>my major gripe right now is that we still have bug reports that say that
>systems hang when using nmi_watchdog=1 and work if nmi_watchdog=0.
>Changing the NMI watchdog to be 1 Hz will make these bugreports "Linux
>hangs once a week" instead of a "Linux hangs after 1-2 hours", which is
>clearly hiding things and making debugging harder.
All reports I've seen have been for SMP kernels on MP machines using
the IO-APIC to drive the watchdog. Are you saying there are also cases
where UP boxes fail with nmi_watchdog non-zero? (My 1Hz change only
affected the local APIC-driven watchdog which MP boxes normally don't use.)
>(and driving kernel-profiling from the NMI interrupt is a short-term
>patch, so there is just no point in going to 1 Hz right now just to go
>back to 100 Hz a few days later.)
Another "constructive" use of the perfctrs. Ok, this I can see wants
a higher rate.
How far in the future is this? I'm concerned that the conflicting
uses of the perfctrs (watchdog, kernel profiling, my perfctr driver
for user-space performance measurements) is going to require some
low-level request/release API.
>the rest of the changes are excellent - it's only the 100 Hz NMI issue i
>have a problem with.
Ok.
Alan beat me to it for -ac20, so I'm not including a new patch now
with the 1Hz bit backed out. Ingo, I guess this means the kernel
profiling patch will have to "fix" the 1Hz thing by itself.
/Mikael
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/