.. | .. |
---|
| 1 | +// SPDX-License-Identifier: GPL-2.0-only |
---|
1 | 2 | /* |
---|
2 | 3 | * common.c - C code for kernel entry and exit |
---|
3 | 4 | * Copyright (c) 2015 Andrew Lutomirski |
---|
4 | | - * GPL v2 |
---|
5 | 5 | * |
---|
6 | 6 | * Based on asm and ptrace code by many authors. The code here originated |
---|
7 | 7 | * in ptrace.c and signal.c. |
---|
.. | .. |
---|
10 | 10 | #include <linux/kernel.h> |
---|
11 | 11 | #include <linux/sched.h> |
---|
12 | 12 | #include <linux/sched/task_stack.h> |
---|
| 13 | +#include <linux/entry-common.h> |
---|
13 | 14 | #include <linux/mm.h> |
---|
14 | 15 | #include <linux/smp.h> |
---|
15 | 16 | #include <linux/errno.h> |
---|
16 | 17 | #include <linux/ptrace.h> |
---|
17 | | -#include <linux/tracehook.h> |
---|
18 | | -#include <linux/audit.h> |
---|
19 | | -#include <linux/seccomp.h> |
---|
20 | | -#include <linux/signal.h> |
---|
21 | 18 | #include <linux/export.h> |
---|
22 | | -#include <linux/context_tracking.h> |
---|
23 | | -#include <linux/user-return-notifier.h> |
---|
24 | 19 | #include <linux/nospec.h> |
---|
25 | | -#include <linux/uprobes.h> |
---|
26 | | -#include <linux/livepatch.h> |
---|
27 | 20 | #include <linux/syscalls.h> |
---|
| 21 | +#include <linux/uaccess.h> |
---|
| 22 | + |
---|
| 23 | +#ifdef CONFIG_XEN_PV |
---|
| 24 | +#include <xen/xen-ops.h> |
---|
| 25 | +#include <xen/events.h> |
---|
| 26 | +#endif |
---|
28 | 27 | |
---|
29 | 28 | #include <asm/desc.h> |
---|
30 | 29 | #include <asm/traps.h> |
---|
31 | 30 | #include <asm/vdso.h> |
---|
32 | | -#include <linux/uaccess.h> |
---|
33 | 31 | #include <asm/cpufeature.h> |
---|
| 32 | +#include <asm/fpu/api.h> |
---|
34 | 33 | #include <asm/nospec-branch.h> |
---|
35 | | - |
---|
36 | | -#define CREATE_TRACE_POINTS |
---|
37 | | -#include <trace/events/syscalls.h> |
---|
38 | | - |
---|
39 | | -#ifdef CONFIG_CONTEXT_TRACKING |
---|
40 | | -/* Called on entry from user mode with IRQs off. */ |
---|
41 | | -__visible inline void enter_from_user_mode(void) |
---|
42 | | -{ |
---|
43 | | - CT_WARN_ON(ct_state() != CONTEXT_USER); |
---|
44 | | - user_exit_irqoff(); |
---|
45 | | -} |
---|
46 | | -#else |
---|
47 | | -static inline void enter_from_user_mode(void) {} |
---|
48 | | -#endif |
---|
49 | | - |
---|
50 | | -static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch) |
---|
51 | | -{ |
---|
52 | | -#ifdef CONFIG_X86_64 |
---|
53 | | - if (arch == AUDIT_ARCH_X86_64) { |
---|
54 | | - audit_syscall_entry(regs->orig_ax, regs->di, |
---|
55 | | - regs->si, regs->dx, regs->r10); |
---|
56 | | - } else |
---|
57 | | -#endif |
---|
58 | | - { |
---|
59 | | - audit_syscall_entry(regs->orig_ax, regs->bx, |
---|
60 | | - regs->cx, regs->dx, regs->si); |
---|
61 | | - } |
---|
62 | | -} |
---|
63 | | - |
---|
64 | | -/* |
---|
65 | | - * Returns the syscall nr to run (which should match regs->orig_ax) or -1 |
---|
66 | | - * to skip the syscall. |
---|
67 | | - */ |
---|
68 | | -static long syscall_trace_enter(struct pt_regs *regs) |
---|
69 | | -{ |
---|
70 | | - u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64; |
---|
71 | | - |
---|
72 | | - struct thread_info *ti = current_thread_info(); |
---|
73 | | - unsigned long ret = 0; |
---|
74 | | - bool emulated = false; |
---|
75 | | - u32 work; |
---|
76 | | - |
---|
77 | | - if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) |
---|
78 | | - BUG_ON(regs != task_pt_regs(current)); |
---|
79 | | - |
---|
80 | | - work = READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY; |
---|
81 | | - |
---|
82 | | - if (unlikely(work & _TIF_SYSCALL_EMU)) |
---|
83 | | - emulated = true; |
---|
84 | | - |
---|
85 | | - if ((emulated || (work & _TIF_SYSCALL_TRACE)) && |
---|
86 | | - tracehook_report_syscall_entry(regs)) |
---|
87 | | - return -1L; |
---|
88 | | - |
---|
89 | | - if (emulated) |
---|
90 | | - return -1L; |
---|
91 | | - |
---|
92 | | -#ifdef CONFIG_SECCOMP |
---|
93 | | - /* |
---|
94 | | - * Do seccomp after ptrace, to catch any tracer changes. |
---|
95 | | - */ |
---|
96 | | - if (work & _TIF_SECCOMP) { |
---|
97 | | - struct seccomp_data sd; |
---|
98 | | - |
---|
99 | | - sd.arch = arch; |
---|
100 | | - sd.nr = regs->orig_ax; |
---|
101 | | - sd.instruction_pointer = regs->ip; |
---|
102 | | -#ifdef CONFIG_X86_64 |
---|
103 | | - if (arch == AUDIT_ARCH_X86_64) { |
---|
104 | | - sd.args[0] = regs->di; |
---|
105 | | - sd.args[1] = regs->si; |
---|
106 | | - sd.args[2] = regs->dx; |
---|
107 | | - sd.args[3] = regs->r10; |
---|
108 | | - sd.args[4] = regs->r8; |
---|
109 | | - sd.args[5] = regs->r9; |
---|
110 | | - } else |
---|
111 | | -#endif |
---|
112 | | - { |
---|
113 | | - sd.args[0] = regs->bx; |
---|
114 | | - sd.args[1] = regs->cx; |
---|
115 | | - sd.args[2] = regs->dx; |
---|
116 | | - sd.args[3] = regs->si; |
---|
117 | | - sd.args[4] = regs->di; |
---|
118 | | - sd.args[5] = regs->bp; |
---|
119 | | - } |
---|
120 | | - |
---|
121 | | - ret = __secure_computing(&sd); |
---|
122 | | - if (ret == -1) |
---|
123 | | - return ret; |
---|
124 | | - } |
---|
125 | | -#endif |
---|
126 | | - |
---|
127 | | - if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT))) |
---|
128 | | - trace_sys_enter(regs, regs->orig_ax); |
---|
129 | | - |
---|
130 | | - do_audit_syscall_entry(regs, arch); |
---|
131 | | - |
---|
132 | | - return ret ?: regs->orig_ax; |
---|
133 | | -} |
---|
134 | | - |
---|
135 | | -#define EXIT_TO_USERMODE_LOOP_FLAGS \ |
---|
136 | | - (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \ |
---|
137 | | - _TIF_NEED_RESCHED_MASK | _TIF_USER_RETURN_NOTIFY | _TIF_PATCH_PENDING) |
---|
138 | | - |
---|
139 | | -static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags) |
---|
140 | | -{ |
---|
141 | | - /* |
---|
142 | | - * In order to return to user mode, we need to have IRQs off with |
---|
143 | | - * none of EXIT_TO_USERMODE_LOOP_FLAGS set. Several of these flags |
---|
144 | | - * can be set at any time on preemptable kernels if we have IRQs on, |
---|
145 | | - * so we need to loop. Disabling preemption wouldn't help: doing the |
---|
146 | | - * work to clear some of the flags can sleep. |
---|
147 | | - */ |
---|
148 | | - while (true) { |
---|
149 | | - /* We have work to do. */ |
---|
150 | | - local_irq_enable(); |
---|
151 | | - |
---|
152 | | - if (cached_flags & _TIF_NEED_RESCHED_MASK) |
