]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | /* |
3 | * linux/arch/x86_64/entry.S | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs | |
7 | * Copyright (C) 2000 Pavel Machek <pavel@suse.cz> | |
4d732138 | 8 | * |
1da177e4 LT |
9 | * entry.S contains the system-call and fault low-level handling routines. |
10 | * | |
8b4777a4 AL |
11 | * Some of this is documented in Documentation/x86/entry_64.txt |
12 | * | |
0bd7b798 | 13 | * A note on terminology: |
4d732138 IM |
14 | * - iret frame: Architecture defined interrupt frame from SS to RIP |
15 | * at the top of the kernel process stack. | |
2e91a17b AK |
16 | * |
17 | * Some macro usage: | |
4d732138 IM |
18 | * - ENTRY/END: Define functions in the symbol table. |
19 | * - TRACE_IRQ_*: Trace hardirq state for lock debugging. | |
20 | * - idtentry: Define exception entry points. | |
1da177e4 | 21 | */ |
1da177e4 LT |
22 | #include <linux/linkage.h> |
23 | #include <asm/segment.h> | |
1da177e4 LT |
24 | #include <asm/cache.h> |
25 | #include <asm/errno.h> | |
e2d5df93 | 26 | #include <asm/asm-offsets.h> |
1da177e4 LT |
27 | #include <asm/msr.h> |
28 | #include <asm/unistd.h> | |
29 | #include <asm/thread_info.h> | |
30 | #include <asm/hw_irq.h> | |
0341c14d | 31 | #include <asm/page_types.h> |
2601e64d | 32 | #include <asm/irqflags.h> |
72fe4858 | 33 | #include <asm/paravirt.h> |
9939ddaf | 34 | #include <asm/percpu.h> |
d7abc0fa | 35 | #include <asm/asm.h> |
63bcff2a | 36 | #include <asm/smap.h> |
3891a04a | 37 | #include <asm/pgtable_types.h> |
784d5699 | 38 | #include <asm/export.h> |
8c1f7558 | 39 | #include <asm/frame.h> |
2641f08b | 40 | #include <asm/nospec-branch.h> |
d7e7528b | 41 | #include <linux/err.h> |
1da177e4 | 42 | |
6fd166aa PZ |
43 | #include "calling.h" |
44 | ||
4d732138 IM |
45 | .code64 |
46 | .section .entry.text, "ax" | |
16444a8a | 47 | |
72fe4858 | 48 | #ifdef CONFIG_PARAVIRT |
2be29982 | 49 | ENTRY(native_usergs_sysret64) |
8c1f7558 | 50 | UNWIND_HINT_EMPTY |
72fe4858 GOC |
51 | swapgs |
52 | sysretq | |
8c1f7558 | 53 | END(native_usergs_sysret64) |
72fe4858 GOC |
54 | #endif /* CONFIG_PARAVIRT */ |
55 | ||
ca37e57b | 56 | .macro TRACE_IRQS_FLAGS flags:req |
2601e64d | 57 | #ifdef CONFIG_TRACE_IRQFLAGS |
a368d7fd | 58 | btl $9, \flags /* interrupts off? */ |
4d732138 | 59 | jnc 1f |
2601e64d IM |
60 | TRACE_IRQS_ON |
61 | 1: | |
62 | #endif | |
63 | .endm | |
64 | ||
ca37e57b AL |
65 | .macro TRACE_IRQS_IRETQ |
66 | TRACE_IRQS_FLAGS EFLAGS(%rsp) | |
67 | .endm | |
68 | ||
5963e317 SR |
69 | /* |
70 | * When dynamic function tracer is enabled it will add a breakpoint | |
71 | * to all locations that it is about to modify, sync CPUs, update | |
72 | * all the code, sync CPUs, then remove the breakpoints. In this time | |
73 | * if lockdep is enabled, it might jump back into the debug handler | |
74 | * outside the updating of the IST protection. (TRACE_IRQS_ON/OFF). | |
75 | * | |
76 | * We need to change the IDT table before calling TRACE_IRQS_ON/OFF to | |
77 | * make sure the stack pointer does not get reset back to the top | |
78 | * of the debug stack, and instead just reuses the current stack. | |
79 | */ | |
80 | #if defined(CONFIG_DYNAMIC_FTRACE) && defined(CONFIG_TRACE_IRQFLAGS) | |
81 | ||
82 | .macro TRACE_IRQS_OFF_DEBUG | |
4d732138 | 83 | call debug_stack_set_zero |
5963e317 | 84 | TRACE_IRQS_OFF |
4d732138 | 85 | call debug_stack_reset |
5963e317 SR |
86 | .endm |
87 | ||
88 | .macro TRACE_IRQS_ON_DEBUG | |
4d732138 | 89 | call debug_stack_set_zero |
5963e317 | 90 | TRACE_IRQS_ON |
4d732138 | 91 | call debug_stack_reset |
5963e317 SR |
92 | .endm |
93 | ||
f2db9382 | 94 | .macro TRACE_IRQS_IRETQ_DEBUG |
6709812f | 95 | btl $9, EFLAGS(%rsp) /* interrupts off? */ |
4d732138 | 96 | jnc 1f |
5963e317 SR |
97 | TRACE_IRQS_ON_DEBUG |
98 | 1: | |
99 | .endm | |
100 | ||
101 | #else | |
4d732138 IM |
102 | # define TRACE_IRQS_OFF_DEBUG TRACE_IRQS_OFF |
103 | # define TRACE_IRQS_ON_DEBUG TRACE_IRQS_ON | |
104 | # define TRACE_IRQS_IRETQ_DEBUG TRACE_IRQS_IRETQ | |
5963e317 SR |
105 | #endif |
106 | ||
1da177e4 | 107 | /* |
4d732138 | 108 | * 64-bit SYSCALL instruction entry. Up to 6 arguments in registers. |
1da177e4 | 109 | * |
fda57b22 AL |
110 | * This is the only entry point used for 64-bit system calls. The |
111 | * hardware interface is reasonably well designed and the register to | |
112 | * argument mapping Linux uses fits well with the registers that are | |
113 | * available when SYSCALL is used. | |
114 | * | |
115 | * SYSCALL instructions can be found inlined in libc implementations as | |
116 | * well as some other programs and libraries. There are also a handful | |
117 | * of SYSCALL instructions in the vDSO used, for example, as a | |
118 | * clock_gettimeofday fallback. | |
119 | * | |
4d732138 | 120 | * 64-bit SYSCALL saves rip to rcx, clears rflags.RF, then saves rflags to r11, |
b87cf63e DV |
121 | * then loads new ss, cs, and rip from previously programmed MSRs. |
122 | * rflags gets masked by a value from another MSR (so CLD and CLAC | |
123 | * are not needed). SYSCALL does not save anything on the stack | |
124 | * and does not change rsp. | |
125 | * | |
126 | * Registers on entry: | |
1da177e4 | 127 | * rax system call number |
b87cf63e DV |
128 | * rcx return address |
129 | * r11 saved rflags (note: r11 is callee-clobbered register in C ABI) | |
1da177e4 | 130 | * rdi arg0 |
1da177e4 | 131 | * rsi arg1 |
0bd7b798 | 132 | * rdx arg2 |
b87cf63e | 133 | * r10 arg3 (needs to be moved to rcx to conform to C ABI) |
1da177e4 LT |
134 | * r8 arg4 |
135 | * r9 arg5 | |
4d732138 | 136 | * (note: r12-r15, rbp, rbx are callee-preserved in C ABI) |
0bd7b798 | 137 | * |
1da177e4 LT |
138 | * Only called from user space. |
139 | * | |
7fcb3bc3 | 140 | * When user can change pt_regs->foo always force IRET. That is because |
7bf36bbc AK |
141 | * it deals with uncanonical addresses better. SYSRET has trouble |
142 | * with them due to bugs in both AMD and Intel CPUs. | |
0bd7b798 | 143 | */ |
1da177e4 | 144 | |
3386bc8a AL |
145 | .pushsection .entry_trampoline, "ax" |
146 | ||
147 | /* | |
148 | * The code in here gets remapped into cpu_entry_area's trampoline. This means | |
149 | * that the assembler and linker have the wrong idea as to where this code | |
150 | * lives (and, in fact, it's mapped more than once, so it's not even at a | |
151 | * fixed address). So we can't reference any symbols outside the entry | |
152 | * trampoline and expect it to work. | |
153 | * | |
154 | * Instead, we carefully abuse %rip-relative addressing. | |
155 | * _entry_trampoline(%rip) refers to the start of the remapped) entry | |
156 | * trampoline. We can thus find cpu_entry_area with this macro: | |
157 | */ | |
158 | ||
159 | #define CPU_ENTRY_AREA \ | |
160 | _entry_trampoline - CPU_ENTRY_AREA_entry_trampoline(%rip) | |
161 | ||
162 | /* The top word of the SYSENTER stack is hot and is usable as scratch space. */ | |
4fe2d8b1 DH |
163 | #define RSP_SCRATCH CPU_ENTRY_AREA_entry_stack + \ |
164 | SIZEOF_entry_stack - 8 + CPU_ENTRY_AREA | |
3386bc8a AL |
165 | |
166 | ENTRY(entry_SYSCALL_64_trampoline) | |
167 | UNWIND_HINT_EMPTY | |
168 | swapgs | |
169 | ||
170 | /* Stash the user RSP. */ | |
171 | movq %rsp, RSP_SCRATCH | |
172 | ||
8a09317b DH |
173 | /* Note: using %rsp as a scratch reg. */ |
174 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rsp | |
175 | ||
3386bc8a AL |
176 | /* Load the top of the task stack into RSP */ |
177 | movq CPU_ENTRY_AREA_tss + TSS_sp1 + CPU_ENTRY_AREA, %rsp | |
178 | ||
179 | /* Start building the simulated IRET frame. */ | |
180 | pushq $__USER_DS /* pt_regs->ss */ | |
181 | pushq RSP_SCRATCH /* pt_regs->sp */ | |
182 | pushq %r11 /* pt_regs->flags */ | |
183 | pushq $__USER_CS /* pt_regs->cs */ | |
184 | pushq %rcx /* pt_regs->ip */ | |
185 | ||
186 | /* | |
187 | * x86 lacks a near absolute jump, and we can't jump to the real | |
188 | * entry text with a relative jump. We could push the target | |
189 | * address and then use retq, but this destroys the pipeline on | |
190 | * many CPUs (wasting over 20 cycles on Sandy Bridge). Instead, | |
191 | * spill RDI and restore it in a second-stage trampoline. | |
192 | */ | |
193 | pushq %rdi | |
194 | movq $entry_SYSCALL_64_stage2, %rdi | |
2641f08b | 195 | JMP_NOSPEC %rdi |
3386bc8a AL |
196 | END(entry_SYSCALL_64_trampoline) |
197 | ||
198 | .popsection | |
199 | ||
200 | ENTRY(entry_SYSCALL_64_stage2) | |
201 | UNWIND_HINT_EMPTY | |
202 | popq %rdi | |
203 | jmp entry_SYSCALL_64_after_hwframe | |
204 | END(entry_SYSCALL_64_stage2) | |
205 | ||
b2502b41 | 206 | ENTRY(entry_SYSCALL_64) |
8c1f7558 | 207 | UNWIND_HINT_EMPTY |
9ed8e7d8 DV |
208 | /* |
209 | * Interrupts are off on entry. | |
210 | * We do not frame this tiny irq-off block with TRACE_IRQS_OFF/ON, | |
211 | * it is too small to ever cause noticeable irq latency. | |
212 | */ | |
72fe4858 | 213 | |
8a9949bc | 214 | swapgs |
8a09317b | 215 | /* |
