stack already switched */
839 * SYSENTER doesn't filter flags, so we need to clear NT, AC
840 * and TF ourselves. To save a few cycles, we can check whether
841 * either was set instead of doing an unconditional popfq.
842 * This needs to happen before enabling interrupts so that
843 * we don't get preempted with NT set.
845 * If TF is set, we will single-step all the way to here -- do_debug
846 * will ignore all the traps. (Yes, this is slow, but so is
847 * single-stepping in general. This allows us to avoid having
848 * a more complicated code to handle the case where a user program
849 * forces us to single-step through the SYSENTER entry code.)
851 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
852 * out-of-line as an optimization: NT is unlikely to be set in the
853 * majority of the cases and instead of polluting the I$ unnecessarily,
854 * we're keeping that code behind a branch which will predict as
855 * not-taken and therefore its instructions won't be fetched.
857 testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
858 jnz .Lsysenter_fix_flags
859 .Lsysenter_flags_fixed:
862 * User mode is traced as though IRQs are on, and SYSENTER
868 call do_fast_syscall_32
869 /* XEN PV guests always use IRET path */
870 ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
871 "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
873 /* Opportunistic SYSEXIT */
874 TRACE_IRQS_ON /* User mode traces as IRQs on. */
877 * Setup entry stack - we keep the pointer in %eax and do the
878 * switch after almost all user-state is restored.
881 /* Load entry stack pointer and allocate frame for eflags/eax */
882 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
885 /* Copy eflags and eax to entry stack */
886 movl PT_EFLAGS(%esp), %edi
887 movl PT_EAX(%esp), %esi
891 /* Restore user registers and segments */
892 movl PT_EIP(%esp), %edx /* pt_regs->ip */
893 movl PT_OLDESP(%esp), %ecx /* pt_regs->sp */
894 1: mov PT_FS(%esp), %fs
897 popl %ebx /* pt_regs->bx */
898 addl $2*4, %esp /* skip pt_regs->cx and pt_regs->dx */
899 popl %esi /* pt_regs->si */
900 popl %edi /* pt_regs->di */
901 popl %ebp /* pt_regs->bp */
903 /* Switch to entry stack */
906 /* Now ready to switch the cr3 */
907 SWITCH_TO_USER_CR3 scratch_reg=%eax
910 * Restore all flags except IF. (We restore IF separately because
911 * STI gives a one-instruction window in which we won't be interrupted,
912 * whereas POPF does not.)
914 btrl $X86_EFLAGS_IF_BIT, (%esp)
915 BUG_IF_WRONG_CR3 no_user_check=1
920 * Return back to the vDSO, which will pop ecx and edx.
921 * Don't bother with DS and ES (they already contain __USER_DS).
926 .pushsection .fixup, "ax"
927 2: movl $0, PT_FS(%esp)
933 .Lsysenter_fix_flags:
934 pushl $X86_EFLAGS_FIXED
936 jmp .Lsysenter_flags_fixed
937 GLOBAL(__end_SYSENTER_singlestep_region)
938 ENDPROC(entry_SYSENTER_32)
941 * 32-bit legacy system call entry.
943 * 32-bit x86 Linux system calls traditionally used the INT $0x80
944 * instruction. INT $0x80 lands here.
946 * This entry point can be used by any 32-bit perform system calls.
947 * Instances of INT $0x80 can be found inline in various programs and
948 * libraries. It is also used by the vDSO's __kernel_vsyscall
949 * fallback for hardware that doesn't support a faster entry method.
950 * Restarted 32-bit system calls also fall back to INT $0x80
951 * regardless of what instruction was originally used to do the system
952 * call. (64-bit programs can use INT $0x80 as well, but they can
953 * only run on 64-bit kernels and therefore land in
954 * entry_INT80_compat.)
956 * This is considered a slow path. It is not used by most libc
957 * implementations on modern hardware except during process startup.