---|
153 | | - schedule(); |
---|
154 | | - |
---|
155 | | -#ifdef ARCH_RT_DELAYS_SIGNAL_SEND |
---|
156 | | - if (unlikely(current->forced_info.si_signo)) { |
---|
157 | | - struct task_struct *t = current; |
---|
158 | | - force_sig_info(t->forced_info.si_signo, &t->forced_info, t); |
---|
159 | | - t->forced_info.si_signo = 0; |
---|
160 | | - } |
---|
161 | | -#endif |
---|
162 | | - if (cached_flags & _TIF_UPROBE) |
---|
163 | | - uprobe_notify_resume(regs); |
---|
164 | | - |
---|
165 | | - if (cached_flags & _TIF_PATCH_PENDING) |
---|
166 | | - klp_update_patch_state(current); |
---|
167 | | - |
---|
168 | | - /* deal with pending signal delivery */ |
---|
169 | | - if (cached_flags & _TIF_SIGPENDING) |
---|
170 | | - do_signal(regs); |
---|
171 | | - |
---|
172 | | - if (cached_flags & _TIF_NOTIFY_RESUME) { |
---|
173 | | - clear_thread_flag(TIF_NOTIFY_RESUME); |
---|
174 | | - tracehook_notify_resume(regs); |
---|
175 | | - rseq_handle_notify_resume(NULL, regs); |
---|
176 | | - } |
---|
177 | | - |
---|
178 | | - if (cached_flags & _TIF_USER_RETURN_NOTIFY) |
---|
179 | | - fire_user_return_notifiers(); |
---|
180 | | - |
---|
181 | | - /* Disable IRQs and retry */ |
---|
182 | | - local_irq_disable(); |
---|
183 | | - |
---|
184 | | - cached_flags = READ_ONCE(current_thread_info()->flags); |
---|
185 | | - |
---|
186 | | - if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) |
---|
187 | | - break; |
---|
188 | | - } |
---|
189 | | -} |
---|
190 | | - |
---|
191 | | -/* Called with IRQs disabled. */ |
---|
192 | | -__visible inline void prepare_exit_to_usermode(struct pt_regs *regs) |
---|
193 | | -{ |
---|
194 | | - struct thread_info *ti = current_thread_info(); |
---|
195 | | - u32 cached_flags; |
---|
196 | | - |
---|
197 | | - addr_limit_user_check(); |
---|
198 | | - |
---|
199 | | - lockdep_assert_irqs_disabled(); |
---|
200 | | - lockdep_sys_exit(); |
---|
201 | | - |
---|
202 | | - cached_flags = READ_ONCE(ti->flags); |
---|
203 | | - |
---|
204 | | - if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) |
---|
205 | | - exit_to_usermode_loop(regs, cached_flags); |
---|
206 | | - |
---|
207 | | -#ifdef CONFIG_COMPAT |
---|
208 | | - /* |
---|
209 | | - * Compat syscalls set TS_COMPAT. Make sure we clear it before |
---|
210 | | - * returning to user mode. We need to clear it *after* signal |
---|
211 | | - * handling, because syscall restart has a fixup for compat |
---|
212 | | - * syscalls. The fixup is exercised by the ptrace_syscall_32 |
---|
213 | | - * selftest. |
---|
214 | | - * |
---|
215 | | - * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer |
---|
216 | | - * special case only applies after poking regs and before the |
---|
217 | | - * very next return to user mode. |
---|
218 | | - */ |
---|
219 | | - ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED); |
---|
220 | | -#endif |
---|
221 | | - |
---|
222 | | - user_enter_irqoff(); |
---|
223 | | - |
---|
224 | | - mds_user_clear_cpu_buffers(); |
---|
225 | | -} |
---|
226 | | - |
---|
227 | | -#define SYSCALL_EXIT_WORK_FLAGS \ |
---|
228 | | - (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \ |
---|
229 | | - _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT) |
---|
230 | | - |
---|
231 | | -static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags) |
---|
232 | | -{ |
---|
233 | | - bool step; |
---|
234 | | - |
---|
235 | | - audit_syscall_exit(regs); |
---|
236 | | - |
---|
237 | | - if (cached_flags & _TIF_SYSCALL_TRACEPOINT) |
---|
238 | | - trace_sys_exit(regs, regs->ax); |
---|
239 | | - |
---|
240 | | - /* |
---|
241 | | - * If TIF_SYSCALL_EMU is set, we only get here because of |
---|
242 | | - * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP). |
---|
243 | | - * We already reported this syscall instruction in |
---|
244 | | - * syscall_trace_enter(). |
---|
245 | | - */ |
---|
246 | | - step = unlikely( |
---|
247 | | - (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU)) |
---|
248 | | - == _TIF_SINGLESTEP); |
---|
249 | | - if (step || cached_flags & _TIF_SYSCALL_TRACE) |
---|
250 | | - tracehook_report_syscall_exit(regs, step); |
---|
251 | | -} |
---|
252 | | - |
---|
253 | | -/* |
---|
254 | | - * Called with IRQs on and fully valid regs. Returns with IRQs off in a |
---|
255 | | - * state such that we can immediately switch to user mode. |
---|
256 | | - */ |
---|
257 | | -__visible inline void syscall_return_slowpath(struct pt_regs *regs) |
---|
258 | | -{ |
---|
259 | | - struct thread_info *ti = current_thread_info(); |
---|
260 | | - u32 cached_flags = READ_ONCE(ti->flags); |
---|
261 | | - |
---|
262 | | - CT_WARN_ON(ct_state() != CONTEXT_KERNEL); |
---|
263 | | - |
---|
264 | | - if (IS_ENABLED(CONFIG_PROVE_LOCKING) && |
---|
265 | | - WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax)) |
---|
266 | | - local_irq_enable(); |
---|
267 | | - |
---|
268 | | - rseq_syscall(regs); |
---|
269 | | - |
---|
270 | | - /* |
---|
271 | | - * First do one-time work. If these work items are enabled, we |
---|
272 | | - * want to run them exactly once per syscall exit with IRQs on. |
---|
273 | | - */ |
---|
274 | | - if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS)) |
---|
275 | | - syscall_slow_exit_work(regs, cached_flags); |
---|
276 | | - |
---|
277 | | - local_irq_disable(); |
---|
278 | | - prepare_exit_to_usermode(regs); |
---|
279 | | -} |
---|
| 34 | +#include <asm/io_bitmap.h> |
---|
| 35 | +#include <asm/syscall.h> |
---|
| 36 | +#include <asm/irq_stack.h> |
---|
280 | 37 | |
---|
281 | 38 | #ifdef CONFIG_X86_64 |
---|
282 | | -__visible void do_syscall_64(unsigned long nr, struct pt_regs *regs) |
---|
| 39 | +__visible noinstr void do_syscall_64(unsigned long nr, struct pt_regs *regs) |
---|
283 | 40 | { |
---|
284 | | - struct thread_info *ti; |
---|
| 41 | + nr = syscall_enter_from_user_mode(regs, nr); |
---|
285 | 42 | |
---|
286 | | - enter_from_user_mode(); |
---|
287 | | - local_irq_enable(); |
---|
288 | | - ti = current_thread_info(); |
---|
289 | | - if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) |
---|
290 | | - nr = syscall_trace_enter(regs); |
---|
291 | | - |
---|
292 | | - /* |
---|
293 | | - * NB: Native and x32 syscalls are dispatched from the same |
---|
294 | | - * table. The only functional difference is the x32 bit in |
