1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_ENTRYCOMMON_H 2 #ifndef __LINUX_ENTRYCOMMON_H
3 #define __LINUX_ENTRYCOMMON_H 3 #define __LINUX_ENTRYCOMMON_H
4 4
5 #include <linux/static_call_types.h> 5 #include <linux/static_call_types.h>
6 #include <linux/ptrace.h> !! 6 #include <linux/tracehook.h>
7 #include <linux/syscalls.h> 7 #include <linux/syscalls.h>
8 #include <linux/seccomp.h> 8 #include <linux/seccomp.h>
9 #include <linux/sched.h> 9 #include <linux/sched.h>
10 #include <linux/context_tracking.h> <<
11 #include <linux/livepatch.h> <<
12 #include <linux/resume_user_mode.h> <<
13 #include <linux/tick.h> <<
14 #include <linux/kmsan.h> <<
15 10
16 #include <asm/entry-common.h> 11 #include <asm/entry-common.h>
17 12
18 /* 13 /*
19 * Define dummy _TIF work flags if not defined 14 * Define dummy _TIF work flags if not defined by the architecture or for
20 * disabled functionality. 15 * disabled functionality.
21 */ 16 */
22 #ifndef _TIF_PATCH_PENDING 17 #ifndef _TIF_PATCH_PENDING
23 # define _TIF_PATCH_PENDING (0) 18 # define _TIF_PATCH_PENDING (0)
24 #endif 19 #endif
25 20
26 #ifndef _TIF_UPROBE 21 #ifndef _TIF_UPROBE
27 # define _TIF_UPROBE (0) 22 # define _TIF_UPROBE (0)
28 #endif 23 #endif
29 24
30 /* 25 /*
31 * SYSCALL_WORK flags handled in syscall_enter 26 * SYSCALL_WORK flags handled in syscall_enter_from_user_mode()
32 */ 27 */
33 #ifndef ARCH_SYSCALL_WORK_ENTER 28 #ifndef ARCH_SYSCALL_WORK_ENTER
34 # define ARCH_SYSCALL_WORK_ENTER (0) 29 # define ARCH_SYSCALL_WORK_ENTER (0)
35 #endif 30 #endif
36 31
37 /* 32 /*
38 * SYSCALL_WORK flags handled in syscall_exit_ 33 * SYSCALL_WORK flags handled in syscall_exit_to_user_mode()
39 */ 34 */
40 #ifndef ARCH_SYSCALL_WORK_EXIT 35 #ifndef ARCH_SYSCALL_WORK_EXIT
41 # define ARCH_SYSCALL_WORK_EXIT (0) 36 # define ARCH_SYSCALL_WORK_EXIT (0)
42 #endif 37 #endif
43 38
44 #define SYSCALL_WORK_ENTER (SYSCALL_WORK_ 39 #define SYSCALL_WORK_ENTER (SYSCALL_WORK_SECCOMP | \
45 SYSCALL_WORK_ 40 SYSCALL_WORK_SYSCALL_TRACEPOINT | \
46 SYSCALL_WORK_ 41 SYSCALL_WORK_SYSCALL_TRACE | \
47 SYSCALL_WORK_ 42 SYSCALL_WORK_SYSCALL_EMU | \
48 SYSCALL_WORK_ 43 SYSCALL_WORK_SYSCALL_AUDIT | \
49 SYSCALL_WORK_ 44 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
50 ARCH_SYSCALL_ 45 ARCH_SYSCALL_WORK_ENTER)
51 #define SYSCALL_WORK_EXIT (SYSCALL_WORK_ 46 #define SYSCALL_WORK_EXIT (SYSCALL_WORK_SYSCALL_TRACEPOINT | \
52 SYSCALL_WORK_ 47 SYSCALL_WORK_SYSCALL_TRACE | \
53 SYSCALL_WORK_ 48 SYSCALL_WORK_SYSCALL_AUDIT | \
54 SYSCALL_WORK_ 49 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
55 SYSCALL_WORK_ 50 SYSCALL_WORK_SYSCALL_EXIT_TRAP | \
56 ARCH_SYSCALL_ 51 ARCH_SYSCALL_WORK_EXIT)
57 52
58 /* 53 /*
59 * TIF flags handled in exit_to_user_mode_loop 54 * TIF flags handled in exit_to_user_mode_loop()
60 */ 55 */
61 #ifndef ARCH_EXIT_TO_USER_MODE_WORK 56 #ifndef ARCH_EXIT_TO_USER_MODE_WORK
62 # define ARCH_EXIT_TO_USER_MODE_WORK 57 # define ARCH_EXIT_TO_USER_MODE_WORK (0)
63 #endif 58 #endif
64 59
65 #define EXIT_TO_USER_MODE_WORK 60 #define EXIT_TO_USER_MODE_WORK \
66 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME 61 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
67 _TIF_NEED_RESCHED | _TIF_PATCH_PENDIN 62 _TIF_NEED_RESCHED | _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL | \
68 ARCH_EXIT_TO_USER_MODE_WORK) 63 ARCH_EXIT_TO_USER_MODE_WORK)
69 64
70 /** 65 /**
71 * arch_enter_from_user_mode - Architecture sp !! 66 * arch_check_user_regs - Architecture specific sanity check for user mode regs
72 * @regs: Pointer to currents pt_regs 67 * @regs: Pointer to currents pt_regs
73 * 68 *
74 * Defaults to an empty implementation. Can be 69 * Defaults to an empty implementation. Can be replaced by architecture
75 * specific code. 70 * specific code.
