[lkml]   [2010]   [Oct]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 0/5] Proper kernel irq time reporting -v0
On 10/20/2010 05:48 PM, Venkatesh Pallipadi wrote:
> This is part 2 of
> "Proper kernel irq time accounting -v4"
> and applies over those changes.

I applied both sets on top of 2.6.36 and tested on x86 32-bit and 64-bit
machines with some telephony cards that interrupt at a 1000Hz rate. The
time reported in top matched the expected value as measured with the
function graph tracer. i.e, if /sys/kernel/debug/tracing/trace
indicated that the interrupt handler was averaging 13us, then top was
reporting 1.3% CPU time. I changed the smp_affinity and the "hi" time
moved around as I would have expected. I also scheduled work items on
all the CPUs that just disable interrupt and spin for a selectable
period. Scheduling 100ms of non-interruptible delay at 1 second
intervals resulted in 10% time in "hi" as expected.

Hopefully I can check it out on a Powermac G3 B&W next week sometime,
but FWIW:

Tested-by: Shaun Ruffell <>

 \ /
  Last update: 2010-10-21 19:31    [W:0.253 / U:1.992 seconds]
©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site