xref: /linux/include/linux/irq-entry-common.h (revision 6fb5ff63b35b7e849cc8510957f25753f87f63d2)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_IRQENTRYCOMMON_H
3 #define __LINUX_IRQENTRYCOMMON_H
4 
5 #include <linux/static_call_types.h>
6 #include <linux/syscalls.h>
7 #include <linux/context_tracking.h>
8 #include <linux/tick.h>
9 #include <linux/kmsan.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 /*
22  * TIF flags handled in exit_to_user_mode_loop()
23  */
24 #ifndef ARCH_EXIT_TO_USER_MODE_WORK
25 # define ARCH_EXIT_TO_USER_MODE_WORK		(0)
26 #endif
27 
28 #define EXIT_TO_USER_MODE_WORK						\
29 	(_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE |		\
30 	 _TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY |			\
31 	 _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL |			\
32 	 ARCH_EXIT_TO_USER_MODE_WORK)
33 
34 /**
35  * arch_enter_from_user_mode - Architecture specific sanity check for user mode regs
36  * @regs:	Pointer to currents pt_regs
37  *
38  * Defaults to an empty implementation. Can be replaced by architecture
39  * specific code.
40  *
41  * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
42  * section. Use __always_inline so the compiler cannot push it out of line
43  * and make it instrumentable.
44  */
45 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs);
46 
47 #ifndef arch_enter_from_user_mode
48 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs) {}
49 #endif
50 
51 /**
52  * enter_from_user_mode - Establish state when coming from user mode
53  *
54  * Syscall/interrupt entry disables interrupts, but user mode is traced as
55  * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
56  *
57  * 1) Tell lockdep that interrupts are disabled
58  * 2) Invoke context tracking if enabled to reactivate RCU
59  * 3) Trace interrupts off state
60  *
61  * Invoked from architecture specific syscall entry code with interrupts
62  * disabled. The calling code has to be non-instrumentable. When the
63  * function returns all state is correct and interrupts are still
64  * disabled. The subsequent functions can be instrumented.
65  *
66  * This is invoked when there is architecture specific functionality to be
67  * done between establishing state and enabling interrupts. The caller must
68  * enable interrupts before invoking syscall_enter_from_user_mode_work().
69  */
70 static __always_inline void enter_from_user_mode(struct pt_regs *regs)
71 {
72 	arch_enter_from_user_mode(regs);
73 	lockdep_hardirqs_off(CALLER_ADDR0);
74 
75 	CT_WARN_ON(__ct_state() != CT_STATE_USER);
76 	user_exit_irqoff();
77 
78 	instrumentation_begin();
79 	kmsan_unpoison_entry_regs(regs);
80 	trace_hardirqs_off_finish();
81 	instrumentation_end();
82 }
83 
84 /**
85  * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
86  * @ti_work:	Cached TIF flags gathered with interrupts disabled
87  *
88  * Defaults to local_irq_enable(). Can be supplied by architecture specific
89  * code.
90  */
91 static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
92 
93 #ifndef local_irq_enable_exit_to_user
94 static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
95 {
96 	local_irq_enable();
97 }
98 #endif
99 
100 /**
101  * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
102  *
103  * Defaults to local_irq_disable(). Can be supplied by architecture specific
104  * code.
105  */
106 static inline void local_irq_disable_exit_to_user(void);
107 
108 #ifndef local_irq_disable_exit_to_user
109 static inline void local_irq_disable_exit_to_user(void)
110 {
111 	local_irq_disable();
112 }
113 #endif
114 
115 /**
116  * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
117  *				 to user mode.
118  * @regs:	Pointer to currents pt_regs
119  * @ti_work:	Cached TIF flags gathered with interrupts disabled
120  *
121  * Invoked from exit_to_user_mode_loop() with interrupt enabled
122  *
123  * Defaults to NOOP. Can be supplied by architecture specific code.
124  */
125 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
126 					       unsigned long ti_work);
127 
128 #ifndef arch_exit_to_user_mode_work
129 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
130 					       unsigned long ti_work)
131 {
132 }
133 #endif
134 
135 /**
136  * arch_exit_to_user_mode_prepare - Architecture specific preparation for
137  *				    exit to user mode.
138  * @regs:	Pointer to currents pt_regs
139  * @ti_work:	Cached TIF flags gathered with interrupts disabled
140  *
141  * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
142  * function before return. Defaults to NOOP.
143  */
144 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
145 						  unsigned long ti_work);
146 
147 #ifndef arch_exit_to_user_mode_prepare
148 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
149 						  unsigned long ti_work)
150 {
151 }
152 #endif
153 
154 /**
155  * arch_exit_to_user_mode - Architecture specific final work before
156  *			    exit to user mode.