76 * 71 *
77 * Invoked from syscall_enter_from_user_mode() 72 * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
78 * section. Use __always_inline so the compile 73 * section. Use __always_inline so the compiler cannot push it out of line
79 * and make it instrumentable. 74 * and make it instrumentable.
80 */ 75 */
81 static __always_inline void arch_enter_from_us !! 76 static __always_inline void arch_check_user_regs(struct pt_regs *regs);
82 77
83 #ifndef arch_enter_from_user_mode !! 78 #ifndef arch_check_user_regs
84 static __always_inline void arch_enter_from_us !! 79 static __always_inline void arch_check_user_regs(struct pt_regs *regs) {}
>> 80 #endif
>> 81
>> 82 /**
>> 83 * arch_syscall_enter_tracehook - Wrapper around tracehook_report_syscall_entry()
>> 84 * @regs: Pointer to currents pt_regs
>> 85 *
>> 86 * Returns: 0 on success or an error code to skip the syscall.
>> 87 *
>> 88 * Defaults to tracehook_report_syscall_entry(). Can be replaced by
>> 89 * architecture specific code.
>> 90 *
>> 91 * Invoked from syscall_enter_from_user_mode()
>> 92 */
>> 93 static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs);
>> 94
>> 95 #ifndef arch_syscall_enter_tracehook
>> 96 static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs)
>> 97 {
>> 98 return tracehook_report_syscall_entry(regs);
>> 99 }
85 #endif 100 #endif
86 101
87 /** 102 /**
88 * enter_from_user_mode - Establish state when 103 * enter_from_user_mode - Establish state when coming from user mode
89 * 104 *
90 * Syscall/interrupt entry disables interrupts 105 * Syscall/interrupt entry disables interrupts, but user mode is traced as
91 * interrupts enabled. Also with NO_HZ_FULL RC 106 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
92 * 107 *
93 * 1) Tell lockdep that interrupts are disable 108 * 1) Tell lockdep that interrupts are disabled
94 * 2) Invoke context tracking if enabled to re 109 * 2) Invoke context tracking if enabled to reactivate RCU
95 * 3) Trace interrupts off state 110 * 3) Trace interrupts off state
96 * 111 *
97 * Invoked from architecture specific syscall 112 * Invoked from architecture specific syscall entry code with interrupts
98 * disabled. The calling code has to be non-in 113 * disabled. The calling code has to be non-instrumentable. When the
99 * function returns all state is correct and i 114 * function returns all state is correct and interrupts are still
100 * disabled. The subsequent functions can be i 115 * disabled. The subsequent functions can be instrumented.
101 * 116 *
102 * This is invoked when there is architecture 117 * This is invoked when there is architecture specific functionality to be
103 * done between establishing state and enablin 118 * done between establishing state and enabling interrupts. The caller must
104 * enable interrupts before invoking syscall_e 119 * enable interrupts before invoking syscall_enter_from_user_mode_work().
105 */ 120 */
106 static __always_inline void enter_from_user_mo !! 121 void enter_from_user_mode(struct pt_regs *regs);
107 { <<
108 arch_enter_from_user_mode(regs); <<
109 lockdep_hardirqs_off(CALLER_ADDR0); <<
110 <<
111 CT_WARN_ON(__ct_state() != CT_STATE_US <<
112 user_exit_irqoff(); <<
113 <<
114 instrumentation_begin(); <<
115 kmsan_unpoison_entry_regs(regs); <<
116 trace_hardirqs_off_finish(); <<
117 instrumentation_end(); <<
118 } <<
119 122
120 /** 123 /**
121 * syscall_enter_from_user_mode_prepare - Esta 124 * syscall_enter_from_user_mode_prepare - Establish state and enable interrupts
122 * @regs: Pointer to currents pt_regs 125 * @regs: Pointer to currents pt_regs
123 * 126 *
124 * Invoked from architecture specific syscall 127 * Invoked from architecture specific syscall entry code with interrupts
125 * disabled. The calling code has to be non-in 128 * disabled. The calling code has to be non-instrumentable. When the
126 * function returns all state is correct, inte 129 * function returns all state is correct, interrupts are enabled and the
127 * subsequent functions can be instrumented. 130 * subsequent functions can be instrumented.
