Re: Occasional wedging entering DDB via serial break

From: Kris Kennaway (kris_at_obsecurity.org)
Date: 12/27/04

  • Next message: Chuck Swiger: "Re: BIND9 performance issues with SMP"
    Date: Sun, 26 Dec 2004 16:47:59 -0800
    To: Robert Watson <rwatson@FreeBSD.org>
    
    
    

    On Sun, Dec 26, 2004 at 10:37:32PM +0000, Robert Watson wrote:
    >
    > I'm seeing occasional wedging attempting to enter DDB via a serial break
    > on a dual-Xeon box (4 logical processors). The symptoms usually look
    > something like this:
    >
    > hippy# ./tmp.csh
    > ~KDB: enter: Line break on console
    > [thread pid 560 tid 100202 ]
    > Stopped at kdb_enter+0x2c: leave
    > db> show alllocks
    > db> cont
    > load: 0.04 cmd: super-smack 619 [runnable] 0.02u 0.51s 2% 2216k
    > ~KDB: enter: Line break on console
    > [thread pid 560 tid 100201 ]
    > Stopped at kdb_enter+0x2c: leave
    > db> show alllocks
    > Process 560 (mysqld)
    > Thread 0x18769
    > exclusive sleep mutex so_rcv r = 0 (0xc2a25bcc) locked @
    > kern/uipc_usrreq.c:464
    > exclusive sleep mutex unp r = 0 (0xc091a740) locked @
    > kern/uipc_usrreq.c:392
    > db> cont
    > load: 0.92 cmd: super-smack 619 [runnable] 0.05u 1.36s 6% 2216k
    > ~KDB: enter: Line break on console
    > [thread pid 560 tid 100199 ]
    > Stopped at kdb_enter+0x2c: leave
    > db> show alllocks
    > db> cont
    > load: 2.12 cmd: super-smack 616 [running] 0.08u 2.90s 10% 2216k
    > ~KDB: enter: Line break on console
    > <wedge>
    >
    > After that point, a serial break will no longer drop to DDB, respond to
    > pings, etc. The box is running 6.x-CURRENT from this morning. Is anyone
    > else seeing this, or does anyone else have ideas about what might cause
    > this?

    I see much the same..it seems to be just one more way in which DDB is
    broken on SMP machines. Setting debug.kdb.stop_cpus=0 tends to fix
    this for me, but then I get the joy of overlapping panics from
    multiple CPUs.

    Kris

    
    



  • Next message: Chuck Swiger: "Re: BIND9 performance issues with SMP"

    Relevant Pages

    • Re: 5.2-BETA lockup
      ... It doesn't panic or drop into DDB. ... > It's currently booted with the older kernel from the 16th and this is ... Any chance you could hook up a serial console, ... Under some circumstances a serial break can be more effective getting into ...
      (freebsd-current)
    • Re: CFT: vr(4)
      ... Robert Watson wrote: ... I was not able to break into DDB. ... It may well be that a serial break doesn't get into DDB for ... After continue, I am able to break ping, but no more packet could be ...
      (freebsd-current)
    • Re: CFT: vr(4)
      ... I was not able to break into DDB. ... serial break doesn't get into DDB for you, ... This time it hangs right after ping, ... So no data to check, actually, only hard hang. ...
      (freebsd-current)
    • Re: CFT: vr(4)
      ... would you break into DDB and show me the output of 'show alllocks' and 'ps'? ... Hang was really hard, ... This is because syscons's interrupt handler acquires the Giant lock in an ithread, requiring a lot more things to be happy to succeed. ... It may well be that a serial break doesn't get into DDB for you, ...
      (freebsd-current)
    • Re: lockup upon resuming from DDB
      ... > I broke to DDB on a healthy quad-CPU system, then did 'cont'. ... > and hung. ...
      (freebsd-current)