xref: /linux/include/linux/entry-common.h (revision 07f0148aafe8c95a3a76cd59e9e75b4d78d1d31d)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_ENTRYCOMMON_H
3 #define __LINUX_ENTRYCOMMON_H
4 
5 #include <linux/static_call_types.h>
6 #include <linux/ptrace.h>
7 #include <linux/syscalls.h>
8 #include <linux/seccomp.h>
9 #include <linux/sched.h>
10 
11 #include <asm/entry-common.h>
12 
13 /*
14  * Define dummy _TIF work flags if not defined by the architecture or for
15  * disabled functionality.
16  */
17 #ifndef _TIF_PATCH_PENDING
18 # define _TIF_PATCH_PENDING		(0)
19 #endif
20 
21 #ifndef _TIF_UPROBE
22 # define _TIF_UPROBE			(0)
23 #endif
24 
25 /*
26  * SYSCALL_WORK flags handled in syscall_enter_from_user_mode()
27  */
28 #ifndef ARCH_SYSCALL_WORK_ENTER
29 # define ARCH_SYSCALL_WORK_ENTER	(0)
30 #endif
31 
32 /*
33  * SYSCALL_WORK flags handled in syscall_exit_to_user_mode()
34  */
35 #ifndef ARCH_SYSCALL_WORK_EXIT
36 # define ARCH_SYSCALL_WORK_EXIT		(0)
37 #endif
38 
39 #define SYSCALL_WORK_ENTER	(SYSCALL_WORK_SECCOMP |			\
40 				 SYSCALL_WORK_SYSCALL_TRACEPOINT |	\
41 				 SYSCALL_WORK_SYSCALL_TRACE |		\
42 				 SYSCALL_WORK_SYSCALL_EMU |		\
43 				 SYSCALL_WORK_SYSCALL_AUDIT |		\
44 				 SYSCALL_WORK_SYSCALL_USER_DISPATCH |	\
45 				 ARCH_SYSCALL_WORK_ENTER)
46 #define SYSCALL_WORK_EXIT	(SYSCALL_WORK_SYSCALL_TRACEPOINT |	\
47 				 SYSCALL_WORK_SYSCALL_TRACE |		\
48 				 SYSCALL_WORK_SYSCALL_AUDIT |		\
49 				 SYSCALL_WORK_SYSCALL_USER_DISPATCH |	\
50 				 SYSCALL_WORK_SYSCALL_EXIT_TRAP	|	\
51 				 ARCH_SYSCALL_WORK_EXIT)
52 
53 /*
54  * TIF flags handled in exit_to_user_mode_loop()
55  */
56 #ifndef ARCH_EXIT_TO_USER_MODE_WORK
57 # define ARCH_EXIT_TO_USER_MODE_WORK		(0)
58 #endif
59 
60 #define EXIT_TO_USER_MODE_WORK						\
61 	(_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE |		\
62 	 _TIF_NEED_RESCHED | _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL |	\
63 	 ARCH_EXIT_TO_USER_MODE_WORK)
64 
65 /**
66  * arch_enter_from_user_mode - Architecture specific sanity check for user mode regs
67  * @regs:	Pointer to currents pt_regs
68  *
69  * Defaults to an empty implementation. Can be replaced by architecture
70  * specific code.
71  *
72  * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
73  * section. Use __always_inline so the compiler cannot push it out of line
74  * and make it instrumentable.
75  */
76 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs);
77 
78 #ifndef arch_enter_from_user_mode
79 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs) {}
80 #endif
81 
82 /**
83  * enter_from_user_mode - Establish state when coming from user mode
84  *
85  * Syscall/interrupt entry disables interrupts, but user mode is traced as
86  * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
87  *
88  * 1) Tell lockdep that interrupts are disabled
89  * 2) Invoke context tracking if enabled to reactivate RCU
90  * 3) Trace interrupts off state
91  *
92  * Invoked from architecture specific syscall entry code with interrupts
93  * disabled. The calling code has to be non-instrumentable. When the
94  * function returns all state is correct and interrupts are still
95  * disabled. The subsequent functions can be instrumented.
96  *
97  * This is invoked when there is architecture specific functionality to be
98  * done between establishing state and enabling interrupts. The caller must
99  * enable interrupts before invoking syscall_enter_from_user_mode_work().
100  */
101 void enter_from_user_mode(struct pt_regs *regs);
102 
103 /**
104  * syscall_enter_from_user_mode_prepare - Establish state and enable interrupts
105  * @regs:	Pointer to currents pt_regs
106  *
107  * Invoked from architecture specific syscall entry code with interrupts
108  * disabled. The calling code has to be non-instrumentable. When the
109  * function returns all state is correct, interrupts are enabled and the
110  * subsequent functions can be instrumented.