128 * 131 *
129 * This handles lockdep, RCU (context tracking 132 * This handles lockdep, RCU (context tracking) and tracing state, i.e.
130 * the functionality provided by enter_from_us 133 * the functionality provided by enter_from_user_mode().
131 * 134 *
132 * This is invoked when there is extra archite 135 * This is invoked when there is extra architecture specific functionality
133 * to be done between establishing state and h 136 * to be done between establishing state and handling user mode entry work.
134 */ 137 */
135 void syscall_enter_from_user_mode_prepare(stru 138 void syscall_enter_from_user_mode_prepare(struct pt_regs *regs);
136 139
137 long syscall_trace_enter(struct pt_regs *regs, <<
138 unsigned long work); <<
139 <<
140 /** 140 /**
141 * syscall_enter_from_user_mode_work - Check a 141 * syscall_enter_from_user_mode_work - Check and handle work before invoking
142 * a sysca 142 * a syscall
143 * @regs: Pointer to currents pt_regs 143 * @regs: Pointer to currents pt_regs
144 * @syscall: The syscall number 144 * @syscall: The syscall number
145 * 145 *
146 * Invoked from architecture specific syscall 146 * Invoked from architecture specific syscall entry code with interrupts
147 * enabled after invoking syscall_enter_from_u 147 * enabled after invoking syscall_enter_from_user_mode_prepare() and extra
148 * architecture specific work. 148 * architecture specific work.
149 * 149 *
150 * Returns: The original or a modified syscall 150 * Returns: The original or a modified syscall number
151 * 151 *
152 * If the returned syscall number is -1 then t 152 * If the returned syscall number is -1 then the syscall should be
153 * skipped. In this case the caller may invoke 153 * skipped. In this case the caller may invoke syscall_set_error() or
154 * syscall_set_return_value() first. If neith 154 * syscall_set_return_value() first. If neither of those are called and -1
155 * is returned, then the syscall will fail wit 155 * is returned, then the syscall will fail with ENOSYS.
156 * 156 *
157 * It handles the following work items: 157 * It handles the following work items:
158 * 158 *
159 * 1) syscall_work flag dependent invocations 159 * 1) syscall_work flag dependent invocations of
160 * ptrace_report_syscall_entry(), __secure !! 160 * arch_syscall_enter_tracehook(), __secure_computing(), trace_sys_enter()
161 * 2) Invocation of audit_syscall_entry() 161 * 2) Invocation of audit_syscall_entry()
162 */ 162 */
163 static __always_inline long syscall_enter_from !! 163 long syscall_enter_from_user_mode_work(struct pt_regs *regs, long syscall);
164 { <<
165 unsigned long work = READ_ONCE(current <<
166 <<
167 if (work & SYSCALL_WORK_ENTER) <<
168 syscall = syscall_trace_enter( <<
169 <<
170 return syscall; <<
171 } <<
172 164
173 /** 165 /**
174 * syscall_enter_from_user_mode - Establish st 166 * syscall_enter_from_user_mode - Establish state and check and handle work
175 * before invok 167 * before invoking a syscall
176 * @regs: Pointer to currents pt_regs 168 * @regs: Pointer to currents pt_regs
177 * @syscall: The syscall number 169 * @syscall: The syscall number
178 * 170 *
179 * Invoked from architecture specific syscall 171 * Invoked from architecture specific syscall entry code with interrupts
180 * disabled. The calling code has to be non-in 172 * disabled. The calling code has to be non-instrumentable. When the
181 * function returns all state is correct, inte 173 * function returns all state is correct, interrupts are enabled and the
182 * subsequent functions can be instrumented. 174 * subsequent functions can be instrumented.
183 * 175 *
184 * This is combination of syscall_enter_from_u 176 * This is combination of syscall_enter_from_user_mode_prepare() and
185 * syscall_enter_from_user_mode_work(). 177 * syscall_enter_from_user_mode_work().
186 * 178 *
187 * Returns: The original or a modified syscall 179 * Returns: The original or a modified syscall number. See
188 * syscall_enter_from_user_mode_work() for fur 180 * syscall_enter_from_user_mode_work() for further explanation.
