| To: | netdev@xxxxxxxxxxx |
|---|---|
| Subject: | High interrupt rate on tg3 driver |
| From: | Steve Modica <modica@xxxxxxx> |
| Date: | Wed, 05 Nov 2003 14:41:46 -0600 |
| Organization: | SGI |
| Sender: | netdev-bounce@xxxxxxxxxxx |
| User-agent: | Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4b) Gecko/20030425 |
Hi All, It turns out that there's a hardware bug that can cause a missed interrupt with the 570X chipset. So the tg3 driver will poke the card every 10msec to make sure we don't miss one. Each of these pokes generates an interrupt. However this still doesn't explain the high interrupt rate I'm seeing. I'm seeing about 22k/sec when moving 500kBytes/sec of data. I've got 5 cards ports on the system, so I can account for 500/sec, and I could even see each NAPI interrupt/poll cycle generating a few hundred a sec, but 22000 seems like way too many. Can anyone think of a real good way to capture this? Profiling just shows me spending a lot of time in tg3_poll, do_csum and memcpy. Steve
|
| <Prev in Thread] | Current Thread | [Next in Thread> |
|---|---|---|
| ||
| Previous by Date: | Re: [PATCH] IPv6: Flowinfo fix in ip6_tunnel.c, David S. Miller |
|---|---|
| Next by Date: | Re: [PATCH] IPv6: Autoconfig link-local address on ip6-ip6 tunnel device, David S. Miller |
| Previous by Thread: | rp_filter dropping things it shouldn't, Brian White |
| Next by Thread: | Re: Oops in __xfrm4_state_lookup when setting up an IPSEC tunnel (fwd), James Morris |
| Indexes: | [Date] [Thread] [Top] [All Lists] |