Re: Pin-pointing the root of unusual application latencies

[John Sigler <linux.kernel@free.fr>]

John Sigler wrote:

> ( check_dektec_in-1095 |#0): new 271 us user-latency.
> ( check_dektec_in-1095 |#0): new 275 us user-latency.
> ( check_dektec_in-1095 |#0): new 290 us user-latency.
> ( check_dektec_in-1095 |#0): new 297 us user-latency.
> ( check_dektec_in-1095 |#0): new 345 us user-latency.
> ( check_dektec_in-1095 |#0): new 358 us user-latency.
> ( check_dektec_in-1095 |#0): new 384 us user-latency.
> ( check_dektec_in-1095 |#0): new 392 us user-latency.
> ( check_dektec_in-1095 |#0): new 395 us user-latency.
> ( check_dektec_in-1095 |#0): new 396 us user-latency.
> ( check_dektec_in-1095 |#0): new 1031 us user-latency.
> ( check_dektec_in-1095 |#0): new 1100 us user-latency.
> ( check_dektec_in-1095 |#0): new 1105 us user-latency.
> ( check_dektec_in-1095 |#0): new 1106 us user-latency.
>
> Here's the function trace for the 1106-µs latency:
>
> http://linux.kernel.free.fr/latency/1106-us-trace.txt

The function trace for 400-µs latencies is different:

( check_dektec_in-1145 |#0): new 275 us user-latency.
( check_dektec_in-1145 |#0): new 276 us user-latency.
( check_dektec_in-1145 |#0): new 288 us user-latency.
( check_dektec_in-1145 |#0): new 289 us user-latency.
( check_dektec_in-1145 |#0): new 289 us user-latency.
( check_dektec_in-1145 |#0): new 290 us user-latency.
( check_dektec_in-1145 |#0): new 297 us user-latency.
( check_dektec_in-1145 |#0): new 345 us user-latency.
( check_dektec_in-1145 |#0): new 354 us user-latency.
( check_dektec_in-1145 |#0): new 377 us user-latency.
( check_dektec_in-1145 |#0): new 393 us user-latency.

http://linux.kernel.free.fr/latency/393-us-trace.txt

There are ~200 calls to ioread32 from mdio_read from speedo_timer.

http://lxr.linux.no/source/drivers/net/eepro100.c#L1159
http://lxr.linux.no/source/drivers/net/eepro100.c#L928

In this case, and as far as I understand, the culprit is the eepro100
driver talking to one of the NICs (which one?). Is that correct?

What is the consequence of IRQ10 being shared by eth2 and
by my I/O board?

How can I force Linux to assign different IRQs to every peripheral
if I have free IRQs lines?

Regards.
-
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/