[Crash-utility] netdump netpoll bug?

[Rohan Mutagi]

Excellent Jeff!! You are a life saver!
I couldn't find a quick way to make Vmware change it drivers, however,
adding arp entry to my server worked! Now I need to figure out a way
to make this working on a more permanent basis.
thanks once again.
Rohan Mutagi

On 5/15/06, Jeff Moyer <jmoyer at redhat.com> wrote:
> ==> Regarding Re: [Crash-utility] netdump netpoll bug?; "Rohan Mutagi" <rohan208 at gmail.com> adds:
>
> rohan208> My kernel version is 2.6.9-34.EL (my system is "up to date" from
> rohan208> RHN)... I am running this inside VMWare and the eth0 is seen as
> rohan208> "PCnet/PCI II 79C97OA"
>
> rohan208> I am wondering how do I provide the stack trace? As its on the
> rohan208> screen of my laptop. This trace doesnt apper in the "log" file
> rohan208> generated by the netdump-server.
>
> Typically, I would recommend a serial console.  I guess that's not possible
> with VMWare, though.  ;)
>
> rohan208> Any leads would be appricaited.
>
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=142269
>
> That bug seems to be private... *sigh*  Here is the stack trace that you
> are likely getting:
>
> Call Trace:
>  [<f882604d>] pcnet32_start_xmit+0x23/0xf8 [pcnet32]
>  [<c0274c78>] netpoll_send_skb+0x6a/0x98
>  [<c02751d6>] arp_reply+0x2f7/0x2ff
>  [<c027521f>] netpoll_rx+0x41/0x2e4
>  [<c026a5c0>] netif_rx+0x1c/0x1b2
>  [<f88267e8>] pcnet32_rx+0x2c8/0x325 [pcnet32]
>  [<f88261ed>] pcnet32_interrupt+0xcb/0x3fe [pcnet32]
>  [<f8824132>] pcnet32_poll_controller+0x16/0x1f [pcnet32]
>  [<c0274aa2>] netpoll_poll+0x32/0x8f
>  [<f89ef69d>] netpoll_netdump+0xa3/0x478 [netdump]
>  [<c01f9a96>] sysrq_handle_crash+0x0/0x8
>  [<c01f9a96>] sysrq_handle_crash+0x0/0x8
>  [<f89ef5fa>] netpoll_netdump+0x0/0x478 [netdump]
>  [<f89ef5f1>] netpoll_start_netdump+0xf1/0xfa [netdump]
>  =======================
>  ...
>
> Basically, the network transmit queue is busy, and so we call the
> poll_controller routine to free up some TX descriptors.  If, in the
> process, we end up with ARP requests, we need to service them.  This ends
> up in a call to the hard_start_xmit routine inside the pcnet32 driver, and
> deadlocks.
>
> Is there possibly another driver that VMWare will mimic?  If not, you can
> circumvent this problem by adding an ARP entry for the crashing system to
> the netdump server.
>
> -Jeff
>
> --
> Crash-utility mailing list
> Crash-utility at redhat.com
> https://www.redhat.com/mailman/listinfo/crash-utility
>