14b1fcc6 | 216 | * This path is only taken when PAGE_TABLE_ISOLATION is disabled so it |
8a09317b | 217 | * is not required to switch CR3. |
98f05b51 AL |
218 | * |
219 | * tss.sp2 is scratch space. | |
8a09317b | 220 | */ |
98f05b51 | 221 | movq %rsp, PER_CPU_VAR(cpu_tss_rw + TSS_sp2) |
4d732138 | 222 | movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp |
9ed8e7d8 DV |
223 | |
224 | /* Construct struct pt_regs on stack */ | |
98f05b51 AL |
225 | pushq $__USER_DS /* pt_regs->ss */ |
226 | pushq PER_CPU_VAR(cpu_tss_rw + TSS_sp2) /* pt_regs->sp */ | |
227 | pushq %r11 /* pt_regs->flags */ | |
228 | pushq $__USER_CS /* pt_regs->cs */ | |
229 | pushq %rcx /* pt_regs->ip */ | |
8a9949bc | 230 | GLOBAL(entry_SYSCALL_64_after_hwframe) |
98f05b51 | 231 | pushq %rax /* pt_regs->orig_ax */ |
30907fd1 DB |
232 | |
233 | PUSH_AND_CLEAR_REGS rax=$-ENOSYS | |
4d732138 | 234 | |
548c3050 AL |
235 | TRACE_IRQS_OFF |
236 | ||
1e423bff | 237 | /* IRQs are off. */ |
dfe64506 LT |
238 | movq %rax, %rdi |
239 | movq %rsp, %rsi | |
1e423bff AL |
240 | call do_syscall_64 /* returns with IRQs disabled */ |
241 | ||
29ea1b25 | 242 | TRACE_IRQS_IRETQ /* we're about to change IF */ |
fffbb5dc DV |
243 | |
244 | /* | |
245 | * Try to use SYSRET instead of IRET if we're returning to | |
8a055d7f AL |
246 | * a completely clean 64-bit userspace context. If we're not, |
247 | * go to the slow exit path. | |
fffbb5dc | 248 | */ |
4d732138 IM |
249 | movq RCX(%rsp), %rcx |
250 | movq RIP(%rsp), %r11 | |
8a055d7f AL |
251 | |
252 | cmpq %rcx, %r11 /* SYSRET requires RCX == RIP */ | |
253 | jne swapgs_restore_regs_and_return_to_usermode | |
fffbb5dc DV |
254 | |
255 | /* | |
256 | * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP | |
257 | * in kernel space. This essentially lets the user take over | |
17be0aec | 258 | * the kernel, since userspace controls RSP. |
fffbb5dc | 259 | * |
17be0aec | 260 | * If width of "canonical tail" ever becomes variable, this will need |
fffbb5dc | 261 | * to be updated to remain correct on both old and new CPUs. |
361b4b58 | 262 | * |
cbe0317b KS |
263 | * Change top bits to match most significant bit (47th or 56th bit |
264 | * depending on paging mode) in the address. | |
fffbb5dc | 265 | */ |
09e61a77 | 266 | #ifdef CONFIG_X86_5LEVEL |
39b95522 KS |
267 | ALTERNATIVE "shl $(64 - 48), %rcx; sar $(64 - 48), %rcx", \ |
268 | "shl $(64 - 57), %rcx; sar $(64 - 57), %rcx", X86_FEATURE_LA57 | |
09e61a77 | 269 | #else |
17be0aec DV |
270 | shl $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx |
271 | sar $(64 - (__VIRTUAL_MASK_SHIFT+1)), %rcx | |
09e61a77 | 272 | #endif |
4d732138 | 273 | |
17be0aec DV |
274 | /* If this changed %rcx, it was not canonical */ |
275 | cmpq %rcx, %r11 | |
8a055d7f | 276 | jne swapgs_restore_regs_and_return_to_usermode |
fffbb5dc | 277 | |
4d732138 | 278 | cmpq $__USER_CS, CS(%rsp) /* CS must match SYSRET */ |
8a055d7f | 279 | jne swapgs_restore_regs_and_return_to_usermode |
fffbb5dc | 280 | |
4d732138 IM |
281 | movq R11(%rsp), %r11 |
282 | cmpq %r11, EFLAGS(%rsp) /* R11 == RFLAGS */ | |
8a055d7f | 283 | jne swapgs_restore_regs_and_return_to_usermode |
fffbb5dc DV |
284 | |
285 | /* | |
3e035305 BP |
286 | * 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: | |
fffbb5dc | 295 | * |
4d732138 | 296 | * movq $stuck_here, %rcx |
fffbb5dc DV |
297 | * pushfq |
298 | * popq %r11 | |
299 | * stuck_here: | |
300 | * | |
301 | * would never get past 'stuck_here'. | |
302 | */ | |
4d732138 | 303 | testq $(X86_EFLAGS_RF|X86_EFLAGS_TF), %r11 |
8a055d7f | 304 | jnz swapgs_restore_regs_and_return_to_usermode |
fffbb5dc DV |
305 | |
306 | /* nothing to check for RSP */ | |
307 | ||
4d732138 | 308 | cmpq $__USER_DS, SS(%rsp) /* SS must match SYSRET */ |
8a055d7f | 309 | jne swapgs_restore_regs_and_return_to_usermode |
fffbb5dc DV |
310 | |
311 | /* | |
4d732138 IM |
312 | * We win! This label is here just for ease of understanding |
313 | * perf profiles. Nothing jumps here. | |
fffbb5dc DV |
314 | */ |
315 | syscall_return_via_sysret: | |
17be0aec | 316 | /* rcx and r11 are already restored (see code above) */ |
8c1f7558 | 317 | UNWIND_HINT_EMPTY |
502af0d7 | 318 | POP_REGS pop_rdi=0 skip_r11rcx=1 |
3e3b9293 AL |
319 | |
320 | /* | |
321 | * Now all regs are restored except RSP and RDI. | |
322 | * Save old stack pointer and switch to trampoline stack. | |
323 | */ | |
324 | movq %rsp, %rdi | |
c482feef | 325 | movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp |
3e3b9293 AL |
326 | |
327 | pushq RSP-RDI(%rdi) /* RSP */ | |
328 | pushq (%rdi) /* RDI */ | |
329 | ||
330 | /* | |
331 | * We are on the trampoline stack. All regs except RDI are live. | |
332 | * We can do future final exit work right here. | |
333 | */ | |
6fd166aa | 334 | SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi |
3e3b9293 | 335 | |
4fbb3910 | 336 | popq %rdi |
3e3b9293 | 337 | popq %rsp |
fffbb5dc | 338 | USERGS_SYSRET64 |
b2502b41 | 339 | END(entry_SYSCALL_64) |
0bd7b798 | 340 | |
0100301b BG |
341 | /* |
342 | * %rdi: prev task | |
343 | * %rsi: next task | |
344 | */ | |
345 | ENTRY(__switch_to_asm) | |
8c1f7558 | 346 | UNWIND_HINT_FUNC |
0100301b BG |
347 | /* |
348 | * Save callee-saved registers | |
349 | * This must match the order in inactive_task_frame | |
350 | */ | |
351 | pushq %rbp | |
352 | pushq %rbx | |
353 | pushq %r12 | |
354 | pushq %r13 | |
355 | pushq %r14 | |
356 | pushq %r15 | |
357 | ||
358 | /* switch stack */ | |
359 | movq %rsp, TASK_threadsp(%rdi) | |
360 | movq TASK_threadsp(%rsi), %rsp | |
361 | ||
050e9baa | 362 | #ifdef CONFIG_STACKPROTECTOR |
0100301b BG |
363 | movq TASK_stack_canary(%rsi), %rbx |
364 | movq %rbx, PER_CPU_VAR(irq_stack_union)+stack_canary_offset | |
365 | #endif | |
366 | ||
c995efd5 DW |
367 | #ifdef CONFIG_RETPOLINE |
368 | /* | |
369 | * When switching from a shallower to a deeper call stack | |
370 | * the RSB may either underflow or use entries populated | |
371 | * with userspace addresses. On CPUs where those concerns | |
372 | * exist, overwrite the RSB with entries which capture | |
373 | * speculative execution to prevent attack. | |
374 | */ | |
d1c99108 | 375 | FILL_RETURN_BUFFER %r12, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_CTXSW |
c995efd5 DW |
376 | #endif |
377 | ||
0100301b BG |
378 | /* restore callee-saved registers */ |
379 | popq %r15 | |
380 | popq %r14 | |
381 | popq %r13 | |
382 | popq %r12 | |
383 | popq %rbx | |
384 | popq %rbp | |
385 | ||
386 | jmp __switch_to | |
387 | END(__switch_to_asm) | |
388 | ||
1eeb207f DV |
389 | /* |
390 | * A newly forked process directly context switches into this address. | |
391 | * | |
0100301b | 392 | * rax: prev task we switched from |
616d2483 BG |
393 | * rbx: kernel thread func (NULL for user thread) |
394 | * r12: kernel thread arg | |
1eeb207f DV |
395 | */ |
396 | ENTRY(ret_from_fork) | |
8c1f7558 | 397 | UNWIND_HINT_EMPTY |
0100301b | 398 | movq %rax, %rdi |
ebd57499 | 399 | call schedule_tail /* rdi: 'prev' task parameter */ |
1eeb207f | 400 | |
ebd57499 JP |
401 | testq %rbx, %rbx /* from kernel_thread? */ |
402 | jnz 1f /* kernel threads are uncommon */ | |
24d978b7 | 403 | |
616d2483 | 404 | 2: |
8c1f7558 | 405 | UNWIND_HINT_REGS |
ebd57499 | 406 | movq %rsp, %rdi |
24d978b7 AL |
407 | call syscall_return_slowpath /* returns with IRQs disabled */ |
408 | TRACE_IRQS_ON /* user mode is traced as IRQS on */ | |
8a055d7f | 409 | jmp swapgs_restore_regs_and_return_to_usermode |
616d2483 BG |
410 | |
411 | 1: | |
412 | /* kernel thread */ | |
d31a5802 | 413 | UNWIND_HINT_EMPTY |
616d2483 | 414 | movq %r12, %rdi |
2641f08b | 415 | CALL_NOSPEC %rbx |
616d2483 BG |
416 | /* |
417 | * A kernel thread is allowed to return here after successfully | |
418 | * calling do_execve(). Exit to userspace to complete the execve() | |
419 | * syscall. | |
420 | */ | |
421 | movq $0, RAX(%rsp) | |
422 | jmp 2b | |
1eeb207f DV |
423 | END(ret_from_fork) |
424 | ||
939b7871 | 425 | /* |
3304c9c3 DV |
426 | * Build the entry stubs with some assembler magic. |
427 | * We pack 1 stub into every 8-byte block. | |
939b7871 | 428 | */ |
3304c9c3 | 429 | .align 8 |
939b7871 | 430 | ENTRY(irq_entries_start) |
3304c9c3 DV |
431 | vector=FIRST_EXTERNAL_VECTOR |
432 | .rept (FIRST_SYSTEM_VECTOR - FIRST_EXTERNAL_VECTOR) | |
8c1f7558 | 433 | UNWIND_HINT_IRET_REGS |
4d732138 | 434 | pushq $(~vector+0x80) /* Note: always in signed byte range */ |
3304c9c3 | 435 | jmp common_interrupt |
3304c9c3 | 436 | .align 8 |
8c1f7558 | 437 | vector=vector+1 |
3304c9c3 | 438 | .endr |
939b7871 PA |
439 | END(irq_entries_start) |
440 | ||
1d3e53e8 AL |
441 | .macro DEBUG_ENTRY_ASSERT_IRQS_OFF |
442 | #ifdef CONFIG_DEBUG_ENTRY | |
e17f8234 BO |
443 | pushq %rax |
444 | SAVE_FLAGS(CLBR_RAX) | |
445 | testl $X86_EFLAGS_IF, %eax | |
1d3e53e8 AL |
446 | jz .Lokay_\@ |
447 | ud2 | |
448 | .Lokay_\@: | |
e17f8234 | 449 | popq %rax |
1d3e53e8 AL |
450 | #endif |
451 | .endm | |
452 | ||
453 | /* | |
