3 Using RCU to Protect Dynamic NMI Handlers
4 =========================================
7 Although RCU is usually used to protect read-mostly data structures,
8 it is possible to use RCU to provide dynamic non-maskable interrupt
9 handlers, as well as dynamic irq handlers. This document describes
10 how to do this, drawing loosely from Zwane Mwaikambo's NMI-timer
11 work in "arch/x86/oprofile/nmi_timer_int.c" and in
12 "arch/x86/kernel/traps.c".
14 The relevant pieces of code are listed below, each followed by a
17 static int dummy_nmi_callback(struct pt_regs *regs, int cpu)
22 The dummy_nmi_callback() function is a "dummy" NMI handler that does
23 nothing, but returns zero, thus saying that it did nothing, allowing
24 the NMI handler to take the default machine-specific action::
26 static nmi_callback_t nmi_callback = dummy_nmi_callback;
28 This nmi_callback variable is a global function pointer to the current
31 void do_nmi(struct pt_regs * regs, long error_code)
37 cpu = smp_processor_id();
40 if (!rcu_dereference_sched(nmi_callback)(regs, cpu))
46 The do_nmi() function processes each NMI. It first disables preemption
47 in the same way that a hardware irq would, then increments the per-CPU
48 count of NMIs. It then invokes the NMI handler stored in the nmi_callback
49 function pointer. If this handler returns zero, do_nmi() invokes the
50 default_do_nmi() function to handle a machine-specific NMI. Finally,
51 preemption is restored.
53 In theory, rcu_dereference_sched() is not needed, since this code runs
54 only on i386, which in theory does not need rcu_dereference_sched()
55 anyway. However, in practice it is a good documentation aid, particularly
56 for anyone attempting to do something similar on Alpha or on systems
57 with aggressive optimizing compilers.
60 Why might the rcu_dereference_sched() be necessary on Alpha, given that the code referenced by the pointer is read-only?
62 :ref:`Answer to Quick Quiz <answer_quick_quiz_NMI>`
64 Back to the discussion of NMI and RCU::
66 void set_nmi_callback(nmi_callback_t callback)
68 rcu_assign_pointer(nmi_callback, callback);
71 The set_nmi_callback() function registers an NMI handler. Note that any
72 data that is to be used by the callback must be initialized up -before-
73 the call to set_nmi_callback(). On architectures that do not order
74 writes, the rcu_assign_pointer() ensures that the NMI handler sees the
77 void unset_nmi_callback(void)
79 rcu_assign_pointer(nmi_callback, dummy_nmi_callback);
82 This function unregisters an NMI handler, restoring the original
83 dummy_nmi_handler(). However, there may well be an NMI handler
84 currently executing on some other CPU. We therefore cannot free
85 up any data structures used by the old NMI handler until execution
86 of it completes on all other CPUs.
88 One way to accomplish this is via synchronize_rcu(), perhaps as
95 This works because (as of v4.20) synchronize_rcu() blocks until all
96 CPUs complete any preemption-disabled segments of code that they were
98 Since NMI handlers disable preemption, synchronize_rcu() is guaranteed
99 not to return until all ongoing NMI handlers exit. It is therefore safe
100 to free up the handler's data as soon as synchronize_rcu() returns.
102 Important note: for this to work, the architecture in question must
103 invoke nmi_enter() and nmi_exit() on NMI entry and exit, respectively.
105 .. _answer_quick_quiz_NMI:
107 Answer to Quick Quiz:
108 Why might the rcu_dereference_sched() be necessary on Alpha, given that the code referenced by the pointer is read-only?
110 The caller to set_nmi_callback() might well have
111 initialized some data that is to be used by the new NMI
112 handler. In this case, the rcu_dereference_sched() would
113 be needed, because otherwise a CPU that received an NMI
114 just after the new handler was set might see the pointer
115 to the new NMI handler, but the old pre-initialized
116 version of the handler's data.
118 This same sad story can happen on other CPUs when using
119 a compiler with aggressive pointer-value speculation
122 More important, the rcu_dereference_sched() makes it
123 clear to someone reading the code that the pointer is
124 being protected by RCU-sched.