OK, I ended with these changes after changing my mind few times and hopefully better understanding the aspects of preemption, interrupts, locking, and scheduling.
1st patch fixes the race in ring_buffer_swap_cpu(). It is the one from previous versions. I have only added checks for irqs_disabled() as suggested by Steven. 2nd patch should be easy. It only makes sure that we are on the same CPU when handling the critical timings. 3rd patch was the most problematic for me. It creates two variants of smp_call_function_single() that could be called either with or without disabled interrupts. It is what Steven suggested but I wonder if it could be done more elegant way. I am in somewhat strange mode after the twins were born on Thursday. Petr Mladek (3): ring-buffer: Race when writing and swapping cpu buffer in parallel trace: process entire stop_critical_timing() on the same CPU trace: Allow to call update_max_tr_single() from another CPU kernel/trace/ring_buffer.c | 115 ++++++++++++++++++++++++++++++------------- kernel/trace/trace.c | 60 +++++++++++++++++++--- kernel/trace/trace.h | 1 + kernel/trace/trace_irqsoff.c | 17 +++++-- 4 files changed, 146 insertions(+), 47 deletions(-) -- 1.8.4 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