---|
295 | | - * regs->orig_ax, which changes the behavior of some syscalls. |
---|
296 | | - */ |
---|
297 | | - nr &= __SYSCALL_MASK; |
---|
| 43 | + instrumentation_begin(); |
---|
298 | 44 | if (likely(nr < NR_syscalls)) { |
---|
299 | 45 | nr = array_index_nospec(nr, NR_syscalls); |
---|
300 | 46 | regs->ax = sys_call_table[nr](regs); |
---|
| 47 | +#ifdef CONFIG_X86_X32_ABI |
---|
| 48 | + } else if (likely((nr & __X32_SYSCALL_BIT) && |
---|
| 49 | + (nr & ~__X32_SYSCALL_BIT) < X32_NR_syscalls)) { |
---|
| 50 | + nr = array_index_nospec(nr & ~__X32_SYSCALL_BIT, |
---|
| 51 | + X32_NR_syscalls); |
---|
| 52 | + regs->ax = x32_sys_call_table[nr](regs); |
---|
| 53 | +#endif |
---|
301 | 54 | } |
---|
302 | | - |
---|
303 | | - syscall_return_slowpath(regs); |
---|
| 55 | + instrumentation_end(); |
---|
| 56 | + syscall_exit_to_user_mode(regs); |
---|
304 | 57 | } |
---|
305 | 58 | #endif |
---|
306 | 59 | |
---|
307 | 60 | #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) |
---|
308 | | -/* |
---|
309 | | - * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does |
---|
310 | | - * all entry and exit work and returns with IRQs off. This function is |
---|
311 | | - * extremely hot in workloads that use it, and it's usually called from |
---|
312 | | - * do_fast_syscall_32, so forcibly inline it to improve performance. |
---|
313 | | - */ |
---|
314 | | -static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs) |
---|
| 61 | +static __always_inline unsigned int syscall_32_enter(struct pt_regs *regs) |
---|
315 | 62 | { |
---|
316 | | - struct thread_info *ti = current_thread_info(); |
---|
317 | | - unsigned int nr = (unsigned int)regs->orig_ax; |
---|
| 63 | + if (IS_ENABLED(CONFIG_IA32_EMULATION)) |
---|
| 64 | + current_thread_info()->status |= TS_COMPAT; |
---|
318 | 65 | |
---|
319 | | -#ifdef CONFIG_IA32_EMULATION |
---|
320 | | - ti->status |= TS_COMPAT; |
---|
321 | | -#endif |
---|
| 66 | + return (unsigned int)regs->orig_ax; |
---|
| 67 | +} |
---|
322 | 68 | |
---|
323 | | - if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) { |
---|
324 | | - /* |
---|
325 | | - * Subtlety here: if ptrace pokes something larger than |
---|
326 | | - * 2^32-1 into orig_ax, this truncates it. This may or |
---|
327 | | - * may not be necessary, but it matches the old asm |
---|
328 | | - * behavior. |
---|
329 | | - */ |
---|
330 | | - nr = syscall_trace_enter(regs); |
---|
331 | | - } |
---|
332 | | - |
---|
| 69 | +/* |
---|
| 70 | + * Invoke a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. |
---|
| 71 | + */ |
---|
| 72 | +static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs, |
---|
| 73 | + unsigned int nr) |
---|
| 74 | +{ |
---|
333 | 75 | if (likely(nr < IA32_NR_syscalls)) { |
---|
334 | 76 | nr = array_index_nospec(nr, IA32_NR_syscalls); |
---|
335 | | -#ifdef CONFIG_IA32_EMULATION |
---|
336 | 77 | regs->ax = ia32_sys_call_table[nr](regs); |
---|
337 | | -#else |
---|
338 | | - /* |
---|
339 | | - * It's possible that a 32-bit syscall implementation |
---|
340 | | - * takes a 64-bit parameter but nonetheless assumes that |
---|
341 | | - * the high bits are zero. Make sure we zero-extend all |
