blob: 607d72c4a485cf25a3e582804cea0b7c031e1d32 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/arch/x86_64/entry.S
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
6 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
Ingo Molnar4d732132015-06-08 20:43:07 +02007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * entry.S contains the system-call and fault low-level handling routines.
9 *
Andy Lutomirski8b4777a2011-06-05 13:50:18 -040010 * Some of this is documented in Documentation/x86/entry_64.txt
11 *
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +010012 * A note on terminology:
Ingo Molnar4d732132015-06-08 20:43:07 +020013 * - iret frame: Architecture defined interrupt frame from SS to RIP
14 * at the top of the kernel process stack.
Andi Kleen2e91a172006-09-26 10:52:29 +020015 *
16 * Some macro usage:
Ingo Molnar4d732132015-06-08 20:43:07 +020017 * - ENTRY/END: Define functions in the symbol table.
18 * - TRACE_IRQ_*: Trace hardirq state for lock debugging.
19 * - idtentry: Define exception entry points.
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/linkage.h>
22#include <asm/segment.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <asm/cache.h>
24#include <asm/errno.h>
Ingo Molnard36f9472015-06-03 18:29:26 +020025#include "calling.h"
Sam Ravnborge2d5df92005-09-09 21:28:48 +020026#include <asm/asm-offsets.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/msr.h>
28#include <asm/unistd.h>
29#include <asm/thread_info.h>
30#include <asm/hw_irq.h>
Jeremy Fitzhardinge0341c142009-02-13 11:14:01 -080031#include <asm/page_types.h>
Ingo Molnar2601e642006-07-03 00:24:45 -070032#include <asm/irqflags.h>
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +010033#include <asm/paravirt.h>
Tejun Heo9939dda2009-01-13 20:41:35 +090034#include <asm/percpu.h>
H. Peter Anvind7abc0f2012-04-20 12:19:50 -070035#include <asm/asm.h>
H. Peter Anvin63bcff22012-09-21 12:43:12 -070036#include <asm/smap.h>
H. Peter Anvin3891a042014-04-29 16:46:09 -070037#include <asm/pgtable_types.h>
Al Viro784d5692016-01-11 11:04:34 -050038#include <asm/export.h>
Josh Poimboeufff3f7e22017-01-09 12:00:25 -060039#include <asm/frame.h>
Eric Parisd7e7528b2012-01-03 14:23:06 -050040#include <linux/err.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Ingo Molnar4d732132015-06-08 20:43:07 +020042.code64
43.section .entry.text, "ax"
Arnaldo Carvalho de Melo16444a82008-05-12 21:20:42 +020044
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +010045#ifdef CONFIG_PARAVIRT
Jeremy Fitzhardinge2be29982008-06-25 00:19:28 -040046ENTRY(native_usergs_sysret64)
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +010047 swapgs
48 sysretq
Cyrill Gorcunovb3baaa12009-02-23 22:57:00 +030049ENDPROC(native_usergs_sysret64)
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +010050#endif /* CONFIG_PARAVIRT */
51
Denys Vlasenkof2db9382015-02-26 14:40:30 -080052.macro TRACE_IRQS_IRETQ
Ingo Molnar2601e642006-07-03 00:24:45 -070053#ifdef CONFIG_TRACE_IRQFLAGS
Ingo Molnar4d732132015-06-08 20:43:07 +020054 bt $9, EFLAGS(%rsp) /* interrupts off? */
55 jnc 1f
Ingo Molnar2601e642006-07-03 00:24:45 -070056 TRACE_IRQS_ON
571:
58#endif
59.endm
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
Steven Rostedt5963e312012-05-30 11:54:53 -040062 * When dynamic function tracer is enabled it will add a breakpoint
63 * to all locations that it is about to modify, sync CPUs, update
64 * all the code, sync CPUs, then remove the breakpoints. In this time
65 * if lockdep is enabled, it might jump back into the debug handler
66 * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF).
67 *
68 * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to
69 * make sure the stack pointer does not get reset back to the top
70 * of the debug stack, and instead just reuses the current stack.
71 */
72#if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS)
73
74.macro TRACE_IRQS_OFF_DEBUG
Ingo Molnar4d732132015-06-08 20:43:07 +020075 call debug_stack_set_zero
Steven Rostedt5963e312012-05-30 11:54:53 -040076 TRACE_IRQS_OFF
Ingo Molnar4d732132015-06-08 20:43:07 +020077 call debug_stack_reset
Steven Rostedt5963e312012-05-30 11:54:53 -040078.endm
79
80.macro TRACE_IRQS_ON_DEBUG
Ingo Molnar4d732132015-06-08 20:43:07 +020081 call debug_stack_set_zero
Steven Rostedt5963e312012-05-30 11:54:53 -040082 TRACE_IRQS_ON
Ingo Molnar4d732132015-06-08 20:43:07 +020083 call debug_stack_reset
Steven Rostedt5963e312012-05-30 11:54:53 -040084.endm
85
Denys Vlasenkof2db9382015-02-26 14:40:30 -080086.macro TRACE_IRQS_IRETQ_DEBUG
Ingo Molnar4d732132015-06-08 20:43:07 +020087 bt $9, EFLAGS(%rsp) /* interrupts off? */
88 jnc 1f
Steven Rostedt5963e312012-05-30 11:54:53 -040089 TRACE_IRQS_ON_DEBUG
901:
91.endm
92
93#else
Ingo Molnar4d732132015-06-08 20:43:07 +020094# define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF
95# define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON
96# define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ
Steven Rostedt5963e312012-05-30 11:54:53 -040097#endif
98
99/*
Ingo Molnar4d732132015-06-08 20:43:07 +0200100 * 64-bit SYSCALL instruction entry. Up to 6 arguments in registers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 *
Andy Lutomirskifda57b22016-03-09 19:00:35 -0800102 * This is the only entry point used for 64-bit system calls. The
103 * hardware interface is reasonably well designed and the register to
104 * argument mapping Linux uses fits well with the registers that are
105 * available when SYSCALL is used.
106 *
107 * SYSCALL instructions can be found inlined in libc implementations as
108 * well as some other programs and libraries. There are also a handful
109 * of SYSCALL instructions in the vDSO used, for example, as a
110 * clock_gettimeofday fallback.
111 *
Ingo Molnar4d732132015-06-08 20:43:07 +0200112 * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11,
Denys Vlasenkob87cf632015-02-26 14:40:32 -0800113 * then loads new ss, cs, and rip from previously programmed MSRs.
114 * rflags gets masked by a value from another MSR (so CLD and CLAC
115 * are not needed). SYSCALL does not save anything on the stack
116 * and does not change rsp.
117 *
118 * Registers on entry:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * rax system call number
Denys Vlasenkob87cf632015-02-26 14:40:32 -0800120 * rcx return address
121 * r11 saved rflags (note: r11 is callee-clobbered register in C ABI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 * rdi arg0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 * rsi arg1
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100124 * rdx arg2
Denys Vlasenkob87cf632015-02-26 14:40:32 -0800125 * r10 arg3 (needs to be moved to rcx to conform to C ABI)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * r8 arg4
127 * r9 arg5
Ingo Molnar4d732132015-06-08 20:43:07 +0200128 * (note: r12-r15, rbp, rbx are callee-preserved in C ABI)
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100129 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 * Only called from user space.
131 *
Ingo Molnar7fcb3bc2015-03-17 14:42:59 +0100132 * When user can change pt_regs->foo always force IRET. That is because
Andi Kleen7bf36bb2006-04-07 19:50:00 +0200133 * it deals with uncanonical addresses better. SYSRET has trouble
134 * with them due to bugs in both AMD and Intel CPUs.
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100135 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Ingo Molnarb2502b42015-06-08 08:42:03 +0200137ENTRY(entry_SYSCALL_64)
Denys Vlasenko9ed8e7d2015-03-19 18:17:47 +0100138 /*
139 * Interrupts are off on entry.
140 * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON,
141 * it is too small to ever cause noticeable irq latency.
142 */
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +0100143 SWAPGS_UNSAFE_STACK
144 /*
145 * A hypervisor implementation might want to use a label
146 * after the swapgs, so that it can do the swapgs
147 * for the guest and jump here on syscall.
148 */
Ingo Molnarb2502b42015-06-08 08:42:03 +0200149GLOBAL(entry_SYSCALL_64_after_swapgs)
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +0100150
Ingo Molnar4d732132015-06-08 20:43:07 +0200151 movq %rsp, PER_CPU_VAR(rsp_scratch)
152 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
Denys Vlasenko9ed8e7d2015-03-19 18:17:47 +0100153
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800154 TRACE_IRQS_OFF
155
Denys Vlasenko9ed8e7d2015-03-19 18:17:47 +0100156 /* Construct struct pt_regs on stack */
Ingo Molnar4d732132015-06-08 20:43:07 +0200157 pushq $__USER_DS /* pt_regs->ss */
158 pushq PER_CPU_VAR(rsp_scratch) /* pt_regs->sp */
Ingo Molnar4d732132015-06-08 20:43:07 +0200159 pushq %r11 /* pt_regs->flags */
160 pushq $__USER_CS /* pt_regs->cs */
161 pushq %rcx /* pt_regs->ip */
162 pushq %rax /* pt_regs->orig_ax */
163 pushq %rdi /* pt_regs->di */
164 pushq %rsi /* pt_regs->si */
165 pushq %rdx /* pt_regs->dx */
166 pushq %rcx /* pt_regs->cx */
167 pushq $-ENOSYS /* pt_regs->ax */
168 pushq %r8 /* pt_regs->r8 */
169 pushq %r9 /* pt_regs->r9 */
170 pushq %r10 /* pt_regs->r10 */
171 pushq %r11 /* pt_regs->r11 */
172 sub $(6*8), %rsp /* pt_regs->bp, bx, r12-15 not saved */
Denys Vlasenko9ed8e7d2015-03-19 18:17:47 +0100173
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800174 /*
175 * If we need to do entry work or if we guess we'll need to do
176 * exit work, go straight to the slow path.
177 */
Andy Lutomirski15f4eae2016-09-13 14:29:25 -0700178 movq PER_CPU_VAR(current_task), %r11
179 testl $_TIF_WORK_SYSCALL_ENTRY|_TIF_ALLWORK_MASK, TASK_TI_flags(%r11)
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800180 jnz entry_SYSCALL64_slow_path
181
Ingo Molnarb2502b42015-06-08 08:42:03 +0200182entry_SYSCALL_64_fastpath:
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800183 /*
184 * Easy case: enable interrupts and issue the syscall. If the syscall
185 * needs pt_regs, we'll call a stub that disables interrupts again
186 * and jumps to the slow path.
187 */
188 TRACE_IRQS_ON
189 ENABLE_INTERRUPTS(CLBR_NONE)
H. Peter Anvinfca460f2012-02-19 07:56:26 -0800190#if __SYSCALL_MASK == ~0
Ingo Molnar4d732132015-06-08 20:43:07 +0200191 cmpq $__NR_syscall_max, %rax
H. Peter Anvinfca460f2012-02-19 07:56:26 -0800192#else
Ingo Molnar4d732132015-06-08 20:43:07 +0200193 andl $__SYSCALL_MASK, %eax
194 cmpl $__NR_syscall_max, %eax
H. Peter Anvinfca460f2012-02-19 07:56:26 -0800195#endif
Ingo Molnar4d732132015-06-08 20:43:07 +0200196 ja 1f /* return -ENOSYS (already in pt_regs->ax) */
197 movq %r10, %rcx
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800198
199 /*
200 * This call instruction is handled specially in stub_ptregs_64.
