1 Entry/exit handling for exceptions, interrupts, syscalls and KVM
2 ================================================================
3
4 All transitions between execution domains require state updates which are
5 subject to strict ordering constraints. State updates are required for the
6 following:
7
8 * Lockdep
9 * RCU / Context tracking
10 * Preemption counter
11 * Tracing
12 * Time accounting
13
14 The update order depends on the transition type and is explained below in
15 the transition type sections: `Syscalls`_, `KVM`_, `Interrupts and regular
16 exceptions`_, `NMI and NMI-like exceptions`_.
17
18 Non-instrumentable code - noinstr
19 ---------------------------------
20
21 Most instrumentation facilities depend on RCU, so instrumentation is prohibited
22 for entry code before RCU starts watching and exit code after RCU stops
23 watching. In addition, many architectures must save and restore register state,
24 which means that (for example) a breakpoint in the breakpoint entry code would
25 overwrite the debug registers of the initial breakpoint.
26
27 Such code must be marked with the 'noinstr' attribute, placing that code into a
28 special section inaccessible to instrumentation and debug facilities. Some
29 functions are partially instrumentable, which is handled by marking them
30 noinstr and using instrumentation_begin() and instrumentation_end() to flag the
31 instrumentable ranges of code:
32
33 .. code-block:: c
34
35 noinstr void entry(void)
36 {
37 handle_entry(); // <-- must be 'noinstr' or '__always_inline'
38 ...
39
40 instrumentation_begin();
41 handle_context(); // <-- instrumentable code
42 instrumentation_end();
43
44 ...
45 handle_exit(); // <-- must be 'noinstr' or '__always_inline'
46 }
47
48 This allows verification of the 'noinstr' restrictions via objtool on
49 supported architectures.
50
51 Invoking non-instrumentable functions from instrumentable context has no
52 restrictions and is useful to protect e.g. state switching which would
53 cause malfunction if instrumented.
54
55 All non-instrumentable entry/exit code sections before and after the RCU
56 state transitions must run with interrupts disabled.
57
58 Syscalls
59 --------
60
61 Syscall-entry code starts in assembly code and calls out into low-level C code
62 after establishing low-level architecture-specific state and stack frames. This
63 low-level C code must not be instrumented. A typical syscall handling function
64 invoked from low-level assembly code looks like this:
65
66 .. code-block:: c
67
68 noinstr void syscall(struct pt_regs *regs, int nr)
69 {
70 arch_syscall_enter(regs);
71 nr = syscall_enter_from_user_mode(regs, nr);
72
73 instrumentation_begin();
74 if (!invoke_syscall(regs, nr) && nr != -1)
75 result_reg(regs) = __sys_ni_syscall(regs);
76 instrumentation_end();
77
78 syscall_exit_to_user_mode(regs);
79 }
80
81 syscall_enter_from_user_mode() first invokes enter_from_user_mode() which
82 establishes state in the following order:
83
84 * Lockdep
85 * RCU / Context tracking
86 * Tracing
87
88 and then invokes the various entry work functions like ptrace, seccomp, audit,
89 syscall tracing, etc. After all that is done, the instrumentable invoke_syscall
90 function can be invoked. The instrumentable code section then ends, after which
91 syscall_exit_to_user_mode() is invoked.
92
93 syscall_exit_to_user_mode() handles all work which needs to be done before
94 returning to user space like tracing, audit, signals, task work etc. After
95 that it invokes exit_to_user_mode() which again handles the state
96 transition in the reverse order:
97
98 * Tracing
99 * RCU / Context tracking
100 * Lockdep
101
102 syscall_enter_from_user_mode() and syscall_exit_to_user_mode() are also
103 available as fine grained subfunctions in cases where the architecture code
104 has to do extra work between the various steps. In such cases it has to
105 ensure that enter_from_user_mode() is called first on entry and
106 exit_to_user_mode() is called last on exit.
107
108 Do not nest syscalls. Nested systcalls will cause RCU and/or context tracking
109 to print a warning.
110
111 KVM
112 ---
113
114 Entering or exiting guest mode is very similar to syscalls. From the host
115 kernel point of view the CPU goes off into user space when entering the
116 guest and returns to the kernel on exit.
117
118 kvm_guest_enter_irqoff() is a KVM-specific variant of exit_to_user_mode()
119 and kvm_guest_exit_irqoff() is the KVM variant of enter_from_user_mode().
120 The state operations have the same ordering.
121
122 Task work handling is done separately for guest at the boundary of the
123 vcpu_run() loop via xfer_to_guest_mode_handle_work() which is a subset of
124 the work handled on return to user space.
125
126 Do not nest KVM entry/exit transitions because doing so is nonsensical.
127
128 Interrupts and regular exceptions
129 ---------------------------------
130
131 Interrupts entry and exit handling is slightly more complex than syscalls
132 and KVM transitions.
133
134 If an interrupt is raised while the CPU executes in user space, the entry
135 and exit handling is exactly the same as for syscalls.
136
137 If the interrupt is raised while the CPU executes in kernel space the entry and
138 exit handling is slightly different. RCU state is only updated when the
139 interrupt is raised in the context of the CPU's idle task. Otherwise, RCU will
140 already be watching. Lockdep and tracing have to be updated unconditionally.