---|
342 | | - * of the args. |
---|
343 | | - */ |
---|
344 | | - regs->ax = ia32_sys_call_table[nr]( |
---|
345 | | - (unsigned int)regs->bx, (unsigned int)regs->cx, |
---|
346 | | - (unsigned int)regs->dx, (unsigned int)regs->si, |
---|
347 | | - (unsigned int)regs->di, (unsigned int)regs->bp); |
---|
348 | | -#endif /* CONFIG_IA32_EMULATION */ |
---|
349 | 78 | } |
---|
350 | | - |
---|
351 | | - syscall_return_slowpath(regs); |
---|
352 | 79 | } |
---|
353 | 80 | |
---|
354 | 81 | /* Handles int $0x80 */ |
---|
355 | | -__visible void do_int80_syscall_32(struct pt_regs *regs) |
---|
| 82 | +__visible noinstr void do_int80_syscall_32(struct pt_regs *regs) |
---|
356 | 83 | { |
---|
357 | | - enter_from_user_mode(); |
---|
358 | | - local_irq_enable(); |
---|
359 | | - do_syscall_32_irqs_on(regs); |
---|
| 84 | + unsigned int nr = syscall_32_enter(regs); |
---|
| 85 | + |
---|
| 86 | + /* |
---|
| 87 | + * Subtlety here: if ptrace pokes something larger than 2^32-1 into |
---|
| 88 | + * orig_ax, the unsigned int return value truncates it. This may |
---|
| 89 | + * or may not be necessary, but it matches the old asm behavior. |
---|
| 90 | + */ |
---|
| 91 | + nr = (unsigned int)syscall_enter_from_user_mode(regs, nr); |
---|
| 92 | + instrumentation_begin(); |
---|
| 93 | + |
---|
| 94 | + do_syscall_32_irqs_on(regs, nr); |
---|
| 95 | + |
---|
| 96 | + instrumentation_end(); |
---|
| 97 | + syscall_exit_to_user_mode(regs); |
---|
| 98 | +} |
---|
| 99 | + |
---|
| 100 | +static noinstr bool __do_fast_syscall_32(struct pt_regs *regs) |
---|
| 101 | +{ |
---|
| 102 | + unsigned int nr = syscall_32_enter(regs); |
---|
| 103 | + int res; |
---|
| 104 | + |
---|
| 105 | + /* |
---|
| 106 | + * This cannot use syscall_enter_from_user_mode() as it has to |
---|
| 107 | + * fetch EBP before invoking any of the syscall entry work |
---|
| 108 | + * functions. |
---|
| 109 | + */ |
---|
| 110 | + syscall_enter_from_user_mode_prepare(regs); |
---|
| 111 | + |
---|
| 112 | + instrumentation_begin(); |
---|
| 113 | + /* Fetch EBP from where the vDSO stashed it. */ |
---|
| 114 | + if (IS_ENABLED(CONFIG_X86_64)) { |
---|
| 115 | + /* |
---|
| 116 | + * Micro-optimization: the pointer we're following is |
---|
| 117 | + * explicitly 32 bits, so it can't be out of range. |
---|
| 118 | + */ |
---|
| 119 | + res = __get_user(*(u32 *)®s->bp, |
---|
| 120 | + (u32 __user __force *)(unsigned long)(u32)regs->sp); |
---|
| 121 | + } else { |
---|
| 122 | + res = get_user(*(u32 *)®s->bp, |
---|
| 123 | + (u32 __user __force *)(unsigned long)(u32)regs->sp); |
---|
| 124 | + } |
---|
| 125 | + |
---|
| 126 | + if (res) { |
---|
| 127 | + /* User code screwed up. */ |
---|
| 128 | + regs->ax = -EFAULT; |
---|
| 129 | + |
---|
| 130 | + local_irq_disable(); |
---|
| 131 | + instrumentation_end(); |
---|
| 132 | + irqentry_exit_to_user_mode(regs); |
---|
| 133 | + return false; |
---|
| 134 | + } |
---|
| 135 | + |
---|
| 136 | + /* The case truncates any ptrace induced syscall nr > 2^32 -1 */ |
---|
| 137 | + nr = (unsigned int)syscall_enter_from_user_mode_work(regs, nr); |
---|
| 138 | + |
---|