Andy Lutomirskib7765082016-01-31 09:33:26 -0800201 * It might end up jumping to the slow path. If it jumps, RAX
202 * and all argument registers are clobbered.
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800203 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200204 call *sys_call_table(, %rax, 8)
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800205.Lentry_SYSCALL_64_after_fastpath_call:
206
Ingo Molnar4d732132015-06-08 20:43:07 +0200207 movq %rax, RAX(%rsp)
Denys Vlasenko146b2b02015-03-25 18:18:13 +01002081:
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800209
Denys Vlasenko4416c5a2015-03-31 19:00:03 +0200210 /*
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800211 * If we get here, then we know that pt_regs is clean for SYSRET64.
212 * If we see that no exit work is required (which we are required
213 * to check with IRQs off), then we can go straight to SYSRET64.
Denys Vlasenko4416c5a2015-03-31 19:00:03 +0200214 */
Jan Beulich2140a992017-02-03 02:03:25 -0700215 DISABLE_INTERRUPTS(CLBR_ANY)
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800216 TRACE_IRQS_OFF
Andy Lutomirski15f4eae2016-09-13 14:29:25 -0700217 movq PER_CPU_VAR(current_task), %r11
218 testl $_TIF_ALLWORK_MASK, TASK_TI_flags(%r11)
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800219 jnz 1f
Andy Lutomirskib3494a42015-03-23 12:32:54 -0700220
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800221 LOCKDEP_SYS_EXIT
222 TRACE_IRQS_ON /* user mode is traced as IRQs on */
Andy Lutomirskieb2a54c2016-01-31 09:33:27 -0800223 movq RIP(%rsp), %rcx
224 movq EFLAGS(%rsp), %r11
225 RESTORE_C_REGS_EXCEPT_RCX_R11
Ingo Molnar4d732132015-06-08 20:43:07 +0200226 movq RSP(%rsp), %rsp
Jeremy Fitzhardinge2be29982008-06-25 00:19:28 -0400227 USERGS_SYSRET64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Andy Lutomirski1e423bf2016-01-28 15:11:28 -08002291:
230 /*
231 * The fast path looked good when we started, but something changed
232 * along the way and we need to switch to the slow path. Calling
233 * raise(3) will trigger this, for example. IRQs are off.
234 */
Andy Lutomirski29ea1b22015-07-03 12:44:28 -0700235 TRACE_IRQS_ON
Jan Beulich2140a992017-02-03 02:03:25 -0700236 ENABLE_INTERRUPTS(CLBR_ANY)
Denys Vlasenko76f5df42015-02-26 14:40:27 -0800237 SAVE_EXTRA_REGS
Andy Lutomirski29ea1b22015-07-03 12:44:28 -0700238 movq %rsp, %rdi
239 call syscall_return_slowpath /* returns with IRQs disabled */
Andy Lutomirski1e423bf2016-01-28 15:11:28 -0800240 jmp return_from_SYSCALL_64
241
242entry_SYSCALL64_slow_path:
243 /* IRQs are off. */
244 SAVE_EXTRA_REGS
245 movq %rsp, %rdi
246 call do_syscall_64 /* returns with IRQs disabled */
247
248return_from_SYSCALL_64:
Denys Vlasenko76f5df42015-02-26 14:40:27 -0800249 RESTORE_EXTRA_REGS
Andy Lutomirski29ea1b22015-07-03 12:44:28 -0700250 TRACE_IRQS_IRETQ /* we're about to change IF */
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200251
252 /*
253 * Try to use SYSRET instead of IRET if we're returning to
254 * a completely clean 64-bit userspace context.
255 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200256 movq RCX(%rsp), %rcx
257 movq RIP(%rsp), %r11
258 cmpq %rcx, %r11 /* RCX == RIP */
259 jne opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200260
261 /*
262 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
263 * in kernel space. This essentially lets the user take over
Denys Vlasenko17be0ae2015-04-21 18:27:29 +0200264 * the kernel, since userspace controls RSP.
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200265 *
Denys Vlasenko17be0ae2015-04-21 18:27:29 +0200266 * If width of "canonical tail" ever becomes variable, this will need
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200267 * to be updated to remain correct on both old and new CPUs.
Kirill A. Shutemov361b4b52017-03-30 11:07:26 +0300268 *
269 * Change top 16 bits to be the sign-extension of 47th bit
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200270 */
Denys Vlasenko17be0ae2015-04-21 18:27:29 +0200271 shl $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
272 sar $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx
Ingo Molnar4d732132015-06-08 20:43:07 +0200273
Denys Vlasenko17be0ae2015-04-21 18:27:29 +0200274 /* If this changed %rcx, it was not canonical */
275 cmpq %rcx, %r11
276 jne opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200277
Ingo Molnar4d732132015-06-08 20:43:07 +0200278 cmpq $__USER_CS, CS(%rsp) /* CS must match SYSRET */
279 jne opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200280
Ingo Molnar4d732132015-06-08 20:43:07 +0200281 movq R11(%rsp), %r11
282 cmpq %r11, EFLAGS(%rsp) /* R11 == RFLAGS */
283 jne opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200284
285 /*
Borislav Petkov3e035302016-08-03 19:14:29 +0200286 * SYSCALL clears RF when it saves RFLAGS in R11 and SYSRET cannot
287 * restore RF properly. If the slowpath sets it for whatever reason, we
288 * need to restore it correctly.
289 *
290 * SYSRET can restore TF, but unlike IRET, restoring TF results in a
291 * trap from userspace immediately after SYSRET. This would cause an
292 * infinite loop whenever #DB happens with register state that satisfies
293 * the opportunistic SYSRET conditions. For example, single-stepping
294 * this user code:
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200295 *
Ingo Molnar4d732132015-06-08 20:43:07 +0200296 * movq $stuck_here, %rcx
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200297 * pushfq
298 * popq %r11
299 * stuck_here:
300 *
301 * would never get past 'stuck_here'.
302 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200303 testq $(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11
304 jnz opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200305
306 /* nothing to check for RSP */
307
Ingo Molnar4d732132015-06-08 20:43:07 +0200308 cmpq $__USER_DS, SS(%rsp) /* SS must match SYSRET */
309 jne opportunistic_sysret_failed
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200310
311 /*
Ingo Molnar4d732132015-06-08 20:43:07 +0200312 * We win! This label is here just for ease of understanding
313 * perf profiles. Nothing jumps here.
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200314 */
315syscall_return_via_sysret:
Denys Vlasenko17be0ae2015-04-21 18:27:29 +0200316 /* rcx and r11 are already restored (see code above) */
317 RESTORE_C_REGS_EXCEPT_RCX_R11
Ingo Molnar4d732132015-06-08 20:43:07 +0200318 movq RSP(%rsp), %rsp
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200319 USERGS_SYSRET64
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200320
321opportunistic_sysret_failed:
322 SWAPGS
323 jmp restore_c_regs_and_iret
Ingo Molnarb2502b42015-06-08 08:42:03 +0200324END(entry_SYSCALL_64)
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100325
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800326ENTRY(stub_ptregs_64)
327 /*
328 * Syscalls marked as needing ptregs land here.
Andy Lutomirskib7765082016-01-31 09:33:26 -0800329 * If we are on the fast path, we need to save the extra regs,
330 * which we achieve by trying again on the slow path. If we are on
331 * the slow path, the extra regs are already saved.
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800332 *
333 * RAX stores a pointer to the C function implementing the syscall.
Andy Lutomirskib7765082016-01-31 09:33:26 -0800334 * IRQs are on.
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800335 */
336 cmpq $.Lentry_SYSCALL_64_after_fastpath_call, (%rsp)
337 jne 1f
338
Andy Lutomirskib7765082016-01-31 09:33:26 -0800339 /*
340 * Called from fast path -- disable IRQs again, pop return address
341 * and jump to slow path
342 */
Jan Beulich2140a992017-02-03 02:03:25 -0700343 DISABLE_INTERRUPTS(CLBR_ANY)
Andy Lutomirskib7765082016-01-31 09:33:26 -0800344 TRACE_IRQS_OFF
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800345 popq %rax
Andy Lutomirskib7765082016-01-31 09:33:26 -0800346 jmp entry_SYSCALL64_slow_path
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800347
3481:
Borislav Petkovb3830e82016-08-01 12:05:02 +0200349 jmp *%rax /* Called from C */
Andy Lutomirski302f5b22016-01-28 15:11:25 -0800350END(stub_ptregs_64)
351
352.macro ptregs_stub func
353ENTRY(ptregs_\func)
354 leaq \func(%rip), %rax
355 jmp stub_ptregs_64
356END(ptregs_\func)
357.endm
358
359/* Instantiate ptregs_stub for each ptregs-using syscall */
360#define __SYSCALL_64_QUAL_(sym)
361#define __SYSCALL_64_QUAL_ptregs(sym) ptregs_stub sym
362#define __SYSCALL_64(nr, sym, qual) __SYSCALL_64_QUAL_##qual(sym)
363#include <asm/syscalls_64.h>
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200364
Jan Beulich7effaa82005-09-12 18:49:24 +0200365/*
Brian Gerst01003012016-08-13 12:38:19 -0400366 * %rdi: prev task
367 * %rsi: next task
368 */
369ENTRY(__switch_to_asm)
370 /*
371 * Save callee-saved registers
372 * This must match the order in inactive_task_frame
373 */
374 pushq %rbp
375 pushq %rbx
376 pushq %r12
377 pushq %r13
378 pushq %r14
379 pushq %r15
380
381 /* switch stack */
382 movq %rsp, TASK_threadsp(%rdi)
383 movq TASK_threadsp(%rsi), %rsp
384
385#ifdef CONFIG_CC_STACKPROTECTOR
386 movq TASK_stack_canary(%rsi), %rbx
387 movq %rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset
388#endif
389
390 /* restore callee-saved registers */
391 popq %r15
392 popq %r14
393 popq %r13
394 popq %r12
395 popq %rbx
396 popq %rbp
397
398 jmp __switch_to
399END(__switch_to_asm)
400
401/*
Denys Vlasenko1eeb2072015-02-26 14:40:33 -0800402 * A newly forked process directly context switches into this address.
403 *
Brian Gerst01003012016-08-13 12:38:19 -0400404 * rax: prev task we switched from
Brian Gerst616d2482016-08-13 12:38:20 -0400405 * rbx: kernel thread func (NULL for user thread)
406 * r12: kernel thread arg
Denys Vlasenko1eeb2072015-02-26 14:40:33 -0800407 */
408ENTRY(ret_from_fork)
Josh Poimboeufff3f7e22017-01-09 12:00:25 -0600409 FRAME_BEGIN /* help unwinder find end of stack */
Brian Gerst01003012016-08-13 12:38:19 -0400410 movq %rax, %rdi
Josh Poimboeufff3f7e22017-01-09 12:00:25 -0600411 call schedule_tail /* rdi: 'prev' task parameter */
Denys Vlasenko1eeb2072015-02-26 14:40:33 -0800412
Josh Poimboeufff3f7e22017-01-09 12:00:25 -0600413 testq %rbx, %rbx /* from kernel_thread? */
414 jnz 1f /* kernel threads are uncommon */
Denys Vlasenko1eeb2072015-02-26 14:40:33 -0800415
Brian Gerst616d2482016-08-13 12:38:20 -04004162:
Josh Poimboeufff3f7e22017-01-09 12:00:25 -0600417 leaq FRAME_OFFSET(%rsp),%rdi /* pt_regs pointer */
Andy Lutomirski24d978b2016-01-28 15:11:27 -0800418 call syscall_return_slowpath /* returns with IRQs disabled */
419 TRACE_IRQS_ON /* user mode is traced as IRQS on */
420 SWAPGS
Josh Poimboeufff3f7e22017-01-09 12:00:25 -0600421 FRAME_END
Andy Lutomirski24d978b2016-01-28 15:11:27 -0800422 jmp restore_regs_and_iret
Brian Gerst616d2482016-08-13 12:38:20 -0400423
4241:
425 /* kernel thread */
426 movq %r12, %rdi
427 call *%rbx
428 /*
429 * A kernel thread is allowed to return here after successfully
430 * calling do_execve(). Exit to userspace to complete the execve()
431 * syscall.
432 */
433 movq $0, RAX(%rsp)
434 jmp 2b
Denys Vlasenko1eeb2072015-02-26 14:40:33 -0800435END(ret_from_fork)
436
437/*
Denys Vlasenko3304c9c2015-04-03 21:49:13 +0200438 * Build the entry stubs with some assembler magic.
439 * We pack 1 stub into every 8-byte block.
H. Peter Anvin939b7872008-11-11 13:51:52 -0800440 */
Denys Vlasenko3304c9c2015-04-03 21:49:13 +0200441 .align 8
H. Peter Anvin939b7872008-11-11 13:51:52 -0800442ENTRY(irq_entries_start)
Denys Vlasenko3304c9c2015-04-03 21:49:13 +0200443 vector=FIRST_EXTERNAL_VECTOR
444 .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR)
Ingo Molnar4d732132015-06-08 20:43:07 +0200445 pushq $(~vector+0x80) /* Note: always in signed byte range */
Denys Vlasenko3304c9c2015-04-03 21:49:13 +0200446 vector=vector+1
447 jmp common_interrupt
Denys Vlasenko3304c9c2015-04-03 21:49:13 +0200448 .align 8
449 .endr
H. Peter Anvin939b7872008-11-11 13:51:52 -0800450END(irq_entries_start)
451
Alexander van Heukelumd99015b2008-11-19 01:18:11 +0100452/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * Interrupt entry/exit.
454 *
455 * Interrupt entry points save only callee clobbered registers in fast path.
Alexander van Heukelumd99015b2008-11-19 01:18:11 +0100456 *
457 * Entry runs with interrupts off.
458 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Alexander van Heukelum722024d2008-11-13 13:50:20 +0100460/* 0(%rsp): ~(interrupt number) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 .macro interrupt func
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100462 cld
Andy Lutomirskiff467592015-07-03 12:44:29 -0700463 ALLOC_PT_GPREGS_ON_STACK
464 SAVE_C_REGS
465 SAVE_EXTRA_REGS
Josh Poimboeuf946c1912016-10-20 11:34:40 -0500466 ENCODE_FRAME_POINTER
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100467
Andy Lutomirskiff467592015-07-03 12:44:29 -0700468 testb $3, CS(%rsp)
Denys Vlasenkodde74f22015-04-27 15:21:51 +0200469 jz 1f
Andy Lutomirski02bc7762015-07-03 12:44:31 -0700470
471 /*
472 * IRQ from user mode. Switch to kernel gsbase and inform context
473 * tracking that we're in kernel mode.
474 */
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100475 SWAPGS
Andy Lutomirskif1075052015-11-12 12:59:00 -0800476
477 /*
478 * We need to tell lockdep that IRQs are off. We can't do this until
479 * we fix gsbase, and we should do it before enter_from_user_mode
480 * (which can take locks). Since TRACE_IRQS_OFF idempotent,
481 * the simplest way to handle it is to just call it twice if
482 * we enter from user mode. There's no reason to optimize this since
483 * TRACE_IRQS_OFF is a no-op if lockdep is off.
484 */
485 TRACE_IRQS_OFF
486
Andy Lutomirski478dc892015-11-12 12:59:04 -0800487 CALL_enter_from_user_mode
Andy Lutomirski02bc7762015-07-03 12:44:31 -0700488
Denys Vlasenko76f5df42015-02-26 14:40:27 -08004891:
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100490 /*
Denys Vlasenkoe90e1472015-02-26 14:40:28 -0800491 * Save previous stack pointer, optionally switch to interrupt stack.
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100492 * irq_count is used to check if a CPU is already on an interrupt stack
493 * or not. While this is essentially redundant with preempt_count it is
494 * a little cheaper to use a separate counter in the PDA (short of
495 * moving irq_enter into assembly, which would be too much work)
496 */
Andy Lutomirskia586f982015-07-03 12:44:30 -0700497 movq %rsp, %rdi
Ingo Molnar4d732132015-06-08 20:43:07 +0200498 incl PER_CPU_VAR(irq_count)
499 cmovzq PER_CPU_VAR(irq_stack_ptr), %rsp
Andy Lutomirskia586f982015-07-03 12:44:30 -0700500 pushq %rdi
Denys Vlasenkof6f64682015-01-08 17:25:15 +0100501 /* We entered an interrupt context - irqs are off: */
502 TRACE_IRQS_OFF
503
Andy Lutomirskia586f982015-07-03 12:44:30 -0700504 call \func /* rdi points to pt_regs */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 .endm
506
Alexander van Heukelum722024d2008-11-13 13:50:20 +0100507 /*
508 * The interrupt stubs push (~vector+0x80) onto the stack and
509 * then jump to common_interrupt.
510 */
H. Peter Anvin939b7872008-11-11 13:51:52 -0800511 .p2align CONFIG_X86_L1_CACHE_SHIFT
512common_interrupt:
Jan Beulichee4eb872012-11-02 11:18:39 +0000513 ASM_CLAC
Ingo Molnar4d732132015-06-08 20:43:07 +0200514 addq $-0x80, (%rsp) /* Adjust vector to [-256, -1] range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 interrupt do_IRQ
Denys Vlasenko34061f12015-03-23 14:03:59 +0100516 /* 0(%rsp): old RSP */
Jan Beulich7effaa82005-09-12 18:49:24 +0200517ret_from_intr:
Jan Beulich2140a992017-02-03 02:03:25 -0700518 DISABLE_INTERRUPTS(CLBR_ANY)
Ingo Molnar2601e642006-07-03 00:24:45 -0700519 TRACE_IRQS_OFF
Ingo Molnar4d732132015-06-08 20:43:07 +0200520 decl PER_CPU_VAR(irq_count)
Frederic Weisbecker625dbc3b2011-01-06 15:22:47 +0100521
Frederic Weisbeckera2bbe752011-07-02 16:52:45 +0200522 /* Restore saved previous stack */
Andy Lutomirskiff467592015-07-03 12:44:29 -0700523 popq %rsp
Frederic Weisbecker625dbc3b2011-01-06 15:22:47 +0100524
Denys Vlasenko03335e92015-04-27 15:21:52 +0200525 testb $3, CS(%rsp)
Denys Vlasenkodde74f22015-04-27 15:21:51 +0200526 jz retint_kernel
Andy Lutomirski02bc7762015-07-03 12:44:31 -0700527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 /* Interrupt came from user space */
Andy Lutomirski02bc7762015-07-03 12:44:31 -0700529GLOBAL(retint_user)
530 mov %rsp,%rdi
531 call prepare_exit_to_usermode
Ingo Molnar2601e642006-07-03 00:24:45 -0700532 TRACE_IRQS_IRETQ
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +0100533 SWAPGS
Andy Lutomirskiff467592015-07-03 12:44:29 -0700534 jmp restore_regs_and_iret
Ingo Molnar2601e642006-07-03 00:24:45 -0700535
Denys Vlasenko627276c2015-03-30 20:09:31 +0200536/* Returning to kernel space */
Denys Vlasenko6ba71b72015-03-31 19:00:05 +0200537retint_kernel:
Denys Vlasenko627276c2015-03-30 20:09:31 +0200538#ifdef CONFIG_PREEMPT
539 /* Interrupts are off */
540 /* Check if we need preemption */
Ingo Molnar4d732132015-06-08 20:43:07 +0200541 bt $9, EFLAGS(%rsp) /* were interrupts off? */
Denys Vlasenko6ba71b72015-03-31 19:00:05 +0200542 jnc 1f
Ingo Molnar4d732132015-06-08 20:43:07 +02005430: cmpl $0, PER_CPU_VAR(__preempt_count)
Denys Vlasenko36acef22015-03-31 19:00:07 +0200544 jnz 1f
Denys Vlasenko627276c2015-03-30 20:09:31 +0200545 call preempt_schedule_irq
Denys Vlasenko36acef22015-03-31 19:00:07 +0200546 jmp 0b
Denys Vlasenko6ba71b72015-03-31 19:00:05 +02005471:
Denys Vlasenko627276c2015-03-30 20:09:31 +0200548#endif
Ingo Molnar2601e642006-07-03 00:24:45 -0700549 /*
550 * The iretq could re-enable interrupts:
551 */
552 TRACE_IRQS_IRETQ
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200553
554/*
555 * At this label, code paths which return to kernel and to user,
556 * which come from interrupts/exception and from syscalls, merge.
557 */
Andy Lutomirskiee08c6b2015-10-05 17:48:09 -0700558GLOBAL(restore_regs_and_iret)
Andy Lutomirskiff467592015-07-03 12:44:29 -0700559 RESTORE_EXTRA_REGS
Denys Vlasenkofffbb5d2015-04-02 18:46:59 +0200560restore_c_regs_and_iret:
Denys Vlasenko76f5df42015-02-26 14:40:27 -0800561 RESTORE_C_REGS
562 REMOVE_PT_GPREGS_FROM_STACK 8
Andy Lutomirski7209a752014-07-23 08:34:11 -0700563 INTERRUPT_RETURN
564
565ENTRY(native_iret)
H. Peter Anvin3891a042014-04-29 16:46:09 -0700566 /*
567 * Are we returning to a stack segment from the LDT? Note: in
568 * 64-bit mode SS:RSP on the exception stack is always valid.
