Recent SUP. Installed on Dell PowerEdge 4600. Getting tons of: Sep 30 12:23:16 zige /kernel: bge0: gigabit link up Sep 30 12:23:46 zige last message repeated 98 times Also getting some TCP retransmits. This causes noticable delays in pretty much everything I do now. Not a problem prior to the SUP. I was on 4.8-STABLE before. When pulling/replacing cables, it went into a mode where it would drop link, renegotiate, drop link, reneg, ... for a few minutes and then it stopped. Had to power cycle machine to recover.
Damon Anton Permezel:>Recent SUP. >Installed on Dell PowerEdge 4600. >Getting tons of: > >Sep 30 12:23:16 zige /kernel: bge0: gigabit link up >Sep 30 12:23:46 zige last message repeated 98 times > > >Also getting some TCP retransmits. This causes noticable delays in >pretty much everything I do now. Not a problem prior to the SUP. >I was on 4.8-STABLE before. > >When pulling/replacing cables, it went into a mode where it would >drop link, renegotiate, drop link, reneg, ... for a few minutes and >then it stopped. Had to power cycle machine to recover.I am using a Dell 2650 whose bge identifies as: bge0: <Broadcom BCM5701 Gigabit Ethernet, ASIC rev. 0x105> mem 0xfcf10000-0xfcf1ffff irq 11 at device 6.0 on pci3 bge1: <Broadcom BCM5701 Gigabit Ethernet, ASIC rev. 0x105> mem 0xfcf00000-0xfcf0ffff irq 13 at device 8.0 on pci3 I acknowledge that I have seen many of the "gigabit link up" messages formerly, but they disappeared some weeks ago. They definitely don't show up with a recent kernel (built just after the sendmail commits last week). I don't see tcp retransmits either (total of 29, to be precise :-)). Maybe you are still using an older kernel, or there is something broken with your network infrastructure? I am running with the bge's connected to different Cisco switches (4507R and 3550). And yes, this is a heavy traffic system. Helge
On 30-Sep-2003 Damon Anton Permezel wrote:> Recent SUP. > Installed on Dell PowerEdge 4600. > Getting tons of: > > Sep 30 12:23:16 zige /kernel: bge0: gigabit link up > Sep 30 12:23:46 zige last message repeated 98 timesCheck your dmesg output, and I bet you'll see that bge0 is sharing an IRQ with another device. (If not, I'd like to hear about it.) This bug was fixed in -current in revision 1.41 of if_bge.c, but it never got merged to -stable. I have a patch for -stable that I am planning to commit after the source freeze lifts. It is attached to this mail. Please let me know whether it solves your problem or not. John -------------- next part -------------- Index: if_bge.c ==================================================================RCS file: /home/ncvs/src/sys/dev/bge/if_bge.c,v retrieving revision 1.3.2.28 diff -u -r1.3.2.28 if_bge.c --- if_bge.c 26 Sep 2003 16:02:04 -0000 1.3.2.28 +++ if_bge.c 30 Sep 2003 15:15:45 -0000 @@ -2222,11 +2222,12 @@ { struct bge_softc *sc; struct ifnet *ifp; + u_int32_t statusword; u_int32_t status; - sc = xsc; ifp = &sc->arpcom.ac_if; + statusword = loadandclear(&sc->bge_rdata->bge_status_block.bge_status); #ifdef notdef /* Avoid this for now -- checking this register is expensive. */ @@ -2263,13 +2264,7 @@ BRGPHY_INTRS); } } else { - if ((sc->bge_rdata->bge_status_block.bge_status & - BGE_STATFLAG_UPDATED) && - (sc->bge_rdata->bge_status_block.bge_status & - BGE_STATFLAG_LINKSTATE_CHANGED)) { - sc->bge_rdata->bge_status_block.bge_status &- ~(BGE_STATFLAG_UPDATED| - BGE_STATFLAG_LINKSTATE_CHANGED); + if (statusword & BGE_STATFLAG_LINKSTATE_CHANGED) { /* * Sometimes PCS encoding errors are detected in * TBI mode (on fiber NICs), and for some reason