111  *
112  * This handles lockdep, RCU (context tracking) and tracing state, i.e.
113  * the functionality provided by enter_from_user_mode().
114  *
115  * This is invoked when there is extra architecture specific functionality
116  * to be done between establishing state and handling user mode entry work.
117  */
118 void syscall_enter_from_user_mode_prepare(struct pt_regs *regs);
119 
120 /**
121  * syscall_enter_from_user_mode_work - Check and handle work before invoking
122  *				       a syscall
123  * @regs:	Pointer to currents pt_regs
124  * @syscall:	The syscall number
125  *
126  * Invoked from architecture specific syscall entry code with interrupts
127  * enabled after invoking syscall_enter_from_user_mode_prepare() and extra
128  * architecture specific work.
129  *
130  * Returns: The original or a modified syscall number
131  *
132  * If the returned syscall number is -1 then the syscall should be
133  * skipped. In this case the caller may invoke syscall_set_error() or
134  * syscall_set_return_value() first.  If neither of those are called and -1
135  * is returned, then the syscall will fail with ENOSYS.
136  *
137  * It handles the following work items:
138  *
139  *  1) syscall_work flag dependent invocations of
140  *     ptrace_report_syscall_entry(), __secure_computing(), trace_sys_enter()
141  *  2) Invocation of audit_syscall_entry()
142  */
143 long syscall_enter_from_user_mode_work(struct pt_regs *regs, long syscall);
144 
145 /**
146  * syscall_enter_from_user_mode - Establish state and check and handle work
147  *				  before invoking a syscall
148  * @regs:	Pointer to currents pt_regs
149  * @syscall:	The syscall number
150  *
151  * Invoked from architecture specific syscall entry code with interrupts
152  * disabled. The calling code has to be non-instrumentable. When the
153  * function returns all state is correct, interrupts are enabled and the
154  * subsequent functions can be instrumented.
155  *
156  * This is combination of syscall_enter_from_user_mode_prepare() and
157  * syscall_enter_from_user_mode_work().
158  *
159  * Returns: The original or a modified syscall number. See
160  * syscall_enter_from_user_mode_work() for further explanation.
161  */
162 long syscall_enter_from_user_mode(struct pt_regs *regs, long syscall);
163 
164 /**
165  * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
166  * @ti_work:	Cached TIF flags gathered with interrupts disabled
167  *
168  * Defaults to local_irq_enable(). Can be supplied by architecture specific
169  * code.
170  */
171 static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
172 
173 #ifndef local_irq_enable_exit_to_user
174 static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
175 {
176 	local_irq_enable();
177 }
178 #endif
179 
180 /**
181  * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
182  *
183  * Defaults to local_irq_disable(). Can be supplied by architecture specific
184  * code.
185  */
186 static inline void local_irq_disable_exit_to_user(void);
187 
188 #ifndef local_irq_disable_exit_to_user
189 static inline void local_irq_disable_exit_to_user(void)
190 {
191 	local_irq_disable();
192 }
193 #endif
194 
195 /**
196  * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
197  *				 to user mode.
198  * @regs:	Pointer to currents pt_regs
199  * @ti_work:	Cached TIF flags gathered with interrupts disabled
200  *
201  * Invoked from exit_to_user_mode_loop() with interrupt enabled
202  *
203  * Defaults to NOOP. Can be supplied by architecture specific code.
204  */
205 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
206 					       unsigned long ti_work);
207 
208 #ifndef arch_exit_to_user_mode_work
209 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
210 					       unsigned long ti_work)
211 {
212 }
213 #endif
214 
215 /**
216  * arch_exit_to_user_mode_prepare - Architecture specific preparation for
217  *				    exit to user mode.
218  * @regs:	Pointer to currents pt_regs
219  * @ti_work:	Cached TIF flags gathered with interrupts disabled
220  *
221  * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
222  * function before return. Defaults to NOOP.
223  */
224 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
225 						  unsigned long ti_work);
226 
227 #ifndef arch_exit_to_user_mode_prepare
228 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
229 						  unsigned long ti_work)
230 {
231 }
232 #endif
233 
234 /**
235  * arch_exit_to_user_mode - Architecture specific final work before
236  *			    exit to user mode.
237  *
238  * Invoked from exit_to_user_mode() with interrupt disabled as the last
239  * function before return. Defaults to NOOP.
240  *
241  * This needs to be __always_inline because it is non-instrumentable code
242  * invoked after context tracking switched to user mode.