454 | * Enters the IRQ stack if we're not already using it. NMI-safe. Clobbers | |
455 | * flags and puts old RSP into old_rsp, and leaves all other GPRs alone. | |
456 | * Requires kernel GSBASE. | |
457 | * | |
458 | * The invariant is that, if irq_count != -1, then the IRQ stack is in use. | |
459 | */ | |
2ba64741 | 460 | .macro ENTER_IRQ_STACK regs=1 old_rsp save_ret=0 |
1d3e53e8 | 461 | DEBUG_ENTRY_ASSERT_IRQS_OFF |
2ba64741 DB |
462 | |
463 | .if \save_ret | |
464 | /* | |
465 | * If save_ret is set, the original stack contains one additional | |
466 | * entry -- the return address. Therefore, move the address one | |
467 | * entry below %rsp to \old_rsp. | |
468 | */ | |
469 | leaq 8(%rsp), \old_rsp | |
470 | .else | |
1d3e53e8 | 471 | movq %rsp, \old_rsp |
2ba64741 | 472 | .endif |
8c1f7558 JP |
473 | |
474 | .if \regs | |
475 | UNWIND_HINT_REGS base=\old_rsp | |
476 | .endif | |
477 | ||
1d3e53e8 | 478 | incl PER_CPU_VAR(irq_count) |
29955909 | 479 | jnz .Lirq_stack_push_old_rsp_\@ |
1d3e53e8 AL |
480 | |
481 | /* | |
482 | * Right now, if we just incremented irq_count to zero, we've | |
483 | * claimed the IRQ stack but we haven't switched to it yet. | |
484 | * | |
485 | * If anything is added that can interrupt us here without using IST, | |
486 | * it must be *extremely* careful to limit its stack usage. This | |
487 | * could include kprobes and a hypothetical future IST-less #DB | |
488 | * handler. | |
29955909 AL |
489 | * |
490 | * The OOPS unwinder relies on the word at the top of the IRQ | |
491 | * stack linking back to the previous RSP for the entire time we're | |
492 | * on the IRQ stack. For this to work reliably, we need to write | |
493 | * it before we actually move ourselves to the IRQ stack. | |
494 | */ | |
495 | ||
496 | movq \old_rsp, PER_CPU_VAR(irq_stack_union + IRQ_STACK_SIZE - 8) | |
497 | movq PER_CPU_VAR(irq_stack_ptr), %rsp | |
498 | ||
499 | #ifdef CONFIG_DEBUG_ENTRY | |
500 | /* | |
501 | * If the first movq above becomes wrong due to IRQ stack layout | |
502 | * changes, the only way we'll notice is if we try to unwind right | |
503 | * here. Assert that we set up the stack right to catch this type | |
504 | * of bug quickly. | |
1d3e53e8 | 505 | */ |
29955909 AL |
506 | cmpq -8(%rsp), \old_rsp |
507 | je .Lirq_stack_okay\@ | |
508 | ud2 | |
509 | .Lirq_stack_okay\@: | |
510 | #endif | |
1d3e53e8 | 511 | |
29955909 | 512 | .Lirq_stack_push_old_rsp_\@: |
1d3e53e8 | 513 | pushq \old_rsp |
8c1f7558 JP |
514 | |
515 | .if \regs | |
516 | UNWIND_HINT_REGS indirect=1 | |
517 | .endif | |
2ba64741 DB |
518 | |
519 | .if \save_ret | |
520 | /* | |
521 | * Push the return address to the stack. This return address can | |
522 | * be found at the "real" original RSP, which was offset by 8 at | |
523 | * the beginning of this macro. | |
524 | */ | |
525 | pushq -8(\old_rsp) | |
526 | .endif | |
1d3e53e8 AL |
527 | .endm |
528 | ||
529 | /* | |
530 | * Undoes ENTER_IRQ_STACK. | |
531 | */ | |
8c1f7558 | 532 | .macro LEAVE_IRQ_STACK regs=1 |
1d3e53e8 AL |
533 | DEBUG_ENTRY_ASSERT_IRQS_OFF |
534 | /* We need to be off the IRQ stack before decrementing irq_count. */ | |
535 | popq %rsp | |
536 | ||
8c1f7558 JP |
537 | .if \regs |
538 | UNWIND_HINT_REGS | |
539 | .endif | |
540 | ||
1d3e53e8 AL |
541 | /* |
542 | * As in ENTER_IRQ_STACK, irq_count == 0, we are still claiming | |
543 | * the irq stack but we're not on it. | |
544 | */ | |
545 | ||
546 | decl PER_CPU_VAR(irq_count) | |
547 | .endm | |
548 | ||
d99015b1 | 549 | /* |
f3d415ea | 550 | * Interrupt entry helper function. |
d99015b1 | 551 | * |
f3d415ea DB |
552 | * Entry runs with interrupts off. Stack layout at entry: |
553 | * +----------------------------------------------------+ | |
554 | * | regs->ss | | |
555 | * | regs->rsp | | |
556 | * | regs->eflags | | |
557 | * | regs->cs | | |
558 | * | regs->ip | | |
559 | * +----------------------------------------------------+ | |
560 | * | regs->orig_ax = ~(interrupt number) | | |
561 | * +----------------------------------------------------+ | |
562 | * | return address | | |
563 | * +----------------------------------------------------+ | |
d99015b1 | 564 | */ |
f3d415ea DB |
565 | ENTRY(interrupt_entry) |
566 | UNWIND_HINT_FUNC | |
567 | ASM_CLAC | |
f6f64681 | 568 | cld |
7f2590a1 | 569 | |
f3d415ea | 570 | testb $3, CS-ORIG_RAX+8(%rsp) |
7f2590a1 AL |
571 | jz 1f |
572 | SWAPGS | |
f3d415ea DB |
573 | |
574 | /* | |
575 | * Switch to the thread stack. The IRET frame and orig_ax are | |
576 | * on the stack, as well as the return address. RDI..R12 are | |
577 | * not (yet) on the stack and space has not (yet) been | |
578 | * allocated for them. | |
579 | */ | |
90a6acc4 | 580 | pushq %rdi |
f3d415ea | 581 | |
90a6acc4 DB |
582 | /* Need to switch before accessing the thread stack. */ |
583 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi | |
584 | movq %rsp, %rdi | |
585 | movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp | |
f3d415ea DB |
586 | |
587 | /* | |
588 | * We have RDI, return address, and orig_ax on the stack on | |
589 | * top of the IRET frame. That means offset=24 | |
590 | */ | |
591 | UNWIND_HINT_IRET_REGS base=%rdi offset=24 | |
90a6acc4 DB |
592 | |
593 | pushq 7*8(%rdi) /* regs->ss */ | |
594 | pushq 6*8(%rdi) /* regs->rsp */ | |
595 | pushq 5*8(%rdi) /* regs->eflags */ | |
596 | pushq 4*8(%rdi) /* regs->cs */ | |
597 | pushq 3*8(%rdi) /* regs->ip */ | |
598 | pushq 2*8(%rdi) /* regs->orig_ax */ | |
599 | pushq 8(%rdi) /* return address */ | |
600 | UNWIND_HINT_FUNC | |
601 | ||
602 | movq (%rdi), %rdi | |
7f2590a1 AL |
603 | 1: |
604 | ||
0e34d226 DB |
605 | PUSH_AND_CLEAR_REGS save_ret=1 |
606 | ENCODE_FRAME_POINTER 8 | |
76f5df43 | 607 | |
2ba64741 | 608 | testb $3, CS+8(%rsp) |
dde74f2e | 609 | jz 1f |
02bc7768 AL |
610 | |
611 | /* | |
7f2590a1 AL |
612 | * IRQ from user mode. |
613 | * | |
f1075053 AL |
614 | * We need to tell lockdep that IRQs are off. We can't do this until |
615 | * we fix gsbase, and we should do it before enter_from_user_mode | |
f3d415ea | 616 | * (which can take locks). Since TRACE_IRQS_OFF is idempotent, |
f1075053 AL |
617 | * the simplest way to handle it is to just call it twice if |
618 | * we enter from user mode. There's no reason to optimize this since | |
619 | * TRACE_IRQS_OFF is a no-op if lockdep is off. | |
620 | */ | |
621 | TRACE_IRQS_OFF | |
622 | ||
478dc89c | 623 | CALL_enter_from_user_mode |
02bc7768 | 624 | |
76f5df43 | 625 | 1: |
2ba64741 | 626 | ENTER_IRQ_STACK old_rsp=%rdi save_ret=1 |
f6f64681 DV |
627 | /* We entered an interrupt context - irqs are off: */ |
628 | TRACE_IRQS_OFF | |
629 | ||
2ba64741 DB |
630 | ret |
631 | END(interrupt_entry) | |
632 | ||
f3d415ea DB |
633 | |
634 | /* Interrupt entry/exit. */ | |
1da177e4 | 635 | |
722024db AH |
636 | /* |
637 | * The interrupt stubs push (~vector+0x80) onto the stack and | |
638 | * then jump to common_interrupt. | |
639 | */ | |
939b7871 PA |
640 | .p2align CONFIG_X86_L1_CACHE_SHIFT |
641 | common_interrupt: | |
4d732138 | 642 | addq $-0x80, (%rsp) /* Adjust vector to [-256, -1] range */ |
3aa99fc3 DB |
643 | call interrupt_entry |
644 | UNWIND_HINT_REGS indirect=1 | |
645 | call do_IRQ /* rdi points to pt_regs */ | |
34061f13 | 646 | /* 0(%rsp): old RSP */ |
7effaa88 | 647 | ret_from_intr: |
2140a994 | 648 | DISABLE_INTERRUPTS(CLBR_ANY) |
2601e64d | 649 | TRACE_IRQS_OFF |
625dbc3b | 650 | |
1d3e53e8 | 651 | LEAVE_IRQ_STACK |
625dbc3b | 652 | |
03335e95 | 653 | testb $3, CS(%rsp) |
dde74f2e | 654 | jz retint_kernel |
4d732138 | 655 | |
02bc7768 | 656 | /* Interrupt came from user space */ |
02bc7768 AL |
657 | GLOBAL(retint_user) |
658 | mov %rsp,%rdi | |
659 | call prepare_exit_to_usermode | |
2601e64d | 660 | TRACE_IRQS_IRETQ |
26c4ef9c | 661 | |
8a055d7f | 662 | GLOBAL(swapgs_restore_regs_and_return_to_usermode) |
26c4ef9c AL |
663 | #ifdef CONFIG_DEBUG_ENTRY |
664 | /* Assert that pt_regs indicates user mode. */ | |
1e4c4f61 | 665 | testb $3, CS(%rsp) |
26c4ef9c AL |
666 | jnz 1f |
667 | ud2 | |
668 | 1: | |
669 | #endif | |
502af0d7 | 670 | POP_REGS pop_rdi=0 |
3e3b9293 AL |
671 | |
672 | /* | |
673 | * The stack is now user RDI, orig_ax, RIP, CS, EFLAGS, RSP, SS. | |
674 | * Save old stack pointer and switch to trampoline stack. | |
675 | */ | |
676 | movq %rsp, %rdi | |
c482feef | 677 | movq PER_CPU_VAR(cpu_tss_rw + TSS_sp0), %rsp |
3e3b9293 AL |
678 | |
679 | /* Copy the IRET frame to the trampoline stack. */ | |
680 | pushq 6*8(%rdi) /* SS */ | |
681 | pushq 5*8(%rdi) /* RSP */ | |
682 | pushq 4*8(%rdi) /* EFLAGS */ | |
683 | pushq 3*8(%rdi) /* CS */ | |
684 | pushq 2*8(%rdi) /* RIP */ | |
685 | ||
686 | /* Push user RDI on the trampoline stack. */ | |
687 | pushq (%rdi) | |
688 | ||
689 | /* | |
690 | * We are on the trampoline stack. All regs except RDI are live. | |
691 | * We can do future final exit work right here. | |
692 | */ | |
693 | ||
6fd166aa | 694 | SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi |
8a09317b | 695 | |
3e3b9293 AL |