960 * eax system call number
968 ENTRY(entry_INT80_32)
970 pushl %eax /* pt_regs->orig_ax */
972 SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1 /* save rest */
975 * User mode is traced as though IRQs are on, and the interrupt gate
981 call do_int80_syscall_32
986 SWITCH_TO_ENTRY_STACK
987 .Lrestore_all_notrace:
988 CHECK_AND_APPLY_ESPFIX
990 /* Switch back to user CR3 */
991 SWITCH_TO_USER_CR3 scratch_reg=%eax
995 /* Restore user state */
996 RESTORE_REGS pop=4 # skip orig_eax/error_code
999 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1000 * when returning from IPI handler and when returning from
1001 * scheduler to user-space.
1007 PARANOID_EXIT_TO_KERNEL_MODE
1012 .section .fixup, "ax"
1014 pushl $0 # no error code
1015 pushl $do_iret_error
1017 #ifdef CONFIG_DEBUG_ENTRY
1019 * The stack-frame here is the one that iret faulted on, so its a
1020 * return-to-user frame. We are on kernel-cr3 because we come here from
1021 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1022 * as the checker expects it.
1025 SWITCH_TO_USER_CR3 scratch_reg=%eax
1029 jmp common_exception
1031 _ASM_EXTABLE(.Lirq_return, iret_exc)
1032 ENDPROC(entry_INT80_32)
1034 .macro FIXUP_ESPFIX_STACK
1036 * Switch back for ESPFIX stack to the normal zerobased stack
1038 * We can't call C functions using the ESPFIX stack. This code reads
1039 * the high word of the segment base from the GDT and swiches to the
1040 * normal stack and adjusts ESP with the matching offset.
1042 #ifdef CONFIG_X86_ESPFIX32
1043 /* fixup the stack */
1044 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1045 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1047 addl %esp, %eax /* the adjusted stack pointer */
1050 lss (%esp), %esp /* switch to the normal stack segment */
1053 .macro UNWIND_ESPFIX_STACK
1054 #ifdef CONFIG_X86_ESPFIX32
1056 /* see if on espfix stack */
1057 cmpw $__ESPFIX_SS, %ax
1059 movl $__KERNEL_DS, %eax
1062 /* switch to normal stack */
1069 * Build the entry stubs with some assembler magic.
1070 * We pack 1 stub into every 8-byte block.
1073 ENTRY(irq_entries_start)
1074 vector=FIRST_EXTERNAL_VECTOR
1075 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1076 pushl $(~vector+0x80) /* Note: always in signed byte range */
1078 jmp common_interrupt
1081 END(irq_entries_start)
1083 #ifdef CONFIG_X86_LOCAL_APIC
1085 ENTRY(spurious_entries_start)
1086 vector=FIRST_SYSTEM_VECTOR
1087 .rept (NR_VECTORS - FIRST_SYSTEM_VECTOR)
1088 pushl $(~vector+0x80) /* Note: always in signed byte range */
1093 END(spurious_entries_start)
1097 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1098 SAVE_ALL switch_stacks=1
1099 ENCODE_FRAME_POINTER
1102 call smp_spurious_interrupt
1104 ENDPROC(common_spurious)
1108 * the CPU automatically disables interrupts when executing an IRQ vector,
1109 * so IRQ-flags tracing has to follow that:
1111 .p2align CONFIG_X86_L1_CACHE_SHIFT
1114 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1116 SAVE_ALL switch_stacks=1
1117 ENCODE_FRAME_POINTER
1122 ENDPROC(common_interrupt)
1124 #define BUILD_INTERRUPT3(name, nr, fn) \
1128 SAVE_ALL switch_stacks=1; \