189 */ 181 */
190 static __always_inline long syscall_enter_from !! 182 long syscall_enter_from_user_mode(struct pt_regs *regs, long syscall);
191 { <<
192 long ret; <<
193 <<
194 enter_from_user_mode(regs); <<
195 <<
196 instrumentation_begin(); <<
197 local_irq_enable(); <<
198 ret = syscall_enter_from_user_mode_wor <<
199 instrumentation_end(); <<
200 <<
201 return ret; <<
202 } <<
203 183
204 /** 184 /**
205 * local_irq_enable_exit_to_user - Exit to use 185 * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
206 * @ti_work: Cached TIF flags gathered with 186 * @ti_work: Cached TIF flags gathered with interrupts disabled
207 * 187 *
208 * Defaults to local_irq_enable(). Can be supp 188 * Defaults to local_irq_enable(). Can be supplied by architecture specific
209 * code. 189 * code.
210 */ 190 */
211 static inline void local_irq_enable_exit_to_us 191 static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
212 192
213 #ifndef local_irq_enable_exit_to_user 193 #ifndef local_irq_enable_exit_to_user
214 static inline void local_irq_enable_exit_to_us 194 static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
215 { 195 {
216 local_irq_enable(); 196 local_irq_enable();
217 } 197 }
218 #endif 198 #endif
219 199
220 /** 200 /**
221 * local_irq_disable_exit_to_user - Exit to us 201 * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
222 * 202 *
223 * Defaults to local_irq_disable(). Can be sup 203 * Defaults to local_irq_disable(). Can be supplied by architecture specific
224 * code. 204 * code.
225 */ 205 */
226 static inline void local_irq_disable_exit_to_u 206 static inline void local_irq_disable_exit_to_user(void);
227 207
228 #ifndef local_irq_disable_exit_to_user 208 #ifndef local_irq_disable_exit_to_user
229 static inline void local_irq_disable_exit_to_u 209 static inline void local_irq_disable_exit_to_user(void)
230 { 210 {
231 local_irq_disable(); 211 local_irq_disable();
232 } 212 }
233 #endif 213 #endif
234 214
235 /** 215 /**
236 * arch_exit_to_user_mode_work - Architecture 216 * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
237 * to user mode. 217 * to user mode.
238 * @regs: Pointer to currents pt_regs 218 * @regs: Pointer to currents pt_regs
239 * @ti_work: Cached TIF flags gathered with 219 * @ti_work: Cached TIF flags gathered with interrupts disabled
240 * 220 *
241 * Invoked from exit_to_user_mode_loop() with 221 * Invoked from exit_to_user_mode_loop() with interrupt enabled
242 * 222 *
243 * Defaults to NOOP. Can be supplied by archit 223 * Defaults to NOOP. Can be supplied by architecture specific code.
244 */ 224 */
245 static inline void arch_exit_to_user_mode_work 225 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
246 226 unsigned long ti_work);
247 227
248 #ifndef arch_exit_to_user_mode_work 228 #ifndef arch_exit_to_user_mode_work
249 static inline void arch_exit_to_user_mode_work 229 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
250 230 unsigned long ti_work)
251 { 231 {
252 } 232 }
253 #endif 233 #endif
254 234
255 /** 235 /**
256 * arch_exit_to_user_mode_prepare - Architectu 236 * arch_exit_to_user_mode_prepare - Architecture specific preparation for
257 * exit to us 237 * exit to user mode.
258 * @regs: Pointer to currents pt_regs 238 * @regs: Pointer to currents pt_regs
259 * @ti_work: Cached TIF flags gathered with 239 * @ti_work: Cached TIF flags gathered with interrupts disabled
260 * 240 *
261 * Invoked from exit_to_user_mode_prepare() wi 241 * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
262 * function before return. Defaults to NOOP. 242 * function before return. Defaults to NOOP.
263 */ 243 */
264 static inline void arch_exit_to_user_mode_prep 244 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
265 245 unsigned long ti_work);
266 246
267 #ifndef arch_exit_to_user_mode_prepare 247 #ifndef arch_exit_to_user_mode_prepare
268 static inline void arch_exit_to_user_mode_prep 248 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
269 249 unsigned long ti_work)
270 { 250 {
271 } 251 }
272 #endif 252 #endif
273 253
274 /** 254 /**
275 * arch_exit_to_user_mode - Architecture speci 255 * arch_exit_to_user_mode - Architecture specific final work before
276 * exit to user mode. 256 * exit to user mode.
277 * 257 *
278 * Invoked from exit_to_user_mode() with inter 258 * Invoked from exit_to_user_mode() with interrupt disabled as the last
279 * function before return. Defaults to NOOP. 259 * function before return. Defaults to NOOP.
280 * 260 *
281 * This needs to be __always_inline because it 261 * This needs to be __always_inline because it is non-instrumentable code
282 * invoked after context tracking switched to 262 * invoked after context tracking switched to user mode.
