1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_ENTRYCOMMON_H
3 #define __LINUX_ENTRYCOMMON_H
5 #include <linux/tracehook.h>
6 #include <linux/syscalls.h>
7 #include <linux/seccomp.h>
8 #include <linux/sched.h>
10 #include <asm/entry-common.h>
13 * Define dummy _TIF work flags if not defined by the architecture or for
14 * disabled functionality.
16 #ifndef _TIF_PATCH_PENDING
17 # define _TIF_PATCH_PENDING (0)
21 # define _TIF_UPROBE (0)
25 * SYSCALL_WORK flags handled in syscall_enter_from_user_mode()
27 #ifndef ARCH_SYSCALL_WORK_ENTER
28 # define ARCH_SYSCALL_WORK_ENTER (0)
32 * SYSCALL_WORK flags handled in syscall_exit_to_user_mode()
34 #ifndef ARCH_SYSCALL_WORK_EXIT
35 # define ARCH_SYSCALL_WORK_EXIT (0)
38 #define SYSCALL_WORK_ENTER (SYSCALL_WORK_SECCOMP | \
39 SYSCALL_WORK_SYSCALL_TRACEPOINT | \
40 SYSCALL_WORK_SYSCALL_TRACE | \
41 SYSCALL_WORK_SYSCALL_EMU | \
42 SYSCALL_WORK_SYSCALL_AUDIT | \
43 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
44 ARCH_SYSCALL_WORK_ENTER)
45 #define SYSCALL_WORK_EXIT (SYSCALL_WORK_SYSCALL_TRACEPOINT | \
46 SYSCALL_WORK_SYSCALL_TRACE | \
47 SYSCALL_WORK_SYSCALL_AUDIT | \
48 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
49 SYSCALL_WORK_SYSCALL_EXIT_TRAP | \
50 ARCH_SYSCALL_WORK_EXIT)
53 * TIF flags handled in exit_to_user_mode_loop()
55 #ifndef ARCH_EXIT_TO_USER_MODE_WORK
56 # define ARCH_EXIT_TO_USER_MODE_WORK (0)
59 #define EXIT_TO_USER_MODE_WORK \
60 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
61 _TIF_NEED_RESCHED | _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL | \
62 ARCH_EXIT_TO_USER_MODE_WORK)
65 * arch_check_user_regs - Architecture specific sanity check for user mode regs
66 * @regs: Pointer to currents pt_regs
68 * Defaults to an empty implementation. Can be replaced by architecture
71 * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
72 * section. Use __always_inline so the compiler cannot push it out of line
73 * and make it instrumentable.
75 static __always_inline void arch_check_user_regs(struct pt_regs *regs);
77 #ifndef arch_check_user_regs
78 static __always_inline void arch_check_user_regs(struct pt_regs *regs) {}
82 * arch_syscall_enter_tracehook - Wrapper around tracehook_report_syscall_entry()
83 * @regs: Pointer to currents pt_regs
85 * Returns: 0 on success or an error code to skip the syscall.
87 * Defaults to tracehook_report_syscall_entry(). Can be replaced by
88 * architecture specific code.
90 * Invoked from syscall_enter_from_user_mode()
92 static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs);
94 #ifndef arch_syscall_enter_tracehook
95 static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs)
97 return tracehook_report_syscall_entry(regs);
102 * enter_from_user_mode - Establish state when coming from user mode
104 * Syscall/interrupt entry disables interrupts, but user mode is traced as
105 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
107 * 1) Tell lockdep that interrupts are disabled
108 * 2) Invoke context tracking if enabled to reactivate RCU
109 * 3) Trace interrupts off state
111 * Invoked from architecture specific syscall entry code with interrupts
112 * disabled. The calling code has to be non-instrumentable. When the
113 * function returns all state is correct and interrupts are still
114 * disabled. The subsequent functions can be instrumented.
116 * This is invoked when there is architecture specific functionality to be
117 * done between establishing state and enabling interrupts. The caller must
118 * enable interrupts before invoking syscall_enter_from_user_mode_work().
