Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __LINUX_ENTRYCOMMON_H
3#define __LINUX_ENTRYCOMMON_H
4
5#include <linux/tracehook.h>
6#include <linux/syscalls.h>
7#include <linux/seccomp.h>
8#include <linux/sched.h>
9
10#include <asm/entry-common.h>
11
12/*
13 * Define dummy _TIF work flags if not defined by the architecture or for
14 * disabled functionality.
15 */
16#ifndef _TIF_SYSCALL_EMU
17# define _TIF_SYSCALL_EMU (0)
18#endif
19
20#ifndef _TIF_SYSCALL_TRACEPOINT
21# define _TIF_SYSCALL_TRACEPOINT (0)
22#endif
23
24#ifndef _TIF_SECCOMP
25# define _TIF_SECCOMP (0)
26#endif
27
28#ifndef _TIF_SYSCALL_AUDIT
29# define _TIF_SYSCALL_AUDIT (0)
30#endif
31
32#ifndef _TIF_PATCH_PENDING
33# define _TIF_PATCH_PENDING (0)
34#endif
35
36#ifndef _TIF_UPROBE
37# define _TIF_UPROBE (0)
38#endif
39
40/*
41 * TIF flags handled in syscall_enter_from_usermode()
42 */
43#ifndef ARCH_SYSCALL_ENTER_WORK
44# define ARCH_SYSCALL_ENTER_WORK (0)
45#endif
46
47#define SYSCALL_ENTER_WORK \
48 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | _TIF_SECCOMP | \
49 _TIF_SYSCALL_TRACEPOINT | _TIF_SYSCALL_EMU | \
50 ARCH_SYSCALL_ENTER_WORK)
51
52/*
53 * TIF flags handled in syscall_exit_to_user_mode()
54 */
55#ifndef ARCH_SYSCALL_EXIT_WORK
56# define ARCH_SYSCALL_EXIT_WORK (0)
57#endif
58
59#define SYSCALL_EXIT_WORK \
60 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
61 _TIF_SYSCALL_TRACEPOINT | ARCH_SYSCALL_EXIT_WORK)
62
63/*
64 * TIF flags handled in exit_to_user_mode_loop()
65 */
66#ifndef ARCH_EXIT_TO_USER_MODE_WORK
67# define ARCH_EXIT_TO_USER_MODE_WORK (0)
68#endif
69
70#define EXIT_TO_USER_MODE_WORK \
71 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
72 _TIF_NEED_RESCHED | _TIF_PATCH_PENDING | \
73 ARCH_EXIT_TO_USER_MODE_WORK)
74
75/**
76 * arch_check_user_regs - Architecture specific sanity check for user mode regs
77 * @regs: Pointer to currents pt_regs
78 *
79 * Defaults to an empty implementation. Can be replaced by architecture
80 * specific code.
81 *
82 * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
83 * section. Use __always_inline so the compiler cannot push it out of line
84 * and make it instrumentable.
85 */
86static __always_inline void arch_check_user_regs(struct pt_regs *regs);
87
88#ifndef arch_check_user_regs
89static __always_inline void arch_check_user_regs(struct pt_regs *regs) {}
90#endif
91
92/**
93 * arch_syscall_enter_tracehook - Wrapper around tracehook_report_syscall_entry()
94 * @regs: Pointer to currents pt_regs
95 *
96 * Returns: 0 on success or an error code to skip the syscall.
97 *
98 * Defaults to tracehook_report_syscall_entry(). Can be replaced by
99 * architecture specific code.
100 *
101 * Invoked from syscall_enter_from_user_mode()
102 */
103static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs);
104
105#ifndef arch_syscall_enter_tracehook
106static inline __must_check int arch_syscall_enter_tracehook(struct pt_regs *regs)
107{
108 return tracehook_report_syscall_entry(regs);
109}
110#endif
111
112/**
113 * syscall_enter_from_user_mode - Check and handle work before invoking
114 * a syscall
115 * @regs: Pointer to currents pt_regs
116 * @syscall: The syscall number
117 *
118 * Invoked from architecture specific syscall entry code with interrupts
119 * disabled. The calling code has to be non-instrumentable. When the
120 * function returns all state is correct and the subsequent functions can be
121 * instrumented.