157  *
158  * Invoked from exit_to_user_mode() with interrupt disabled as the last
159  * function before return. Defaults to NOOP.
160  *
161  * This needs to be __always_inline because it is non-instrumentable code
162  * invoked after context tracking switched to user mode.
163  *
164  * An architecture implementation must not do anything complex, no locking
165  * etc. The main purpose is for speculation mitigations.
166  */
167 static __always_inline void arch_exit_to_user_mode(void);
168 
169 #ifndef arch_exit_to_user_mode
170 static __always_inline void arch_exit_to_user_mode(void) { }
171 #endif
172 
173 /**
174  * arch_do_signal_or_restart -  Architecture specific signal delivery function
175  * @regs:	Pointer to currents pt_regs
176  *
177  * Invoked from exit_to_user_mode_loop().
178  */
179 void arch_do_signal_or_restart(struct pt_regs *regs);
180 
181 /**
182  * exit_to_user_mode_loop - do any pending work before leaving to user space
183  */
184 unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
185 				     unsigned long ti_work);
186 
187 /**
188  * exit_to_user_mode_prepare - call exit_to_user_mode_loop() if required
189  * @regs:	Pointer to pt_regs on entry stack
190  *
191  * 1) check that interrupts are disabled
192  * 2) call tick_nohz_user_enter_prepare()
193  * 3) call exit_to_user_mode_loop() if any flags from
194  *    EXIT_TO_USER_MODE_WORK are set
195  * 4) check that interrupts are still disabled
196  */
197 static __always_inline void exit_to_user_mode_prepare(struct pt_regs *regs)
198 {
199 	unsigned long ti_work;
200 
201 	lockdep_assert_irqs_disabled();
202 
203 	/* Flush pending rcuog wakeup before the last need_resched() check */
204 	tick_nohz_user_enter_prepare();
205 
206 	ti_work = read_thread_flags();
207 	if (unlikely(ti_work & EXIT_TO_USER_MODE_WORK))
208 		ti_work = exit_to_user_mode_loop(regs, ti_work);
209 
210 	arch_exit_to_user_mode_prepare(regs, ti_work);
211 
212 	/* Ensure that kernel state is sane for a return to userspace */
213 	kmap_assert_nomap();
214 	lockdep_assert_irqs_disabled();
215 	lockdep_sys_exit();
216 }
217 
218 /**
219  * exit_to_user_mode - Fixup state when exiting to user mode
220  *
221  * Syscall/interrupt exit enables interrupts, but the kernel state is
222  * interrupts disabled when this is invoked. Also tell RCU about it.
223  *
224  * 1) Trace interrupts on state
225  * 2) Invoke context tracking if enabled to adjust RCU state
226  * 3) Invoke architecture specific last minute exit code, e.g. speculation
227  *    mitigations, etc.: arch_exit_to_user_mode()
228  * 4) Tell lockdep that interrupts are enabled
229  *
230  * Invoked from architecture specific code when syscall_exit_to_user_mode()
231  * is not suitable as the last step before returning to userspace. Must be
232  * invoked with interrupts disabled and the caller must be
233  * non-instrumentable.
234  * The caller has to invoke syscall_exit_to_user_mode_work() before this.
235  */
236 static __always_inline void exit_to_user_mode(void)
237 {
238 	instrumentation_begin();
239 	trace_hardirqs_on_prepare();
240 	lockdep_hardirqs_on_prepare();
241 	instrumentation_end();
242 
243 	user_enter_irqoff();
244 	arch_exit_to_user_mode();
245 	lockdep_hardirqs_on(CALLER_ADDR0);
246 }
247 
248 /**
249  * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
250  * @regs:	Pointer to currents pt_regs
251  *
252  * Invoked from architecture specific entry code with interrupts disabled.
253  * Can only be called when the interrupt entry came from user mode. The
254  * calling code must be non-instrumentable.  When the function returns all
255  * state is correct and the subsequent functions can be instrumented.
256  *
257  * The function establishes state (lockdep, RCU (context tracking), tracing)
258  */
259 void irqentry_enter_from_user_mode(struct pt_regs *regs);
260 
261 /**
262  * irqentry_exit_to_user_mode - Interrupt exit work
263  * @regs:	Pointer to current's pt_regs
264  *
265  * Invoked with interrupts disabled and fully valid regs. Returns with all
266  * work handled, interrupts disabled such that the caller can immediately
267  * switch to user mode. Called from architecture specific interrupt
268  * handling code.
269  *
270  * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
271  * Interrupt exit is not invoking #1 which is the syscall specific one time
272  * work.