120 void enter_from_user_mode(struct pt_regs *regs);
123 * syscall_enter_from_user_mode_prepare - Establish state and enable interrupts
124 * @regs: Pointer to currents pt_regs
126 * Invoked from architecture specific syscall entry code with interrupts
127 * disabled. The calling code has to be non-instrumentable. When the
128 * function returns all state is correct, interrupts are enabled and the
129 * subsequent functions can be instrumented.
131 * This handles lockdep, RCU (context tracking) and tracing state, i.e.
132 * the functionality provided by enter_from_user_mode().
134 * This is invoked when there is extra architecture specific functionality
135 * to be done between establishing state and handling user mode entry work.
137 void syscall_enter_from_user_mode_prepare(struct pt_regs *regs);
140 * syscall_enter_from_user_mode_work - Check and handle work before invoking
142 * @regs: Pointer to currents pt_regs
143 * @syscall: The syscall number
145 * Invoked from architecture specific syscall entry code with interrupts
146 * enabled after invoking syscall_enter_from_user_mode_prepare() and extra
147 * architecture specific work.
149 * Returns: The original or a modified syscall number
151 * If the returned syscall number is -1 then the syscall should be
152 * skipped. In this case the caller may invoke syscall_set_error() or
153 * syscall_set_return_value() first. If neither of those are called and -1
154 * is returned, then the syscall will fail with ENOSYS.
156 * It handles the following work items:
158 * 1) syscall_work flag dependent invocations of
159 * arch_syscall_enter_tracehook(), __secure_computing(), trace_sys_enter()
160 * 2) Invocation of audit_syscall_entry()
162 long syscall_enter_from_user_mode_work(struct pt_regs *regs, long syscall);
165 * syscall_enter_from_user_mode - Establish state and check and handle work
166 * before invoking a syscall
167 * @regs: Pointer to currents pt_regs
168 * @syscall: The syscall number
170 * Invoked from architecture specific syscall entry code with interrupts
171 * disabled. The calling code has to be non-instrumentable. When the
172 * function returns all state is correct, interrupts are enabled and the
173 * subsequent functions can be instrumented.
175 * This is combination of syscall_enter_from_user_mode_prepare() and
176 * syscall_enter_from_user_mode_work().
178 * Returns: The original or a modified syscall number. See
179 * syscall_enter_from_user_mode_work() for further explanation.
181 long syscall_enter_from_user_mode(struct pt_regs *regs, long syscall);
184 * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
185 * @ti_work: Cached TIF flags gathered with interrupts disabled
187 * Defaults to local_irq_enable(). Can be supplied by architecture specific
190 static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
192 #ifndef local_irq_enable_exit_to_user
193 static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
200 * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
202 * Defaults to local_irq_disable(). Can be supplied by architecture specific
205 static inline void local_irq_disable_exit_to_user(void);
207 #ifndef local_irq_disable_exit_to_user
208 static inline void local_irq_disable_exit_to_user(void)
215 * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
217 * @regs: Pointer to currents pt_regs
218 * @ti_work: Cached TIF flags gathered with interrupts disabled
220 * Invoked from exit_to_user_mode_loop() with interrupt enabled
222 * Defaults to NOOP. Can be supplied by architecture specific code.
224 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
225 unsigned long ti_work);
227 #ifndef arch_exit_to_user_mode_work
228 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
229 unsigned long ti_work)
235 * arch_exit_to_user_mode_prepare - Architecture specific preparation for
237 * @regs: Pointer to currents pt_regs
238 * @ti_work: Cached TIF flags gathered with interrupts disabled
240 * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
241 * function before return. Defaults to NOOP.
243 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
244 unsigned long ti_work);
246 #ifndef arch_exit_to_user_mode_prepare
247 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
248 unsigned long ti_work)
254 * arch_exit_to_user_mode - Architecture specific final work before
257 * Invoked from exit_to_user_mode() with interrupt disabled as the last
258 * function before return. Defaults to NOOP.