| 139 | + /* Now this is just like a normal syscall. */ |
---|
| 140 | + do_syscall_32_irqs_on(regs, nr); |
---|
| 141 | + |
---|
| 142 | + instrumentation_end(); |
---|
| 143 | + syscall_exit_to_user_mode(regs); |
---|
| 144 | + return true; |
---|
360 | 145 | } |
---|
361 | 146 | |
---|
362 | 147 | /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */ |
---|
363 | | -__visible long do_fast_syscall_32(struct pt_regs *regs) |
---|
| 148 | +__visible noinstr long do_fast_syscall_32(struct pt_regs *regs) |
---|
364 | 149 | { |
---|
365 | 150 | /* |
---|
366 | 151 | * Called using the internal vDSO SYSENTER/SYSCALL32 calling |
---|
367 | 152 | * convention. Adjust regs so it looks like we entered using int80. |
---|
368 | 153 | */ |
---|
369 | | - |
---|
370 | 154 | unsigned long landing_pad = (unsigned long)current->mm->context.vdso + |
---|
371 | | - vdso_image_32.sym_int80_landing_pad; |
---|
| 155 | + vdso_image_32.sym_int80_landing_pad; |
---|
372 | 156 | |
---|
373 | 157 | /* |
---|
374 | 158 | * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward |
---|
.. | .. |
---|
377 | 161 | */ |
---|
378 | 162 | regs->ip = landing_pad; |
---|
379 | 163 | |
---|
380 | | - enter_from_user_mode(); |
---|
381 | | - |
---|
382 | | - local_irq_enable(); |
---|
383 | | - |
---|
384 | | - /* Fetch EBP from where the vDSO stashed it. */ |
---|
385 | | - if ( |
---|
386 | | -#ifdef CONFIG_X86_64 |
---|
387 | | - /* |
---|
388 | | - * Micro-optimization: the pointer we're following is explicitly |
---|
389 | | - * 32 bits, so it can't be out of range. |
---|
390 | | - */ |
---|
391 | | - __get_user(*(u32 *)®s->bp, |
---|
392 | | - (u32 __user __force *)(unsigned long)(u32)regs->sp) |
---|
393 | | -#else |
---|
394 | | - get_user(*(u32 *)®s->bp, |
---|
395 | | - (u32 __user __force *)(unsigned long)(u32)regs->sp) |
---|
396 | | -#endif |
---|
397 | | - ) { |
---|
398 | | - |
---|
399 | | - /* User code screwed up. */ |
---|
400 | | - local_irq_disable(); |
---|
401 | | - regs->ax = -EFAULT; |
---|
402 | | - prepare_exit_to_usermode(regs); |
---|
403 | | - return 0; /* Keep it simple: use IRET. */ |
---|
404 | | - } |
---|
405 | | - |
---|
406 | | - /* Now this is just like a normal syscall. */ |
---|
407 | | - do_syscall_32_irqs_on(regs); |
---|
| 164 | + /* Invoke the syscall. If it failed, keep it simple: use IRET. */ |
---|
| 165 | + if (!__do_fast_syscall_32(regs)) |
---|
| 166 | + return 0; |
---|
408 | 167 | |
---|
409 | 168 | #ifdef CONFIG_X86_64 |
---|
410 | 169 | /* |
---|
.. | .. |
---|
436 | 195 | (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0; |
---|
437 | 196 | #endif |
---|
438 | 197 | } |
---|
| 198 | + |
---|
| 199 | +/* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */ |
---|
| 200 | +__visible noinstr long do_SYSENTER_32(struct pt_regs *regs) |
---|
| 201 | +{ |
---|
| 202 | + /* SYSENTER loses RSP, but the vDSO saved it in RBP. */ |
---|
| 203 | + regs->sp = regs->bp; |
---|
| 204 | + |
---|
| 205 | + /* SYSENTER clobbers EFLAGS.IF. Assume it was set in usermode. */ |
---|
| 206 | + regs->flags |= X86_EFLAGS_IF; |
---|
| 207 | + |
---|
| 208 | + return do_fast_syscall_32(regs); |
---|
| 209 | +} |
---|