696 | /* Restore RDI. */ |
697 | popq %rdi | |
698 | SWAPGS | |
26c4ef9c AL |
699 | INTERRUPT_RETURN |
700 | ||
2601e64d | 701 | |
627276cb | 702 | /* Returning to kernel space */ |
6ba71b76 | 703 | retint_kernel: |
627276cb DV |
704 | #ifdef CONFIG_PREEMPT |
705 | /* Interrupts are off */ | |
706 | /* Check if we need preemption */ | |
6709812f | 707 | btl $9, EFLAGS(%rsp) /* were interrupts off? */ |
6ba71b76 | 708 | jnc 1f |
4d732138 | 709 | 0: cmpl $0, PER_CPU_VAR(__preempt_count) |
36acef25 | 710 | jnz 1f |
627276cb | 711 | call preempt_schedule_irq |
36acef25 | 712 | jmp 0b |
6ba71b76 | 713 | 1: |
627276cb | 714 | #endif |
2601e64d IM |
715 | /* |
716 | * The iretq could re-enable interrupts: | |
717 | */ | |
718 | TRACE_IRQS_IRETQ | |
fffbb5dc | 719 | |
26c4ef9c AL |
720 | GLOBAL(restore_regs_and_return_to_kernel) |
721 | #ifdef CONFIG_DEBUG_ENTRY | |
722 | /* Assert that pt_regs indicates kernel mode. */ | |
1e4c4f61 | 723 | testb $3, CS(%rsp) |
26c4ef9c AL |
724 | jz 1f |
725 | ud2 | |
726 | 1: | |
727 | #endif | |
502af0d7 | 728 | POP_REGS |
e872045b | 729 | addq $8, %rsp /* skip regs->orig_ax */ |
10bcc80e MD |
730 | /* |
731 | * ARCH_HAS_MEMBARRIER_SYNC_CORE rely on IRET core serialization | |
732 | * when returning from IPI handler. | |
733 | */ | |
7209a75d AL |
734 | INTERRUPT_RETURN |
735 | ||
736 | ENTRY(native_iret) | |
8c1f7558 | 737 | UNWIND_HINT_IRET_REGS |
3891a04a PA |
738 | /* |
739 | * Are we returning to a stack segment from the LDT? Note: in | |
740 | * 64-bit mode SS:RSP on the exception stack is always valid. | |
741 | */ | |
34273f41 | 742 | #ifdef CONFIG_X86_ESPFIX64 |
4d732138 IM |
743 | testb $4, (SS-RIP)(%rsp) |
744 | jnz native_irq_return_ldt | |
34273f41 | 745 | #endif |
3891a04a | 746 | |
af726f21 | 747 | .global native_irq_return_iret |
7209a75d | 748 | native_irq_return_iret: |
b645af2d AL |
749 | /* |
750 | * This may fault. Non-paranoid faults on return to userspace are | |
751 | * handled by fixup_bad_iret. These include #SS, #GP, and #NP. | |
752 | * Double-faults due to espfix64 are handled in do_double_fault. | |
753 | * Other faults here are fatal. | |
754 | */ | |
1da177e4 | 755 | iretq |
3701d863 | 756 | |
34273f41 | 757 | #ifdef CONFIG_X86_ESPFIX64 |
7209a75d | 758 | native_irq_return_ldt: |
85063fac AL |
759 | /* |
760 | * We are running with user GSBASE. All GPRs contain their user | |
761 | * values. We have a percpu ESPFIX stack that is eight slots | |
762 | * long (see ESPFIX_STACK_SIZE). espfix_waddr points to the bottom | |
763 | * of the ESPFIX stack. | |
764 | * | |
765 | * We clobber RAX and RDI in this code. We stash RDI on the | |
766 | * normal stack and RAX on the ESPFIX stack. | |
767 | * | |
768 | * The ESPFIX stack layout we set up looks like this: | |
769 | * | |
770 | * --- top of ESPFIX stack --- | |
771 | * SS | |
772 | * RSP | |
773 | * RFLAGS | |
774 | * CS | |
775 | * RIP <-- RSP points here when we're done | |
776 | * RAX <-- espfix_waddr points here | |
777 | * --- bottom of ESPFIX stack --- | |
778 | */ | |
779 | ||
780 | pushq %rdi /* Stash user RDI */ | |
8a09317b DH |
781 | SWAPGS /* to kernel GS */ |
782 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rdi /* to kernel CR3 */ | |
783 | ||
4d732138 | 784 | movq PER_CPU_VAR(espfix_waddr), %rdi |
85063fac AL |
785 | movq %rax, (0*8)(%rdi) /* user RAX */ |
786 | movq (1*8)(%rsp), %rax /* user RIP */ | |
4d732138 | 787 | movq %rax, (1*8)(%rdi) |
85063fac | 788 | movq (2*8)(%rsp), %rax /* user CS */ |
4d732138 | 789 | movq %rax, (2*8)(%rdi) |
85063fac | 790 | movq (3*8)(%rsp), %rax /* user RFLAGS */ |
4d732138 | 791 | movq %rax, (3*8)(%rdi) |
85063fac | 792 | movq (5*8)(%rsp), %rax /* user SS */ |
4d732138 | 793 | movq %rax, (5*8)(%rdi) |
85063fac | 794 | movq (4*8)(%rsp), %rax /* user RSP */ |
4d732138 | 795 | movq %rax, (4*8)(%rdi) |
85063fac AL |
796 | /* Now RAX == RSP. */ |
797 | ||
798 | andl $0xffff0000, %eax /* RAX = (RSP & 0xffff0000) */ | |
85063fac AL |
799 | |
800 | /* | |
801 | * espfix_stack[31:16] == 0. The page tables are set up such that | |
802 | * (espfix_stack | (X & 0xffff0000)) points to a read-only alias of | |
803 | * espfix_waddr for any X. That is, there are 65536 RO aliases of | |
804 | * the same page. Set up RSP so that RSP[31:16] contains the | |
805 | * respective 16 bits of the /userspace/ RSP and RSP nonetheless | |
806 | * still points to an RO alias of the ESPFIX stack. | |
807 | */ | |
4d732138 | 808 | orq PER_CPU_VAR(espfix_stack), %rax |
8a09317b | 809 | |
6fd166aa | 810 | SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi |
8a09317b DH |
811 | SWAPGS /* to user GS */ |
812 | popq %rdi /* Restore user RDI */ | |
813 | ||
4d732138 | 814 | movq %rax, %rsp |
8c1f7558 | 815 | UNWIND_HINT_IRET_REGS offset=8 |
85063fac AL |
816 | |
817 | /* | |
818 | * At this point, we cannot write to the stack any more, but we can | |
819 | * still read. | |
820 | */ | |
821 | popq %rax /* Restore user RAX */ | |
822 | ||
823 | /* | |
824 | * RSP now points to an ordinary IRET frame, except that the page | |
825 | * is read-only and RSP[31:16] are preloaded with the userspace | |
826 | * values. We can now IRET back to userspace. | |
827 | */ | |
4d732138 | 828 | jmp native_irq_return_iret |
34273f41 | 829 | #endif |
4b787e0b | 830 | END(common_interrupt) |
3891a04a | 831 | |
1da177e4 LT |
832 | /* |
833 | * APIC interrupts. | |
0bd7b798 | 834 | */ |
cf910e83 | 835 | .macro apicinterrupt3 num sym do_sym |
322648d1 | 836 | ENTRY(\sym) |
8c1f7558 | 837 | UNWIND_HINT_IRET_REGS |
4d732138 | 838 | pushq $~(\num) |
39e95433 | 839 | .Lcommon_\sym: |
3aa99fc3 DB |
840 | call interrupt_entry |
841 | UNWIND_HINT_REGS indirect=1 | |
842 | call \do_sym /* rdi points to pt_regs */ | |
4d732138 | 843 | jmp ret_from_intr |
322648d1 AH |
844 | END(\sym) |
845 | .endm | |
1da177e4 | 846 | |
469f0023 | 847 | /* Make sure APIC interrupt handlers end up in the irqentry section: */ |
229a7186 MH |
848 | #define PUSH_SECTION_IRQENTRY .pushsection .irqentry.text, "ax" |
849 | #define POP_SECTION_IRQENTRY .popsection | |
469f0023 | 850 | |
cf910e83 | 851 | .macro apicinterrupt num sym do_sym |
469f0023 | 852 | PUSH_SECTION_IRQENTRY |
cf910e83 | 853 | apicinterrupt3 \num \sym \do_sym |
469f0023 | 854 | POP_SECTION_IRQENTRY |
cf910e83 SA |
855 | .endm |
856 | ||
322648d1 | 857 | #ifdef CONFIG_SMP |
4d732138 IM |
858 | apicinterrupt3 IRQ_MOVE_CLEANUP_VECTOR irq_move_cleanup_interrupt smp_irq_move_cleanup_interrupt |
859 | apicinterrupt3 REBOOT_VECTOR reboot_interrupt smp_reboot_interrupt | |
322648d1 | 860 | #endif |
1da177e4 | 861 | |
03b48632 | 862 | #ifdef CONFIG_X86_UV |
4d732138 | 863 | apicinterrupt3 UV_BAU_MESSAGE uv_bau_message_intr1 uv_bau_message_interrupt |
03b48632 | 864 | #endif |
4d732138 IM |
865 | |
866 | apicinterrupt LOCAL_TIMER_VECTOR apic_timer_interrupt smp_apic_timer_interrupt | |
867 | apicinterrupt X86_PLATFORM_IPI_VECTOR x86_platform_ipi smp_x86_platform_ipi | |
89b831ef | 868 | |
d78f2664 | 869 | #ifdef CONFIG_HAVE_KVM |
4d732138 IM |
870 | apicinterrupt3 POSTED_INTR_VECTOR kvm_posted_intr_ipi smp_kvm_posted_intr_ipi |
871 | apicinterrupt3 POSTED_INTR_WAKEUP_VECTOR kvm_posted_intr_wakeup_ipi smp_kvm_posted_intr_wakeup_ipi | |
210f84b0 | 872 | apicinterrupt3 POSTED_INTR_NESTED_VECTOR kvm_posted_intr_nested_ipi smp_kvm_posted_intr_nested_ipi |
d78f2664 YZ |
873 | #endif |
874 | ||
33e5ff63 | 875 | #ifdef CONFIG_X86_MCE_THRESHOLD |
4d732138 | 876 | apicinterrupt THRESHOLD_APIC_VECTOR threshold_interrupt smp_threshold_interrupt |
33e5ff63 SA |
877 | #endif |
878 | ||
24fd78a8 | 879 | #ifdef CONFIG_X86_MCE_AMD |
4d732138 | 880 | apicinterrupt DEFERRED_ERROR_VECTOR deferred_error_interrupt smp_deferred_error_interrupt |
24fd78a8 AG |
881 | #endif |
882 | ||
33e5ff63 | 883 | #ifdef CONFIG_X86_THERMAL_VECTOR |
4d732138 | 884 | apicinterrupt THERMAL_APIC_VECTOR thermal_interrupt smp_thermal_interrupt |
33e5ff63 | 885 | #endif |
1812924b | 886 | |
322648d1 | 887 | #ifdef CONFIG_SMP |
4d732138 IM |
888 | apicinterrupt CALL_FUNCTION_SINGLE_VECTOR call_function_single_interrupt smp_call_function_single_interrupt |
889 | apicinterrupt CALL_FUNCTION_VECTOR call_function_interrupt smp_call_function_interrupt | |
890 | apicinterrupt RESCHEDULE_VECTOR reschedule_interrupt smp_reschedule_interrupt | |
322648d1 | 891 | #endif |
1da177e4 | 892 | |
4d732138 IM |
893 | apicinterrupt ERROR_APIC_VECTOR error_interrupt smp_error_interrupt |
894 | apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt | |
0bd7b798 | 895 | |
e360adbe | 896 | #ifdef CONFIG_IRQ_WORK |
4d732138 | 897 | apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt |
241771ef IM |
898 | #endif |
899 | ||
1da177e4 LT |
900 | /* |
901 | * Exception entry points. | |
0bd7b798 | 902 | */ |
c482feef | 903 | #define CPU_TSS_IST(x) PER_CPU_VAR(cpu_tss_rw) + (TSS_ist + ((x) - 1) * 8) |
577ed45e | 904 | |
bd7b1f7c AL |
905 | /** |
906 | * idtentry - Generate an IDT entry stub | |
907 | * @sym: Name of the generated entry point | |
