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>
51 .section .entry.text, "ax"
54 * We use macros for low-level operations which need to be overridden
55 * for paravirtualization. The following will never clobber any registers:
56 * INTERRUPT_RETURN (aka. "iret")
57 * GET_CR0_INTO_EAX (aka. "movl %cr0, %eax")
58 * ENABLE_INTERRUPTS_SYSEXIT (aka "sti; sysexit").
60 * For DISABLE_INTERRUPTS/ENABLE_INTERRUPTS (aka "cli"/"sti"), you must
61 * specify what registers can be overwritten (CLBR_NONE, CLBR_EAX/EDX/ECX/ANY).
62 * Allowing a register to be clobbered can shrink the paravirt replacement
63 * enough to patch inline, increasing performance.
66 #ifdef CONFIG_PREEMPTION
67 # define preempt_stop(clobbers) DISABLE_INTERRUPTS(clobbers); TRACE_IRQS_OFF
69 # define preempt_stop(clobbers)
72 .macro TRACE_IRQS_IRET
73 #ifdef CONFIG_TRACE_IRQFLAGS
74 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off?
81 #define PTI_SWITCH_MASK (1 << PAGE_SHIFT)
84 * User gs save/restore
86 * %gs is used for userland TLS and kernel only uses it for stack
87 * canary which is required to be at %gs:20 by gcc. Read the comment
88 * at the top of stackprotector.h for more info.
90 * Local labels 98 and 99 are used.
92 #ifdef CONFIG_X86_32_LAZY_GS
94 /* unfortunately push/pop can't be no-op */
99 addl $(4 + \pop), %esp
104 /* all the rest are no-op */
111 .macro REG_TO_PTGS reg
113 .macro SET_KERNEL_GS reg
116 #else /* CONFIG_X86_32_LAZY_GS */
129 .pushsection .fixup, "ax"
133 _ASM_EXTABLE(98b, 99b)
137 98: mov PT_GS(%esp), %gs
140 .pushsection .fixup, "ax"
141 99: movl $0, PT_GS(%esp)
144 _ASM_EXTABLE(98b, 99b)
150 .macro REG_TO_PTGS reg
151 movl \reg, PT_GS(%esp)
153 .macro SET_KERNEL_GS reg
154 movl $(__KERNEL_STACK_CANARY), \reg
158 #endif /* CONFIG_X86_32_LAZY_GS */
160 /* Unconditionally switch to user cr3 */
161 .macro SWITCH_TO_USER_CR3 scratch_reg:req
162 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
164 movl %cr3, \scratch_reg
165 orl $PTI_SWITCH_MASK, \scratch_reg
166 movl \scratch_reg, %cr3
170 .macro BUG_IF_WRONG_CR3 no_user_check=0
171 #ifdef CONFIG_DEBUG_ENTRY
172 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
173 .if \no_user_check == 0
174 /* coming from usermode? */
175 testl $SEGMENT_RPL_MASK, PT_CS(%esp)
180 testl $PTI_SWITCH_MASK, %eax
182 /* From userspace with kernel cr3 - BUG */
189 * Switch to kernel cr3 if not already loaded and return current cr3 in
192 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req
193 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI
194 movl %cr3, \scratch_reg
195 /* Test if we are already on kernel CR3 */
196 testl $PTI_SWITCH_MASK, \scratch_reg
198 andl $(~PTI_SWITCH_MASK), \scratch_reg
199 movl \scratch_reg, %cr3
200 /* Return original CR3 in \scratch_reg */
201 orl $PTI_SWITCH_MASK, \scratch_reg
205 #define CS_FROM_ENTRY_STACK (1 << 31)
206 #define CS_FROM_USER_CR3 (1 << 30)
207 #define CS_FROM_KERNEL (1 << 29)
211 * The high bits of the CS dword (__csh) are used for CS_FROM_*.
212 * Clear them in case hardware didn't do this for us.
214 andl $0x0000ffff, 3*4(%esp)
217 testl $X86_EFLAGS_VM, 4*4(%esp)
218 jnz .Lfrom_usermode_no_fixup_\@
220 testl $SEGMENT_RPL_MASK, 3*4(%esp)
221 jnz .Lfrom_usermode_no_fixup_\@
223 orl $CS_FROM_KERNEL, 3*4(%esp)
226 * When we're here from kernel mode; the (exception) stack looks like:
228 * 5*4(%esp) - <previous context>
232 * 1*4(%esp) - orig_eax
233 * 0*4(%esp) - gs / function
235 * Lets build a 5 entry IRET frame after that, such that struct pt_regs
236 * is complete and in particular regs->sp is correct. This gives us
237 * the original 5 enties as gap:
239 * 12*4(%esp) - <previous context>
240 * 11*4(%esp) - gap / flags
241 * 10*4(%esp) - gap / cs
242 * 9*4(%esp) - gap / ip
243 * 8*4(%esp) - gap / orig_eax
244 * 7*4(%esp) - gap / gs / function
250 * 1*4(%esp) - orig_eax
251 * 0*4(%esp) - gs / function
255 pushl %esp # sp (points at ss)
256 addl $6*4, (%esp) # point sp back at the previous context
257 pushl 6*4(%esp) # flags
260 pushl 6*4(%esp) # orig_eax
261 pushl 6*4(%esp) # gs / function
262 .Lfrom_usermode_no_fixup_\@:
266 testl $CS_FROM_KERNEL, 1*4(%esp)
267 jz .Lfinished_frame_\@
270 * Reconstruct the 3 entry IRET frame right after the (modified)
271 * regs->sp without lowering %esp in between, such that an NMI in the
272 * middle doesn't scribble our stack.