569 */
H. Peter Anvin34273f42014-05-04 10:36:22 -0700570#ifdef CONFIG_X86_ESPFIX64
Ingo Molnar4d732132015-06-08 20:43:07 +0200571 testb $4, (SS-RIP)(%rsp)
572 jnz native_irq_return_ldt
H. Peter Anvin34273f42014-05-04 10:36:22 -0700573#endif
H. Peter Anvin3891a042014-04-29 16:46:09 -0700574
Andy Lutomirskiaf726f22014-11-22 18:00:31 -0800575.global native_irq_return_iret
Andy Lutomirski7209a752014-07-23 08:34:11 -0700576native_irq_return_iret:
Andy Lutomirskib645af22014-11-22 18:00:33 -0800577 /*
578 * This may fault. Non-paranoid faults on return to userspace are
579 * handled by fixup_bad_iret. These include #SS, #GP, and #NP.
580 * Double-faults due to espfix64 are handled in do_double_fault.
581 * Other faults here are fatal.
582 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 iretq
Ingo Molnar3701d8632008-02-09 23:24:08 +0100584
H. Peter Anvin34273f42014-05-04 10:36:22 -0700585#ifdef CONFIG_X86_ESPFIX64
Andy Lutomirski7209a752014-07-23 08:34:11 -0700586native_irq_return_ldt:
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700587 /*
588 * We are running with user GSBASE. All GPRs contain their user
589 * values. We have a percpu ESPFIX stack that is eight slots
590 * long (see ESPFIX_STACK_SIZE). espfix_waddr points to the bottom
591 * of the ESPFIX stack.
592 *
593 * We clobber RAX and RDI in this code. We stash RDI on the
594 * normal stack and RAX on the ESPFIX stack.
595 *
596 * The ESPFIX stack layout we set up looks like this:
597 *
598 * --- top of ESPFIX stack ---
599 * SS
600 * RSP
601 * RFLAGS
602 * CS
603 * RIP <-- RSP points here when we're done
604 * RAX <-- espfix_waddr points here
605 * --- bottom of ESPFIX stack ---
606 */
607
608 pushq %rdi /* Stash user RDI */
H. Peter Anvin3891a042014-04-29 16:46:09 -0700609 SWAPGS
Ingo Molnar4d732132015-06-08 20:43:07 +0200610 movq PER_CPU_VAR(espfix_waddr), %rdi
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700611 movq %rax, (0*8)(%rdi) /* user RAX */
612 movq (1*8)(%rsp), %rax /* user RIP */
Ingo Molnar4d732132015-06-08 20:43:07 +0200613 movq %rax, (1*8)(%rdi)
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700614 movq (2*8)(%rsp), %rax /* user CS */
Ingo Molnar4d732132015-06-08 20:43:07 +0200615 movq %rax, (2*8)(%rdi)
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700616 movq (3*8)(%rsp), %rax /* user RFLAGS */
Ingo Molnar4d732132015-06-08 20:43:07 +0200617 movq %rax, (3*8)(%rdi)
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700618 movq (5*8)(%rsp), %rax /* user SS */
Ingo Molnar4d732132015-06-08 20:43:07 +0200619 movq %rax, (5*8)(%rdi)
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700620 movq (4*8)(%rsp), %rax /* user RSP */
Ingo Molnar4d732132015-06-08 20:43:07 +0200621 movq %rax, (4*8)(%rdi)
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700622 /* Now RAX == RSP. */
623
624 andl $0xffff0000, %eax /* RAX = (RSP & 0xffff0000) */
625 popq %rdi /* Restore user RDI */
626
627 /*
628 * espfix_stack[31:16] == 0. The page tables are set up such that
629 * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of
630 * espfix_waddr for any X. That is, there are 65536 RO aliases of
631 * the same page. Set up RSP so that RSP[31:16] contains the
632 * respective 16 bits of the /userspace/ RSP and RSP nonetheless
633 * still points to an RO alias of the ESPFIX stack.
634 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200635 orq PER_CPU_VAR(espfix_stack), %rax
H. Peter Anvin3891a042014-04-29 16:46:09 -0700636 SWAPGS
Ingo Molnar4d732132015-06-08 20:43:07 +0200637 movq %rax, %rsp
Andy Lutomirski85063fa2016-09-12 15:05:51 -0700638
639 /*
640 * At this point, we cannot write to the stack any more, but we can
641 * still read.
642 */
643 popq %rax /* Restore user RAX */
644
645 /*
646 * RSP now points to an ordinary IRET frame, except that the page
647 * is read-only and RSP[31:16] are preloaded with the userspace
648 * values. We can now IRET back to userspace.
649 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200650 jmp native_irq_return_iret
H. Peter Anvin34273f42014-05-04 10:36:22 -0700651#endif
Jan Beulich4b787e02006-06-26 13:56:55 +0200652END(common_interrupt)
H. Peter Anvin3891a042014-04-29 16:46:09 -0700653
Masami Hiramatsu8222d712009-08-27 13:23:25 -0400654/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 * APIC interrupts.
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100656 */
Seiji Aguchicf910e82013-06-20 11:46:53 -0400657.macro apicinterrupt3 num sym do_sym
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100658ENTRY(\sym)
Jan Beulichee4eb872012-11-02 11:18:39 +0000659 ASM_CLAC
Ingo Molnar4d732132015-06-08 20:43:07 +0200660 pushq $~(\num)
Jan Beulich39e95432011-11-29 11:03:46 +0000661.Lcommon_\sym:
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100662 interrupt \do_sym
Ingo Molnar4d732132015-06-08 20:43:07 +0200663 jmp ret_from_intr
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100664END(\sym)
665.endm
Jacob Shin89b831e2005-11-05 17:25:53 +0100666
Seiji Aguchicf910e82013-06-20 11:46:53 -0400667#ifdef CONFIG_TRACING
668#define trace(sym) trace_##sym
669#define smp_trace(sym) smp_trace_##sym
670
671.macro trace_apicinterrupt num sym
672apicinterrupt3 \num trace(\sym) smp_trace(\sym)
673.endm
674#else
675.macro trace_apicinterrupt num sym do_sym
676.endm
677#endif
678
Alexander Potapenko469f0022016-07-15 11:42:43 +0200679/* Make sure APIC interrupt handlers end up in the irqentry section: */
680#if defined(CONFIG_FUNCTION_GRAPH_TRACER) || defined(CONFIG_KASAN)
681# define PUSH_SECTION_IRQENTRY .pushsection .irqentry.text, "ax"
682# define POP_SECTION_IRQENTRY .popsection
683#else
684# define PUSH_SECTION_IRQENTRY
685# define POP_SECTION_IRQENTRY
686#endif
687
Seiji Aguchicf910e82013-06-20 11:46:53 -0400688.macro apicinterrupt num sym do_sym
Alexander Potapenko469f0022016-07-15 11:42:43 +0200689PUSH_SECTION_IRQENTRY
Seiji Aguchicf910e82013-06-20 11:46:53 -0400690apicinterrupt3 \num \sym \do_sym
691trace_apicinterrupt \num \sym
Alexander Potapenko469f0022016-07-15 11:42:43 +0200692POP_SECTION_IRQENTRY
Seiji Aguchicf910e82013-06-20 11:46:53 -0400693.endm
694
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100695#ifdef CONFIG_SMP
Ingo Molnar4d732132015-06-08 20:43:07 +0200696apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt
697apicinterrupt3 REBOOT_VECTOR reboot_interrupt smp_reboot_interrupt
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698#endif
699
Nick Piggin03b48632009-01-20 04:36:04 +0100700#ifdef CONFIG_X86_UV
Ingo Molnar4d732132015-06-08 20:43:07 +0200701apicinterrupt3 UV_BAU_MESSAGE uv_bau_message_intr1 uv_bau_message_interrupt
Nick Piggin03b48632009-01-20 04:36:04 +0100702#endif
Ingo Molnar4d732132015-06-08 20:43:07 +0200703
704apicinterrupt LOCAL_TIMER_VECTOR apic_timer_interrupt smp_apic_timer_interrupt
705apicinterrupt X86_PLATFORM_IPI_VECTOR x86_platform_ipi smp_x86_platform_ipi
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Yang Zhangd78f2662013-04-11 19:25:11 +0800707#ifdef CONFIG_HAVE_KVM
Ingo Molnar4d732132015-06-08 20:43:07 +0200708apicinterrupt3 POSTED_INTR_VECTOR kvm_posted_intr_ipi smp_kvm_posted_intr_ipi
709apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR kvm_posted_intr_wakeup_ipi smp_kvm_posted_intr_wakeup_ipi
Yang Zhangd78f2662013-04-11 19:25:11 +0800710#endif
711
Seiji Aguchi33e5ff62013-06-22 07:33:30 -0400712#ifdef CONFIG_X86_MCE_THRESHOLD
Ingo Molnar4d732132015-06-08 20:43:07 +0200713apicinterrupt THRESHOLD_APIC_VECTOR threshold_interrupt smp_threshold_interrupt
Seiji Aguchi33e5ff62013-06-22 07:33:30 -0400714#endif
715
Aravind Gopalakrishnan24fd78a2015-05-06 06:58:56 -0500716#ifdef CONFIG_X86_MCE_AMD
Ingo Molnar4d732132015-06-08 20:43:07 +0200717apicinterrupt DEFERRED_ERROR_VECTOR deferred_error_interrupt smp_deferred_error_interrupt
Aravind Gopalakrishnan24fd78a2015-05-06 06:58:56 -0500718#endif
719
Seiji Aguchi33e5ff62013-06-22 07:33:30 -0400720#ifdef CONFIG_X86_THERMAL_VECTOR
Ingo Molnar4d732132015-06-08 20:43:07 +0200721apicinterrupt THERMAL_APIC_VECTOR thermal_interrupt smp_thermal_interrupt
Seiji Aguchi33e5ff62013-06-22 07:33:30 -0400722#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100724#ifdef CONFIG_SMP
Ingo Molnar4d732132015-06-08 20:43:07 +0200725apicinterrupt CALL_FUNCTION_SINGLE_VECTOR call_function_single_interrupt smp_call_function_single_interrupt
726apicinterrupt CALL_FUNCTION_VECTOR call_function_interrupt smp_call_function_interrupt
727apicinterrupt RESCHEDULE_VECTOR reschedule_interrupt smp_reschedule_interrupt
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100728#endif
729
Ingo Molnar4d732132015-06-08 20:43:07 +0200730apicinterrupt ERROR_APIC_VECTOR error_interrupt smp_error_interrupt
731apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100732
Peter Zijlstrae360adb2010-10-14 14:01:34 +0800733#ifdef CONFIG_IRQ_WORK
Ingo Molnar4d732132015-06-08 20:43:07 +0200734apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt
Ingo Molnar241771e2008-12-03 10:39:53 +0100735#endif
736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737/*
738 * Exception entry points.