283 * 263 *
284 * An architecture implementation must not do 264 * An architecture implementation must not do anything complex, no locking
285 * etc. The main purpose is for speculation mi 265 * etc. The main purpose is for speculation mitigations.
286 */ 266 */
287 static __always_inline void arch_exit_to_user_ 267 static __always_inline void arch_exit_to_user_mode(void);
288 268
289 #ifndef arch_exit_to_user_mode 269 #ifndef arch_exit_to_user_mode
290 static __always_inline void arch_exit_to_user_ 270 static __always_inline void arch_exit_to_user_mode(void) { }
291 #endif 271 #endif
292 272
293 /** 273 /**
294 * arch_do_signal_or_restart - Architecture s 274 * arch_do_signal_or_restart - Architecture specific signal delivery function
295 * @regs: Pointer to currents pt_regs 275 * @regs: Pointer to currents pt_regs
>> 276 * @has_signal: actual signal to handle
296 * 277 *
297 * Invoked from exit_to_user_mode_loop(). 278 * Invoked from exit_to_user_mode_loop().
298 */ 279 */
299 void arch_do_signal_or_restart(struct pt_regs !! 280 void arch_do_signal_or_restart(struct pt_regs *regs, bool has_signal);
300 281
301 /** 282 /**
302 * exit_to_user_mode_loop - do any pending wor !! 283 * arch_syscall_exit_tracehook - Wrapper around tracehook_report_syscall_exit()
303 */ !! 284 * @regs: Pointer to currents pt_regs
304 unsigned long exit_to_user_mode_loop(struct pt !! 285 * @step: Indicator for single step
305 unsigned !! 286 *
306 !! 287 * Defaults to tracehook_report_syscall_exit(). Can be replaced by
307 /** !! 288 * architecture specific code.
308 * exit_to_user_mode_prepare - call exit_to_us <<
309 * @regs: Pointer to pt_regs on entry st <<
310 * 289 *
311 * 1) check that interrupts are disabled !! 290 * Invoked from syscall_exit_to_user_mode()
312 * 2) call tick_nohz_user_enter_prepare() <<
313 * 3) call exit_to_user_mode_loop() if any fla <<
314 * EXIT_TO_USER_MODE_WORK are set <<
315 * 4) check that interrupts are still disabled <<
316 */ 291 */
317 static __always_inline void exit_to_user_mode_ !! 292 static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step);
318 { <<
319 unsigned long ti_work; <<
320 <<
321 lockdep_assert_irqs_disabled(); <<
322 <<
323 /* Flush pending rcuog wakeup before t <<
324 tick_nohz_user_enter_prepare(); <<
325 <<
326 ti_work = read_thread_flags(); <<
327 if (unlikely(ti_work & EXIT_TO_USER_MO <<
328 ti_work = exit_to_user_mode_lo <<
329 <<
330 arch_exit_to_user_mode_prepare(regs, t <<
331 293
332 /* Ensure that kernel state is sane fo !! 294 #ifndef arch_syscall_exit_tracehook
333 kmap_assert_nomap(); !! 295 static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step)
334 lockdep_assert_irqs_disabled(); !! 296 {
335 lockdep_sys_exit(); !! 297 tracehook_report_syscall_exit(regs, step);
336 } 298 }
>> 299 #endif
337 300
338 /** 301 /**
339 * exit_to_user_mode - Fixup state when exitin 302 * exit_to_user_mode - Fixup state when exiting to user mode
340 * 303 *
341 * Syscall/interrupt exit enables interrupts, 304 * Syscall/interrupt exit enables interrupts, but the kernel state is
342 * interrupts disabled when this is invoked. A 305 * interrupts disabled when this is invoked. Also tell RCU about it.
343 * 306 *
344 * 1) Trace interrupts on state 307 * 1) Trace interrupts on state
345 * 2) Invoke context tracking if enabled to ad 308 * 2) Invoke context tracking if enabled to adjust RCU state
346 * 3) Invoke architecture specific last minute 309 * 3) Invoke architecture specific last minute exit code, e.g. speculation
347 * mitigations, etc.: arch_exit_to_user_mod 310 * mitigations, etc.: arch_exit_to_user_mode()
348 * 4) Tell lockdep that interrupts are enabled 311 * 4) Tell lockdep that interrupts are enabled
349 * 312 *
350 * Invoked from architecture specific code whe 313 * Invoked from architecture specific code when syscall_exit_to_user_mode()
351 * is not suitable as the last step before ret 314 * is not suitable as the last step before returning to userspace. Must be
352 * invoked with interrupts disabled and the ca 315 * invoked with interrupts disabled and the caller must be
353 * non-instrumentable. 316 * non-instrumentable.
354 * The caller has to invoke syscall_exit_to_us 317 * The caller has to invoke syscall_exit_to_user_mode_work() before this.