276 movl 5*4(%esp), %eax # (modified) regs->sp
278 movl 4*4(%esp), %ecx # flags
281 movl 3*4(%esp), %ecx # cs
282 andl $0x0000ffff, %ecx
285 movl 2*4(%esp), %ecx # ip
288 movl 1*4(%esp), %ecx # eax
297 .macro SAVE_ALL pt_regs_ax=%eax switch_stacks=0 skip_gs=0
313 movl $(__USER_DS), %edx
316 movl $(__KERNEL_PERCPU), %edx
321 /* Switch to kernel stack if necessary */
322 .if \switch_stacks > 0
323 SWITCH_TO_KERNEL_STACK
327 .macro SAVE_ALL_NMI cr3_reg:req
333 * Now switch the CR3 when PTI is enabled.
335 * We can enter with either user or kernel cr3, the code will
336 * store the old cr3 in \cr3_reg and switches to the kernel cr3
339 SWITCH_TO_KERNEL_CR3 scratch_reg=\cr3_reg
344 .macro RESTORE_INT_REGS
354 .macro RESTORE_REGS pop=0
360 .pushsection .fixup, "ax"
374 .macro RESTORE_ALL_NMI cr3_reg:req pop=0
376 * Now switch the CR3 when PTI is enabled.
378 * We enter with kernel cr3 and switch the cr3 to the value
379 * stored on \cr3_reg, which is either a user or a kernel cr3.
381 ALTERNATIVE "jmp .Lswitched_\@", "", X86_FEATURE_PTI
383 testl $PTI_SWITCH_MASK, \cr3_reg
386 /* User cr3 in \cr3_reg - write it to hardware cr3 */
393 RESTORE_REGS pop=\pop
396 .macro CHECK_AND_APPLY_ESPFIX
397 #ifdef CONFIG_X86_ESPFIX32
398 #define GDT_ESPFIX_SS PER_CPU_VAR(gdt_page) + (GDT_ENTRY_ESPFIX_SS * 8)
400 ALTERNATIVE "jmp .Lend_\@", "", X86_BUG_ESPFIX
402 movl PT_EFLAGS(%esp), %eax # mix EFLAGS, SS and CS
404 * Warning: PT_OLDSS(%esp) contains the wrong/random values if we
405 * are returning to the kernel.
406 * See comments in process.c:copy_thread() for details.
408 movb PT_OLDSS(%esp), %ah
409 movb PT_CS(%esp), %al
410 andl $(X86_EFLAGS_VM | (SEGMENT_TI_MASK << 8) | SEGMENT_RPL_MASK), %eax
411 cmpl $((SEGMENT_LDT << 8) | USER_RPL), %eax
412 jne .Lend_\@ # returning to user-space with LDT SS
415 * Setup and switch to ESPFIX stack
417 * We're returning to userspace with a 16 bit stack. The CPU will not
418 * restore the high word of ESP for us on executing iret... This is an
419 * "official" bug of all the x86-compatible CPUs, which we can work
420 * around to make dosemu and wine happy. We do this by preloading the
421 * high word of ESP with the high word of the userspace ESP while
422 * compensating for the offset by changing to the ESPFIX segment with
423 * a base address that matches for the difference.
425 mov %esp, %edx /* load kernel esp */
426 mov PT_OLDESP(%esp), %eax /* load userspace esp */
427 mov %dx, %ax /* eax: new kernel esp */
428 sub %eax, %edx /* offset (low word is 0) */
430 mov %dl, GDT_ESPFIX_SS + 4 /* bits 16..23 */
431 mov %dh, GDT_ESPFIX_SS + 7 /* bits 24..31 */
433 pushl %eax /* new kernel esp */
435 * Disable interrupts, but do not irqtrace this section: we
436 * will soon execute iret and the tracer was already set to
437 * the irqstate after the IRET:
439 DISABLE_INTERRUPTS(CLBR_ANY)
440 lss (%esp), %esp /* switch to espfix segment */
442 #endif /* CONFIG_X86_ESPFIX32 */
446 * Called with pt_regs fully populated and kernel segments loaded,
447 * so we can access PER_CPU and use the integer registers.
449 * We need to be very careful here with the %esp switch, because an NMI
450 * can happen everywhere. If the NMI handler finds itself on the
451 * entry-stack, it will overwrite the task-stack and everything we
452 * copied there. So allocate the stack-frame on the task-stack and
453 * switch to it before we do any copying.
