1=======================
2IRQ-flags state tracing
3=======================
4
5:Author: started by Ingo Molnar <mingo@redhat.com>
6
7The "irq-flags tracing" feature "traces" hardirq and softirq state, in
8that it gives interested subsystems an opportunity to be notified of
9every hardirqs-off/hardirqs-on, softirqs-off/softirqs-on event that
10happens in the kernel.
11
12CONFIG_TRACE_IRQFLAGS_SUPPORT is needed for CONFIG_PROVE_SPIN_LOCKING
13and CONFIG_PROVE_RW_LOCKING to be offered by the generic lock debugging
14code. Otherwise only CONFIG_PROVE_MUTEX_LOCKING and
15CONFIG_PROVE_RWSEM_LOCKING will be offered on an architecture - these
16are locking APIs that are not used in IRQ context. (the one exception
17for rwsems is worked around)
18
19Architecture support for this is certainly not in the "trivial"
20category, because lots of lowlevel assembly code deal with irq-flags
21state changes. But an architecture can be irq-flags-tracing enabled in a
22rather straightforward and risk-free manner.
23
24Architectures that want to support this need to do a couple of
25code-organizational changes first:
26
27- add and enable TRACE_IRQFLAGS_SUPPORT in their arch level Kconfig file
28
29and then a couple of functional changes are needed as well to implement
30irq-flags-tracing support:
31
32- in lowlevel entry code add (build-conditional) calls to the
33  trace_hardirqs_off()/trace_hardirqs_on() functions. The lock validator
34  closely guards whether the 'real' irq-flags matches the 'virtual'
35  irq-flags state, and complains loudly (and turns itself off) if the
36  two do not match. Usually most of the time for arch support for
37  irq-flags-tracing is spent in this state: look at the lockdep
38  complaint, try to figure out the assembly code we did not cover yet,
39  fix and repeat. Once the system has booted up and works without a
40  lockdep complaint in the irq-flags-tracing functions arch support is
41  complete.
42- if the architecture has non-maskable interrupts then those need to be
43  excluded from the irq-tracing [and lock validation] mechanism via
44  lockdep_off()/lockdep_on().
45
46In general there is no risk from having an incomplete irq-flags-tracing
47implementation in an architecture: lockdep will detect that and will
48turn itself off. I.e. the lock validator will still be reliable. There
49should be no crashes due to irq-tracing bugs. (except if the assembly
50changes break other code by modifying conditions or registers that
51shouldn't be)
52
53