Date:	Mon, 26 Jan 2009 23:58:34 +0000
From:	Ben Dooks <ben-linux@...ff.org>
To:	"Narnakaje, Snehaprabha" <nsnehaprabha@...com>
Cc:	"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [patch 2.6.28-rc8-davinci-git] dm355: dm9000

On Mon, Jan 26, 2009 at 04:36:50PM -0600, Narnakaje, Snehaprabha wrote:
> Hi,
> 
> I am having an issue with the dm9000 driver on a DaVinci DM355 based platform.
> 
> I configure dm9000 to get an IP address through dhcp and then mount via NFS. I get the following dump, just after getting an IP Address and trying to connect to the NFS server.
> 
> ------------[ cut here ]------------
> WARNING: at net/sched/sch_generic.c:226 dev_watchdog+0x188/0x2a0()
> NETDEV WATCHDOG: eth0 (dm9000): transmit timed out
> Modules linked in:
> [<c002c890>] (dump_stack+0x0/0x14) from [<c003b81c>] (warn_slowpath+0x64/0x80)
> [<c003b7b8>] (warn_slowpath+0x0/0x80) from [<c01bf934>] (dev_watchdog+0x188/0x2a0)
>  r3:c6c81800 r2:c02af9a4
>  r6:c02ce000 r5:c6c81800 r4:c030d710
> [<c01bf7ac>] (dev_watchdog+0x0/0x2a0) from [<c00459b4>] (run_timer_softirq+0x1bc/0x278)
> [<c00457f8>] (run_timer_softirq+0x0/0x278) from [<c0040a44>] (__do_softirq+0x74/0x110)
> [<c00409d0>] (__do_softirq+0x0/0x110) from [<c0040e74>] (irq_exit+0x50/0xa0)
> [<c0040e24>] (irq_exit+0x0/0xa0) from [<c0028068>] (__exception_text_start+0x68/0x84)
>  r4:00000020
> [<c0028000>] (__exception_text_start+0x0/0x84) from [<c0028a28>] (__irq_svc+0x48/0x88)
> Exception stack(0xc02cff40 to 0xc02cff88)
> ff40: 00000000 0005317f 0005217f 60000013 c0029e80 c02ce000 c02d1e44 c0029e80
> ff60: c02e82a4 41069265 80020e54 c02cff94 600000d3 c02cff88 c0029ec4 c0029ed0
> ff80: 60000013 ffffffff
>  r5:fec48000 r4:ffffffff
> [<c0029e80>] (default_idle+0x0/0x58) from [<c0029d4c>] (cpu_idle+0x58/0x98)
> [<c0029cf4>] (cpu_idle+0x0/0x98) from [<c022cb9c>] (rest_init+0x68/0x7c)
>  r8:80020ef0 r7:c02d1cd8 r6:c0022f0c r5:c02e7e4c r4:c02fd180
> [<c022cb34>] (rest_init+0x0/0x7c) from [<c00089f0>] (start_kernel+0x260/0x2c0)
> [<c0008790>] (start_kernel+0x0/0x2c0) from [<80008034>] (0x80008034)
>  r5:c02e8354 r4:00053175
> ---[ end trace a0edb9581ce81991 ]---
> 
> 
> As per the details below, this happens only if I use spinlock/spinunlock sequence in dm9000_start_xmit(). If I change this to dm9000 specific disable/enable interrupts, I do see the above dump and platform comes up fine with NFS.
> 
> Any ideas?

The spin lock sequence already disables the IRQs over the transmit. We
cannot allow the code to be re-entered either by another interrupt
such as a watchdog timeout or receive interrupt.
 
> Thanks a lot,
> Sneha
> 
> 
> -----Original Message-----
> From: davinci-linux-open-source-bounces+nsnehaprabha=ti.com@...ux.davincidsp.com [mailto:davinci-linux-open-source-bounces+nsnehaprabha=ti.com@...ux.davincidsp.com] On Behalf Of Narnakaje, Snehaprabha
> Sent: Tuesday, January 20, 2009 7:38 PM
> To: Kevin Hilman; David Brownell
> Cc: davinci-linux-open-source@...ux.davincidsp.com
> Subject: RE: [patch 2.6.28-rc8-davinci-git] dm355: mmc0 works for pio
> 
> David, Kevin,
> 
> Here is an update on the DM9000 issue.
> 
> I spent some time on debugging DM9000 issue - "tx timeout and a kernel crash".
> 
> In the current version of dm9000 driver (1.3), the hard_start_xmit API for DM9000 - dm9000_start_xmiit() uses spin_lock_irqsave(&db->lock, flags); to disable the interrupts, before accessing/setting DM9000 TX registers. It also calls spin_unlock_irqrestore(&db->lock, flags); after initiating a tx. 
> 
> In an older version of the dm9000.c (1.2 from kernel 2.6.18 to 2.6.22) we just disabled dm9000 rx/tx interrupts by writing to the dm9000 register and enabled them after initiating the tx.
> 
> It does look like, dm9000 stopped receiving the tx interrupts (at least dm9000_interrupt is not getting called for tx completion) and thus we starting seeing tx timeout from the netdev.
> 
> I can get NFS working, if I change the spin_lock/spin_unlock function calls to dm9000 specific rx/tx interrupt disable/enable functions.
> 
> I believe, dm9000 specific rx/tx interrupt functions were replaced by the spin_lock/spin_unlock functions to a fix a hang problem under heavy load.
> 
> I will continue to debug this further, wanted to see, if you have any inputs in this area.
> 
> I did not see this issue being reported on the netdev kernel.org mailing list, so not sure, if there is a DM355 specific irq initialization we have missed.
> 
> Thanks
> Sneha
> 
> > -----Original Message-----
> > From: davinci-linux-open-source-bounces@...ux.davincidsp.com
> > [mailto:davinci-linux-open-source-bounces@...ux.davincidsp.com] On Behalf
> > Of Kevin Hilman
> > Sent: Friday, December 19, 2008 11:01 AM
> > To: David Brownell
> > Cc: davinci-linux-open-source@...ux.davincidsp.com
> > Subject: Re: [patch 2.6.28-rc8-davinci-git] dm355: mmc0 works for pio
> > 
> > David Brownell <david-b@...bell.net> writes:
> > 
> > > On Thursday 18 December 2008, Kevin Hilman wrote:
> > >> Strange. ?I get NETDEV WATCHDOG timeouts whether I try to mount over
> > >> NFS, or whether I simply ssh from a MMC rootfs to linux box. ?This is
> > >> with current davinci git head. ?Have I missed any of your patches?
> > >
> > > Nothing related to the dm9000 or AEMIF, it seems.  Does using
> > >
> > >   CONFIG_DM9000_FORCE_SIMPLE_PHY_POLL=y
> > >
> > > change things?
> > >
> > 
> > Nope.
> > 
> > Kevin
> > 
> > 
> > 
> > 
> > _______________________________________________
> > Davinci-linux-open-source mailing list
> > Davinci-linux-open-source@...ux.davincidsp.com
> > http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
> _______________________________________________
> Davinci-linux-open-source mailing list
> Davinci-linux-open-source@...ux.davincidsp.com
> http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@...r.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Ben (ben@...ff.org, http://www.fluff.org/)

  'a smiley only costs 4 bytes'
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists