Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 2 Aug 2005 21:18:06 -0400
From:      Ed Maste <emaste@phaedrus.sandvine.ca>
To:        Robert Watson <rwatson@FreeBSD.org>
Cc:        freebsd-net@freebsd.org, gnn@FreeBSD.org
Subject:   Re: what to replace splnet in FreeBSD 5.x?
Message-ID:  <20050803011806.GA32338@sandvine.com>
In-Reply-To: <20050801190211.Q61500@fledge.watson.org>
References:  <42CEF0EB.4000107@borderware.com> <42D006DB.8080108@errno.com> <20050712150224.GA38249@sandvine.com> <20050712162332.Q79478@fledge.watson.org> <20050712220452.GB38249@sandvine.com> <20050723012046.K40216@fledge.watson.org> <20050723013902.L40216@fledge.watson.org> <20050801190211.Q61500@fledge.watson.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Aug 01, 2005 at 07:16:58PM +0100, Robert Watson wrote:

> >Updated version of the patch at:
> >
> >http://www.watson.org/~robert/freebsd/netperf/20050523-multicast.diff
> >
> >Spl-related cleanups (no longer useful documentation of old 
> >synchronization in many places), lock order hard-coded into WITNESS 
> >order rather than dynamically detected.
> 
> Ed,
> 
> Have you had a chance to run with this patch as yet?  If possible, I'd 
> like to get it merged to 7.x-CURRENT so we can backport it for 6.0.

I finally got our local CVS repository back into shape and gave this
patch a try.  Unfortunately got a panic as soon as our multicast app
started up:

panic: _mtx_lock_sleep: recursed on non-recursive mutex in_multi_mtx @ /d2/emaste/cvs_mcast/src/sys/netinet/ip_output.c:298

cpuid = 0
KDB: stack backtrace:
kdb_backtrace(100,a2894300,a07eba00,a07eba00,12a) at 0xa0586915 = kdb_backtrace+0x29
panic(a071a883,a071f26e,a0729680,12a,a07eba00) at 0xa056eeb8 = panic+0x114
_mtx_lock_sleep(a07eba00,a2894300,0,a0729680,12a) at 0xa05669cb = _mtx_lock_sleep+0x33
_mtx_lock_flags(a07eba00,0,a0729680,12a,0) at 0xa05664ba = _mtx_lock_flags+0x8a
ip_output(a2645e00,a2649100,a07eb9b0,0,c8635b08) at 0xa05f7b0e = ip_output+0x3fe
igmp_sendpkt(a24b9de0,16,0,a24b9de0,c8635bb8) at 0xa05e292e = igmp_sendpkt+0x106
igmp_joingroup(a24b9de0,a2504240,210,240600ef,0) at 0xa05e2642 = igmp_joingroup+0x92
in_addmulti(c8635bcc,a23a1400,a23a1400,240600ef,1000001) at 0xa05e437e = in_addmulti+0x10e
ip_setmoptions(a272c7f8,c8635c90,a071a701,12c,0) at 0xa05f9278 = ip_setmoptions+0x3a0
ip_ctloutput(a27c2650,c8635c90,a283a22c,0,a0722810) at 0xa05f8a57 = ip_ctloutput+0x36b
sosetopt(a27c2650,c8635c90,a26af798,1,0) at 0xa05a91ac = sosetopt+0x2c
kern_setsockopt(a2894300,b,0,c,9fbfcd60) at 0xa05ada71 = kern_setsockopt+0xb5
setsockopt(a2894300,c8635d04,5,3,246) at 0xa05ad9ba = setsockopt+0x1e
syscall(81e003b,81e003b,9fbf003b,9fbfcd70,841fc28) at 0xa06c72fb = syscall+0x25b

in_addmulti holds the in_multi_mtx when it calls _joingroup.

I haven't yet looked over the whole patch to understand the
specifics.  I'll post again once I do if I have any ideas.

--
Ed Maste, Sandvine Incorporated.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050803011806.GA32338>