456 .macro SWITCH_TO_KERNEL_STACK
458 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
462 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
465 * %eax now contains the entry cr3 and we carry it forward in
466 * that register for the time this macro runs
469 /* Are we on the entry stack? Bail out if not! */
470 movl PER_CPU_VAR(cpu_entry_area), %ecx
471 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
472 subl %esp, %ecx /* ecx = (end of entry_stack) - esp */
473 cmpl $SIZEOF_entry_stack, %ecx
476 /* Load stack pointer into %esi and %edi */
480 /* Move %edi to the top of the entry stack */
481 andl $(MASK_entry_stack), %edi
482 addl $(SIZEOF_entry_stack), %edi
484 /* Load top of task-stack into %edi */
485 movl TSS_entry2task_stack(%edi), %edi
487 /* Special case - entry from kernel mode via entry stack */
489 movl PT_EFLAGS(%esp), %ecx # mix EFLAGS and CS
490 movb PT_CS(%esp), %cl
491 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %ecx
493 movl PT_CS(%esp), %ecx
494 andl $SEGMENT_RPL_MASK, %ecx
497 jb .Lentry_from_kernel_\@
500 movl $PTREGS_SIZE, %ecx
503 testl $X86_EFLAGS_VM, PT_EFLAGS(%esi)
507 * Stack-frame contains 4 additional segment registers when
508 * coming from VM86 mode
515 /* Allocate frame on task-stack */
518 /* Switch to task-stack */
522 * We are now on the task-stack and can safely copy over the
531 .Lentry_from_kernel_\@:
534 * This handles the case when we enter the kernel from
535 * kernel-mode and %esp points to the entry-stack. When this
536 * happens we need to switch to the task-stack to run C code,
537 * but switch back to the entry-stack again when we approach
538 * iret and return to the interrupted code-path. This usually
539 * happens when we hit an exception while restoring user-space
540 * segment registers on the way back to user-space or when the
541 * sysenter handler runs with eflags.tf set.
543 * When we switch to the task-stack here, we can't trust the
544 * contents of the entry-stack anymore, as the exception handler
545 * might be scheduled out or moved to another CPU. Therefore we
546 * copy the complete entry-stack to the task-stack and set a
547 * marker in the iret-frame (bit 31 of the CS dword) to detect
548 * what we've done on the iret path.
550 * On the iret path we copy everything back and switch to the
551 * entry-stack, so that the interrupted kernel code-path
552 * continues on the same stack it was interrupted with.
554 * Be aware that an NMI can happen anytime in this code.
556 * %esi: Entry-Stack pointer (same as %esp)
557 * %edi: Top of the task stack
558 * %eax: CR3 on kernel entry
561 /* Calculate number of bytes on the entry stack in %ecx */
564 /* %ecx to the top of entry-stack */
565 andl $(MASK_entry_stack), %ecx
566 addl $(SIZEOF_entry_stack), %ecx
568 /* Number of bytes on the entry stack to %ecx */
571 /* Mark stackframe as coming from entry stack */
572 orl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
575 * Test the cr3 used to enter the kernel and add a marker
576 * so that we can switch back to it before iret.
578 testl $PTI_SWITCH_MASK, %eax
580 orl $CS_FROM_USER_CR3, PT_CS(%esp)
583 * %esi and %edi are unchanged, %ecx contains the number of
584 * bytes to copy. The code at .Lcopy_pt_regs_\@ will allocate
585 * the stack-frame on task-stack and copy everything over
587 jmp .Lcopy_pt_regs_\@
593 * Switch back from the kernel stack to the entry stack.
595 * The %esp register must point to pt_regs on the task stack. It will
596 * first calculate the size of the stack-frame to copy, depending on
597 * whether we return to VM86 mode or not. With that it uses 'rep movsl'
598 * to copy the contents of the stack over to the entry stack.
600 * We must be very careful here, as we can't trust the contents of the
601 * task-stack once we switched to the entry-stack. When an NMI happens
602 * while on the entry-stack, the NMI handler will switch back to the top
603 * of the task stack, overwriting our stack-frame we are about to copy.
604 * Therefore we switch the stack only after everything is copied over.
606 .macro SWITCH_TO_ENTRY_STACK
608 ALTERNATIVE "", "jmp .Lend_\@", X86_FEATURE_XENPV
611 movl $PTREGS_SIZE, %ecx
614 testl $(X86_EFLAGS_VM), PT_EFLAGS(%esp)
617 /* Additional 4 registers to copy when returning to VM86 mode */
623 /* Initialize source and destination for movsl */
624 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
628 /* Save future stack pointer in %ebx */
631 /* Copy over the stack-frame */
637 * Switch to entry-stack - needs to happen after everything is
638 * copied because the NMI handler will overwrite the task-stack
639 * when on entry-stack
647 * This macro handles the case when we return to kernel-mode on the iret
648 * path and have to switch back to the entry stack and/or user-cr3
650 * See the comments below the .Lentry_from_kernel_\@ label in the
651 * SWITCH_TO_KERNEL_STACK macro for more details.
653 .macro PARANOID_EXIT_TO_KERNEL_MODE
656 * Test if we entered the kernel with the entry-stack. Most
657 * likely we did not, because this code only runs on the
658 * return-to-kernel path.
660 testl $CS_FROM_ENTRY_STACK, PT_CS(%esp)
663 /* Unlikely slow-path */
665 /* Clear marker from stack-frame */
666 andl $(~CS_FROM_ENTRY_STACK), PT_CS(%esp)
668 /* Copy the remaining task-stack contents to entry-stack */
670 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %edi
672 /* Bytes on the task-stack to ecx */
673 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp1), %ecx
676 /* Allocate stack-frame on entry-stack */
680 * Save future stack-pointer, we must not switch until the
681 * copy is done, otherwise the NMI handler could destroy the
682 * contents of the task-stack we are about to copy.
691 /* Safe to switch to entry-stack now */
695 * We came from entry-stack and need to check if we also need to
696 * switch back to user cr3.
698 testl $CS_FROM_USER_CR3, PT_CS(%esp)
701 /* Clear marker from stack-frame */
702 andl $(~CS_FROM_USER_CR3), PT_CS(%esp)
704 SWITCH_TO_USER_CR3 scratch_reg=%eax
712 ENTRY(__switch_to_asm)
714 * Save callee-saved registers
715 * This must match the order in struct inactive_task_frame
724 movl %esp, TASK_threadsp(%eax)
725 movl TASK_threadsp(%edx), %esp
727 #ifdef CONFIG_STACKPROTECTOR
728 movl TASK_stack_canary(%edx), %ebx
729 movl %ebx, PER_CPU_VAR(stack_canary)+stack_canary_offset
732 #ifdef CONFIG_RETPOLINE
734 * When switching from a shallower to a deeper call stack
735 * the RSB may either underflow or use entries populated
736 * with userspace addresses. On CPUs where those concerns
737 * exist, overwrite the RSB with entries which capture
738 * speculative execution to prevent attack.
740 FILL_RETURN_BUFFER %ebx, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW
743 /* restore callee-saved registers */
754 * The unwinder expects the last frame on the stack to always be at the same
755 * offset from the end of the page, which allows it to validate the stack.
756 * Calling schedule_tail() directly would break that convention because its an
757 * asmlinkage function so its argument has to be pushed on the stack. This
758 * wrapper creates a proper "end of stack" frame header before the call.
760 ENTRY(schedule_tail_wrapper)
769 ENDPROC(schedule_tail_wrapper)
771 * A newly forked process directly context switches into this address.
773 * eax: prev task we switched from
774 * ebx: kernel thread func (NULL for user thread)
775 * edi: kernel thread arg
778 call schedule_tail_wrapper
781 jnz 1f /* kernel threads are uncommon */
784 /* When we fork, we trace the syscall return in the child, too. */
786 call syscall_return_slowpath
794 * A kernel thread is allowed to return here after successfully
795 * calling do_execve(). Exit to userspace to complete the execve()
798 movl $0, PT_EAX(%esp)
803 * Return to user mode is not as complex as all this looks,
804 * but we want the default path for a system call return to
805 * go as quickly as possible which is why some of this is
806 * less clear than it otherwise should be.
809 # userspace resumption stub bypassing syscall exit tracing
812 preempt_stop(CLBR_ANY)
815 movl PT_EFLAGS(%esp), %eax # mix EFLAGS and CS
816 movb PT_CS(%esp), %al
817 andl $(X86_EFLAGS_VM | SEGMENT_RPL_MASK), %eax
820 * We can be coming here from child spawned by kernel_thread().
822 movl PT_CS(%esp), %eax
823 andl $SEGMENT_RPL_MASK, %eax
826 jb restore_all_kernel # not returning to v8086 or userspace
828 ENTRY(resume_userspace)
829 DISABLE_INTERRUPTS(CLBR_ANY)
832 call prepare_exit_to_usermode
834 END(ret_from_exception)
836 GLOBAL(__begin_SYSENTER_singlestep_region)
838 * All code from here through __end_SYSENTER_singlestep_region is subject
839 * to being single-stepped if a user program sets TF and executes SYSENTER.
840 * There is absolutely nothing that we can do to prevent this from happening
841 * (thanks Intel!). To keep our handling of this situation as simple as
842 * possible, we handle TF just like AC and NT, except that our #DB handler
843 * will ignore all of the single-step traps generated in this range.
848 * Xen doesn't set %esp to be precisely what the normal SYSENTER
849 * entry point expects, so fix it up before using the normal path.
851 ENTRY(xen_sysenter_target)
852 addl $5*4, %esp /* remove xen-provided frame */
853 jmp .Lsysenter_past_esp
857 * 32-bit SYSENTER entry.
859 * 32-bit system calls through the vDSO's __kernel_vsyscall enter here
860 * if X86_FEATURE_SEP is available. This is the preferred system call
861 * entry on 32-bit systems.
863 * The SYSENTER instruction, in principle, should *only* occur in the
864 * vDSO. In practice, a small number of Android devices were shipped
865 * with a copy of Bionic that inlined a SYSENTER instruction. This
866 * never happened in any of Google's Bionic versions -- it only happened
867 * in a narrow range of Intel-provided versions.