355 */ 318 */
356 static __always_inline void exit_to_user_mode( !! 319 void exit_to_user_mode(void);
357 { <<
358 instrumentation_begin(); <<
359 trace_hardirqs_on_prepare(); <<
360 lockdep_hardirqs_on_prepare(); <<
361 instrumentation_end(); <<
362 <<
363 user_enter_irqoff(); <<
364 arch_exit_to_user_mode(); <<
365 lockdep_hardirqs_on(CALLER_ADDR0); <<
366 } <<
367 320
368 /** 321 /**
369 * syscall_exit_to_user_mode_work - Handle wor 322 * syscall_exit_to_user_mode_work - Handle work before returning to user mode
370 * @regs: Pointer to currents pt_regs 323 * @regs: Pointer to currents pt_regs
371 * 324 *
372 * Same as step 1 and 2 of syscall_exit_to_use 325 * Same as step 1 and 2 of syscall_exit_to_user_mode() but without calling
373 * exit_to_user_mode() to perform the final tr 326 * exit_to_user_mode() to perform the final transition to user mode.
374 * 327 *
375 * Calling convention is the same as for sysca 328 * Calling convention is the same as for syscall_exit_to_user_mode() and it
376 * returns with all work handled and interrupt 329 * returns with all work handled and interrupts disabled. The caller must
377 * invoke exit_to_user_mode() before actually 330 * invoke exit_to_user_mode() before actually switching to user mode to
378 * make the final state transitions. Interrupt 331 * make the final state transitions. Interrupts must stay disabled between
379 * return from this function and the invocatio 332 * return from this function and the invocation of exit_to_user_mode().
380 */ 333 */
381 void syscall_exit_to_user_mode_work(struct pt_ 334 void syscall_exit_to_user_mode_work(struct pt_regs *regs);
382 335
383 /** 336 /**
384 * syscall_exit_to_user_mode - Handle work bef 337 * syscall_exit_to_user_mode - Handle work before returning to user mode
385 * @regs: Pointer to currents pt_regs 338 * @regs: Pointer to currents pt_regs
386 * 339 *
387 * Invoked with interrupts enabled and fully v 340 * Invoked with interrupts enabled and fully valid regs. Returns with all
388 * work handled, interrupts disabled such that 341 * work handled, interrupts disabled such that the caller can immediately
389 * switch to user mode. Called from architectu 342 * switch to user mode. Called from architecture specific syscall and ret
390 * from fork code. 343 * from fork code.
391 * 344 *
392 * The call order is: 345 * The call order is:
393 * 1) One-time syscall exit work: 346 * 1) One-time syscall exit work:
394 * - rseq syscall exit 347 * - rseq syscall exit
395 * - audit 348 * - audit
396 * - syscall tracing 349 * - syscall tracing
397 * - ptrace (single stepping) !! 350 * - tracehook (single stepping)
398 * 351 *
399 * 2) Preparatory work 352 * 2) Preparatory work
400 * - Exit to user mode loop (common TIF h 353 * - Exit to user mode loop (common TIF handling). Invokes
401 * arch_exit_to_user_mode_work() for ar 354 * arch_exit_to_user_mode_work() for architecture specific TIF work
402 * - Architecture specific one time work 355 * - Architecture specific one time work arch_exit_to_user_mode_prepare()
403 * - Address limit and lockdep checks 356 * - Address limit and lockdep checks
404 * 357 *
405 * 3) Final transition (lockdep, tracing, con 358 * 3) Final transition (lockdep, tracing, context tracking, RCU), i.e. the
406 * functionality in exit_to_user_mode(). 359 * functionality in exit_to_user_mode().
407 * 360 *
408 * This is a combination of syscall_exit_to_us 361 * This is a combination of syscall_exit_to_user_mode_work() (1,2) and
409 * exit_to_user_mode(). This function is prefe 362 * exit_to_user_mode(). This function is preferred unless there is a
410 * compelling architectural reason to use the 363 * compelling architectural reason to use the separate functions.
411 */ 364 */
412 void syscall_exit_to_user_mode(struct pt_regs 365 void syscall_exit_to_user_mode(struct pt_regs *regs);
413 366
414 /** 367 /**
415 * irqentry_enter_from_user_mode - Establish s 368 * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
416 * @regs: Pointer to currents pt_regs 369 * @regs: Pointer to currents pt_regs
417 * 370 *
418 * Invoked from architecture specific entry co 371 * Invoked from architecture specific entry code with interrupts disabled.
419 * Can only be called when the interrupt entry 372 * Can only be called when the interrupt entry came from user mode. The
420 * calling code must be non-instrumentable. W 373 * calling code must be non-instrumentable. When the function returns all
421 * state is correct and the subsequent functio 374 * state is correct and the subsequent functions can be instrumented.
422 * 375 *
423 * The function establishes state (lockdep, RC 376 * The function establishes state (lockdep, RCU (context tracking), tracing)
424 */ 377 */
425 void irqentry_enter_from_user_mode(struct pt_r 378 void irqentry_enter_from_user_mode(struct pt_regs *regs);
426 379
427 /** 380 /**
428 * irqentry_exit_to_user_mode - Interrupt exit 381 * irqentry_exit_to_user_mode - Interrupt exit work
429 * @regs: Pointer to current's pt_regs 382 * @regs: Pointer to current's pt_regs
430 * 383 *
431 * Invoked with interrupts disabled and fully 384 * Invoked with interrupts disabled and fully valid regs. Returns with all
432 * work handled, interrupts disabled such that 385 * work handled, interrupts disabled such that the caller can immediately
433 * switch to user mode. Called from architectu 386 * switch to user mode. Called from architecture specific interrupt
434 * handling code. 387 * handling code.
435 * 388 *
436 * The call order is #2 and #3 as described in 389 * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
437 * Interrupt exit is not invoking #1 which is 390 * Interrupt exit is not invoking #1 which is the syscall specific one time
438 * work. 391 * work.
439 */ 392 */
440 void irqentry_exit_to_user_mode(struct pt_regs 393 void irqentry_exit_to_user_mode(struct pt_regs *regs);
441 394
442 #ifndef irqentry_state 395 #ifndef irqentry_state
443 /** 396 /**
444 * struct irqentry_state - Opaque object for e 397 * struct irqentry_state - Opaque object for exception state storage
445 * @exit_rcu: Used exclusively in the irqentry 398 * @exit_rcu: Used exclusively in the irqentry_*() calls; signals whether the
446 * exit path has to invoke ct_irq_e !! 399 * exit path has to invoke rcu_irq_exit().
447 * @lockdep: Used exclusively in the irqentry_ 400 * @lockdep: Used exclusively in the irqentry_nmi_*() calls; ensures that
448 * lockdep state is restored correct 401 * lockdep state is restored correctly on exit from nmi.
449 * 402 *
450 * This opaque object is filled in by the irqe 403 * This opaque object is filled in by the irqentry_*_enter() functions and
451 * must be passed back into the corresponding 404 * must be passed back into the corresponding irqentry_*_exit() functions
452 * when the exception is complete. 405 * when the exception is complete.
453 * 406 *
454 * Callers of irqentry_*_[enter|exit]() must c 407 * Callers of irqentry_*_[enter|exit]() must consider this structure opaque
455 * and all members private. Descriptions of t 408 * and all members private. Descriptions of the members are provided to aid in
456 * the maintenance of the irqentry_*() functio 409 * the maintenance of the irqentry_*() functions.
457 */ 410 */
458 typedef struct irqentry_state { 411 typedef struct irqentry_state {
459 union { 412 union {
460 bool exit_rcu; 413 bool exit_rcu;
461 bool lockdep; 414 bool lockdep;
462 }; 415 };
463 } irqentry_state_t; 416 } irqentry_state_t;
464 #endif 417 #endif
465 418
466 /** 419 /**
467 * irqentry_enter - Handle state tracking on o 420 * irqentry_enter - Handle state tracking on ordinary interrupt entries
468 * @regs: Pointer to pt_regs of interrup 421 * @regs: Pointer to pt_regs of interrupted context
469 * 422 *
470 * Invokes: 423 * Invokes:
471 * - lockdep irqflag state tracking as low le 424 * - lockdep irqflag state tracking as low level ASM entry disabled
472 * interrupts. 425 * interrupts.
473 * 426 *
474 * - Context tracking if the exception hit us 427 * - Context tracking if the exception hit user mode.
475 * 428 *
476 * - The hardirq tracer to keep the state con 429 * - The hardirq tracer to keep the state consistent as low level ASM
477 * entry disabled interrupts. 430 * entry disabled interrupts.
478 * 431 *
479 * As a precondition, this requires that the e 432 * As a precondition, this requires that the entry came from user mode,
480 * idle, or a kernel context in which RCU is w 433 * idle, or a kernel context in which RCU is watching.
481 * 434 *
482 * For kernel mode entries RCU handling is don 435 * For kernel mode entries RCU handling is done conditional. If RCU is
483 * watching then the only RCU requirement is t 436 * watching then the only RCU requirement is to check whether the tick has
484 * to be restarted. If RCU is not watching the !! 437 * to be restarted. If RCU is not watching then rcu_irq_enter() has to be
485 * invoked on entry and ct_irq_exit() on exit. !! 438 * invoked on entry and rcu_irq_exit() on exit.