243  *
244  * An architecture implementation must not do anything complex, no locking
245  * etc. The main purpose is for speculation mitigations.
246  */
247 static __always_inline void arch_exit_to_user_mode(void);
248 
249 #ifndef arch_exit_to_user_mode
250 static __always_inline void arch_exit_to_user_mode(void) { }
251 #endif
252 
253 /**
254  * arch_do_signal_or_restart -  Architecture specific signal delivery function
255  * @regs:	Pointer to currents pt_regs
256  *
257  * Invoked from exit_to_user_mode_loop().
258  */
259 void arch_do_signal_or_restart(struct pt_regs *regs);
260 
261 /**
262  * exit_to_user_mode - Fixup state when exiting to user mode
263  *
264  * Syscall/interrupt exit enables interrupts, but the kernel state is
265  * interrupts disabled when this is invoked. Also tell RCU about it.
266  *
267  * 1) Trace interrupts on state
268  * 2) Invoke context tracking if enabled to adjust RCU state
269  * 3) Invoke architecture specific last minute exit code, e.g. speculation
270  *    mitigations, etc.: arch_exit_to_user_mode()
271  * 4) Tell lockdep that interrupts are enabled
272  *
273  * Invoked from architecture specific code when syscall_exit_to_user_mode()
274  * is not suitable as the last step before returning to userspace. Must be
275  * invoked with interrupts disabled and the caller must be
276  * non-instrumentable.
277  * The caller has to invoke syscall_exit_to_user_mode_work() before this.
278  */
279 void exit_to_user_mode(void);
280 
281 /**
282  * syscall_exit_to_user_mode_work - Handle work before returning to user mode
283  * @regs:	Pointer to currents pt_regs
284  *
285  * Same as step 1 and 2 of syscall_exit_to_user_mode() but without calling
286  * exit_to_user_mode() to perform the final transition to user mode.
287  *
288  * Calling convention is the same as for syscall_exit_to_user_mode() and it
289  * returns with all work handled and interrupts disabled. The caller must
290  * invoke exit_to_user_mode() before actually switching to user mode to
291  * make the final state transitions. Interrupts must stay disabled between
292  * return from this function and the invocation of exit_to_user_mode().
293  */
294 void syscall_exit_to_user_mode_work(struct pt_regs *regs);
295 
296 /**
297  * syscall_exit_to_user_mode - Handle work before returning to user mode
298  * @regs:	Pointer to currents pt_regs
299  *
300  * Invoked with interrupts enabled and fully valid regs. Returns with all
301  * work handled, interrupts disabled such that the caller can immediately
302  * switch to user mode. Called from architecture specific syscall and ret
303  * from fork code.
304  *
305  * The call order is:
306  *  1) One-time syscall exit work:
307  *	- rseq syscall exit
308  *      - audit
309  *	- syscall tracing
310  *	- ptrace (single stepping)
311  *
312  *  2) Preparatory work
313  *	- Exit to user mode loop (common TIF handling). Invokes
314  *	  arch_exit_to_user_mode_work() for architecture specific TIF work
315  *	- Architecture specific one time work arch_exit_to_user_mode_prepare()
316  *	- Address limit and lockdep checks
317  *
318  *  3) Final transition (lockdep, tracing, context tracking, RCU), i.e. the
319  *     functionality in exit_to_user_mode().
320  *
321  * This is a combination of syscall_exit_to_user_mode_work() (1,2) and
322  * exit_to_user_mode(). This function is preferred unless there is a
323  * compelling architectural reason to use the separate functions.
324  */
325 void syscall_exit_to_user_mode(struct pt_regs *regs);
326 
327 /**
328  * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
329  * @regs:	Pointer to currents pt_regs
330  *
331  * Invoked from architecture specific entry code with interrupts disabled.
332  * Can only be called when the interrupt entry came from user mode. The
333  * calling code must be non-instrumentable.  When the function returns all
334  * state is correct and the subsequent functions can be instrumented.
335  *
336  * The function establishes state (lockdep, RCU (context tracking), tracing)
337  */
338 void irqentry_enter_from_user_mode(struct pt_regs *regs);
339 
340 /**
341  * irqentry_exit_to_user_mode - Interrupt exit work
342  * @regs:	Pointer to current's pt_regs
343  *
344  * Invoked with interrupts disabled and fully valid regs. Returns with all
345  * work handled, interrupts disabled such that the caller can immediately
346  * switch to user mode. Called from architecture specific interrupt
347  * handling code.
348  *
349  * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
350  * Interrupt exit is not invoking #1 which is the syscall specific one time
351  * work.