908 | * @do_sym: C function to be called | |
909 | * @has_error_code: True if this IDT vector has an error code on the stack | |
910 | * @paranoid: non-zero means that this vector may be invoked from | |
911 | * kernel mode with user GSBASE and/or user CR3. | |
912 | * 2 is special -- see below. | |
913 | * @shift_ist: Set to an IST index if entries from kernel mode should | |
914 | * decrement the IST stack so that nested entries get a | |
915 | * fresh stack. (This is for #DB, which has a nasty habit | |
916 | * of recursing.) | |
917 | * | |
918 | * idtentry generates an IDT stub that sets up a usable kernel context, | |
919 | * creates struct pt_regs, and calls @do_sym. The stub has the following | |
920 | * special behaviors: | |
921 | * | |
922 | * On an entry from user mode, the stub switches from the trampoline or | |
923 | * IST stack to the normal thread stack. On an exit to user mode, the | |
924 | * normal exit-to-usermode path is invoked. | |
925 | * | |
926 | * On an exit to kernel mode, if @paranoid == 0, we check for preemption, | |
927 | * whereas we omit the preemption check if @paranoid != 0. This is purely | |
928 | * because the implementation is simpler this way. The kernel only needs | |
929 | * to check for asynchronous kernel preemption when IRQ handlers return. | |
930 | * | |
931 | * If @paranoid == 0, then the stub will handle IRET faults by pretending | |
932 | * that the fault came from user mode. It will handle gs_change faults by | |
933 | * pretending that the fault happened with kernel GSBASE. Since this handling | |
934 | * is omitted for @paranoid != 0, the #GP, #SS, and #NP stubs must have | |
935 | * @paranoid == 0. This special handling will do the wrong thing for | |
936 | * espfix-induced #DF on IRET, so #DF must not use @paranoid == 0. | |
937 | * | |
938 | * @paranoid == 2 is special: the stub will never switch stacks. This is for | |
939 | * #DF: if the thread stack is somehow unusable, we'll still get a useful OOPS. | |
940 | */ | |
577ed45e | 941 | .macro idtentry sym do_sym has_error_code:req paranoid=0 shift_ist=-1 |
322648d1 | 942 | ENTRY(\sym) |
98990a33 | 943 | UNWIND_HINT_IRET_REGS offset=\has_error_code*8 |
8c1f7558 | 944 | |
577ed45e AL |
945 | /* Sanity check */ |
946 | .if \shift_ist != -1 && \paranoid == 0 | |
947 | .error "using shift_ist requires paranoid=1" | |
948 | .endif | |
949 | ||
ee4eb87b | 950 | ASM_CLAC |
cb5dd2c5 | 951 | |
82c62fa0 | 952 | .if \has_error_code == 0 |
4d732138 | 953 | pushq $-1 /* ORIG_RAX: no syscall to restart */ |
cb5dd2c5 AL |
954 | .endif |
955 | ||
071ccc96 | 956 | .if \paranoid == 1 |
9e809d15 | 957 | testb $3, CS-ORIG_RAX(%rsp) /* If coming from userspace, switch stacks */ |
7f2590a1 | 958 | jnz .Lfrom_usermode_switch_stack_\@ |
48e08d0f | 959 | .endif |
7f2590a1 AL |
960 | |
961 | .if \paranoid | |
4d732138 | 962 | call paranoid_entry |
cb5dd2c5 | 963 | .else |
4d732138 | 964 | call error_entry |
cb5dd2c5 | 965 | .endif |
8c1f7558 | 966 | UNWIND_HINT_REGS |
ebfc453e | 967 | /* returned flag: ebx=0: need swapgs on exit, ebx=1: don't need it */ |
cb5dd2c5 | 968 | |
cb5dd2c5 | 969 | .if \paranoid |
577ed45e | 970 | .if \shift_ist != -1 |
4d732138 | 971 | TRACE_IRQS_OFF_DEBUG /* reload IDT in case of recursion */ |
577ed45e | 972 | .else |
b8b1d08b | 973 | TRACE_IRQS_OFF |
cb5dd2c5 | 974 | .endif |
577ed45e | 975 | .endif |
cb5dd2c5 | 976 | |
4d732138 | 977 | movq %rsp, %rdi /* pt_regs pointer */ |
cb5dd2c5 AL |
978 | |
979 | .if \has_error_code | |
4d732138 IM |
980 | movq ORIG_RAX(%rsp), %rsi /* get error code */ |
981 | movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */ | |
cb5dd2c5 | 982 | .else |
4d732138 | 983 | xorl %esi, %esi /* no error code */ |
cb5dd2c5 AL |
984 | .endif |
985 | ||
577ed45e | 986 | .if \shift_ist != -1 |
4d732138 | 987 | subq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist) |
577ed45e AL |
988 | .endif |
989 | ||
4d732138 | 990 | call \do_sym |
cb5dd2c5 | 991 | |
577ed45e | 992 | .if \shift_ist != -1 |
4d732138 | 993 | addq $EXCEPTION_STKSZ, CPU_TSS_IST(\shift_ist) |
577ed45e AL |
994 | .endif |
995 | ||
ebfc453e | 996 | /* these procedures expect "no swapgs" flag in ebx */ |
cb5dd2c5 | 997 | .if \paranoid |
4d732138 | 998 | jmp paranoid_exit |
cb5dd2c5 | 999 | .else |
4d732138 | 1000 | jmp error_exit |
cb5dd2c5 AL |
1001 | .endif |
1002 | ||
071ccc96 | 1003 | .if \paranoid == 1 |
48e08d0f | 1004 | /* |
7f2590a1 | 1005 | * Entry from userspace. Switch stacks and treat it |
48e08d0f AL |
1006 | * as a normal entry. This means that paranoid handlers |
1007 | * run in real process context if user_mode(regs). | |
1008 | */ | |
7f2590a1 | 1009 | .Lfrom_usermode_switch_stack_\@: |
4d732138 | 1010 | call error_entry |
48e08d0f | 1011 | |
4d732138 | 1012 | movq %rsp, %rdi /* pt_regs pointer */ |
48e08d0f AL |
1013 | |
1014 | .if \has_error_code | |
4d732138 IM |
1015 | movq ORIG_RAX(%rsp), %rsi /* get error code */ |
1016 | movq $-1, ORIG_RAX(%rsp) /* no syscall to restart */ | |
48e08d0f | 1017 | .else |
4d732138 | 1018 | xorl %esi, %esi /* no error code */ |
48e08d0f AL |
1019 | .endif |
1020 | ||
4d732138 | 1021 | call \do_sym |
48e08d0f | 1022 | |
b3681dd5 | 1023 | jmp error_exit |
48e08d0f | 1024 | .endif |
ddeb8f21 | 1025 | END(\sym) |
322648d1 | 1026 | .endm |
b8b1d08b | 1027 | |
4d732138 IM |
1028 | idtentry divide_error do_divide_error has_error_code=0 |
1029 | idtentry overflow do_overflow has_error_code=0 | |
1030 | idtentry bounds do_bounds has_error_code=0 | |
1031 | idtentry invalid_op do_invalid_op has_error_code=0 | |
1032 | idtentry device_not_available do_device_not_available has_error_code=0 | |
1033 | idtentry double_fault do_double_fault has_error_code=1 paranoid=2 | |
1034 | idtentry coprocessor_segment_overrun do_coprocessor_segment_overrun has_error_code=0 | |
1035 | idtentry invalid_TSS do_invalid_TSS has_error_code=1 | |
1036 | idtentry segment_not_present do_segment_not_present has_error_code=1 | |
1037 | idtentry spurious_interrupt_bug do_spurious_interrupt_bug has_error_code=0 | |
1038 | idtentry coprocessor_error do_coprocessor_error has_error_code=0 | |
1039 | idtentry alignment_check do_alignment_check has_error_code=1 | |
1040 | idtentry simd_coprocessor_error do_simd_coprocessor_error has_error_code=0 | |
1041 | ||
1042 | ||
1043 | /* | |
1044 | * Reload gs selector with exception handling | |
1045 | * edi: new selector | |
1046 | */ | |
9f9d489a | 1047 | ENTRY(native_load_gs_index) |
8c1f7558 | 1048 | FRAME_BEGIN |
131484c8 | 1049 | pushfq |
b8aa287f | 1050 | DISABLE_INTERRUPTS(CLBR_ANY & ~CLBR_RDI) |
ca37e57b | 1051 | TRACE_IRQS_OFF |
9f1e87ea | 1052 | SWAPGS |
42c748bb | 1053 | .Lgs_change: |
4d732138 | 1054 | movl %edi, %gs |
96e5d28a | 1055 | 2: ALTERNATIVE "", "mfence", X86_BUG_SWAPGS_FENCE |
72fe4858 | 1056 | SWAPGS |
ca37e57b | 1057 | TRACE_IRQS_FLAGS (%rsp) |
131484c8 | 1058 | popfq |
8c1f7558 | 1059 | FRAME_END |
9f1e87ea | 1060 | ret |
8c1f7558 | 1061 | ENDPROC(native_load_gs_index) |
784d5699 | 1062 | EXPORT_SYMBOL(native_load_gs_index) |
0bd7b798 | 1063 | |
42c748bb | 1064 | _ASM_EXTABLE(.Lgs_change, bad_gs) |
4d732138 | 1065 | .section .fixup, "ax" |
1da177e4 | 1066 | /* running with kernelgs */ |
0bd7b798 | 1067 | bad_gs: |
4d732138 | 1068 | SWAPGS /* switch back to user gs */ |
b038c842 AL |
1069 | .macro ZAP_GS |
1070 | /* This can't be a string because the preprocessor needs to see it. */ | |
1071 | movl $__USER_DS, %eax | |
1072 | movl %eax, %gs | |
1073 | .endm | |
1074 | ALTERNATIVE "", "ZAP_GS", X86_BUG_NULL_SEG | |
4d732138 IM |
1075 | xorl %eax, %eax |
1076 | movl %eax, %gs | |
1077 | jmp 2b | |
9f1e87ea | 1078 | .previous |
0bd7b798 | 1079 | |
2699500b | 1080 | /* Call softirq on interrupt stack. Interrupts are off. */ |
7d65f4a6 | 1081 | ENTRY(do_softirq_own_stack) |
4d732138 IM |
1082 | pushq %rbp |
1083 | mov %rsp, %rbp | |
8c1f7558 | 1084 | ENTER_IRQ_STACK regs=0 old_rsp=%r11 |
4d732138 | 1085 | call __do_softirq |
8c1f7558 | 1086 | LEAVE_IRQ_STACK regs=0 |
2699500b | 1087 | leaveq |
ed6b676c | 1088 | ret |
8c1f7558 | 1089 | ENDPROC(do_softirq_own_stack) |
75154f40 | 1090 | |
3d75e1b8 | 1091 | #ifdef CONFIG_XEN |
5878d5d6 | 1092 | idtentry hypervisor_callback xen_do_hypervisor_callback has_error_code=0 |
3d75e1b8 JF |
1093 | |
1094 | /* | |
9f1e87ea CG |
1095 | * A note on the "critical region" in our callback handler. |
1096 | * We want to avoid stacking callback handlers due to events occurring | |
1097 | * during handling of the last event. To do this, we keep events disabled | |
1098 | * until we've done all processing. HOWEVER, we must enable events before | |
1099 | * popping the stack frame (can't be done atomically) and so it would still | |
1100 | * be possible to get enough handler activations to overflow the stack. | |
1101 | * Although unlikely, bugs of that kind are hard to track down, so we'd | |
1102 | * like to avoid the possibility. | |
1103 | * So, on entry to the handler we detect whether we interrupted an | |