273  */
274 void irqentry_exit_to_user_mode(struct pt_regs *regs);
275 
276 #ifndef irqentry_state
277 /**
278  * struct irqentry_state - Opaque object for exception state storage
279  * @exit_rcu: Used exclusively in the irqentry_*() calls; signals whether the
280  *            exit path has to invoke ct_irq_exit().
281  * @lockdep: Used exclusively in the irqentry_nmi_*() calls; ensures that
282  *           lockdep state is restored correctly on exit from nmi.
283  *
284  * This opaque object is filled in by the irqentry_*_enter() functions and
285  * must be passed back into the corresponding irqentry_*_exit() functions
286  * when the exception is complete.
287  *
288  * Callers of irqentry_*_[enter|exit]() must consider this structure opaque
289  * and all members private.  Descriptions of the members are provided to aid in
290  * the maintenance of the irqentry_*() functions.
291  */
292 typedef struct irqentry_state {
293 	union {
294 		bool	exit_rcu;
295 		bool	lockdep;
296 	};
297 } irqentry_state_t;
298 #endif
299 
300 /**
301  * irqentry_enter - Handle state tracking on ordinary interrupt entries
302  * @regs:	Pointer to pt_regs of interrupted context
303  *
304  * Invokes:
305  *  - lockdep irqflag state tracking as low level ASM entry disabled
306  *    interrupts.
307  *
308  *  - Context tracking if the exception hit user mode.
309  *
310  *  - The hardirq tracer to keep the state consistent as low level ASM
311  *    entry disabled interrupts.
312  *
313  * As a precondition, this requires that the entry came from user mode,
314  * idle, or a kernel context in which RCU is watching.
315  *
316  * For kernel mode entries RCU handling is done conditional. If RCU is
317  * watching then the only RCU requirement is to check whether the tick has
318  * to be restarted. If RCU is not watching then ct_irq_enter() has to be
319  * invoked on entry and ct_irq_exit() on exit.
320  *
321  * Avoiding the ct_irq_enter/exit() calls is an optimization but also
322  * solves the problem of kernel mode pagefaults which can schedule, which
323  * is not possible after invoking ct_irq_enter() without undoing it.
324  *
325  * For user mode entries irqentry_enter_from_user_mode() is invoked to
326  * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
327  * would not be possible.
328  *
329  * Returns: An opaque object that must be passed to idtentry_exit()
330  */
331 irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
332 
333 /**
334  * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
335  *
336  * Conditional reschedule with additional sanity checks.
337  */
338 void raw_irqentry_exit_cond_resched(void);
339 #ifdef CONFIG_PREEMPT_DYNAMIC
340 #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
341 #define irqentry_exit_cond_resched_dynamic_enabled	raw_irqentry_exit_cond_resched
342 #define irqentry_exit_cond_resched_dynamic_disabled	NULL
343 DECLARE_STATIC_CALL(irqentry_exit_cond_resched, raw_irqentry_exit_cond_resched);
344 #define irqentry_exit_cond_resched()	static_call(irqentry_exit_cond_resched)()
345 #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
346 DECLARE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched);
347 void dynamic_irqentry_exit_cond_resched(void);
348 #define irqentry_exit_cond_resched()	dynamic_irqentry_exit_cond_resched()
349 #endif
350 #else /* CONFIG_PREEMPT_DYNAMIC */
351 #define irqentry_exit_cond_resched()	raw_irqentry_exit_cond_resched()
352 #endif /* CONFIG_PREEMPT_DYNAMIC */
353 
354 /**
355  * irqentry_exit - Handle return from exception that used irqentry_enter()
356  * @regs:	Pointer to pt_regs (exception entry regs)
357  * @state:	Return value from matching call to irqentry_enter()
358  *
359  * Depending on the return target (kernel/user) this runs the necessary
360  * preemption and work checks if possible and required and returns to
361  * the caller with interrupts disabled and no further work pending.
362  *
363  * This is the last action before returning to the low level ASM code which
364  * just needs to return to the appropriate context.
365  *
366  * Counterpart to irqentry_enter().
367  */
368 void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
369 
370 /**
371  * irqentry_nmi_enter - Handle NMI entry
372  * @regs:	Pointer to currents pt_regs
373  *
374  * Similar to irqentry_enter() but taking care of the NMI constraints.
375  */
376 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs);
377 
378 /**
379  * irqentry_nmi_exit - Handle return from NMI handling
380  * @regs:	Pointer to pt_regs (NMI entry regs)
381  * @irq_state:	Return value from matching call to irqentry_nmi_enter()
382  *
383  * Last action before returning to the low level assembly code.
384  *
385  * Counterpart to irqentry_nmi_enter().
386  */
387 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state);
388 
389 #endif
390