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100739 */
Andy Lutomirski9b476682015-03-05 19:19:07 -0800740#define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss) + (TSS_ist + ((x) - 1) * 8)
Andy Lutomirski577ed452014-05-21 15:07:09 -0700741
742.macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100743ENTRY(\sym)
Andy Lutomirski577ed452014-05-21 15:07:09 -0700744 /* Sanity check */
745 .if \shift_ist != -1 && \paranoid == 0
746 .error "using shift_ist requires paranoid=1"
747 .endif
748
Jan Beulichee4eb872012-11-02 11:18:39 +0000749 ASM_CLAC
Alexander van Heukelumb8b1d082008-11-21 16:44:28 +0100750 PARAVIRT_ADJUST_EXCEPTION_FRAME
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700751
752 .ifeq \has_error_code
Ingo Molnar4d732132015-06-08 20:43:07 +0200753 pushq $-1 /* ORIG_RAX: no syscall to restart */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700754 .endif
755
Denys Vlasenko76f5df42015-02-26 14:40:27 -0800756 ALLOC_PT_GPREGS_ON_STACK
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700757
758 .if \paranoid
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800759 .if \paranoid == 1
Ingo Molnar4d732132015-06-08 20:43:07 +0200760 testb $3, CS(%rsp) /* If coming from userspace, switch stacks */
761 jnz 1f
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800762 .endif
Ingo Molnar4d732132015-06-08 20:43:07 +0200763 call paranoid_entry
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700764 .else
Ingo Molnar4d732132015-06-08 20:43:07 +0200765 call error_entry
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700766 .endif
Denys Vlasenkoebfc4532015-02-26 14:40:34 -0800767 /* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700768
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700769 .if \paranoid
Andy Lutomirski577ed452014-05-21 15:07:09 -0700770 .if \shift_ist != -1
Ingo Molnar4d732132015-06-08 20:43:07 +0200771 TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */
Andy Lutomirski577ed452014-05-21 15:07:09 -0700772 .else
Alexander van Heukelumb8b1d082008-11-21 16:44:28 +0100773 TRACE_IRQS_OFF
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700774 .endif
Andy Lutomirski577ed452014-05-21 15:07:09 -0700775 .endif
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700776
Ingo Molnar4d732132015-06-08 20:43:07 +0200777 movq %rsp, %rdi /* pt_regs pointer */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700778
779 .if \has_error_code
Ingo Molnar4d732132015-06-08 20:43:07 +0200780 movq ORIG_RAX(%rsp), %rsi /* get error code */
781 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700782 .else
Ingo Molnar4d732132015-06-08 20:43:07 +0200783 xorl %esi, %esi /* no error code */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700784 .endif
785
Andy Lutomirski577ed452014-05-21 15:07:09 -0700786 .if \shift_ist != -1
Ingo Molnar4d732132015-06-08 20:43:07 +0200787 subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
Andy Lutomirski577ed452014-05-21 15:07:09 -0700788 .endif
789
Ingo Molnar4d732132015-06-08 20:43:07 +0200790 call \do_sym
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700791
Andy Lutomirski577ed452014-05-21 15:07:09 -0700792 .if \shift_ist != -1
Ingo Molnar4d732132015-06-08 20:43:07 +0200793 addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist)
Andy Lutomirski577ed452014-05-21 15:07:09 -0700794 .endif
795
Denys Vlasenkoebfc4532015-02-26 14:40:34 -0800796 /* these procedures expect "no swapgs" flag in ebx */
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700797 .if \paranoid
Ingo Molnar4d732132015-06-08 20:43:07 +0200798 jmp paranoid_exit
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700799 .else
Ingo Molnar4d732132015-06-08 20:43:07 +0200800 jmp error_exit
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700801 .endif
802
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800803 .if \paranoid == 1
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800804 /*
805 * Paranoid entry from userspace. Switch stacks and treat it
806 * as a normal entry. This means that paranoid handlers
807 * run in real process context if user_mode(regs).
808 */
8091:
Ingo Molnar4d732132015-06-08 20:43:07 +0200810 call error_entry
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800811
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800812
Ingo Molnar4d732132015-06-08 20:43:07 +0200813 movq %rsp, %rdi /* pt_regs pointer */
814 call sync_regs
815 movq %rax, %rsp /* switch stack */
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800816
Ingo Molnar4d732132015-06-08 20:43:07 +0200817 movq %rsp, %rdi /* pt_regs pointer */
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800818
819 .if \has_error_code
Ingo Molnar4d732132015-06-08 20:43:07 +0200820 movq ORIG_RAX(%rsp), %rsi /* get error code */
821 movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800822 .else
Ingo Molnar4d732132015-06-08 20:43:07 +0200823 xorl %esi, %esi /* no error code */
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800824 .endif
825
Ingo Molnar4d732132015-06-08 20:43:07 +0200826 call \do_sym
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800827
Ingo Molnar4d732132015-06-08 20:43:07 +0200828 jmp error_exit /* %ebx: no swapgs flag */
Andy Lutomirski48e08d02014-11-11 12:49:41 -0800829 .endif
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +0100830END(\sym)
Alexander van Heukelum322648d12008-11-23 10:08:28 +0100831.endm
Alexander van Heukelumb8b1d082008-11-21 16:44:28 +0100832
Seiji Aguchi25c74b102013-10-30 16:37:00 -0400833#ifdef CONFIG_TRACING
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700834.macro trace_idtentry sym do_sym has_error_code:req
835idtentry trace(\sym) trace(\do_sym) has_error_code=\has_error_code
836idtentry \sym \do_sym has_error_code=\has_error_code
Seiji Aguchi25c74b102013-10-30 16:37:00 -0400837.endm
838#else
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700839.macro trace_idtentry sym do_sym has_error_code:req
840idtentry \sym \do_sym has_error_code=\has_error_code
Seiji Aguchi25c74b102013-10-30 16:37:00 -0400841.endm
842#endif
843
Ingo Molnar4d732132015-06-08 20:43:07 +0200844idtentry divide_error do_divide_error has_error_code=0
845idtentry overflow do_overflow has_error_code=0
846idtentry bounds do_bounds has_error_code=0
847idtentry invalid_op do_invalid_op has_error_code=0
848idtentry device_not_available do_device_not_available has_error_code=0
849idtentry double_fault do_double_fault has_error_code=1 paranoid=2
850idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0
851idtentry invalid_TSS do_invalid_TSS has_error_code=1
852idtentry segment_not_present do_segment_not_present has_error_code=1
853idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0
854idtentry coprocessor_error do_coprocessor_error has_error_code=0
855idtentry alignment_check do_alignment_check has_error_code=1
856idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0
Andy Lutomirski5cec93c2011-06-05 13:50:24 -0400857
Ingo Molnar2601e642006-07-03 00:24:45 -0700858
Ingo Molnar4d732132015-06-08 20:43:07 +0200859 /*
860 * Reload gs selector with exception handling
861 * edi: new selector
862 */
Jeremy Fitzhardinge9f9d4892008-06-25 00:19:32 -0400863ENTRY(native_load_gs_index)
Ingo Molnar131484c2015-05-28 12:21:47 +0200864 pushfq
Jeremy Fitzhardingeb8aa2872009-01-28 14:35:03 -0800865 DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI)
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300866 SWAPGS
Borislav Petkov42c748bb2016-04-07 17:31:50 -0700867.Lgs_change:
Ingo Molnar4d732132015-06-08 20:43:07 +0200868 movl %edi, %gs
Borislav Petkov96e5d282016-04-07 17:31:49 -07008692: ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE
Glauber de Oliveira Costa72fe4852008-01-30 13:32:08 +0100870 SWAPGS
Ingo Molnar131484c2015-05-28 12:21:47 +0200871 popfq
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300872 ret
Alexander van Heukelum6efdcfaf2008-11-23 10:15:32 +0100873END(native_load_gs_index)
Al Viro784d5692016-01-11 11:04:34 -0500874EXPORT_SYMBOL(native_load_gs_index)
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100875
Borislav Petkov42c748bb2016-04-07 17:31:50 -0700876 _ASM_EXTABLE(.Lgs_change, bad_gs)
Ingo Molnar4d732132015-06-08 20:43:07 +0200877 .section .fixup, "ax"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 /* running with kernelgs */
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100879bad_gs:
Ingo Molnar4d732132015-06-08 20:43:07 +0200880 SWAPGS /* switch back to user gs */
Andy Lutomirskib038c842016-04-26 12:23:27 -0700881.macro ZAP_GS
882 /* This can't be a string because the preprocessor needs to see it. */
883 movl $__USER_DS, %eax
884 movl %eax, %gs
885.endm
886 ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG
Ingo Molnar4d732132015-06-08 20:43:07 +0200887 xorl %eax, %eax
888 movl %eax, %gs
889 jmp 2b
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300890 .previous
Alexander van Heukelum0bd7b792008-11-16 15:29:00 +0100891
Andi Kleen26995002006-08-02 22:37:28 +0200892/* Call softirq on interrupt stack. Interrupts are off. */
Frederic Weisbecker7d65f4a2013-09-05 15:49:45 +0200893ENTRY(do_softirq_own_stack)
Ingo Molnar4d732132015-06-08 20:43:07 +0200894 pushq %rbp
895 mov %rsp, %rbp
896 incl PER_CPU_VAR(irq_count)
897 cmove PER_CPU_VAR(irq_stack_ptr), %rsp
898 push %rbp /* frame pointer backlink */
899 call __do_softirq
Andi Kleen26995002006-08-02 22:37:28 +0200900 leaveq
Ingo Molnar4d732132015-06-08 20:43:07 +0200901 decl PER_CPU_VAR(irq_count)
Andi Kleened6b6762005-07-28 21:15:49 -0700902 ret
Frederic Weisbecker7d65f4a2013-09-05 15:49:45 +0200903END(do_softirq_own_stack)
Andi Kleen75154f42007-06-23 02:29:25 +0200904
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700905#ifdef CONFIG_XEN
Andy Lutomirskicb5dd2c2014-05-21 15:07:08 -0700906idtentry xen_hypervisor_callback xen_do_hypervisor_callback has_error_code=0
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700907
908/*
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300909 * A note on the "critical region" in our callback handler.
910 * We want to avoid stacking callback handlers due to events occurring
911 * during handling of the last event. To do this, we keep events disabled
912 * until we've done all processing. HOWEVER, we must enable events before
913 * popping the stack frame (can't be done atomically) and so it would still
914 * be possible to get enough handler activations to overflow the stack.
915 * Although unlikely, bugs of that kind are hard to track down, so we'd
916 * like to avoid the possibility.
917 * So, on entry to the handler we detect whether we interrupted an
918 * existing activation in its critical region -- if so, we pop the current
919 * activation and restart the handler using the previous one.
920 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200921ENTRY(xen_do_hypervisor_callback) /* do_hypervisor_callback(struct *pt_regs) */
922
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300923/*
924 * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will
925 * see the correct pointer to the pt_regs
926 */
Ingo Molnar4d732132015-06-08 20:43:07 +0200927 movq %rdi, %rsp /* we don't return, adjust the stack frame */
92811: incl PER_CPU_VAR(irq_count)
929 movq %rsp, %rbp
930 cmovzq PER_CPU_VAR(irq_stack_ptr), %rsp
931 pushq %rbp /* frame pointer backlink */
932 call xen_evtchn_do_upcall
933 popq %rsp
934 decl PER_CPU_VAR(irq_count)
David Vrabelfdfd8112015-02-19 15:23:17 +0000935#ifndef CONFIG_PREEMPT
Ingo Molnar4d732132015-06-08 20:43:07 +0200936 call xen_maybe_preempt_hcall
David Vrabelfdfd8112015-02-19 15:23:17 +0000937#endif
Ingo Molnar4d732132015-06-08 20:43:07 +0200938 jmp error_exit
Alexander van Heukelum371c3942011-03-11 21:59:38 +0100939END(xen_do_hypervisor_callback)
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700940
941/*
Cyrill Gorcunov9f1e87e2008-11-27 21:10:08 +0300942 * Hypervisor uses this for application faults while it executes.