1104 | * existing activation in its critical region -- if so, we pop the current | |
1105 | * activation and restart the handler using the previous one. | |
1106 | */ | |
4d732138 IM |
1107 | ENTRY(xen_do_hypervisor_callback) /* do_hypervisor_callback(struct *pt_regs) */ |
1108 | ||
9f1e87ea CG |
1109 | /* |
1110 | * Since we don't modify %rdi, evtchn_do_upall(struct *pt_regs) will | |
1111 | * see the correct pointer to the pt_regs | |
1112 | */ | |
8c1f7558 | 1113 | UNWIND_HINT_FUNC |
4d732138 | 1114 | movq %rdi, %rsp /* we don't return, adjust the stack frame */ |
8c1f7558 | 1115 | UNWIND_HINT_REGS |
1d3e53e8 AL |
1116 | |
1117 | ENTER_IRQ_STACK old_rsp=%r10 | |
4d732138 | 1118 | call xen_evtchn_do_upcall |
1d3e53e8 AL |
1119 | LEAVE_IRQ_STACK |
1120 | ||
fdfd811d | 1121 | #ifndef CONFIG_PREEMPT |
4d732138 | 1122 | call xen_maybe_preempt_hcall |
fdfd811d | 1123 | #endif |
4d732138 | 1124 | jmp error_exit |
371c394a | 1125 | END(xen_do_hypervisor_callback) |
3d75e1b8 JF |
1126 | |
1127 | /* | |
9f1e87ea CG |
1128 | * Hypervisor uses this for application faults while it executes. |
1129 | * We get here for two reasons: | |
1130 | * 1. Fault while reloading DS, ES, FS or GS | |
1131 | * 2. Fault while executing IRET | |
1132 | * Category 1 we do not need to fix up as Xen has already reloaded all segment | |
1133 | * registers that could be reloaded and zeroed the others. | |
1134 | * Category 2 we fix up by killing the current process. We cannot use the | |
1135 | * normal Linux return path in this case because if we use the IRET hypercall | |
1136 | * to pop the stack frame we end up in an infinite loop of failsafe callbacks. | |
1137 | * We distinguish between categories by comparing each saved segment register | |
1138 | * with its current contents: any discrepancy means we in category 1. | |
1139 | */ | |
3d75e1b8 | 1140 | ENTRY(xen_failsafe_callback) |
8c1f7558 | 1141 | UNWIND_HINT_EMPTY |
4d732138 IM |
1142 | movl %ds, %ecx |
1143 | cmpw %cx, 0x10(%rsp) | |
1144 | jne 1f | |
1145 | movl %es, %ecx | |
1146 | cmpw %cx, 0x18(%rsp) | |
1147 | jne 1f | |
1148 | movl %fs, %ecx | |
1149 | cmpw %cx, 0x20(%rsp) | |
1150 | jne 1f | |
1151 | movl %gs, %ecx | |
1152 | cmpw %cx, 0x28(%rsp) | |
1153 | jne 1f | |
3d75e1b8 | 1154 | /* All segments match their saved values => Category 2 (Bad IRET). */ |
4d732138 IM |
1155 | movq (%rsp), %rcx |
1156 | movq 8(%rsp), %r11 | |
1157 | addq $0x30, %rsp | |
1158 | pushq $0 /* RIP */ | |
8c1f7558 | 1159 | UNWIND_HINT_IRET_REGS offset=8 |
4d732138 | 1160 | jmp general_protection |
3d75e1b8 | 1161 | 1: /* Segment mismatch => Category 1 (Bad segment). Retry the IRET. */ |
4d732138 IM |
1162 | movq (%rsp), %rcx |
1163 | movq 8(%rsp), %r11 | |
1164 | addq $0x30, %rsp | |
8c1f7558 | 1165 | UNWIND_HINT_IRET_REGS |
4d732138 | 1166 | pushq $-1 /* orig_ax = -1 => not a system call */ |
3f01daec | 1167 | PUSH_AND_CLEAR_REGS |
946c1911 | 1168 | ENCODE_FRAME_POINTER |
4d732138 | 1169 | jmp error_exit |
3d75e1b8 JF |
1170 | END(xen_failsafe_callback) |
1171 | ||
cf910e83 | 1172 | apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \ |
38e20b07 SY |
1173 | xen_hvm_callback_vector xen_evtchn_do_upcall |
1174 | ||
3d75e1b8 | 1175 | #endif /* CONFIG_XEN */ |
ddeb8f21 | 1176 | |
bc2b0331 | 1177 | #if IS_ENABLED(CONFIG_HYPERV) |
cf910e83 | 1178 | apicinterrupt3 HYPERVISOR_CALLBACK_VECTOR \ |
bc2b0331 | 1179 | hyperv_callback_vector hyperv_vector_handler |
93286261 VK |
1180 | |
1181 | apicinterrupt3 HYPERV_REENLIGHTENMENT_VECTOR \ | |
1182 | hyperv_reenlightenment_vector hyperv_reenlightenment_intr | |
248e742a MK |
1183 | |
1184 | apicinterrupt3 HYPERV_STIMER0_VECTOR \ | |
1185 | hv_stimer0_callback_vector hv_stimer0_vector_handler | |
bc2b0331 S |
1186 | #endif /* CONFIG_HYPERV */ |
1187 | ||
4d732138 | 1188 | idtentry debug do_debug has_error_code=0 paranoid=1 shift_ist=DEBUG_STACK |
d8ba61ba | 1189 | idtentry int3 do_int3 has_error_code=0 |
4d732138 IM |
1190 | idtentry stack_segment do_stack_segment has_error_code=1 |
1191 | ||
6cac5a92 | 1192 | #ifdef CONFIG_XEN |
43e41110 | 1193 | idtentry xennmi do_nmi has_error_code=0 |
5878d5d6 JG |
1194 | idtentry xendebug do_debug has_error_code=0 |
1195 | idtentry xenint3 do_int3 has_error_code=0 | |
6cac5a92 | 1196 | #endif |
4d732138 IM |
1197 | |
1198 | idtentry general_protection do_general_protection has_error_code=1 | |
11a7ffb0 | 1199 | idtentry page_fault do_page_fault has_error_code=1 |
4d732138 | 1200 | |
631bc487 | 1201 | #ifdef CONFIG_KVM_GUEST |
4d732138 | 1202 | idtentry async_page_fault do_async_page_fault has_error_code=1 |
631bc487 | 1203 | #endif |
4d732138 | 1204 | |
ddeb8f21 | 1205 | #ifdef CONFIG_X86_MCE |
6f41c34d | 1206 | idtentry machine_check do_mce has_error_code=0 paranoid=1 |
ddeb8f21 AH |
1207 | #endif |
1208 | ||
ebfc453e | 1209 | /* |
9e809d15 | 1210 | * Save all registers in pt_regs, and switch gs if needed. |
ebfc453e DV |
1211 | * Use slow, but surefire "are we in kernel?" check. |
1212 | * Return: ebx=0: need swapgs on exit, ebx=1: otherwise | |
1213 | */ | |
1214 | ENTRY(paranoid_entry) | |
8c1f7558 | 1215 | UNWIND_HINT_FUNC |
1eeb207f | 1216 | cld |
9e809d15 DB |
1217 | PUSH_AND_CLEAR_REGS save_ret=1 |
1218 | ENCODE_FRAME_POINTER 8 | |
4d732138 IM |
1219 | movl $1, %ebx |
1220 | movl $MSR_GS_BASE, %ecx | |
1eeb207f | 1221 | rdmsr |
4d732138 IM |
1222 | testl %edx, %edx |
1223 | js 1f /* negative -> in kernel */ | |
1eeb207f | 1224 | SWAPGS |
4d732138 | 1225 | xorl %ebx, %ebx |
8a09317b DH |
1226 | |
1227 | 1: | |
1228 | SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg=%rax save_reg=%r14 | |
1229 | ||
1230 | ret | |
ebfc453e | 1231 | END(paranoid_entry) |
ddeb8f21 | 1232 | |
ebfc453e DV |
1233 | /* |
1234 | * "Paranoid" exit path from exception stack. This is invoked | |
1235 | * only on return from non-NMI IST interrupts that came | |
1236 | * from kernel space. | |
1237 | * | |
1238 | * We may be returning to very strange contexts (e.g. very early | |
1239 | * in syscall entry), so checking for preemption here would | |
1240 | * be complicated. Fortunately, we there's no good reason | |
1241 | * to try to handle preemption here. | |
4d732138 IM |
1242 | * |
1243 | * On entry, ebx is "no swapgs" flag (1: don't need swapgs, 0: need it) | |
ebfc453e | 1244 | */ |
ddeb8f21 | 1245 | ENTRY(paranoid_exit) |
8c1f7558 | 1246 | UNWIND_HINT_REGS |
2140a994 | 1247 | DISABLE_INTERRUPTS(CLBR_ANY) |
5963e317 | 1248 | TRACE_IRQS_OFF_DEBUG |
4d732138 | 1249 | testl %ebx, %ebx /* swapgs needed? */ |
e5317832 | 1250 | jnz .Lparanoid_exit_no_swapgs |
f2db9382 | 1251 | TRACE_IRQS_IRETQ |
21e94459 | 1252 | RESTORE_CR3 scratch_reg=%rbx save_reg=%r14 |
ddeb8f21 | 1253 | SWAPGS_UNSAFE_STACK |
e5317832 AL |
1254 | jmp .Lparanoid_exit_restore |
1255 | .Lparanoid_exit_no_swapgs: | |
f2db9382 | 1256 | TRACE_IRQS_IRETQ_DEBUG |
e4865757 | 1257 | RESTORE_CR3 scratch_reg=%rbx save_reg=%r14 |
e5317832 AL |
1258 | .Lparanoid_exit_restore: |
1259 | jmp restore_regs_and_return_to_kernel | |
ddeb8f21 AH |
1260 | END(paranoid_exit) |
1261 | ||
1262 | /* | |
9e809d15 | 1263 | * Save all registers in pt_regs, and switch GS if needed. |
ddeb8f21 AH |
1264 | */ |
1265 | ENTRY(error_entry) | |
9e809d15 | 1266 | UNWIND_HINT_FUNC |
ddeb8f21 | 1267 | cld |
9e809d15 DB |
1268 | PUSH_AND_CLEAR_REGS save_ret=1 |
1269 | ENCODE_FRAME_POINTER 8 | |
03335e95 | 1270 | testb $3, CS+8(%rsp) |
cb6f64ed | 1271 | jz .Lerror_kernelspace |
539f5113 | 1272 | |
cb6f64ed AL |
1273 | /* |
1274 | * We entered from user mode or we're pretending to have entered | |
1275 | * from user mode due to an IRET fault. | |
1276 | */ | |
ddeb8f21 | 1277 | SWAPGS |
8a09317b DH |
1278 | /* We have user CR3. Change to kernel CR3. */ |
1279 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rax | |
539f5113 | 1280 | |
cb6f64ed | 1281 | .Lerror_entry_from_usermode_after_swapgs: |
7f2590a1 AL |
1282 | /* Put us onto the real thread stack. */ |
1283 | popq %r12 /* save return addr in %12 */ | |
1284 | movq %rsp, %rdi /* arg0 = pt_regs pointer */ | |
1285 | call sync_regs | |
1286 | movq %rax, %rsp /* switch stack */ | |
1287 | ENCODE_FRAME_POINTER | |
1288 | pushq %r12 | |
1289 | ||
f1075053 AL |
1290 | /* |
1291 | * We need to tell lockdep that IRQs are off. We can't do this until | |
1292 | * we fix gsbase, and we should do it before enter_from_user_mode | |
1293 | * (which can take locks). | |
1294 | */ | |
1295 | TRACE_IRQS_OFF | |
478dc89c | 1296 | CALL_enter_from_user_mode |
f1075053 | 1297 | ret |
02bc7768 | 1298 | |
cb6f64ed | 1299 | .Lerror_entry_done: |
ddeb8f21 AH |
1300 | TRACE_IRQS_OFF |
1301 | ret | |
ddeb8f21 | 1302 | |
ebfc453e DV |
1303 | /* |
1304 | * There are two places in the kernel that can potentially fault with | |
1305 | * usergs. Handle them here. B stepping K8s sometimes report a | |
1306 | * truncated RIP for IRET exceptions returning to compat mode. Check | |
1307 | * for these here too. | |
1308 | */ | |
cb6f64ed | 1309 | .Lerror_kernelspace: |