260 * This needs to be __always_inline because it is non-instrumentable code
261 * invoked after context tracking switched to user mode.
263 * An architecture implementation must not do anything complex, no locking
264 * etc. The main purpose is for speculation mitigations.
266 static __always_inline void arch_exit_to_user_mode(void);
268 #ifndef arch_exit_to_user_mode
269 static __always_inline void arch_exit_to_user_mode(void) { }
273 * arch_do_signal_or_restart - Architecture specific signal delivery function
274 * @regs: Pointer to currents pt_regs
275 * @has_signal: actual signal to handle
277 * Invoked from exit_to_user_mode_loop().
279 void arch_do_signal_or_restart(struct pt_regs *regs, bool has_signal);
282 * arch_syscall_exit_tracehook - Wrapper around tracehook_report_syscall_exit()
283 * @regs: Pointer to currents pt_regs
284 * @step: Indicator for single step
286 * Defaults to tracehook_report_syscall_exit(). Can be replaced by
287 * architecture specific code.
289 * Invoked from syscall_exit_to_user_mode()
291 static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step);
293 #ifndef arch_syscall_exit_tracehook
294 static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step)
296 tracehook_report_syscall_exit(regs, step);
301 * exit_to_user_mode - Fixup state when exiting to user mode
303 * Syscall/interrupt exit enables interrupts, but the kernel state is
304 * interrupts disabled when this is invoked. Also tell RCU about it.
306 * 1) Trace interrupts on state
307 * 2) Invoke context tracking if enabled to adjust RCU state
308 * 3) Invoke architecture specific last minute exit code, e.g. speculation
309 * mitigations, etc.: arch_exit_to_user_mode()
310 * 4) Tell lockdep that interrupts are enabled
312 * Invoked from architecture specific code when syscall_exit_to_user_mode()
313 * is not suitable as the last step before returning to userspace. Must be
314 * invoked with interrupts disabled and the caller must be
315 * non-instrumentable.
316 * The caller has to invoke syscall_exit_to_user_mode_work() before this.
318 void exit_to_user_mode(void);
321 * syscall_exit_to_user_mode_work - Handle work before returning to user mode
322 * @regs: Pointer to currents pt_regs
324 * Same as step 1 and 2 of syscall_exit_to_user_mode() but without calling
325 * exit_to_user_mode() to perform the final transition to user mode.
327 * Calling convention is the same as for syscall_exit_to_user_mode() and it
328 * returns with all work handled and interrupts disabled. The caller must
329 * invoke exit_to_user_mode() before actually switching to user mode to
330 * make the final state transitions. Interrupts must stay disabled between
331 * return from this function and the invocation of exit_to_user_mode().
333 void syscall_exit_to_user_mode_work(struct pt_regs *regs);
336 * syscall_exit_to_user_mode - Handle work before returning to user mode
337 * @regs: Pointer to currents pt_regs
339 * Invoked with interrupts enabled and fully valid regs. Returns with all
340 * work handled, interrupts disabled such that the caller can immediately
341 * switch to user mode. Called from architecture specific syscall and ret
345 * 1) One-time syscall exit work:
346 * - rseq syscall exit
349 * - tracehook (single stepping)
351 * 2) Preparatory work
352 * - Exit to user mode loop (common TIF handling). Invokes
353 * arch_exit_to_user_mode_work() for architecture specific TIF work
354 * - Architecture specific one time work arch_exit_to_user_mode_prepare()
355 * - Address limit and lockdep checks
357 * 3) Final transition (lockdep, tracing, context tracking, RCU), i.e. the
358 * functionality in exit_to_user_mode().
360 * This is a combination of syscall_exit_to_user_mode_work() (1,2) and
361 * exit_to_user_mode(). This function is preferred unless there is a
362 * compelling architectural reason to use the seperate functions.
364 void syscall_exit_to_user_mode(struct pt_regs *regs);
367 * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
368 * @regs: Pointer to currents pt_regs
370 * Invoked from architecture specific entry code with interrupts disabled.
