Summary: | [igb] [panic] freebsd 8.4, 9.1 and 9.2 panic Double-Fault with intel 82576 igb driver | ||
---|---|---|---|
Product: | Base System | Reporter: | napTu |
Component: | kern | Assignee: | freebsd-net (Nobody) <net> |
Status: | Closed DUPLICATE | ||
Severity: | Affects Only Me | CC: | eugen, ports, sbruno, shurd |
Priority: | Normal | Keywords: | IntelNetworking, crash |
Version: | Unspecified | ||
Hardware: | Any | ||
OS: | Any |
Description
napTu
2014-01-02 22:40:00 UTC
Responsible Changed From-To: freebsd-bugs->freebsd-net Over to maintainer. (In reply to napTu from comment #0) crash text URL's no longer resolve. The doublefault seems to indicate that something is happening with memory access causing the cpu's to crash. Does this still happen on 10.1 and 10.2r beta? New url to access to the dumps is http://icenet.net.ru/ ... (In reply to napTu from comment #3) Doesn't seem to be resolvable. Couldn't access the site and crashdumps. I`m sorry, access was filtered, try now please. FreeBSD 9.1-RELEASE and 9.2-RELEASE KDB backtraces and info files: http://icenet.net.ru/crash/core.txt.1 http://icenet.net.ru/crash/core.txt.2 http://icenet.net.ru/crash/core.txt.4 http://icenet.net.ru/crash/core.txt.5 http://icenet.net.ru/crash/core.txt.6 http://icenet.net.ru/crash/core.txt.7 http://icenet.net.ru/crash/core.txt.8 http://icenet.net.ru/crash/core.txt.10 http://icenet.net.ru/crash/info.1 http://icenet.net.ru/crash/info.2 http://icenet.net.ru/crash/info.3 http://icenet.net.ru/crash/info.4 http://icenet.net.ru/crash/info.5 http://icenet.net.ru/crash/info.6 http://icenet.net.ru/crash/info.7 http://icenet.net.ru/crash/info.8 http://icenet.net.ru/crash/info.10 FreeBSD 8.4 crash: http://icenet.net.ru/crash/vmcore.0 This may be a 32bit specific failure case here. We'll have to try and build up an i386 system and test. This doesn't happen on any of my amd64 hosts. (In reply to napTu from comment #5) These crashes do not seem igb-specific for me. It seems, you run something like mpd daemon serving lots of users with NAT and Netflow processing, do you? If so, have you applied some non-default system tuning? Specifically, what is your sysctl net.isr.dispatch setting for this router? It is generally not recommended to run busy mpd servers with 32 bit FreeBSD version due to stability problems even if a router has less that 4GB RAM. Have you read my notes on the topic (in Russian) http://dadv.livejournal.com/137221.html ? Sean Bruno, yes, possible 32 bit specific problem. Now, one of hosts work with FreeBSD10.2 amd64 and I can try again igb netcard. Eugen, FreeBSD8.4 host use only routing, without mpd, without NAT, WITH NETFLOW. FreeBSD9.x host use routing with NAT and NetFlow, without mpd also. net.isr.dispatch: direct and simple tuning: rc.local:sysctl net.inet.ip.fw.one_pass=0 rc.local:sysctl net.inet.tcp.blackhole=2 rc.local:sysctl net.inet.udp.blackhole=1 rc.local:sysctl kern.ipc.nmbclusters=32768 loader.conf: ipfw_load="YES" ipfw_nat_load="YES" ipdivert_load="YES" libalias_load="YES" dummynet_load="YES" 8.4 host modules: # kldstat Id Refs Address Size Name 1 37 0x80400000 bb5f84 kernel 2 1 0x80fb6000 50ea0 if_em.ko 3 1 0x86490000 3000 ng_eiface.ko 4 9 0x86493000 b000 netgraph.ko 5 1 0x86523000 4000 ng_bridge.ko 6 1 0x86532000 4000 ng_ether.ko 7 1 0x86536000 7000 ng_netflow.ko 8 1 0x8653d000 5000 ng_ksocket.ko 9 1 0x865c2000 4000 ng_car.ko 10 1 0x865c6000 2000 ng_ipfw.ko 11 1 0x87af7000 4000 ng_socket.ko 12 1 0x87b7e000 3000 ng_tee.ko 13 1 0x93416000 1d000 nfsclient.ko 14 1 0x93433000 3000 nfslock.ko 15 1 0x93436000 2000 nfs_common.ko 16 1 0x93438000 13000 krpc.ko 9.x host modules # kldstat Id Refs Address Size Name 1 39 0xffffffff80200000 179ddb0 kernel 2 1 0xffffffff8199e000 2f9b00 zfs.ko 3 2 0xffffffff81c98000 6048 opensolaris.ko 4 4 0xffffffff81c9f000 21a60 ipfw.ko 5 1 0xffffffff81cc1000 7a40 ipfw_nat.ko 6 2 0xffffffff81cc9000 16f70 libalias.ko 7 1 0xffffffff81ce0000 8578 ipdivert.ko 8 1 0xffffffff81ce9000 19a80 dummynet.ko 9 1 0xffffffff81e11000 7430 ng_netflow.ko 10 6 0xffffffff81e19000 b9db netgraph.ko 11 1 0xffffffff81e25000 4ae1 ng_ksocket.ko 12 1 0xffffffff81e2a000 40fa ng_ether.ko 13 1 0xffffffff81e2f000 3c48 ng_eiface.ko 14 1 0xffffffff81e33000 3832 ng_socket.ko 15 1 0xffffffff81e37000 17d7 ng_tee.ko (In reply to napTu from comment #8) There seem to be some problems with multithreading and/or locks within 9.x kernel/ng_nat code. 82576 just adds extra parallelism as it supports MSI-X and multiple receive interrupts and increase probability of races within NETGRAPH code. 82571 supports only single-threaded receive path. Why do you use ng_nat if you do not run mpd? Try to switch from ng_nat to ipfw nat and eliminate netgraph to simplify code path. Your 8.x crash link is broken (404). I install igb card to FreeBsd10.2 amd64. Uptime is 55hours. Crash links must be changed from host icenet.net.ua to new host name icenet.net.ru Eugen, I do not use ng_nat, but use ipfw_nat. Netgraph is used only to collect netflow data. (In reply to napTu from comment #10) yes, link to FreeBSD 8.4 crash: http://icenet.net.ru/crash/vmcore.0 is broken batch change: For bugs that match the following - Status Is In progress AND - Untouched since 2018-01-01. AND - Affects Base System OR Documentation DO: Reset to open status. Note: I did a quick pass but if you are getting this email it might be worthwhile to double check to see if this bug ought to be closed. i386 system, long back traces and double faults show this is duplicate of the PR 219476. *** This bug has been marked as a duplicate of bug 219476 *** |