869 * SYSENTER loads SS, ESP, CS, and EIP from previously programmed MSRs.
870 * IF and VM in RFLAGS are cleared (IOW: interrupts are off).
871 * SYSENTER does not save anything on the stack,
872 * and does not save old EIP (!!!), ESP, or EFLAGS.
874 * To avoid losing track of EFLAGS.VM (and thus potentially corrupting
875 * user and/or vm86 state), we explicitly disable the SYSENTER
876 * instruction in vm86 mode by reprogramming the MSRs.
879 * eax system call number
888 ENTRY(entry_SYSENTER_32)
890 * On entry-stack with all userspace-regs live - save and
891 * restore eflags and %eax to use it as scratch-reg for the cr3
896 BUG_IF_WRONG_CR3 no_user_check=1
897 SWITCH_TO_KERNEL_CR3 scratch_reg=%eax
901 /* Stack empty again, switch to task stack */
902 movl TSS_entry2task_stack(%esp), %esp
905 pushl $__USER_DS /* pt_regs->ss */
906 pushl %ebp /* pt_regs->sp (stashed in bp) */
907 pushfl /* pt_regs->flags (except IF = 0) */
908 orl $X86_EFLAGS_IF, (%esp) /* Fix IF */
909 pushl $__USER_CS /* pt_regs->cs */
910 pushl $0 /* pt_regs->ip = 0 (placeholder) */
911 pushl %eax /* pt_regs->orig_ax */
912 SAVE_ALL pt_regs_ax=$-ENOSYS /* save rest, stack already switched */
915 * SYSENTER doesn't filter flags, so we need to clear NT, AC
916 * and TF ourselves. To save a few cycles, we can check whether
917 * either was set instead of doing an unconditional popfq.
918 * This needs to happen before enabling interrupts so that
919 * we don't get preempted with NT set.
921 * If TF is set, we will single-step all the way to here -- do_debug
922 * will ignore all the traps. (Yes, this is slow, but so is
923 * single-stepping in general. This allows us to avoid having
924 * a more complicated code to handle the case where a user program
925 * forces us to single-step through the SYSENTER entry code.)
927 * NB.: .Lsysenter_fix_flags is a label with the code under it moved
928 * out-of-line as an optimization: NT is unlikely to be set in the
929 * majority of the cases and instead of polluting the I$ unnecessarily,
930 * we're keeping that code behind a branch which will predict as
931 * not-taken and therefore its instructions won't be fetched.
933 testl $X86_EFLAGS_NT|X86_EFLAGS_AC|X86_EFLAGS_TF, PT_EFLAGS(%esp)
934 jnz .Lsysenter_fix_flags
935 .Lsysenter_flags_fixed:
938 * User mode is traced as though IRQs are on, and SYSENTER
944 call do_fast_syscall_32
945 /* XEN PV guests always use IRET path */
946 ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
947 "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
951 /* Opportunistic SYSEXIT */
952 TRACE_IRQS_ON /* User mode traces as IRQs on. */
955 * Setup entry stack - we keep the pointer in %eax and do the
956 * switch after almost all user-state is restored.
959 /* Load entry stack pointer and allocate frame for eflags/eax */
960 movl PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %eax
963 /* Copy eflags and eax to entry stack */
964 movl PT_EFLAGS(%esp), %edi
965 movl PT_EAX(%esp), %esi
969 /* Restore user registers and segments */
970 movl PT_EIP(%esp), %edx /* pt_regs->ip */
971 movl PT_OLDESP(%esp), %ecx /* pt_regs->sp */
972 1: mov PT_FS(%esp), %fs
975 popl %ebx /* pt_regs->bx */
976 addl $2*4, %esp /* skip pt_regs->cx and pt_regs->dx */
977 popl %esi /* pt_regs->si */
978 popl %edi /* pt_regs->di */
979 popl %ebp /* pt_regs->bp */
981 /* Switch to entry stack */
984 /* Now ready to switch the cr3 */
985 SWITCH_TO_USER_CR3 scratch_reg=%eax
988 * Restore all flags except IF. (We restore IF separately because
989 * STI gives a one-instruction window in which we won't be interrupted,
990 * whereas POPF does not.)
992 btrl $X86_EFLAGS_IF_BIT, (%esp)
993 BUG_IF_WRONG_CR3 no_user_check=1
998 * Return back to the vDSO, which will pop ecx and edx.
999 * Don't bother with DS and ES (they already contain __USER_DS).
1004 .pushsection .fixup, "ax"
1005 2: movl $0, PT_FS(%esp)
1008 _ASM_EXTABLE(1b, 2b)
1011 .Lsysenter_fix_flags:
1012 pushl $X86_EFLAGS_FIXED
1014 jmp .Lsysenter_flags_fixed
1015 GLOBAL(__end_SYSENTER_singlestep_region)
1016 ENDPROC(entry_SYSENTER_32)
1019 * 32-bit legacy system call entry.