943 * We get here for two reasons:
944 * 1. Fault while reloading DS, ES, FS or GS
945 * 2. Fault while executing IRET
946 * Category 1 we do not need to fix up as Xen has already reloaded all segment
947 * registers that could be reloaded and zeroed the others.
948 * Category 2 we fix up by killing the current process. We cannot use the
949 * normal Linux return path in this case because if we use the IRET hypercall
950 * to pop the stack frame we end up in an infinite loop of failsafe callbacks.
951 * We distinguish between categories by comparing each saved segment register
952 * with its current contents: any discrepancy means we in category 1.
953 */
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700954ENTRY(xen_failsafe_callback)
Ingo Molnar4d732132015-06-08 20:43:07 +0200955 movl %ds, %ecx
956 cmpw %cx, 0x10(%rsp)
957 jne 1f
958 movl %es, %ecx
959 cmpw %cx, 0x18(%rsp)
960 jne 1f
961 movl %fs, %ecx
962 cmpw %cx, 0x20(%rsp)
963 jne 1f
964 movl %gs, %ecx
965 cmpw %cx, 0x28(%rsp)
966 jne 1f
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700967 /* All segments match their saved values => Category 2 (Bad IRET). */
Ingo Molnar4d732132015-06-08 20:43:07 +0200968 movq (%rsp), %rcx
969 movq 8(%rsp), %r11
970 addq $0x30, %rsp
971 pushq $0 /* RIP */
972 pushq %r11
973 pushq %rcx
974 jmp general_protection
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -07009751: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */
Ingo Molnar4d732132015-06-08 20:43:07 +0200976 movq (%rsp), %rcx
977 movq 8(%rsp), %r11
978 addq $0x30, %rsp
979 pushq $-1 /* orig_ax = -1 => not a system call */
Denys Vlasenko76f5df42015-02-26 14:40:27 -0800980 ALLOC_PT_GPREGS_ON_STACK
981 SAVE_C_REGS
982 SAVE_EXTRA_REGS
Josh Poimboeuf946c1912016-10-20 11:34:40 -0500983 ENCODE_FRAME_POINTER
Ingo Molnar4d732132015-06-08 20:43:07 +0200984 jmp error_exit
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700985END(xen_failsafe_callback)
986
Seiji Aguchicf910e82013-06-20 11:46:53 -0400987apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
Sheng Yang38e20b02010-05-14 12:40:51 +0100988 xen_hvm_callback_vector xen_evtchn_do_upcall
989
Jeremy Fitzhardinge3d75e1b2008-07-08 15:06:49 -0700990#endif /* CONFIG_XEN */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +0100991
K. Y. Srinivasanbc2b0332013-02-03 17:22:39 -0800992#if IS_ENABLED(CONFIG_HYPERV)
Seiji Aguchicf910e82013-06-20 11:46:53 -0400993apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \
K. Y. Srinivasanbc2b0332013-02-03 17:22:39 -0800994 hyperv_callback_vector hyperv_vector_handler
995#endif /* CONFIG_HYPERV */
996
Ingo Molnar4d732132015-06-08 20:43:07 +0200997idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
998idtentry int3 do_int3 has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK
999idtentry stack_segment do_stack_segment has_error_code=1
1000
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -07001001#ifdef CONFIG_XEN
Ingo Molnar4d732132015-06-08 20:43:07 +02001002idtentry xen_debug do_debug has_error_code=0
1003idtentry xen_int3 do_int3 has_error_code=0
1004idtentry xen_stack_segment do_stack_segment has_error_code=1
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -07001005#endif
Ingo Molnar4d732132015-06-08 20:43:07 +02001006
1007idtentry general_protection do_general_protection has_error_code=1
1008trace_idtentry page_fault do_page_fault has_error_code=1
1009
Gleb Natapov631bc482010-10-14 11:22:52 +02001010#ifdef CONFIG_KVM_GUEST
Ingo Molnar4d732132015-06-08 20:43:07 +02001011idtentry async_page_fault do_async_page_fault has_error_code=1
Gleb Natapov631bc482010-10-14 11:22:52 +02001012#endif
Ingo Molnar4d732132015-06-08 20:43:07 +02001013
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001014#ifdef CONFIG_X86_MCE
Ingo Molnar4d732132015-06-08 20:43:07 +02001015idtentry machine_check has_error_code=0 paranoid=1 do_sym=*machine_check_vector(%rip)
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001016#endif
1017
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001018/*
1019 * Save all registers in pt_regs, and switch gs if needed.
1020 * Use slow, but surefire "are we in kernel?" check.
1021 * Return: ebx=0: need swapgs on exit, ebx=1: otherwise
1022 */
1023ENTRY(paranoid_entry)
Denys Vlasenko1eeb2072015-02-26 14:40:33 -08001024 cld
1025 SAVE_C_REGS 8
1026 SAVE_EXTRA_REGS 8
Josh Poimboeuf946c1912016-10-20 11:34:40 -05001027 ENCODE_FRAME_POINTER 8
Ingo Molnar4d732132015-06-08 20:43:07 +02001028 movl $1, %ebx
1029 movl $MSR_GS_BASE, %ecx
Denys Vlasenko1eeb2072015-02-26 14:40:33 -08001030 rdmsr
Ingo Molnar4d732132015-06-08 20:43:07 +02001031 testl %edx, %edx
1032 js 1f /* negative -> in kernel */
Denys Vlasenko1eeb2072015-02-26 14:40:33 -08001033 SWAPGS
Ingo Molnar4d732132015-06-08 20:43:07 +02001034 xorl %ebx, %ebx
Denys Vlasenko1eeb2072015-02-26 14:40:33 -080010351: ret
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001036END(paranoid_entry)
Denys Vlasenko1eeb2072015-02-26 14:40:33 -08001037
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001038/*
1039 * "Paranoid" exit path from exception stack. This is invoked
1040 * only on return from non-NMI IST interrupts that came
1041 * from kernel space.
1042 *
1043 * We may be returning to very strange contexts (e.g. very early
1044 * in syscall entry), so checking for preemption here would
1045 * be complicated. Fortunately, we there's no good reason
1046 * to try to handle preemption here.
Ingo Molnar4d732132015-06-08 20:43:07 +02001047 *
1048 * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it)
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001049 */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001050ENTRY(paranoid_exit)
Jan Beulich2140a992017-02-03 02:03:25 -07001051 DISABLE_INTERRUPTS(CLBR_ANY)
Steven Rostedt5963e312012-05-30 11:54:53 -04001052 TRACE_IRQS_OFF_DEBUG
Ingo Molnar4d732132015-06-08 20:43:07 +02001053 testl %ebx, %ebx /* swapgs needed? */
1054 jnz paranoid_exit_no_swapgs
Denys Vlasenkof2db9382015-02-26 14:40:30 -08001055 TRACE_IRQS_IRETQ
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001056 SWAPGS_UNSAFE_STACK
Ingo Molnar4d732132015-06-08 20:43:07 +02001057 jmp paranoid_exit_restore
Denys Vlasenko0d550832015-02-26 14:40:29 -08001058paranoid_exit_no_swapgs:
Denys Vlasenkof2db9382015-02-26 14:40:30 -08001059 TRACE_IRQS_IRETQ_DEBUG
Denys Vlasenko0d550832015-02-26 14:40:29 -08001060paranoid_exit_restore:
Denys Vlasenko76f5df42015-02-26 14:40:27 -08001061 RESTORE_EXTRA_REGS
1062 RESTORE_C_REGS
1063 REMOVE_PT_GPREGS_FROM_STACK 8
Andy Lutomirski48e08d02014-11-11 12:49:41 -08001064 INTERRUPT_RETURN
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001065END(paranoid_exit)
1066
1067/*
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001068 * Save all registers in pt_regs, and switch gs if needed.
Andy Lutomirski539f5112015-06-09 12:36:01 -07001069 * Return: EBX=0: came from user mode; EBX=1: otherwise
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001070 */
1071ENTRY(error_entry)
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001072 cld
Denys Vlasenko76f5df42015-02-26 14:40:27 -08001073 SAVE_C_REGS 8
1074 SAVE_EXTRA_REGS 8
Josh Poimboeuf946c1912016-10-20 11:34:40 -05001075 ENCODE_FRAME_POINTER 8
Ingo Molnar4d732132015-06-08 20:43:07 +02001076 xorl %ebx, %ebx
Denys Vlasenko03335e92015-04-27 15:21:52 +02001077 testb $3, CS+8(%rsp)
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001078 jz .Lerror_kernelspace
Andy Lutomirski539f5112015-06-09 12:36:01 -07001079
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001080 /*
1081 * We entered from user mode or we're pretending to have entered
1082 * from user mode due to an IRET fault.
1083 */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001084 SWAPGS
Andy Lutomirski539f5112015-06-09 12:36:01 -07001085
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001086.Lerror_entry_from_usermode_after_swapgs:
Andy Lutomirskif1075052015-11-12 12:59:00 -08001087 /*
1088 * We need to tell lockdep that IRQs are off. We can't do this until
1089 * we fix gsbase, and we should do it before enter_from_user_mode
1090 * (which can take locks).
1091 */
1092 TRACE_IRQS_OFF
Andy Lutomirski478dc892015-11-12 12:59:04 -08001093 CALL_enter_from_user_mode
Andy Lutomirskif1075052015-11-12 12:59:00 -08001094 ret
Andy Lutomirski02bc7762015-07-03 12:44:31 -07001095
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001096.Lerror_entry_done:
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001097 TRACE_IRQS_OFF
1098 ret
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001099
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001100 /*
1101 * There are two places in the kernel that can potentially fault with
1102 * usergs. Handle them here. B stepping K8s sometimes report a
1103 * truncated RIP for IRET exceptions returning to compat mode. Check
1104 * for these here too.
1105 */
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001106.Lerror_kernelspace:
Ingo Molnar4d732132015-06-08 20:43:07 +02001107 incl %ebx
1108 leaq native_irq_return_iret(%rip), %rcx
1109 cmpq %rcx, RIP+8(%rsp)
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001110 je .Lerror_bad_iret
Ingo Molnar4d732132015-06-08 20:43:07 +02001111 movl %ecx, %eax /* zero extend */
1112 cmpq %rax, RIP+8(%rsp)
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001113 je .Lbstep_iret
Borislav Petkov42c748bb2016-04-07 17:31:50 -07001114 cmpq $.Lgs_change, RIP+8(%rsp)
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001115 jne .Lerror_entry_done
Andy Lutomirski539f5112015-06-09 12:36:01 -07001116
1117 /*
Borislav Petkov42c748bb2016-04-07 17:31:50 -07001118 * hack: .Lgs_change can fail with user gsbase. If this happens, fix up
Andy Lutomirski539f5112015-06-09 12:36:01 -07001119 * gsbase and proceed. We'll fix up the exception and land in
Borislav Petkov42c748bb2016-04-07 17:31:50 -07001120 * .Lgs_change's error handler with kernel gsbase.
