Linus, Andrew Morton mentioned this when he sent you the lost tick detection patch. The new lost-tick detection code can be relatively uninformative when something using the common_interrupt->do_IRQ path keeps interrupts disabled for a long time. This patch will record the last interrupt on the current CPU, and spit it out along with the stack trace. I ignore the timer interrupt, because it is always running when this detection occurs, and would overwrite the previous interrupt that took a long time. You can now turn this on with a boot-time option with the same name as x86-64: "report_lost_ticks". I don't like the default of 5, so I let it take a plain report_lost_ticks, or report_lost_ticks=100, if you want. Warning! Detected 4094446 micro-second gap between interrupts. Compensating for 4093 lost ticks. Call Trace: [] handle_IRQ_event+0x28/0x50 [] do_IRQ+0xa0/0x10c [] common_interrupt+0x43/0x58 Last run common irq: 24: eth2 irq.c | 6 ++++++ time.c | 41 ++++++++++++++++++++++++++++++++--------- 2 files changed, 38 insertions(+), 9 deletions(-) -- Dave Hansen haveblue@us.ibm.com