1021 * 32-bit x86 Linux system calls traditionally used the INT $0x80
1022 * instruction. INT $0x80 lands here.
1024 * This entry point can be used by any 32-bit perform system calls.
1025 * Instances of INT $0x80 can be found inline in various programs and
1026 * libraries. It is also used by the vDSO's __kernel_vsyscall
1027 * fallback for hardware that doesn't support a faster entry method.
1028 * Restarted 32-bit system calls also fall back to INT $0x80
1029 * regardless of what instruction was originally used to do the system
1030 * call. (64-bit programs can use INT $0x80 as well, but they can
1031 * only run on 64-bit kernels and therefore land in
1032 * entry_INT80_compat.)
1034 * This is considered a slow path. It is not used by most libc
1035 * implementations on modern hardware except during process startup.
1038 * eax system call number
1046 ENTRY(entry_INT80_32)
1048 pushl %eax /* pt_regs->orig_ax */
1050 SAVE_ALL pt_regs_ax=$-ENOSYS switch_stacks=1 /* save rest */
1053 * User mode is traced as though IRQs are on, and the interrupt gate
1059 call do_int80_syscall_32
1066 SWITCH_TO_ENTRY_STACK
1067 .Lrestore_all_notrace:
1068 CHECK_AND_APPLY_ESPFIX
1070 /* Switch back to user CR3 */
1071 SWITCH_TO_USER_CR3 scratch_reg=%eax
1075 /* Restore user state */
1076 RESTORE_REGS pop=4 # skip orig_eax/error_code
1080 * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization
1081 * when returning from IPI handler and when returning from
1082 * scheduler to user-space.
1087 #ifdef CONFIG_PREEMPTION
1088 DISABLE_INTERRUPTS(CLBR_ANY)
1089 cmpl $0, PER_CPU_VAR(__preempt_count)
1091 testl $X86_EFLAGS_IF, PT_EFLAGS(%esp) # interrupts off (exception path) ?
1093 call preempt_schedule_irq
1097 PARANOID_EXIT_TO_KERNEL_MODE
1102 .section .fixup, "ax"
1104 pushl $0 # no error code
1105 pushl $do_iret_error
1107 #ifdef CONFIG_DEBUG_ENTRY
1109 * The stack-frame here is the one that iret faulted on, so its a
1110 * return-to-user frame. We are on kernel-cr3 because we come here from
1111 * the fixup code. This confuses the CR3 checker, so switch to user-cr3
1112 * as the checker expects it.
1115 SWITCH_TO_USER_CR3 scratch_reg=%eax
1119 jmp common_exception
1121 _ASM_EXTABLE(.Lirq_return, iret_exc)
1122 ENDPROC(entry_INT80_32)
1124 .macro FIXUP_ESPFIX_STACK
1126 * Switch back for ESPFIX stack to the normal zerobased stack
1128 * We can't call C functions using the ESPFIX stack. This code reads
1129 * the high word of the segment base from the GDT and swiches to the
1130 * normal stack and adjusts ESP with the matching offset.
1132 #ifdef CONFIG_X86_ESPFIX32
1133 /* fixup the stack */
1134 mov GDT_ESPFIX_SS + 4, %al /* bits 16..23 */
1135 mov GDT_ESPFIX_SS + 7, %ah /* bits 24..31 */
1137 addl %esp, %eax /* the adjusted stack pointer */
1140 lss (%esp), %esp /* switch to the normal stack segment */
1143 .macro UNWIND_ESPFIX_STACK
1144 #ifdef CONFIG_X86_ESPFIX32
1146 /* see if on espfix stack */
1147 cmpw $__ESPFIX_SS, %ax
1149 movl $__KERNEL_DS, %eax
1152 /* switch to normal stack */
1159 * Build the entry stubs with some assembler magic.
1160 * We pack 1 stub into every 8-byte block.