Andy Lutomirski539f5112015-06-09 12:36:01 -07001121 */
Wanpeng Li2fa5f042016-09-30 09:01:06 +08001122 SWAPGS
1123 jmp .Lerror_entry_done
Brian Gerstae24ffe2009-10-12 10:18:23 -04001124
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001125.Lbstep_iret:
Brian Gerstae24ffe2009-10-12 10:18:23 -04001126 /* Fix truncated RIP */
Ingo Molnar4d732132015-06-08 20:43:07 +02001127 movq %rcx, RIP+8(%rsp)
Andy Lutomirskib645af22014-11-22 18:00:33 -08001128 /* fall through */
1129
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001130.Lerror_bad_iret:
Andy Lutomirski539f5112015-06-09 12:36:01 -07001131 /*
1132 * We came from an IRET to user mode, so we have user gsbase.
1133 * Switch to kernel gsbase:
1134 */
Andy Lutomirskib645af22014-11-22 18:00:33 -08001135 SWAPGS
Andy Lutomirski539f5112015-06-09 12:36:01 -07001136
1137 /*
1138 * Pretend that the exception came from user mode: set up pt_regs
1139 * as if we faulted immediately after IRET and clear EBX so that
1140 * error_exit knows that we will be returning to user mode.
1141 */
Ingo Molnar4d732132015-06-08 20:43:07 +02001142 mov %rsp, %rdi
1143 call fixup_bad_iret
1144 mov %rax, %rsp
Andy Lutomirski539f5112015-06-09 12:36:01 -07001145 decl %ebx
Andy Lutomirskicb6f64e2015-07-03 12:44:27 -07001146 jmp .Lerror_entry_from_usermode_after_swapgs
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001147END(error_entry)
1148
1149
Andy Lutomirski539f5112015-06-09 12:36:01 -07001150/*
Nicolas Iooss75ca5b22016-07-29 13:39:51 +02001151 * On entry, EBX is a "return to kernel mode" flag:
Andy Lutomirski539f5112015-06-09 12:36:01 -07001152 * 1: already in kernel mode, don't need SWAPGS
1153 * 0: user gsbase is loaded, we need SWAPGS and standard preparation for return to usermode
1154 */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001155ENTRY(error_exit)
Jan Beulich2140a992017-02-03 02:03:25 -07001156 DISABLE_INTERRUPTS(CLBR_ANY)
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001157 TRACE_IRQS_OFF
Jan Beulich2140a992017-02-03 02:03:25 -07001158 testl %ebx, %ebx
Ingo Molnar4d732132015-06-08 20:43:07 +02001159 jnz retint_kernel
1160 jmp retint_user
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001161END(error_exit)
1162
Denys Vlasenko0784b362015-04-01 16:50:57 +02001163/* Runs on exception stack */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001164ENTRY(nmi)
Andy Lutomirskifc57a7c2015-09-20 16:32:04 -07001165 /*
1166 * Fix up the exception frame if we're on Xen.
1167 * PARAVIRT_ADJUST_EXCEPTION_FRAME is guaranteed to push at most
1168 * one value to the stack on native, so it may clobber the rdx
1169 * scratch slot, but it won't clobber any of the important
1170 * slots past it.
1171 *
1172 * Xen is a different story, because the Xen frame itself overlaps
1173 * the "NMI executing" variable.
1174 */
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001175 PARAVIRT_ADJUST_EXCEPTION_FRAME
Andy Lutomirskifc57a7c2015-09-20 16:32:04 -07001176
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001177 /*
1178 * We allow breakpoints in NMIs. If a breakpoint occurs, then
1179 * the iretq it performs will take us out of NMI context.
1180 * This means that we can have nested NMIs where the next
1181 * NMI is using the top of the stack of the previous NMI. We
1182 * can't let it execute because the nested NMI will corrupt the
1183 * stack of the previous NMI. NMI handlers are not re-entrant
1184 * anyway.
1185 *
1186 * To handle this case we do the following:
1187 * Check the a special location on the stack that contains
1188 * a variable that is set when NMIs are executing.
1189 * The interrupted task's stack is also checked to see if it
1190 * is an NMI stack.
1191 * If the variable is not set and the stack is not the NMI
1192 * stack then:
1193 * o Set the special variable on the stack
Andy Lutomirski0b229302015-07-15 10:29:36 -07001194 * o Copy the interrupt frame into an "outermost" location on the
1195 * stack
1196 * o Copy the interrupt frame into an "iret" location on the stack
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001197 * o Continue processing the NMI
1198 * If the variable is set or the previous stack is the NMI stack:
Andy Lutomirski0b229302015-07-15 10:29:36 -07001199 * o Modify the "iret" location to jump to the repeat_nmi
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001200 * o return back to the first NMI
1201 *
1202 * Now on exit of the first NMI, we first clear the stack variable
1203 * The NMI stack will tell any nested NMIs at that point that it is
1204 * nested. Then we pop the stack normally with iret, and if there was
1205 * a nested NMI that updated the copy interrupt stack frame, a
1206 * jump will be made to the repeat_nmi code that will handle the second
1207 * NMI.
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001208 *
1209 * However, espfix prevents us from directly returning to userspace
1210 * with a single IRET instruction. Similarly, IRET to user mode
1211 * can fault. We therefore handle NMIs from user space like
1212 * other IST entries.
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001213 */
1214
Denys Vlasenko146b2b02015-03-25 18:18:13 +01001215 /* Use %rdx as our temp variable throughout */
Ingo Molnar4d732132015-06-08 20:43:07 +02001216 pushq %rdx
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001217
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001218 testb $3, CS-RIP+8(%rsp)
1219 jz .Lnmi_from_kernel
Steven Rostedt45d5a162012-02-19 16:43:37 -05001220
1221 /*
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001222 * NMI from user mode. We need to run on the thread stack, but we
1223 * can't go through the normal entry paths: NMIs are masked, and
1224 * we don't want to enable interrupts, because then we'll end
1225 * up in an awkward situation in which IRQs are on but NMIs
1226 * are off.
Andy Lutomirski83c133c2015-09-20 16:32:05 -07001227 *
1228 * We also must not push anything to the stack before switching
1229 * stacks lest we corrupt the "NMI executing" variable.
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001230 */
1231
Andy Lutomirski83c133c2015-09-20 16:32:05 -07001232 SWAPGS_UNSAFE_STACK
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001233 cld
1234 movq %rsp, %rdx
1235 movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp
1236 pushq 5*8(%rdx) /* pt_regs->ss */
1237 pushq 4*8(%rdx) /* pt_regs->rsp */
1238 pushq 3*8(%rdx) /* pt_regs->flags */
1239 pushq 2*8(%rdx) /* pt_regs->cs */
1240 pushq 1*8(%rdx) /* pt_regs->rip */
1241 pushq $-1 /* pt_regs->orig_ax */
1242 pushq %rdi /* pt_regs->di */
1243 pushq %rsi /* pt_regs->si */
1244 pushq (%rdx) /* pt_regs->dx */
1245 pushq %rcx /* pt_regs->cx */
1246 pushq %rax /* pt_regs->ax */
1247 pushq %r8 /* pt_regs->r8 */
1248 pushq %r9 /* pt_regs->r9 */
1249 pushq %r10 /* pt_regs->r10 */
1250 pushq %r11 /* pt_regs->r11 */
1251 pushq %rbx /* pt_regs->rbx */
1252 pushq %rbp /* pt_regs->rbp */
1253 pushq %r12 /* pt_regs->r12 */
1254 pushq %r13 /* pt_regs->r13 */
1255 pushq %r14 /* pt_regs->r14 */
1256 pushq %r15 /* pt_regs->r15 */
Josh Poimboeuf946c1912016-10-20 11:34:40 -05001257 ENCODE_FRAME_POINTER
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001258
1259 /*
1260 * At this point we no longer need to worry about stack damage
1261 * due to nesting -- we're on the normal thread stack and we're
1262 * done with the NMI stack.
1263 */
1264
1265 movq %rsp, %rdi
1266 movq $-1, %rsi
1267 call do_nmi
1268
1269 /*
1270 * Return back to user mode. We must *not* do the normal exit
Josh Poimboeuf946c1912016-10-20 11:34:40 -05001271 * work, because we don't want to enable interrupts.
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001272 */
1273 SWAPGS
Josh Poimboeuf946c1912016-10-20 11:34:40 -05001274 jmp restore_regs_and_iret
Andy Lutomirski9b6e6a82015-07-15 10:29:35 -07001275
1276.Lnmi_from_kernel:
1277 /*
Andy Lutomirski0b229302015-07-15 10:29:36 -07001278 * Here's what our stack frame will look like:
1279 * +---------------------------------------------------------+
1280 * | original SS |
1281 * | original Return RSP |
1282 * | original RFLAGS |
1283 * | original CS |
1284 * | original RIP |
1285 * +---------------------------------------------------------+
1286 * | temp storage for rdx |
1287 * +---------------------------------------------------------+
1288 * | "NMI executing" variable |
1289 * +---------------------------------------------------------+
1290 * | iret SS } Copied from "outermost" frame |
1291 * | iret Return RSP } on each loop iteration; overwritten |
1292 * | iret RFLAGS } by a nested NMI to force another |
1293 * | iret CS } iteration if needed. |
1294 * | iret RIP } |
1295 * +---------------------------------------------------------+
1296 * | outermost SS } initialized in first_nmi; |
1297 * | outermost Return RSP } will not be changed before |
1298 * | outermost RFLAGS } NMI processing is done. |
1299 * | outermost CS } Copied to "iret" frame on each |
1300 * | outermost RIP } iteration. |
1301 * +---------------------------------------------------------+
1302 * | pt_regs |
1303 * +---------------------------------------------------------+
1304 *
1305 * The "original" frame is used by hardware. Before re-enabling
1306 * NMIs, we need to be done with it, and we need to leave enough
1307 * space for the asm code here.
1308 *
1309 * We return by executing IRET while RSP points to the "iret" frame.
1310 * That will either return for real or it will loop back into NMI
1311 * processing.
1312 *
1313 * The "outermost" frame is copied to the "iret" frame on each
1314 * iteration of the loop, so each iteration starts with the "iret"
1315 * frame pointing to the final return target.
1316 */
1317
1318 /*
1319 * Determine whether we're a nested NMI.
1320 *
Andy Lutomirskia27507c2015-07-15 10:29:37 -07001321 * If we interrupted kernel code between repeat_nmi and
1322 * end_repeat_nmi, then we are a nested NMI. We must not
1323 * modify the "iret" frame because it's being written by
1324 * the outer NMI. That's okay; the outer NMI handler is
1325 * about to about to call do_nmi anyway, so we can just
1326 * resume the outer NMI.
1327 */
1328
1329 movq $repeat_nmi, %rdx
1330 cmpq 8(%rsp), %rdx
1331 ja 1f
1332 movq $end_repeat_nmi, %rdx
1333 cmpq 8(%rsp), %rdx
1334 ja nested_nmi_out
13351:
1336
1337 /*
1338 * Now check "NMI executing". If it's set, then we're nested.