4d732138 IM |
1310 | leaq native_irq_return_iret(%rip), %rcx |
1311 | cmpq %rcx, RIP+8(%rsp) | |
cb6f64ed | 1312 | je .Lerror_bad_iret |
4d732138 IM |
1313 | movl %ecx, %eax /* zero extend */ |
1314 | cmpq %rax, RIP+8(%rsp) | |
cb6f64ed | 1315 | je .Lbstep_iret |
42c748bb | 1316 | cmpq $.Lgs_change, RIP+8(%rsp) |
cb6f64ed | 1317 | jne .Lerror_entry_done |
539f5113 AL |
1318 | |
1319 | /* | |
42c748bb | 1320 | * hack: .Lgs_change can fail with user gsbase. If this happens, fix up |
539f5113 | 1321 | * gsbase and proceed. We'll fix up the exception and land in |
42c748bb | 1322 | * .Lgs_change's error handler with kernel gsbase. |
539f5113 | 1323 | */ |
2fa5f04f | 1324 | SWAPGS |
8a09317b | 1325 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rax |
2fa5f04f | 1326 | jmp .Lerror_entry_done |
ae24ffe5 | 1327 | |
cb6f64ed | 1328 | .Lbstep_iret: |
ae24ffe5 | 1329 | /* Fix truncated RIP */ |
4d732138 | 1330 | movq %rcx, RIP+8(%rsp) |
b645af2d AL |
1331 | /* fall through */ |
1332 | ||
cb6f64ed | 1333 | .Lerror_bad_iret: |
539f5113 | 1334 | /* |
8a09317b DH |
1335 | * We came from an IRET to user mode, so we have user |
1336 | * gsbase and CR3. Switch to kernel gsbase and CR3: | |
539f5113 | 1337 | */ |
b645af2d | 1338 | SWAPGS |
8a09317b | 1339 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rax |
539f5113 AL |
1340 | |
1341 | /* | |
1342 | * Pretend that the exception came from user mode: set up pt_regs | |
b3681dd5 | 1343 | * as if we faulted immediately after IRET. |
539f5113 | 1344 | */ |
4d732138 IM |
1345 | mov %rsp, %rdi |
1346 | call fixup_bad_iret | |
1347 | mov %rax, %rsp | |
cb6f64ed | 1348 | jmp .Lerror_entry_from_usermode_after_swapgs |
ddeb8f21 AH |
1349 | END(error_entry) |
1350 | ||
ddeb8f21 | 1351 | ENTRY(error_exit) |
8c1f7558 | 1352 | UNWIND_HINT_REGS |
2140a994 | 1353 | DISABLE_INTERRUPTS(CLBR_ANY) |
ddeb8f21 | 1354 | TRACE_IRQS_OFF |
b3681dd5 AL |
1355 | testb $3, CS(%rsp) |
1356 | jz retint_kernel | |
4d732138 | 1357 | jmp retint_user |
ddeb8f21 AH |
1358 | END(error_exit) |
1359 | ||
929bacec AL |
1360 | /* |
1361 | * Runs on exception stack. Xen PV does not go through this path at all, | |
1362 | * so we can use real assembly here. | |
8a09317b DH |
1363 | * |
1364 | * Registers: | |
1365 | * %r14: Used to save/restore the CR3 of the interrupted context | |
1366 | * when PAGE_TABLE_ISOLATION is in use. Do not clobber. | |
929bacec | 1367 | */ |
ddeb8f21 | 1368 | ENTRY(nmi) |
8c1f7558 | 1369 | UNWIND_HINT_IRET_REGS |
929bacec | 1370 | |
3f3c8b8c SR |
1371 | /* |
1372 | * We allow breakpoints in NMIs. If a breakpoint occurs, then | |
1373 | * the iretq it performs will take us out of NMI context. | |
1374 | * This means that we can have nested NMIs where the next | |
1375 | * NMI is using the top of the stack of the previous NMI. We | |
1376 | * can't let it execute because the nested NMI will corrupt the | |
1377 | * stack of the previous NMI. NMI handlers are not re-entrant | |
1378 | * anyway. | |
1379 | * | |
1380 | * To handle this case we do the following: | |
1381 | * Check the a special location on the stack that contains | |
1382 | * a variable that is set when NMIs are executing. | |
1383 | * The interrupted task's stack is also checked to see if it | |
1384 | * is an NMI stack. | |
1385 | * If the variable is not set and the stack is not the NMI | |
1386 | * stack then: | |
1387 | * o Set the special variable on the stack | |
0b22930e AL |
1388 | * o Copy the interrupt frame into an "outermost" location on the |
1389 | * stack | |
1390 | * o Copy the interrupt frame into an "iret" location on the stack | |
3f3c8b8c SR |
1391 | * o Continue processing the NMI |
1392 | * If the variable is set or the previous stack is the NMI stack: | |
0b22930e | 1393 | * o Modify the "iret" location to jump to the repeat_nmi |
3f3c8b8c SR |
1394 | * o return back to the first NMI |
1395 | * | |
1396 | * Now on exit of the first NMI, we first clear the stack variable | |
1397 | * The NMI stack will tell any nested NMIs at that point that it is | |
1398 | * nested. Then we pop the stack normally with iret, and if there was | |
1399 | * a nested NMI that updated the copy interrupt stack frame, a | |
1400 | * jump will be made to the repeat_nmi code that will handle the second | |
1401 | * NMI. | |
9b6e6a83 AL |
1402 | * |
1403 | * However, espfix prevents us from directly returning to userspace | |
1404 | * with a single IRET instruction. Similarly, IRET to user mode | |
1405 | * can fault. We therefore handle NMIs from user space like | |
1406 | * other IST entries. | |
3f3c8b8c SR |
1407 | */ |
1408 | ||
e93c1730 AL |
1409 | ASM_CLAC |
1410 | ||
146b2b09 | 1411 | /* Use %rdx as our temp variable throughout */ |
4d732138 | 1412 | pushq %rdx |
3f3c8b8c | 1413 | |
9b6e6a83 AL |
1414 | testb $3, CS-RIP+8(%rsp) |
1415 | jz .Lnmi_from_kernel | |
1416 | ||
1417 | /* | |
1418 | * NMI from user mode. We need to run on the thread stack, but we | |
1419 | * can't go through the normal entry paths: NMIs are masked, and | |
1420 | * we don't want to enable interrupts, because then we'll end | |
1421 | * up in an awkward situation in which IRQs are on but NMIs | |
1422 | * are off. | |
83c133cf AL |
1423 | * |
1424 | * We also must not push anything to the stack before switching | |
1425 | * stacks lest we corrupt the "NMI executing" variable. | |
9b6e6a83 AL |
1426 | */ |
1427 | ||
929bacec | 1428 | swapgs |
9b6e6a83 | 1429 | cld |
8a09317b | 1430 | SWITCH_TO_KERNEL_CR3 scratch_reg=%rdx |
9b6e6a83 AL |
1431 | movq %rsp, %rdx |
1432 | movq PER_CPU_VAR(cpu_current_top_of_stack), %rsp | |
8c1f7558 | 1433 | UNWIND_HINT_IRET_REGS base=%rdx offset=8 |
9b6e6a83 AL |
1434 | pushq 5*8(%rdx) /* pt_regs->ss */ |
1435 | pushq 4*8(%rdx) /* pt_regs->rsp */ | |
1436 | pushq 3*8(%rdx) /* pt_regs->flags */ | |
1437 | pushq 2*8(%rdx) /* pt_regs->cs */ | |
1438 | pushq 1*8(%rdx) /* pt_regs->rip */ | |
8c1f7558 | 1439 | UNWIND_HINT_IRET_REGS |
9b6e6a83 | 1440 | pushq $-1 /* pt_regs->orig_ax */ |
30907fd1 | 1441 | PUSH_AND_CLEAR_REGS rdx=(%rdx) |
946c1911 | 1442 | ENCODE_FRAME_POINTER |
9b6e6a83 AL |
1443 | |
1444 | /* | |
1445 | * At this point we no longer need to worry about stack damage | |
1446 | * due to nesting -- we're on the normal thread stack and we're | |
1447 | * done with the NMI stack. | |
1448 | */ | |
1449 | ||
1450 | movq %rsp, %rdi | |
1451 | movq $-1, %rsi | |
1452 | call do_nmi | |
1453 | ||
45d5a168 | 1454 | /* |
9b6e6a83 | 1455 | * Return back to user mode. We must *not* do the normal exit |
946c1911 | 1456 | * work, because we don't want to enable interrupts. |
45d5a168 | 1457 | */ |
8a055d7f | 1458 | jmp swapgs_restore_regs_and_return_to_usermode |
45d5a168 | 1459 | |
9b6e6a83 | 1460 | .Lnmi_from_kernel: |
3f3c8b8c | 1461 | /* |
0b22930e AL |
1462 | * Here's what our stack frame will look like: |
1463 | * +---------------------------------------------------------+ | |
1464 | * | original SS | | |
1465 | * | original Return RSP | | |
1466 | * | original RFLAGS | | |
1467 | * | original CS | | |
1468 | * | original RIP | | |
1469 | * +---------------------------------------------------------+ | |
1470 | * | temp storage for rdx | | |
1471 | * +---------------------------------------------------------+ | |
1472 | * | "NMI executing" variable | | |
1473 | * +---------------------------------------------------------+ | |
1474 | * | iret SS } Copied from "outermost" frame | | |
1475 | * | iret Return RSP } on each loop iteration; overwritten | | |
1476 | * | iret RFLAGS } by a nested NMI to force another | | |
1477 | * | iret CS } iteration if needed. | | |
1478 | * | iret RIP } | | |
1479 | * +---------------------------------------------------------+ | |
1480 | * | outermost SS } initialized in first_nmi; | | |
1481 | * | outermost Return RSP } will not be changed before | | |
1482 | * | outermost RFLAGS } NMI processing is done. | | |
1483 | * | outermost CS } Copied to "iret" frame on each | | |
1484 | * | outermost RIP } iteration. | | |
1485 | * +---------------------------------------------------------+ | |
1486 | * | pt_regs | | |
1487 | * +---------------------------------------------------------+ | |
1488 | * | |
1489 | * The "original" frame is used by hardware. Before re-enabling | |
1490 | * NMIs, we need to be done with it, and we need to leave enough | |
1491 | * space for the asm code here. | |
1492 | * | |
1493 | * We return by executing IRET while RSP points to the "iret" frame. | |
1494 | * That will either return for real or it will loop back into NMI | |
1495 | * processing. | |
1496 | * | |
1497 | * The "outermost" frame is copied to the "iret" frame on each | |
1498 | * iteration of the loop, so each iteration starts with the "iret" | |
1499 | * frame pointing to the final return target. | |
1500 | */ | |
1501 | ||
45d5a168 | 1502 | /* |
0b22930e AL |
1503 | * Determine whether we're a nested NMI. |
1504 | * | |
a27507ca AL |
1505 | * If we interrupted kernel code between repeat_nmi and |