439 | 210 | #endif |
---|
| 211 | + |
---|
| 212 | +SYSCALL_DEFINE0(ni_syscall) |
---|
| 213 | +{ |
---|
| 214 | + return -ENOSYS; |
---|
| 215 | +} |
---|
| 216 | + |
---|
| 217 | +#ifdef CONFIG_XEN_PV |
---|
| 218 | +#ifndef CONFIG_PREEMPTION |
---|
| 219 | +/* |
---|
| 220 | + * Some hypercalls issued by the toolstack can take many 10s of |
---|
| 221 | + * seconds. Allow tasks running hypercalls via the privcmd driver to |
---|
| 222 | + * be voluntarily preempted even if full kernel preemption is |
---|
| 223 | + * disabled. |
---|
| 224 | + * |
---|
| 225 | + * Such preemptible hypercalls are bracketed by |
---|
| 226 | + * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end() |
---|
| 227 | + * calls. |
---|
| 228 | + */ |
---|
| 229 | +DEFINE_PER_CPU(bool, xen_in_preemptible_hcall); |
---|
| 230 | +EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall); |
---|
| 231 | + |
---|
| 232 | +/* |
---|
| 233 | + * In case of scheduling the flag must be cleared and restored after |
---|
| 234 | + * returning from schedule as the task might move to a different CPU. |
---|
| 235 | + */ |
---|
| 236 | +static __always_inline bool get_and_clear_inhcall(void) |
---|
| 237 | +{ |
---|
| 238 | + bool inhcall = __this_cpu_read(xen_in_preemptible_hcall); |
---|
| 239 | + |
---|
| 240 | + __this_cpu_write(xen_in_preemptible_hcall, false); |
---|
| 241 | + return inhcall; |
---|
| 242 | +} |
---|
| 243 | + |
---|
| 244 | +static __always_inline void restore_inhcall(bool inhcall) |
---|
| 245 | +{ |
---|
| 246 | + __this_cpu_write(xen_in_preemptible_hcall, inhcall); |
---|
| 247 | +} |
---|
| 248 | +#else |
---|
| 249 | +static __always_inline bool get_and_clear_inhcall(void) { return false; } |
---|
| 250 | +static __always_inline void restore_inhcall(bool inhcall) { } |
---|
| 251 | +#endif |
---|
| 252 | + |
---|
| 253 | +static void __xen_pv_evtchn_do_upcall(void) |
---|
| 254 | +{ |
---|
| 255 | + irq_enter_rcu(); |
---|
| 256 | + inc_irq_stat(irq_hv_callback_count); |
---|
| 257 | + |
---|
| 258 | + xen_hvm_evtchn_do_upcall(); |
---|
| 259 | + |
---|
| 260 | + irq_exit_rcu(); |
---|
| 261 | +} |
---|
| 262 | + |
---|
| 263 | +__visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs) |
---|
| 264 | +{ |
---|
| 265 | + struct pt_regs *old_regs; |
---|
| 266 | + bool inhcall; |
---|
| 267 | + irqentry_state_t state; |
---|
| 268 | + |
---|
| 269 | + state = irqentry_enter(regs); |
---|
| 270 | + old_regs = set_irq_regs(regs); |
---|
| 271 | + |
---|
| 272 | + instrumentation_begin(); |
---|
| 273 | + run_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs); |
---|
| 274 | + instrumentation_end(); |
---|
| 275 | + |
---|
| 276 | + set_irq_regs(old_regs); |
---|
| 277 | + |
---|
| 278 | + inhcall = get_and_clear_inhcall(); |
---|
| 279 | + if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) { |
---|
| 280 | + instrumentation_begin(); |
---|
| 281 | + irqentry_exit_cond_resched(); |
---|
| 282 | + instrumentation_end(); |
---|
| 283 | + restore_inhcall(inhcall); |
---|
| 284 | + } else { |
---|
| 285 | + irqentry_exit(regs, state); |
---|
| 286 | + } |
---|
| 287 | +} |
---|
| 288 | +#endif /* CONFIG_XEN_PV */ |
---|