486 * 439 *
487 * Avoiding the ct_irq_enter/exit() calls is a !! 440 * Avoiding the rcu_irq_enter/exit() calls is an optimization but also
488 * solves the problem of kernel mode pagefault 441 * solves the problem of kernel mode pagefaults which can schedule, which
489 * is not possible after invoking ct_irq_enter !! 442 * is not possible after invoking rcu_irq_enter() without undoing it.
490 * 443 *
491 * For user mode entries irqentry_enter_from_u 444 * For user mode entries irqentry_enter_from_user_mode() is invoked to
492 * establish the proper context for NOHZ_FULL. 445 * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
493 * would not be possible. 446 * would not be possible.
494 * 447 *
495 * Returns: An opaque object that must be pass 448 * Returns: An opaque object that must be passed to idtentry_exit()
496 */ 449 */
497 irqentry_state_t noinstr irqentry_enter(struct 450 irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
498 451
499 /** 452 /**
500 * irqentry_exit_cond_resched - Conditionally 453 * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
501 * 454 *
502 * Conditional reschedule with additional sani 455 * Conditional reschedule with additional sanity checks.
503 */ 456 */
504 void raw_irqentry_exit_cond_resched(void); !! 457 void irqentry_exit_cond_resched(void);
505 #ifdef CONFIG_PREEMPT_DYNAMIC 458 #ifdef CONFIG_PREEMPT_DYNAMIC
506 #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) !! 459 DECLARE_STATIC_CALL(irqentry_exit_cond_resched, irqentry_exit_cond_resched);
507 #define irqentry_exit_cond_resched_dynamic_ena !! 460 #endif
508 #define irqentry_exit_cond_resched_dynamic_dis <<
509 DECLARE_STATIC_CALL(irqentry_exit_cond_resched <<
510 #define irqentry_exit_cond_resched() static <<
511 #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) <<
512 DECLARE_STATIC_KEY_TRUE(sk_dynamic_irqentry_ex <<
513 void dynamic_irqentry_exit_cond_resched(void); <<
514 #define irqentry_exit_cond_resched() dynami <<
515 #endif <<
516 #else /* CONFIG_PREEMPT_DYNAMIC */ <<
517 #define irqentry_exit_cond_resched() raw_ir <<
518 #endif /* CONFIG_PREEMPT_DYNAMIC */ <<
519 461
520 /** 462 /**
521 * irqentry_exit - Handle return from exceptio 463 * irqentry_exit - Handle return from exception that used irqentry_enter()
522 * @regs: Pointer to pt_regs (exception 464 * @regs: Pointer to pt_regs (exception entry regs)
523 * @state: Return value from matching cal 465 * @state: Return value from matching call to irqentry_enter()
524 * 466 *
525 * Depending on the return target (kernel/user 467 * Depending on the return target (kernel/user) this runs the necessary
526 * preemption and work checks if possible and 468 * preemption and work checks if possible and required and returns to
527 * the caller with interrupts disabled and no 469 * the caller with interrupts disabled and no further work pending.
528 * 470 *
529 * This is the last action before returning to 471 * This is the last action before returning to the low level ASM code which
530 * just needs to return to the appropriate con 472 * just needs to return to the appropriate context.
531 * 473 *
532 * Counterpart to irqentry_enter(). 474 * Counterpart to irqentry_enter().
533 */ 475 */
534 void noinstr irqentry_exit(struct pt_regs *reg 476 void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
535 477
536 /** 478 /**
537 * irqentry_nmi_enter - Handle NMI entry 479 * irqentry_nmi_enter - Handle NMI entry
538 * @regs: Pointer to currents pt_regs 480 * @regs: Pointer to currents pt_regs
539 * 481 *
540 * Similar to irqentry_enter() but taking care 482 * Similar to irqentry_enter() but taking care of the NMI constraints.
541 */ 483 */
542 irqentry_state_t noinstr irqentry_nmi_enter(st 484 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs);
543 485
544 /** 486 /**
545 * irqentry_nmi_exit - Handle return from NMI 487 * irqentry_nmi_exit - Handle return from NMI handling
546 * @regs: Pointer to pt_regs (NMI entry 488 * @regs: Pointer to pt_regs (NMI entry regs)
547 * @irq_state: Return value from matching cal 489 * @irq_state: Return value from matching call to irqentry_nmi_enter()
548 * 490 *
549 * Last action before returning to the low lev 491 * Last action before returning to the low level assembly code.
550 * 492 *
551 * Counterpart to irqentry_nmi_enter(). 493 * Counterpart to irqentry_nmi_enter().
552 */ 494 */
553 void noinstr irqentry_nmi_exit(struct pt_regs 495 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state);
554 496
555 #endif 497 #endif
556 498
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.