1129 ENCODE_FRAME_POINTER; \
1133 jmp ret_from_intr; \
1136 #define BUILD_INTERRUPT(name, nr) \
1137 BUILD_INTERRUPT3(name, nr, smp_##name); \
1139 /* The include is where all of the SMP etc. interrupts come from */
1140 #include <asm/entry_arch.h>
1142 ENTRY(coprocessor_error)
1145 pushl $do_coprocessor_error
1146 jmp common_exception
1147 END(coprocessor_error)
1149 ENTRY(simd_coprocessor_error)
1152 #ifdef CONFIG_X86_INVD_BUG
1153 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1154 ALTERNATIVE "pushl $do_general_protection", \
1155 "pushl $do_simd_coprocessor_error", \
1158 pushl $do_simd_coprocessor_error
1160 jmp common_exception
1161 END(simd_coprocessor_error)
1163 ENTRY(device_not_available)
1165 pushl $-1 # mark this as an int
1166 pushl $do_device_not_available
1167 jmp common_exception
1168 END(device_not_available)
1170 #ifdef CONFIG_PARAVIRT
1173 _ASM_EXTABLE(native_iret, iret_exc)
1181 jmp common_exception
1188 jmp common_exception
1194 pushl $do_invalid_op
1195 jmp common_exception
1198 ENTRY(coprocessor_segment_overrun)
1201 pushl $do_coprocessor_segment_overrun
1202 jmp common_exception
1203 END(coprocessor_segment_overrun)
1207 pushl $do_invalid_TSS
1208 jmp common_exception
1211 ENTRY(segment_not_present)
1213 pushl $do_segment_not_present
1214 jmp common_exception
1215 END(segment_not_present)
1217 ENTRY(stack_segment)
1219 pushl $do_stack_segment
1220 jmp common_exception
1223 ENTRY(alignment_check)
1225 pushl $do_alignment_check
1226 jmp common_exception
1227 END(alignment_check)
1231 pushl $0 # no error code
1232 pushl $do_divide_error
1233 jmp common_exception
1236 #ifdef CONFIG_X86_MCE
1237 ENTRY(machine_check)
1240 pushl machine_check_vector
1241 jmp common_exception
1245 ENTRY(spurious_interrupt_bug)
1248 pushl $do_spurious_interrupt_bug
1249 jmp common_exception
1250 END(spurious_interrupt_bug)
1253 ENTRY(xen_hypervisor_callback)
1254 pushl $-1 /* orig_ax = -1 => not a system call */
1256 ENCODE_FRAME_POINTER
1260 * Check to see if we got the event in the critical
1261 * region in xen_iret_direct, after we've reenabled
1262 * events and checked for pending events. This simulates
1263 * iret instruction's behaviour where it delivers a
1264 * pending interrupt when enabling interrupts:
1266 movl PT_EIP(%esp), %eax
1267 cmpl $xen_iret_start_crit, %eax
1269 cmpl $xen_iret_end_crit, %eax
1272 jmp xen_iret_crit_fixup
1274 ENTRY(xen_do_upcall)
1276 call xen_evtchn_do_upcall
1277 #ifndef CONFIG_PREEMPT
1278 call xen_maybe_preempt_hcall
1281 ENDPROC(xen_hypervisor_callback)
1284 * Hypervisor uses this for application faults while it executes.
1285 * We get here for two reasons:
1286 * 1. Fault while reloading DS, ES, FS or GS
1287 * 2. Fault while executing IRET
1288 * Category 1 we fix up by reattempting the load, and zeroing the segment
1289 * register if the load fails.
1290 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1291 * normal Linux return path in this case because if we use the IRET hypercall
1292 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1293 * We distinguish between categories by maintaining a status value in EAX.