1163 ENTRY(irq_entries_start)
1164 vector=FIRST_EXTERNAL_VECTOR
1165 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
1166 pushl $(~vector+0x80) /* Note: always in signed byte range */
1168 jmp common_interrupt
1171 END(irq_entries_start)
1173 #ifdef CONFIG_X86_LOCAL_APIC
1175 ENTRY(spurious_entries_start)
1176 vector=FIRST_SYSTEM_VECTOR
1177 .rept (NR_VECTORS - FIRST_SYSTEM_VECTOR)
1178 pushl $(~vector+0x80) /* Note: always in signed byte range */
1183 END(spurious_entries_start)
1187 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1188 SAVE_ALL switch_stacks=1
1189 ENCODE_FRAME_POINTER
1192 call smp_spurious_interrupt
1194 ENDPROC(common_spurious)
1198 * the CPU automatically disables interrupts when executing an IRQ vector,
1199 * so IRQ-flags tracing has to follow that:
1201 .p2align CONFIG_X86_L1_CACHE_SHIFT
1204 addl $-0x80, (%esp) /* Adjust vector into the [-256, -1] range */
1206 SAVE_ALL switch_stacks=1
1207 ENCODE_FRAME_POINTER
1212 ENDPROC(common_interrupt)
1214 #define BUILD_INTERRUPT3(name, nr, fn) \
1218 SAVE_ALL switch_stacks=1; \
1219 ENCODE_FRAME_POINTER; \
1223 jmp ret_from_intr; \
1226 #define BUILD_INTERRUPT(name, nr) \
1227 BUILD_INTERRUPT3(name, nr, smp_##name); \
1229 /* The include is where all of the SMP etc. interrupts come from */
1230 #include <asm/entry_arch.h>
1232 ENTRY(coprocessor_error)
1235 pushl $do_coprocessor_error
1236 jmp common_exception
1237 END(coprocessor_error)
1239 ENTRY(simd_coprocessor_error)
1242 #ifdef CONFIG_X86_INVD_BUG
1243 /* AMD 486 bug: invd from userspace calls exception 19 instead of #GP */
1244 ALTERNATIVE "pushl $do_general_protection", \
1245 "pushl $do_simd_coprocessor_error", \
1248 pushl $do_simd_coprocessor_error
1250 jmp common_exception
1251 END(simd_coprocessor_error)
1253 ENTRY(device_not_available)
1255 pushl $-1 # mark this as an int
1256 pushl $do_device_not_available
1257 jmp common_exception
1258 END(device_not_available)
1260 #ifdef CONFIG_PARAVIRT
1263 _ASM_EXTABLE(native_iret, iret_exc)
1271 jmp common_exception
1278 jmp common_exception
1284 pushl $do_invalid_op
1285 jmp common_exception
1288 ENTRY(coprocessor_segment_overrun)
1291 pushl $do_coprocessor_segment_overrun
1292 jmp common_exception
1293 END(coprocessor_segment_overrun)
1297 pushl $do_invalid_TSS
1298 jmp common_exception
1301 ENTRY(segment_not_present)
1303 pushl $do_segment_not_present
1304 jmp common_exception
1305 END(segment_not_present)
1307 ENTRY(stack_segment)
1309 pushl $do_stack_segment
1310 jmp common_exception
1313 ENTRY(alignment_check)
1315 pushl $do_alignment_check
1316 jmp common_exception
1317 END(alignment_check)
1321 pushl $0 # no error code
1322 pushl $do_divide_error
1323 jmp common_exception
1326 #ifdef CONFIG_X86_MCE
1327 ENTRY(machine_check)
1330 pushl machine_check_vector
1331 jmp common_exception
1335 ENTRY(spurious_interrupt_bug)
1338 pushl $do_spurious_interrupt_bug
1339 jmp common_exception
1340 END(spurious_interrupt_bug)
1342 #ifdef CONFIG_XEN_PV
1343 ENTRY(xen_hypervisor_callback)
1344 pushl $-1 /* orig_ax = -1 => not a system call */
1346 ENCODE_FRAME_POINTER
1350 * Check to see if we got the event in the critical
1351 * region in xen_iret_direct, after we've reenabled
1352 * events and checked for pending events. This simulates
1353 * iret instruction's behaviour where it delivers a
1354 * pending interrupt when enabling interrupts:
1356 movl PT_EIP(%esp), %eax
1357 cmpl $xen_iret_start_crit, %eax
1359 cmpl $xen_iret_end_crit, %eax
1362 jmp xen_iret_crit_fixup
1364 ENTRY(xen_do_upcall)
1366 call xen_evtchn_do_upcall
1367 #ifndef CONFIG_PREEMPTION
1368 call xen_maybe_preempt_hcall
1371 ENDPROC(xen_hypervisor_callback)
1374 * Hypervisor uses this for application faults while it executes.
1375 * We get here for two reasons:
1376 * 1. Fault while reloading DS, ES, FS or GS
1377 * 2. Fault while executing IRET
1378 * Category 1 we fix up by reattempting the load, and zeroing the segment
1379 * register if the load fails.
1380 * Category 2 we fix up by jumping to do_iret_error. We cannot use the
1381 * normal Linux return path in this case because if we use the IRET hypercall
1382 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
1383 * We distinguish between categories by maintaining a status value in EAX.