352  */
353 void irqentry_exit_to_user_mode(struct pt_regs *regs);
354 
355 #ifndef irqentry_state
356 /**
357  * struct irqentry_state - Opaque object for exception state storage
358  * @exit_rcu: Used exclusively in the irqentry_*() calls; signals whether the
359  *            exit path has to invoke ct_irq_exit().
360  * @lockdep: Used exclusively in the irqentry_nmi_*() calls; ensures that
361  *           lockdep state is restored correctly on exit from nmi.
362  *
363  * This opaque object is filled in by the irqentry_*_enter() functions and
364  * must be passed back into the corresponding irqentry_*_exit() functions
365  * when the exception is complete.
366  *
367  * Callers of irqentry_*_[enter|exit]() must consider this structure opaque
368  * and all members private.  Descriptions of the members are provided to aid in
369  * the maintenance of the irqentry_*() functions.
370  */
371 typedef struct irqentry_state {
372 	union {
373 		bool	exit_rcu;
374 		bool	lockdep;
375 	};
376 } irqentry_state_t;
377 #endif
378 
379 /**
380  * irqentry_enter - Handle state tracking on ordinary interrupt entries
381  * @regs:	Pointer to pt_regs of interrupted context
382  *
383  * Invokes:
384  *  - lockdep irqflag state tracking as low level ASM entry disabled
385  *    interrupts.
386  *
387  *  - Context tracking if the exception hit user mode.
388  *
389  *  - The hardirq tracer to keep the state consistent as low level ASM
390  *    entry disabled interrupts.
391  *
392  * As a precondition, this requires that the entry came from user mode,
393  * idle, or a kernel context in which RCU is watching.
394  *
395  * For kernel mode entries RCU handling is done conditional. If RCU is
396  * watching then the only RCU requirement is to check whether the tick has
397  * to be restarted. If RCU is not watching then ct_irq_enter() has to be
398  * invoked on entry and ct_irq_exit() on exit.
399  *
400  * Avoiding the ct_irq_enter/exit() calls is an optimization but also
401  * solves the problem of kernel mode pagefaults which can schedule, which
402  * is not possible after invoking ct_irq_enter() without undoing it.
403  *
404  * For user mode entries irqentry_enter_from_user_mode() is invoked to
405  * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
406  * would not be possible.
407  *
408  * Returns: An opaque object that must be passed to idtentry_exit()
409  */
410 irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
411 
412 /**
413  * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
414  *
415  * Conditional reschedule with additional sanity checks.
416  */
417 void raw_irqentry_exit_cond_resched(void);
418 #ifdef CONFIG_PREEMPT_DYNAMIC
419 #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
420 #define irqentry_exit_cond_resched_dynamic_enabled	raw_irqentry_exit_cond_resched
421 #define irqentry_exit_cond_resched_dynamic_disabled	NULL
422 DECLARE_STATIC_CALL(irqentry_exit_cond_resched, raw_irqentry_exit_cond_resched);
423 #define irqentry_exit_cond_resched()	static_call(irqentry_exit_cond_resched)()
424 #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
425 DECLARE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched);
426 void dynamic_irqentry_exit_cond_resched(void);
427 #define irqentry_exit_cond_resched()	dynamic_irqentry_exit_cond_resched()
428 #endif
429 #else /* CONFIG_PREEMPT_DYNAMIC */
430 #define irqentry_exit_cond_resched()	raw_irqentry_exit_cond_resched()
431 #endif /* CONFIG_PREEMPT_DYNAMIC */
432 
433 /**
434  * irqentry_exit - Handle return from exception that used irqentry_enter()
435  * @regs:	Pointer to pt_regs (exception entry regs)
436  * @state:	Return value from matching call to irqentry_enter()
437  *
438  * Depending on the return target (kernel/user) this runs the necessary
439  * preemption and work checks if possible and required and returns to
440  * the caller with interrupts disabled and no further work pending.
441  *
442  * This is the last action before returning to the low level ASM code which
443  * just needs to return to the appropriate context.
444  *
445  * Counterpart to irqentry_enter().
446  */
447 void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
448 
449 /**
450  * irqentry_nmi_enter - Handle NMI entry
451  * @regs:	Pointer to currents pt_regs
452  *
453  * Similar to irqentry_enter() but taking care of the NMI constraints.
454  */
455 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs);
456 
457 /**
458  * irqentry_nmi_exit - Handle return from NMI handling
459  * @regs:	Pointer to pt_regs (NMI entry regs)
460  * @irq_state:	Return value from matching call to irqentry_nmi_enter()
461  *
462  * Last action before returning to the low level assembly code.
463  *
464  * Counterpart to irqentry_nmi_enter().
465  */
466 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state);
467 
468 #endif
469