122 *
123 * Returns: The original or a modified syscall number
124 *
125 * If the returned syscall number is -1 then the syscall should be
126 * skipped. In this case the caller may invoke syscall_set_error() or
127 * syscall_set_return_value() first. If neither of those are called and -1
128 * is returned, then the syscall will fail with ENOSYS.
129 *
130 * The following functionality is handled here:
131 *
132 * 1) Establish state (lockdep, RCU (context tracking), tracing)
133 * 2) TIF flag dependent invocations of arch_syscall_enter_tracehook(),
134 * __secure_computing(), trace_sys_enter()
135 * 3) Invocation of audit_syscall_entry()
136 */
137long syscall_enter_from_user_mode(struct pt_regs *regs, long syscall);
138
139/**
140 * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
141 * @ti_work: Cached TIF flags gathered with interrupts disabled
142 *
143 * Defaults to local_irq_enable(). Can be supplied by architecture specific
144 * code.
145 */
146static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
147
148#ifndef local_irq_enable_exit_to_user
149static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
150{
151 local_irq_enable();
152}
153#endif
154
155/**
156 * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
157 *
158 * Defaults to local_irq_disable(). Can be supplied by architecture specific
159 * code.
160 */
161static inline void local_irq_disable_exit_to_user(void);
162
163#ifndef local_irq_disable_exit_to_user
164static inline void local_irq_disable_exit_to_user(void)
165{
166 local_irq_disable();
167}
168#endif
169
170/**
171 * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
172 * to user mode.
173 * @regs: Pointer to currents pt_regs
174 * @ti_work: Cached TIF flags gathered with interrupts disabled
175 *
176 * Invoked from exit_to_user_mode_loop() with interrupt enabled
177 *
178 * Defaults to NOOP. Can be supplied by architecture specific code.
179 */
180static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
181 unsigned long ti_work);
182
183#ifndef arch_exit_to_user_mode_work
184static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
185 unsigned long ti_work)
186{
187}
188#endif
189
190/**
191 * arch_exit_to_user_mode_prepare - Architecture specific preparation for
192 * exit to user mode.
193 * @regs: Pointer to currents pt_regs
194 * @ti_work: Cached TIF flags gathered with interrupts disabled
195 *
196 * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
197 * function before return. Defaults to NOOP.
198 */
199static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
200 unsigned long ti_work);
201
202#ifndef arch_exit_to_user_mode_prepare
203static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
204 unsigned long ti_work)
205{
206}
207#endif
208
209/**
210 * arch_exit_to_user_mode - Architecture specific final work before
211 * exit to user mode.
212 *
213 * Invoked from exit_to_user_mode() with interrupt disabled as the last
214 * function before return. Defaults to NOOP.
215 *
216 * This needs to be __always_inline because it is non-instrumentable code
217 * invoked after context tracking switched to user mode.
218 *
219 * An architecture implementation must not do anything complex, no locking
220 * etc. The main purpose is for speculation mitigations.
221 */
222static __always_inline void arch_exit_to_user_mode(void);
223
224#ifndef arch_exit_to_user_mode
225static __always_inline void arch_exit_to_user_mode(void) { }
226#endif
227
228/**
229 * arch_do_signal - Architecture specific signal delivery function
230 * @regs: Pointer to currents pt_regs
231 *
232 * Invoked from exit_to_user_mode_loop().
233 */
234void arch_do_signal(struct pt_regs *regs);
235
236/**
237 * arch_syscall_exit_tracehook - Wrapper around tracehook_report_syscall_exit()
238 * @regs: Pointer to currents pt_regs
239 * @step: Indicator for single step
240 *
241 * Defaults to tracehook_report_syscall_exit(). Can be replaced by
242 * architecture specific code.
243 *
244 * Invoked from syscall_exit_to_user_mode()
245 */
246static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step);
247
248#ifndef arch_syscall_exit_tracehook
249static inline void arch_syscall_exit_tracehook(struct pt_regs *regs, bool step)
250{
251 tracehook_report_syscall_exit(regs, step);
252}
253#endif
254
255/**
256 * syscall_exit_to_user_mode - Handle work before returning to user mode
257 * @regs: Pointer to currents pt_regs
258 *
259 * Invoked with interrupts enabled and fully valid regs. Returns with all
260 * work handled, interrupts disabled such that the caller can immediately
261 * switch to user mode. Called from architecture specific syscall and ret
262 * from fork code.
