irq: detect long-running IRQ handlers
If a hard IRQ handler takes a long time to handle an IRQ, it may cause a
soft lockup or RCU stall, but as this will be detected once the handler
has returned it can be difficult to attribute the delay to the specific
IRQ handler.
It's possible to trace IRQ handlers to diagnose this, but that's not a
great fit for automated testing environments (e.g. fuzzers), where
something like the existing lockup/stall detectors works well.
This patch adds a new stall detector for IRQ handlers, which reports
when handlers took longer than a given timeout value (defaulting to 1
second). This won't detect hung IRQ handlers (which requires an NMI, and
should already be caught by hung task detection on systems with NMIs),
but helps on platforms without NMI or where a periodic watchdog is
undesireable.
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Marc Zyngier <maz@kernel.org>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
2 files changed