1385 ENTRY(xen_failsafe_callback)
1390 3: mov 12(%esp), %fs
1391 4: mov 16(%esp), %gs
1392 /* EAX == 0 => Category 1 (Bad segment)
1393 EAX != 0 => Category 2 (Bad IRET) */
1399 5: pushl $-1 /* orig_ax = -1 => not a system call */
1401 ENCODE_FRAME_POINTER
1402 jmp ret_from_exception
1404 .section .fixup, "ax"
1418 _ASM_EXTABLE(1b, 6b)
1419 _ASM_EXTABLE(2b, 7b)
1420 _ASM_EXTABLE(3b, 8b)
1421 _ASM_EXTABLE(4b, 9b)
1422 ENDPROC(xen_failsafe_callback)
1423 #endif /* CONFIG_XEN_PV */
1425 #ifdef CONFIG_XEN_PVHVM
1426 BUILD_INTERRUPT3(xen_hvm_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1427 xen_evtchn_do_upcall)
1431 #if IS_ENABLED(CONFIG_HYPERV)
1433 BUILD_INTERRUPT3(hyperv_callback_vector, HYPERVISOR_CALLBACK_VECTOR,
1434 hyperv_vector_handler)
1436 BUILD_INTERRUPT3(hyperv_reenlightenment_vector, HYPERV_REENLIGHTENMENT_VECTOR,
1437 hyperv_reenlightenment_intr)
1439 BUILD_INTERRUPT3(hv_stimer0_callback_vector, HYPERV_STIMER0_VECTOR,
1440 hv_stimer0_vector_handler)
1442 #endif /* CONFIG_HYPERV */
1446 pushl $do_page_fault
1447 jmp common_exception_read_cr2
1450 common_exception_read_cr2:
1451 /* the function address is in %gs's slot on the stack */
1452 SAVE_ALL switch_stacks=1 skip_gs=1
1454 ENCODE_FRAME_POINTER
1459 movl PT_GS(%esp), %edi
1463 GET_CR2_INTO(%ecx) # might clobber %eax
1465 /* fixup orig %eax */
1466 movl PT_ORIG_EAX(%esp), %edx # get the error code
1467 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1470 movl %esp, %eax # pt_regs pointer
1472 jmp ret_from_exception
1473 END(common_exception_read_cr2)
1476 /* the function address is in %gs's slot on the stack */
1477 SAVE_ALL switch_stacks=1 skip_gs=1
1478 ENCODE_FRAME_POINTER
1483 movl PT_GS(%esp), %edi # get the function address
1487 /* fixup orig %eax */
1488 movl PT_ORIG_EAX(%esp), %edx # get the error code
1489 movl $-1, PT_ORIG_EAX(%esp) # no syscall to restart
1492 movl %esp, %eax # pt_regs pointer
1494 jmp ret_from_exception
1495 END(common_exception)
1499 * Entry from sysenter is now handled in common_exception
1502 pushl $-1 # mark this as an int
1504 jmp common_exception
1508 * NMI is doubly nasty. It can happen on the first instruction of
1509 * entry_SYSENTER_32 (just like #DB), but it can also interrupt the beginning
1510 * of the #DB handler even if that #DB in turn hit before entry_SYSENTER_32
1511 * switched stacks. We handle both conditions by simply checking whether we
1512 * interrupted kernel code running on the SYSENTER stack.
1517 #ifdef CONFIG_X86_ESPFIX32
1520 cmpw $__ESPFIX_SS, %ax
1522 je .Lnmi_espfix_stack
1525 pushl %eax # pt_regs->orig_ax
1526 SAVE_ALL_NMI cr3_reg=%edi
1527 ENCODE_FRAME_POINTER
1528 xorl %edx, %edx # zero error code
1529 movl %esp, %eax # pt_regs pointer
1531 /* Are we currently on the SYSENTER stack? */
1532 movl PER_CPU_VAR(cpu_entry_area), %ecx
1533 addl $CPU_ENTRY_AREA_entry_stack + SIZEOF_entry_stack, %ecx
1534 subl %eax, %ecx /* ecx = (end of entry_stack) - esp */
1535 cmpl $SIZEOF_entry_stack, %ecx
1536 jb .Lnmi_from_sysenter_stack
1538 /* Not on SYSENTER stack. */
1542 .Lnmi_from_sysenter_stack:
1544 * We're on the SYSENTER stack. Switch off. No one (not even debug)
1545 * is using the thread stack right now, so it's safe for us to use it.
1548 movl PER_CPU_VAR(cpu_current_top_of_stack), %esp
1553 CHECK_AND_APPLY_ESPFIX
1554 RESTORE_ALL_NMI cr3_reg=%edi pop=4
1557 #ifdef CONFIG_X86_ESPFIX32
1560 * create the pointer to lss back
1565 /* copy the iret frame of 12 bytes */
1570 SAVE_ALL_NMI cr3_reg=%edi
1571 ENCODE_FRAME_POINTER
1572 FIXUP_ESPFIX_STACK # %eax == %esp
1573 xorl %edx, %edx # zero error code
1575 RESTORE_ALL_NMI cr3_reg=%edi
1576 lss 12+4(%esp), %esp # back to espfix stack
1583 pushl $-1 # mark this as an int
1585 SAVE_ALL switch_stacks=1
1586 ENCODE_FRAME_POINTER
1588 xorl %edx, %edx # zero error code
1589 movl %esp, %eax # pt_regs pointer
1591 jmp ret_from_exception
1594 ENTRY(general_protection)
1595 pushl $do_general_protection
1596 jmp common_exception
1597 END(general_protection)
1599 #ifdef CONFIG_KVM_GUEST
1600 ENTRY(async_page_fault)
1602 pushl $do_async_page_fault
1603 jmp common_exception_read_cr2
1604 END(async_page_fault)
1607 ENTRY(rewind_stack_do_exit)
1608 /* Prevent any naive code from trying to unwind to our caller. */
1611 movl PER_CPU_VAR(cpu_current_top_of_stack), %esi
1612 leal -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%esi), %esp
1616 END(rewind_stack_do_exit)