1506 | * end_repeat_nmi, then we are a nested NMI. We must not | |
1507 | * modify the "iret" frame because it's being written by | |
1508 | * the outer NMI. That's okay; the outer NMI handler is | |
1509 | * about to about to call do_nmi anyway, so we can just | |
1510 | * resume the outer NMI. | |
45d5a168 | 1511 | */ |
a27507ca AL |
1512 | |
1513 | movq $repeat_nmi, %rdx | |
1514 | cmpq 8(%rsp), %rdx | |
1515 | ja 1f | |
1516 | movq $end_repeat_nmi, %rdx | |
1517 | cmpq 8(%rsp), %rdx | |
1518 | ja nested_nmi_out | |
1519 | 1: | |
45d5a168 | 1520 | |
3f3c8b8c | 1521 | /* |
a27507ca | 1522 | * Now check "NMI executing". If it's set, then we're nested. |
0b22930e AL |
1523 | * This will not detect if we interrupted an outer NMI just |
1524 | * before IRET. | |
3f3c8b8c | 1525 | */ |
4d732138 IM |
1526 | cmpl $1, -8(%rsp) |
1527 | je nested_nmi | |
3f3c8b8c SR |
1528 | |
1529 | /* | |
0b22930e AL |
1530 | * Now test if the previous stack was an NMI stack. This covers |
1531 | * the case where we interrupt an outer NMI after it clears | |
810bc075 AL |
1532 | * "NMI executing" but before IRET. We need to be careful, though: |
1533 | * there is one case in which RSP could point to the NMI stack | |
1534 | * despite there being no NMI active: naughty userspace controls | |
1535 | * RSP at the very beginning of the SYSCALL targets. We can | |
1536 | * pull a fast one on naughty userspace, though: we program | |
1537 | * SYSCALL to mask DF, so userspace cannot cause DF to be set | |
1538 | * if it controls the kernel's RSP. We set DF before we clear | |
1539 | * "NMI executing". | |
3f3c8b8c | 1540 | */ |
0784b364 DV |
1541 | lea 6*8(%rsp), %rdx |
1542 | /* Compare the NMI stack (rdx) with the stack we came from (4*8(%rsp)) */ | |
1543 | cmpq %rdx, 4*8(%rsp) | |
1544 | /* If the stack pointer is above the NMI stack, this is a normal NMI */ | |
1545 | ja first_nmi | |
4d732138 | 1546 | |
0784b364 DV |
1547 | subq $EXCEPTION_STKSZ, %rdx |
1548 | cmpq %rdx, 4*8(%rsp) | |
1549 | /* If it is below the NMI stack, it is a normal NMI */ | |
1550 | jb first_nmi | |
810bc075 AL |
1551 | |
1552 | /* Ah, it is within the NMI stack. */ | |
1553 | ||
1554 | testb $(X86_EFLAGS_DF >> 8), (3*8 + 1)(%rsp) | |
1555 | jz first_nmi /* RSP was user controlled. */ | |
1556 | ||
1557 | /* This is a nested NMI. */ | |
0784b364 | 1558 | |
3f3c8b8c SR |
1559 | nested_nmi: |
1560 | /* | |
0b22930e AL |
1561 | * Modify the "iret" frame to point to repeat_nmi, forcing another |
1562 | * iteration of NMI handling. | |
3f3c8b8c | 1563 | */ |
23a781e9 | 1564 | subq $8, %rsp |
4d732138 IM |
1565 | leaq -10*8(%rsp), %rdx |
1566 | pushq $__KERNEL_DS | |
1567 | pushq %rdx | |
131484c8 | 1568 | pushfq |
4d732138 IM |
1569 | pushq $__KERNEL_CS |
1570 | pushq $repeat_nmi | |
3f3c8b8c SR |
1571 | |
1572 | /* Put stack back */ | |
4d732138 | 1573 | addq $(6*8), %rsp |
3f3c8b8c SR |
1574 | |
1575 | nested_nmi_out: | |
4d732138 | 1576 | popq %rdx |
3f3c8b8c | 1577 | |
0b22930e | 1578 | /* We are returning to kernel mode, so this cannot result in a fault. */ |
929bacec | 1579 | iretq |
3f3c8b8c SR |
1580 | |
1581 | first_nmi: | |
0b22930e | 1582 | /* Restore rdx. */ |
4d732138 | 1583 | movq (%rsp), %rdx |
62610913 | 1584 | |
36f1a77b AL |
1585 | /* Make room for "NMI executing". */ |
1586 | pushq $0 | |
3f3c8b8c | 1587 | |
0b22930e | 1588 | /* Leave room for the "iret" frame */ |
4d732138 | 1589 | subq $(5*8), %rsp |
28696f43 | 1590 | |
0b22930e | 1591 | /* Copy the "original" frame to the "outermost" frame */ |
3f3c8b8c | 1592 | .rept 5 |
4d732138 | 1593 | pushq 11*8(%rsp) |
3f3c8b8c | 1594 | .endr |
8c1f7558 | 1595 | UNWIND_HINT_IRET_REGS |
62610913 | 1596 | |
79fb4ad6 SR |
1597 | /* Everything up to here is safe from nested NMIs */ |
1598 | ||
a97439aa AL |
1599 | #ifdef CONFIG_DEBUG_ENTRY |
1600 | /* | |
1601 | * For ease of testing, unmask NMIs right away. Disabled by | |
1602 | * default because IRET is very expensive. | |
1603 | */ | |
1604 | pushq $0 /* SS */ | |
1605 | pushq %rsp /* RSP (minus 8 because of the previous push) */ | |
1606 | addq $8, (%rsp) /* Fix up RSP */ | |
1607 | pushfq /* RFLAGS */ | |
1608 | pushq $__KERNEL_CS /* CS */ | |
1609 | pushq $1f /* RIP */ | |
929bacec | 1610 | iretq /* continues at repeat_nmi below */ |
8c1f7558 | 1611 | UNWIND_HINT_IRET_REGS |
a97439aa AL |
1612 | 1: |
1613 | #endif | |
1614 | ||
0b22930e | 1615 | repeat_nmi: |
62610913 JB |
1616 | /* |
1617 | * If there was a nested NMI, the first NMI's iret will return | |
1618 | * here. But NMIs are still enabled and we can take another | |
1619 | * nested NMI. The nested NMI checks the interrupted RIP to see | |
1620 | * if it is between repeat_nmi and end_repeat_nmi, and if so | |
1621 | * it will just return, as we are about to repeat an NMI anyway. | |
1622 | * This makes it safe to copy to the stack frame that a nested | |
1623 | * NMI will update. | |
0b22930e AL |
1624 | * |
1625 | * RSP is pointing to "outermost RIP". gsbase is unknown, but, if | |
1626 | * we're repeating an NMI, gsbase has the same value that it had on | |
1627 | * the first iteration. paranoid_entry will load the kernel | |
36f1a77b AL |
1628 | * gsbase if needed before we call do_nmi. "NMI executing" |
1629 | * is zero. | |
62610913 | 1630 | */ |
36f1a77b | 1631 | movq $1, 10*8(%rsp) /* Set "NMI executing". */ |
3f3c8b8c | 1632 | |
62610913 | 1633 | /* |
0b22930e AL |
1634 | * Copy the "outermost" frame to the "iret" frame. NMIs that nest |
1635 | * here must not modify the "iret" frame while we're writing to | |
1636 | * it or it will end up containing garbage. | |
62610913 | 1637 | */ |
4d732138 | 1638 | addq $(10*8), %rsp |
3f3c8b8c | 1639 | .rept 5 |
4d732138 | 1640 | pushq -6*8(%rsp) |
3f3c8b8c | 1641 | .endr |
4d732138 | 1642 | subq $(5*8), %rsp |
62610913 | 1643 | end_repeat_nmi: |
3f3c8b8c SR |
1644 | |
1645 | /* | |
0b22930e AL |
1646 | * Everything below this point can be preempted by a nested NMI. |
1647 | * If this happens, then the inner NMI will change the "iret" | |
1648 | * frame to point back to repeat_nmi. | |
3f3c8b8c | 1649 | */ |
4d732138 | 1650 | pushq $-1 /* ORIG_RAX: no syscall to restart */ |
76f5df43 | 1651 | |
1fd466ef | 1652 | /* |
ebfc453e | 1653 | * Use paranoid_entry to handle SWAPGS, but no need to use paranoid_exit |
1fd466ef SR |
1654 | * as we should not be calling schedule in NMI context. |
1655 | * Even with normal interrupts enabled. An NMI should not be | |
1656 | * setting NEED_RESCHED or anything that normal interrupts and | |
1657 | * exceptions might do. | |
1658 | */ | |
4d732138 | 1659 | call paranoid_entry |
8c1f7558 | 1660 | UNWIND_HINT_REGS |
7fbb98c5 | 1661 | |
ddeb8f21 | 1662 | /* paranoidentry do_nmi, 0; without TRACE_IRQS_OFF */ |
4d732138 IM |
1663 | movq %rsp, %rdi |
1664 | movq $-1, %rsi | |
1665 | call do_nmi | |
7fbb98c5 | 1666 | |
21e94459 | 1667 | RESTORE_CR3 scratch_reg=%r15 save_reg=%r14 |
8a09317b | 1668 | |
4d732138 IM |
1669 | testl %ebx, %ebx /* swapgs needed? */ |
1670 | jnz nmi_restore | |
ddeb8f21 AH |
1671 | nmi_swapgs: |
1672 | SWAPGS_UNSAFE_STACK | |
1673 | nmi_restore: | |
502af0d7 | 1674 | POP_REGS |
0b22930e | 1675 | |
471ee483 AL |
1676 | /* |
1677 | * Skip orig_ax and the "outermost" frame to point RSP at the "iret" | |
1678 | * at the "iret" frame. | |
1679 | */ | |
1680 | addq $6*8, %rsp | |
28696f43 | 1681 | |
810bc075 AL |
1682 | /* |
1683 | * Clear "NMI executing". Set DF first so that we can easily | |
1684 | * distinguish the remaining code between here and IRET from | |
929bacec AL |
1685 | * the SYSCALL entry and exit paths. |
1686 | * | |
1687 | * We arguably should just inspect RIP instead, but I (Andy) wrote | |
1688 | * this code when I had the misapprehension that Xen PV supported | |
1689 | * NMIs, and Xen PV would break that approach. | |
810bc075 AL |
1690 | */ |
1691 | std | |
1692 | movq $0, 5*8(%rsp) /* clear "NMI executing" */ | |
0b22930e AL |
1693 | |
1694 | /* | |
929bacec AL |
1695 | * iretq reads the "iret" frame and exits the NMI stack in a |
1696 | * single instruction. We are returning to kernel mode, so this | |
1697 | * cannot result in a fault. Similarly, we don't need to worry | |
1698 | * about espfix64 on the way back to kernel mode. | |
0b22930e | 1699 | */ |
929bacec | 1700 | iretq |
ddeb8f21 AH |
1701 | END(nmi) |
1702 | ||
1703 | ENTRY(ignore_sysret) | |
8c1f7558 | 1704 | UNWIND_HINT_EMPTY |
4d732138 | 1705 | mov $-ENOSYS, %eax |
ddeb8f21 | 1706 | sysret |
ddeb8f21 | 1707 | END(ignore_sysret) |
2deb4be2 AL |
1708 | |
1709 | ENTRY(rewind_stack_do_exit) | |
8c1f7558 | 1710 | UNWIND_HINT_FUNC |
2deb4be2 AL |
1711 | /* Prevent any naive code from trying to unwind to our caller. */ |
1712 | xorl %ebp, %ebp | |
1713 | ||
1714 | movq PER_CPU_VAR(cpu_current_top_of_stack), %rax | |
8c1f7558 JP |
1715 | leaq -PTREGS_SIZE(%rax), %rsp |
1716 | UNWIND_HINT_FUNC sp_offset=PTREGS_SIZE | |
2deb4be2 AL |
1717 | |
1718 | call do_exit | |
2deb4be2 | 1719 | END(rewind_stack_do_exit) |