Re: [E1000-devel] [PATCH] igb: CPU0 latency and updating statistics

On Wed, 2018-08-01 at 07:29 +0000, JABLONSKY Jan wrote:

The Watchdog workqueue in igb driver is scheduled every 2s for each 
network interface. That includes updating a statistics protected by 
spinlock. Function igb_update_stats in this case will be protected 
against preemption. According to number of a statistics registers 
(cca 60), processing this function might cause additional cpu load
 on CPU0.

In case of statistics spinlock may be replaced with mutex, which 
reduce latency on CPU0.


# Measurements
Measurements were performed on Intel Atom E3950 (Quad-Core), 
E3930 (Dual-Core)

# Cyclictest
High latency is always on CPU0 which might be in range 
from 70us to 140us (sometimes higher)


# cyclictest -l1000000 -Smp90 -i200 -q

# E3930
# default
T: 0 ( 1745) P:90 I:200 C:1000000 Min: 2 Act: 3 Avg: 3 Max: 72
T: 1 ( 1746) P:90 I:700 C: 285737 Min: 3 Act: 4 Avg: 3 Max: 13

# with patch
T: 0 (  932) P:90 I:200 C:1000000 Min: 2 Act: 3 Avg: 3 Max: 22
T: 1 (  933) P:90 I:700 C: 285736 Min: 2 Act: 3 Avg: 3 Max: 12


# E3950
# default
T: 0 ( 1103) P:90 I:200 C:1000000 Min: 2 Act: 2 Avg: 2 Max: 89
T: 1 ( 1104) P:90 I:700 C: 285738 Min: 2 Act: 4 Avg: 2 Max: 9
T: 2 ( 1105) P:90 I:1200 C: 166678 Min: 2 Act: 5 Avg: 3 Max: 22
T: 3 ( 1106) P:90 I:1700 C: 117653 Min: 2 Act: 6 Avg: 2 Max: 13

# with patch
T: 0 (  879) P:90 I:200 C:1000000 Min: 2 Act: 3 Avg: 3 Max: 26
T: 1 (  880) P:90 I:700 C: 285729 Min: 3 Act: 4 Avg: 3 Max: 22
T: 2 (  881) P:90 I:1200 C: 166672 Min: 3 Act: 7 Avg: 3 Max: 24
T: 3 (  882) P:90 I:1700 C: 117649 Min: 3 Act: 7 Avg: 3 Max: 8


Additionally I think updating statistics and watchdog task should 
be implemented in 2 separated ways

Patch is created against a linux-4.4

Reviewed also by: Bernhard Kaindl  [off-list ref]

Signed-off-by: Jan Jablonsky <redacted>
---
 drivers/net/ethernet/intel/igb/igb.h         |  2 +-
 drivers/net/ethernet/intel/igb/igb_ethtool.c |  4 ++--
 drivers/net/ethernet/intel/igb/igb_main.c    | 14 +++++++-------
 3 files changed, 10 insertions(+), 10 deletions(-)

NACK, as is.  This patch is malformed, as well as does not apply to the
current net-next tree.

Since the jist of this change is to wrap statistical updates with mutex
lock versus spin locks, I will put together a patch agains the latest
net-next tree for your review Jan.