1295 ENTRY(xen_failsafe_callback)
1300 3: mov 12(%esp), %fs
1301 4: mov 16(%esp), %gs
1302 /* EAX == 0 => Category 1 (Bad segment)
1303 EAX != 0 => Category 2 (Bad IRET) */
1309 5: pushl $-1 /* orig_ax = -1 => not a system call */
1311 ENCODE_FRAME_POINTER
1312 jmp ret_from_exception
1314 .section .fixup, "ax"
1328 _ASM_EXTABLE(1b, 6b)
1329 _ASM_EXTABLE(2b, 7b)
1330 _ASM_EXTABLE(3b, 8b)
1331 _ASM_EXTABLE(4b, 9b)
1332 ENDPROC(xen_failsafe_callback)
1334 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1335 xen_evtchn_do_upcall)
1337 #endif /* CONFIG_XEN */
1339 #if IS_ENABLED(CONFIG_HYPERV)
1341 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1342 hyperv_vector_handler)
1344 BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1345 hyperv_reenlightenment_intr)
1347 BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1348 hv_stimer0_vector_handler)
1350 #endif /* CONFIG_HYPERV */
1354 pushl $do_page_fault
1356 jmp common_exception
1360 /* the function address is in %gs's slot on the stack */
1365 movl $(__USER_DS), %eax
1368 movl $(__KERNEL_PERCPU), %eax
1376 SWITCH_TO_KERNEL_STACK
1377 ENCODE_FRAME_POINTER
1381 movl PT_GS(%esp), %edi # get the function address
1382 movl PT_ORIG_EAX(%esp), %edx # get the error code
1383 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1387 movl %esp, %eax # pt_regs pointer
1389 jmp ret_from_exception
1390 END(common_exception)
1394 * Entry from sysenter is now handled in common_exception
1397 pushl $-1 # mark this as an int
1399 jmp common_exception
1403 * NMI is doubly nasty. It can happen on the first instruction of
1404 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1405 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1406 * switched stacks. We handle both conditions by simply checking whether we
1407 * interrupted kernel code running on the SYSENTER stack.
1412 #ifdef CONFIG_X86_ESPFIX32
1415 cmpw $__ESPFIX_SS, %ax
1417 je .Lnmi_espfix_stack
1420 pushl %eax # pt_regs->orig_ax
1421 SAVE_ALL_NMI cr3_reg=%edi
1422 ENCODE_FRAME_POINTER
1423 xorl %edx, %edx # zero error code
1424 movl %esp, %eax # pt_regs pointer
1426 /* Are we currently on the SYSENTER stack? */
1427 movl PER_CPU_VAR(cpu_entry_area), %ecx
1428 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1429 subl %eax, %ecx /* ecx = (end of entry_stack) - esp */
1430 cmpl $SIZEOF_entry_stack, %ecx
1431 jb .Lnmi_from_sysenter_stack
1433 /* Not on SYSENTER stack. */
1437 .Lnmi_from_sysenter_stack:
1439 * We're on the SYSENTER stack. Switch off. No one (not even debug)
1440 * is using the thread stack right now, so it's safe for us to use it.
1443 movl PER_CPU_VAR(cpu_current_top_of_stack), %esp
1448 CHECK_AND_APPLY_ESPFIX
1449 RESTORE_ALL_NMI cr3_reg=%edi pop=4
1452 #ifdef CONFIG_X86_ESPFIX32
1455 * create the pointer to lss back
1460 /* copy the iret frame of 12 bytes */
1465 SAVE_ALL_NMI cr3_reg=%edi
1466 ENCODE_FRAME_POINTER
1467 FIXUP_ESPFIX_STACK # %eax == %esp
1468 xorl %edx, %edx # zero error code
1470 RESTORE_ALL_NMI cr3_reg=%edi
1471 lss 12+4(%esp), %esp # back to espfix stack
1478 pushl $-1 # mark this as an int
1480 SAVE_ALL switch_stacks=1
1481 ENCODE_FRAME_POINTER
1483 xorl %edx, %edx # zero error code
1484 movl %esp, %eax # pt_regs pointer
1486 jmp ret_from_exception
1489 ENTRY(general_protection)
1491 pushl $do_general_protection
1492 jmp common_exception
1493 END(general_protection)
1495 #ifdef CONFIG_KVM_GUEST
1496 ENTRY(async_page_fault)
1498 pushl $do_async_page_fault
1499 jmp common_exception
1500 END(async_page_fault)
1503 ENTRY(rewind_stack_and_make_dead)
1504 /* Prevent any naive code from trying to unwind to our caller. */
1507 movl PER_CPU_VAR(cpu_current_top_of_stack), %esi
1508 leal -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1512 END(rewind_stack_and_make_dead)