Andy Lutomirski0b229302015-07-15 10:29:36 -07001339 * This will not detect if we interrupted an outer NMI just
1340 * before IRET.
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001341 */
Ingo Molnar4d732132015-06-08 20:43:07 +02001342 cmpl $1, -8(%rsp)
1343 je nested_nmi
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001344
1345 /*
Andy Lutomirski0b229302015-07-15 10:29:36 -07001346 * Now test if the previous stack was an NMI stack. This covers
1347 * the case where we interrupt an outer NMI after it clears
Andy Lutomirski810bc072015-07-15 10:29:38 -07001348 * "NMI executing" but before IRET. We need to be careful, though:
1349 * there is one case in which RSP could point to the NMI stack
1350 * despite there being no NMI active: naughty userspace controls
1351 * RSP at the very beginning of the SYSCALL targets. We can
1352 * pull a fast one on naughty userspace, though: we program
1353 * SYSCALL to mask DF, so userspace cannot cause DF to be set
1354 * if it controls the kernel's RSP. We set DF before we clear
1355 * "NMI executing".
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001356 */
Denys Vlasenko0784b362015-04-01 16:50:57 +02001357 lea 6*8(%rsp), %rdx
1358 /* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */
1359 cmpq %rdx, 4*8(%rsp)
1360 /* If the stack pointer is above the NMI stack, this is a normal NMI */
1361 ja first_nmi
Ingo Molnar4d732132015-06-08 20:43:07 +02001362
Denys Vlasenko0784b362015-04-01 16:50:57 +02001363 subq $EXCEPTION_STKSZ, %rdx
1364 cmpq %rdx, 4*8(%rsp)
1365 /* If it is below the NMI stack, it is a normal NMI */
1366 jb first_nmi
Andy Lutomirski810bc072015-07-15 10:29:38 -07001367
1368 /* Ah, it is within the NMI stack. */
1369
1370 testb $(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp)
1371 jz first_nmi /* RSP was user controlled. */
1372
1373 /* This is a nested NMI. */
Denys Vlasenko0784b362015-04-01 16:50:57 +02001374
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001375nested_nmi:
1376 /*
Andy Lutomirski0b229302015-07-15 10:29:36 -07001377 * Modify the "iret" frame to point to repeat_nmi, forcing another
1378 * iteration of NMI handling.
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001379 */
Andy Lutomirski23a781e2015-07-15 10:29:39 -07001380 subq $8, %rsp
Ingo Molnar4d732132015-06-08 20:43:07 +02001381 leaq -10*8(%rsp), %rdx
1382 pushq $__KERNEL_DS
1383 pushq %rdx
Ingo Molnar131484c2015-05-28 12:21:47 +02001384 pushfq
Ingo Molnar4d732132015-06-08 20:43:07 +02001385 pushq $__KERNEL_CS
1386 pushq $repeat_nmi
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001387
1388 /* Put stack back */
Ingo Molnar4d732132015-06-08 20:43:07 +02001389 addq $(6*8), %rsp
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001390
1391nested_nmi_out:
Ingo Molnar4d732132015-06-08 20:43:07 +02001392 popq %rdx
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001393
Andy Lutomirski0b229302015-07-15 10:29:36 -07001394 /* We are returning to kernel mode, so this cannot result in a fault. */
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001395 INTERRUPT_RETURN
1396
1397first_nmi:
Andy Lutomirski0b229302015-07-15 10:29:36 -07001398 /* Restore rdx. */
Ingo Molnar4d732132015-06-08 20:43:07 +02001399 movq (%rsp), %rdx
Jan Beulich62610912012-02-24 14:54:37 +00001400
Andy Lutomirski36f1a772015-07-15 10:29:40 -07001401 /* Make room for "NMI executing". */
1402 pushq $0
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001403
Andy Lutomirski0b229302015-07-15 10:29:36 -07001404 /* Leave room for the "iret" frame */
Ingo Molnar4d732132015-06-08 20:43:07 +02001405 subq $(5*8), %rsp
Salman Qazi28696f42012-10-01 17:29:25 -07001406
Andy Lutomirski0b229302015-07-15 10:29:36 -07001407 /* Copy the "original" frame to the "outermost" frame */
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001408 .rept 5
Ingo Molnar4d732132015-06-08 20:43:07 +02001409 pushq 11*8(%rsp)
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001410 .endr
Jan Beulich62610912012-02-24 14:54:37 +00001411
Steven Rostedt79fb4ad2012-02-24 15:55:13 -05001412 /* Everything up to here is safe from nested NMIs */
1413
Andy Lutomirskia97439a2015-07-15 10:29:41 -07001414#ifdef CONFIG_DEBUG_ENTRY
1415 /*
1416 * For ease of testing, unmask NMIs right away. Disabled by
1417 * default because IRET is very expensive.
1418 */
1419 pushq $0 /* SS */
1420 pushq %rsp /* RSP (minus 8 because of the previous push) */
1421 addq $8, (%rsp) /* Fix up RSP */
1422 pushfq /* RFLAGS */
1423 pushq $__KERNEL_CS /* CS */
1424 pushq $1f /* RIP */
1425 INTERRUPT_RETURN /* continues at repeat_nmi below */
14261:
1427#endif
1428
Andy Lutomirski0b229302015-07-15 10:29:36 -07001429repeat_nmi:
Jan Beulich62610912012-02-24 14:54:37 +00001430 /*
1431 * If there was a nested NMI, the first NMI's iret will return
1432 * here. But NMIs are still enabled and we can take another
1433 * nested NMI. The nested NMI checks the interrupted RIP to see
1434 * if it is between repeat_nmi and end_repeat_nmi, and if so
1435 * it will just return, as we are about to repeat an NMI anyway.
1436 * This makes it safe to copy to the stack frame that a nested
1437 * NMI will update.
Andy Lutomirski0b229302015-07-15 10:29:36 -07001438 *
1439 * RSP is pointing to "outermost RIP". gsbase is unknown, but, if
1440 * we're repeating an NMI, gsbase has the same value that it had on
1441 * the first iteration. paranoid_entry will load the kernel
Andy Lutomirski36f1a772015-07-15 10:29:40 -07001442 * gsbase if needed before we call do_nmi. "NMI executing"
1443 * is zero.
Jan Beulich62610912012-02-24 14:54:37 +00001444 */
Andy Lutomirski36f1a772015-07-15 10:29:40 -07001445 movq $1, 10*8(%rsp) /* Set "NMI executing". */
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001446
Andy Lutomirski0b229302015-07-15 10:29:36 -07001447 /*
1448 * Copy the "outermost" frame to the "iret" frame. NMIs that nest
1449 * here must not modify the "iret" frame while we're writing to
1450 * it or it will end up containing garbage.
1451 */
Ingo Molnar4d732132015-06-08 20:43:07 +02001452 addq $(10*8), %rsp
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001453 .rept 5
Ingo Molnar4d732132015-06-08 20:43:07 +02001454 pushq -6*8(%rsp)
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001455 .endr
Ingo Molnar4d732132015-06-08 20:43:07 +02001456 subq $(5*8), %rsp
Jan Beulich62610912012-02-24 14:54:37 +00001457end_repeat_nmi:
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001458
1459 /*
Andy Lutomirski0b229302015-07-15 10:29:36 -07001460 * Everything below this point can be preempted by a nested NMI.
1461 * If this happens, then the inner NMI will change the "iret"
1462 * frame to point back to repeat_nmi.
Steven Rostedt3f3c8b82011-12-08 12:36:23 -05001463 */
Ingo Molnar4d732132015-06-08 20:43:07 +02001464 pushq $-1 /* ORIG_RAX: no syscall to restart */
Denys Vlasenko76f5df42015-02-26 14:40:27 -08001465 ALLOC_PT_GPREGS_ON_STACK
1466
Steven Rostedt1fd466e2011-12-08 12:32:27 -05001467 /*
Denys Vlasenkoebfc4532015-02-26 14:40:34 -08001468 * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit
Steven Rostedt1fd466e2011-12-08 12:32:27 -05001469 * as we should not be calling schedule in NMI context.
1470 * Even with normal interrupts enabled. An NMI should not be
1471 * setting NEED_RESCHED or anything that normal interrupts and
1472 * exceptions might do.
1473 */
Ingo Molnar4d732132015-06-08 20:43:07 +02001474 call paranoid_entry
Steven Rostedt7fbb98c2012-06-07 10:21:21 -04001475
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001476 /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */
Ingo Molnar4d732132015-06-08 20:43:07 +02001477 movq %rsp, %rdi
1478 movq $-1, %rsi
1479 call do_nmi
Steven Rostedt7fbb98c2012-06-07 10:21:21 -04001480
Ingo Molnar4d732132015-06-08 20:43:07 +02001481 testl %ebx, %ebx /* swapgs needed? */
1482 jnz nmi_restore
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001483nmi_swapgs:
1484 SWAPGS_UNSAFE_STACK
1485nmi_restore:
Denys Vlasenko76f5df42015-02-26 14:40:27 -08001486 RESTORE_EXTRA_REGS
1487 RESTORE_C_REGS
Andy Lutomirski0b229302015-07-15 10:29:36 -07001488
1489 /* Point RSP at the "iret" frame. */
Denys Vlasenko76f5df42015-02-26 14:40:27 -08001490 REMOVE_PT_GPREGS_FROM_STACK 6*8
Salman Qazi28696f42012-10-01 17:29:25 -07001491
Andy Lutomirski810bc072015-07-15 10:29:38 -07001492 /*
1493 * Clear "NMI executing". Set DF first so that we can easily
1494 * distinguish the remaining code between here and IRET from
1495 * the SYSCALL entry and exit paths. On a native kernel, we
1496 * could just inspect RIP, but, on paravirt kernels,
1497 * INTERRUPT_RETURN can translate into a jump into a
1498 * hypercall page.
1499 */
1500 std
1501 movq $0, 5*8(%rsp) /* clear "NMI executing" */
Andy Lutomirski0b229302015-07-15 10:29:36 -07001502
1503 /*
1504 * INTERRUPT_RETURN reads the "iret" frame and exits the NMI
1505 * stack in a single instruction. We are returning to kernel
1506 * mode, so this cannot result in a fault.
1507 */
Andy Lutomirski5ca6f702015-06-04 13:24:29 -07001508 INTERRUPT_RETURN
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001509END(nmi)
1510
1511ENTRY(ignore_sysret)
Ingo Molnar4d732132015-06-08 20:43:07 +02001512 mov $-ENOSYS, %eax
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001513 sysret
Alexander van Heukelumddeb8f22008-11-24 13:24:28 +01001514END(ignore_sysret)
Andy Lutomirski2deb4be2016-07-14 13:22:55 -07001515
1516ENTRY(rewind_stack_do_exit)
1517 /* Prevent any naive code from trying to unwind to our caller. */
1518 xorl %ebp, %ebp
1519
1520 movq PER_CPU_VAR(cpu_current_top_of_stack), %rax
1521 leaq -TOP_OF_KERNEL_STACK_PADDING-PTREGS_SIZE(%rax), %rsp
1522
1523 call do_exit
15241: jmp 1b
1525END(rewind_stack_do_exit)