minor e1000 bug
I've noticed that the e1000 driver does not update the counters in /proc/net/dev as quickly as several other drivers I've tried, such as e100 (both the Becker driver, and Intel's), sk90lin, and 3c59x. These drivers seem to update the counters in a very timely fashion while the e1000 driver doesn't seem to update them for several seconds. This is apparent in 2.6.0, and 2.4.xx. Is there an update interval that might be modified within the driver to fix this? It screws up realtime bandwidth measurements for these cards.
-Ethan
Hi Ethan,
Quoted 6 lines · Ethan Weinstein
I've noticed that the e1000 driver does not update the counters in
/proc/net/devas quickly as several other drivers I've tried, such as e100 (both the Becker driver, and Intel's), sk90lin, and 3c59x. These drivers seem to update the counters in a very timely fashion while the e1000 driver doesn't seem to update them for several seconds. This is apparent in 2.6.0, and 2.4.xx. Is there an update
Every 2 seconds exactly.
I would also be interested in a statement from intel fellows on the reasoning behind this decision, since every user of gkrellm will notice some strange behaviour (value oscillating between 0 and throughput * 2). (Poor man's real time bandwidth management ;-).
After being tired of cognitive interpretation of these values, I decided to fix it, which was pretty easy:
drivers/net/e1000/e1000_main.c 2003-08-08 13:20:06.000000000 +0200 renamed from drivers/net/e1000/e1000_main.c~ 2003-08-03 00:40:21.000000000 +0200 +1 -1
@@ -1390,7 +1390,7 @@ netif_stop_queue(netdev); /* Reset the timer */ - mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ); + mod_timer(&adapter->watchdog_timer, jiffies + HZ); } #define E1000_TX_FLAGS_CSUM 0x00000001
Quoted 5 lines
interval that might be modified within the driver to fix this? It screws up realtime bandwidth measurements for these cards.
-Ethan
Enjoy,
Pete