263 *
264 * The call order is:
265 * 1) One-time syscall exit work:
266 * - rseq syscall exit
267 * - audit
268 * - syscall tracing
269 * - tracehook (single stepping)
270 *
271 * 2) Preparatory work
272 * - Exit to user mode loop (common TIF handling). Invokes
273 * arch_exit_to_user_mode_work() for architecture specific TIF work
274 * - Architecture specific one time work arch_exit_to_user_mode_prepare()
275 * - Address limit and lockdep checks
276 *
277 * 3) Final transition (lockdep, tracing, context tracking, RCU). Invokes
278 * arch_exit_to_user_mode() to handle e.g. speculation mitigations
279 */
280void syscall_exit_to_user_mode(struct pt_regs *regs);
281
282/**
283 * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
284 * @regs: Pointer to currents pt_regs
285 *
286 * Invoked from architecture specific entry code with interrupts disabled.
287 * Can only be called when the interrupt entry came from user mode. The
288 * calling code must be non-instrumentable. When the function returns all
289 * state is correct and the subsequent functions can be instrumented.
290 *
291 * The function establishes state (lockdep, RCU (context tracking), tracing)
292 */
293void irqentry_enter_from_user_mode(struct pt_regs *regs);
294
295/**
296 * irqentry_exit_to_user_mode - Interrupt exit work
297 * @regs: Pointer to current's pt_regs
298 *
299 * Invoked with interrupts disbled and fully valid regs. Returns with all
300 * work handled, interrupts disabled such that the caller can immediately
301 * switch to user mode. Called from architecture specific interrupt
302 * handling code.
303 *
304 * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
305 * Interrupt exit is not invoking #1 which is the syscall specific one time
306 * work.
307 */
308void irqentry_exit_to_user_mode(struct pt_regs *regs);
309
310#ifndef irqentry_state
311typedef struct irqentry_state {
312 bool exit_rcu;
313} irqentry_state_t;
314#endif
315
316/**
317 * irqentry_enter - Handle state tracking on ordinary interrupt entries
318 * @regs: Pointer to pt_regs of interrupted context
319 *
320 * Invokes:
321 * - lockdep irqflag state tracking as low level ASM entry disabled
322 * interrupts.
323 *
324 * - Context tracking if the exception hit user mode.
325 *
326 * - The hardirq tracer to keep the state consistent as low level ASM
327 * entry disabled interrupts.
328 *
329 * As a precondition, this requires that the entry came from user mode,
330 * idle, or a kernel context in which RCU is watching.
331 *
332 * For kernel mode entries RCU handling is done conditional. If RCU is
333 * watching then the only RCU requirement is to check whether the tick has
334 * to be restarted. If RCU is not watching then rcu_irq_enter() has to be
335 * invoked on entry and rcu_irq_exit() on exit.
336 *
337 * Avoiding the rcu_irq_enter/exit() calls is an optimization but also
338 * solves the problem of kernel mode pagefaults which can schedule, which
339 * is not possible after invoking rcu_irq_enter() without undoing it.
340 *
341 * For user mode entries irqentry_enter_from_user_mode() is invoked to
342 * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
343 * would not be possible.
344 *
345 * Returns: An opaque object that must be passed to idtentry_exit()
346 */
347irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
348
349/**
350 * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
351 *
352 * Conditional reschedule with additional sanity checks.
353 */
354void irqentry_exit_cond_resched(void);
355
356/**
357 * irqentry_exit - Handle return from exception that used irqentry_enter()
358 * @regs: Pointer to pt_regs (exception entry regs)
359 * @state: Return value from matching call to irqentry_enter()
360 *
361 * Depending on the return target (kernel/user) this runs the necessary
362 * preemption and work checks if possible and reguired and returns to
363 * the caller with interrupts disabled and no further work pending.
364 *
365 * This is the last action before returning to the low level ASM code which
366 * just needs to return to the appropriate context.
367 *
368 * Counterpart to irqentry_enter().
369 */
370void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
371
372#endif