141
142 irqentry_enter() and irqentry_exit() provide the implementation for this.
143
144 The architecture-specific part looks similar to syscall handling:
145
146 .. code-block:: c
147
148 noinstr void interrupt(struct pt_regs *regs, int nr)
149 {
150 arch_interrupt_enter(regs);
151 state = irqentry_enter(regs);
152
153 instrumentation_begin();
154
155 irq_enter_rcu();
156 invoke_irq_handler(regs, nr);
157 irq_exit_rcu();
158
159 instrumentation_end();
160
161 irqentry_exit(regs, state);
162 }
163
164 Note that the invocation of the actual interrupt handler is within a
165 irq_enter_rcu() and irq_exit_rcu() pair.
166
167 irq_enter_rcu() updates the preemption count which makes in_hardirq()
168 return true, handles NOHZ tick state and interrupt time accounting. This
169 means that up to the point where irq_enter_rcu() is invoked in_hardirq()
170 returns false.
171
172 irq_exit_rcu() handles interrupt time accounting, undoes the preemption
173 count update and eventually handles soft interrupts and NOHZ tick state.
174
175 In theory, the preemption count could be updated in irqentry_enter(). In
176 practice, deferring this update to irq_enter_rcu() allows the preemption-count
177 code to be traced, while also maintaining symmetry with irq_exit_rcu() and
178 irqentry_exit(), which are described in the next paragraph. The only downside
179 is that the early entry code up to irq_enter_rcu() must be aware that the
180 preemption count has not yet been updated with the HARDIRQ_OFFSET state.
181
182 Note that irq_exit_rcu() must remove HARDIRQ_OFFSET from the preemption count
183 before it handles soft interrupts, whose handlers must run in BH context rather
184 than irq-disabled context. In addition, irqentry_exit() might schedule, which
185 also requires that HARDIRQ_OFFSET has been removed from the preemption count.
186
187 Even though interrupt handlers are expected to run with local interrupts
188 disabled, interrupt nesting is common from an entry/exit perspective. For
189 example, softirq handling happens within an irqentry_{enter,exit}() block with
190 local interrupts enabled. Also, although uncommon, nothing prevents an
191 interrupt handler from re-enabling interrupts.
192
193 Interrupt entry/exit code doesn't strictly need to handle reentrancy, since it
194 runs with local interrupts disabled. But NMIs can happen anytime, and a lot of
195 the entry code is shared between the two.
196
197 NMI and NMI-like exceptions
198 ---------------------------
199
200 NMIs and NMI-like exceptions (machine checks, double faults, debug
201 interrupts, etc.) can hit any context and must be extra careful with
202 the state.
203
204 State changes for debug exceptions and machine-check exceptions depend on
205 whether these exceptions happened in user-space (breakpoints or watchpoints) or
206 in kernel mode (code patching). From user-space, they are treated like
207 interrupts, while from kernel mode they are treated like NMIs.
208
209 NMIs and other NMI-like exceptions handle state transitions without
210 distinguishing between user-mode and kernel-mode origin.
211
212 The state update on entry is handled in irqentry_nmi_enter() which updates
213 state in the following order:
214
215 * Preemption counter
216 * Lockdep
217 * RCU / Context tracking
218 * Tracing
219
220 The exit counterpart irqentry_nmi_exit() does the reverse operation in the
221 reverse order.
222
223 Note that the update of the preemption counter has to be the first
224 operation on enter and the last operation on exit. The reason is that both
225 lockdep and RCU rely on in_nmi() returning true in this case. The
226 preemption count modification in the NMI entry/exit case must not be
227 traced.
228
229 Architecture-specific code looks like this:
230
231 .. code-block:: c
232
233 noinstr void nmi(struct pt_regs *regs)
234 {
235 arch_nmi_enter(regs);
236 state = irqentry_nmi_enter(regs);
237
238 instrumentation_begin();
239 nmi_handler(regs);
240 instrumentation_end();
241
242 irqentry_nmi_exit(regs);
243 }
244
245 and for e.g. a debug exception it can look like this:
246
247 .. code-block:: c
248
249 noinstr void debug(struct pt_regs *regs)
250 {
251 arch_nmi_enter(regs);
252
253 debug_regs = save_debug_regs();
254
255 if (user_mode(regs)) {
256 state = irqentry_enter(regs);
257
258 instrumentation_begin();
259 user_mode_debug_handler(regs, debug_regs);
260 instrumentation_end();
261
262 irqentry_exit(regs, state);
263 } else {
264 state = irqentry_nmi_enter(regs);
265
266 instrumentation_begin();
267 kernel_mode_debug_handler(regs, debug_regs);
268 instrumentation_end();
269
270 irqentry_nmi_exit(regs, state);
271 }
272 }
273
274 There is no combined irqentry_nmi_if_kernel() function available as the
275 above cannot be handled in an exception-agnostic way.
276
277 NMIs can happen in any context. For example, an NMI-like exception triggered
278 while handling an NMI. So NMI entry code has to be reentrant and state updates
279 need to handle nesting.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.