371 * Can only be called when the interrupt entry came from user mode. The
372 * calling code must be non-instrumentable. When the function returns all
373 * state is correct and the subsequent functions can be instrumented.
375 * The function establishes state (lockdep, RCU (context tracking), tracing)
377 void irqentry_enter_from_user_mode(struct pt_regs *regs);
380 * irqentry_exit_to_user_mode - Interrupt exit work
381 * @regs: Pointer to current's pt_regs
383 * Invoked with interrupts disbled and fully valid regs. Returns with all
384 * work handled, interrupts disabled such that the caller can immediately
385 * switch to user mode. Called from architecture specific interrupt
388 * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
389 * Interrupt exit is not invoking #1 which is the syscall specific one time
392 void irqentry_exit_to_user_mode(struct pt_regs *regs);
394 #ifndef irqentry_state
396 * struct irqentry_state - Opaque object for exception state storage
397 * @exit_rcu: Used exclusively in the irqentry_*() calls; signals whether the
398 * exit path has to invoke rcu_irq_exit().
399 * @lockdep: Used exclusively in the irqentry_nmi_*() calls; ensures that
400 * lockdep state is restored correctly on exit from nmi.
402 * This opaque object is filled in by the irqentry_*_enter() functions and
403 * must be passed back into the corresponding irqentry_*_exit() functions
404 * when the exception is complete.
406 * Callers of irqentry_*_[enter|exit]() must consider this structure opaque
407 * and all members private. Descriptions of the members are provided to aid in
408 * the maintenance of the irqentry_*() functions.
410 typedef struct irqentry_state {
419 * irqentry_enter - Handle state tracking on ordinary interrupt entries
420 * @regs: Pointer to pt_regs of interrupted context
423 * - lockdep irqflag state tracking as low level ASM entry disabled
426 * - Context tracking if the exception hit user mode.
428 * - The hardirq tracer to keep the state consistent as low level ASM
429 * entry disabled interrupts.
431 * As a precondition, this requires that the entry came from user mode,
432 * idle, or a kernel context in which RCU is watching.
434 * For kernel mode entries RCU handling is done conditional. If RCU is
435 * watching then the only RCU requirement is to check whether the tick has
436 * to be restarted. If RCU is not watching then rcu_irq_enter() has to be
437 * invoked on entry and rcu_irq_exit() on exit.
439 * Avoiding the rcu_irq_enter/exit() calls is an optimization but also
440 * solves the problem of kernel mode pagefaults which can schedule, which
441 * is not possible after invoking rcu_irq_enter() without undoing it.
443 * For user mode entries irqentry_enter_from_user_mode() is invoked to
444 * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
445 * would not be possible.
447 * Returns: An opaque object that must be passed to idtentry_exit()
449 irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
452 * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
454 * Conditional reschedule with additional sanity checks.
456 void irqentry_exit_cond_resched(void);
459 * irqentry_exit - Handle return from exception that used irqentry_enter()
460 * @regs: Pointer to pt_regs (exception entry regs)
461 * @state: Return value from matching call to irqentry_enter()
463 * Depending on the return target (kernel/user) this runs the necessary
464 * preemption and work checks if possible and required and returns to
465 * the caller with interrupts disabled and no further work pending.
467 * This is the last action before returning to the low level ASM code which
468 * just needs to return to the appropriate context.
470 * Counterpart to irqentry_enter().
472 void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
475 * irqentry_nmi_enter - Handle NMI entry
476 * @regs: Pointer to currents pt_regs
478 * Similar to irqentry_enter() but taking care of the NMI constraints.
480 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs);
483 * irqentry_nmi_exit - Handle return from NMI handling
484 * @regs: Pointer to pt_regs (NMI entry regs)
485 * @irq_state: Return value from matching call to irqentry_nmi_enter()
487 * Last action before returning to the low level assembly code.
489 * Counterpart to irqentry_nmi_enter().
491 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state);