Usually (e.g., in 2.6.27-rc1 & earlier), I see lots of these messages for
some reason:

tg3: eth2: Link is up at 1000 Mbps, full duplex.
tg3: eth2: Flow control is off for TX and off for RX.
tg3: eth3: Link is up at 1000 Mbps, full duplex.
tg3: eth3: Flow control is off for TX and off for RX.
bnx2: eth1: using MSI
bnx2: eth1 NIC SerDes Link is Up, 1000 Mbps full duplex
tg3: eth2: Link is up at 1000 Mbps, full duplex.
tg3: eth2: Flow control is off for TX and off for RX.
tg3: eth3: Link is up at 1000 Mbps, full duplex.
tg3: eth3: Flow control is off for TX and off for RX.
bnx2: eth1: using MSI
bnx2: eth1 NIC SerDes Link is Up, 1000 Mbps full duplex
tg3: eth2: Link is up at 1000 Mbps, full duplex.
tg3: eth2: Flow control is off for TX and off for RX.
tg3: eth3: Link is up at 1000 Mbps, full duplex.
tg3: eth3: Flow control is off for TX and off for RX.
bnx2: eth1: using MSI
bnx2: eth1 NIC SerDes Link is Up, 1000 Mbps full duplex
tg3: eth2: Link is up at 1000 Mbps, full duplex.
tg3: eth2: Flow control is off for TX and off for RX.
tg3: eth3: Link is up at 1000 Mbps, full duplex.
tg3: eth3: Flow control is off for TX and off for RX.
bnx2: eth1: using MSI


In 2.6.27-rc1-git4, I now see this (39 times):

BUG: scheduling while atomic: ifconfig/16971/0x00000100
Modules linked in: parport_pc lp parport tg3 lpfc cciss ehci_hcd ohci_hcd uhci_hcd
Pid: 16971, comm: ifconfig Not tainted 2.6.27-rc1-git4 #1

Call Trace:
 [<ffffffff80283db9>] ? page_add_new_anon_rmap+0x20/0x22
 [<ffffffff802310e9>] __schedule_bug+0x62/0x66
 [<ffffffff80551146>] schedule+0x99/0x759
 [<ffffffff8023dc9f>] ? __mod_timer+0xc1/0xd3
 [<ffffffff8055584a>] ? do_page_fault+0x473/0x7dd
 [<ffffffff80551ce4>] schedule_timeout+0x8d/0xb4
 [<ffffffff8023d99e>] ? process_timeout+0x0/0xb
 [<ffffffff80551cdf>] ? schedule_timeout+0x88/0xb4
 [<ffffffff80551d24>] schedule_timeout_uninterruptible+0x19/0x1b
 [<ffffffff8023dcc5>] msleep+0x14/0x1e
 [<ffffffff80375f2f>] pci_set_power_state+0x1cd/0x292
 [<ffffffff803739fc>] ...
[ message continues ]

On Mon, 4 Aug 2008 09:48:35 -0700

You have a bad cable or broken ethernet switch that is causing

This is a bug.
--

[ message continues ]

I also have an issue like this one using the latest linus's tree (2.6.27-rc1).  
My calltrace is a bit different, follows: 

BUG: scheduling while atomic: ip/3205/0x00000100
Modules linked in: autofs4 hidp rfcomm l2cap bluetooth sunrpc iptable_filter ip_tables ip6t_REJECT xt_tcpudp ip6table_filter ip6_tables x_tables ipv6 dm_mirror dm_log dm_multipath dm_mod snd_powermac snd_seq_dummy snd_seq_oss snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss snd_pcm snd_page_alloc snd_timer snd soundcore parport_pc lp parport sg tg3 libphy windfarm_pid windfarm_smu_sat i2c_core shpchp ipr libata sd_mod scsi_mod ext3 jbd uhci_hcd ohci_hcd ehci_hcd
Call Trace:

[c0000001eb98b160] [c00000000001038c] .show_stack+0x70/0x184 (unreliable)
[c0000001eb98b210] [c00000000006b03c] .__schedule_bug+0x6c/0x88
[c0000001eb98b2a0] [c0000000004020c8] .schedule+0xc4/0x8e4
[c0000001eb98b390] [c000000000402f48] .schedule_timeout+0xa8/0xe8
[c0000001eb98b460] [c00000000007f478] .msleep+0x20/0x38
[c0000001eb98b4d0] [c0000000001e5608] .pci_set_power_state+0x274/0x3cc
[c0000001eb98b580] [d0000000004774f0] .tg3_set_power_state+0x80/0xab4 [tg3]
[c0000001eb98b630] [d000000000480fa4] .tg3_open+0x5c/0x96c [tg3]
[c0000001eb98b6f0] [c00000000036d658] .dev_open+0xe4/0x154
[c0000001eb98b770] [c00000000036baf4] .dev_change_flags+0x104/0x204
[c0000001eb98b810] [c0000000003c9e40] .devinet_ioctl+0x2c4/0x758
[c0000001eb98b910] [c0000000003cadf0] .inet_ioctl+0xd8/0x12c
[c0000001eb98b990] [c00000000035cad8] .sock_ioctl+0x2b8/0x310
[c0000001eb98ba30] [c000000000108c74] .vfs_ioctl+0x5c/0xf0
[c0000001eb98bad0] [c000000000109114] .do_vfs_ioctl+0x40c/0x448
[c0000001eb98bb80] [c0000000001091c0] .sys_ioctl+0x70/0xb4
[c0000001eb98bc30] [c00000000013ddb4] .dev_ifsioc+0x1b0/0x3e4
[c0000001eb98bd40] [c00000000013d374] .compat_sys_ioctl+0x3d4/0x468
[c0000001eb98be30] [c0000000000086b4] syscall_exit+0x0/0x40

This error is easily reproducible, just run "ifup ethX; ifdown ethX".
The cable connection is ok, and the device is working ...
[ message continues ]

This was introduced by:

tg3: adapt tg3 to use reworked PCI PM code

Adapt the tg3 driver to use the reworked PCI PM and make it use the
exported PCI PM core functions instead of accessing the PCI PM registers
directly by itself.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the original tg3 code, we used udelay() when switching to D0 state
since we were inside tg3_full_lock().


--

[ message continues ]

I just reverted this patch on my tree and the problem doesn't happen anymore.
--

[ message continues ]

From: "Michael Chan" <mchan@broadcom.com>

Right, this change was obviously not tested properly.

I'm likely going to revert unless a really good fix shows up fast.
--

[ message continues ]

I'm looking into it right now.  Stay tuned.



--

[ message continues ]

--

[ message continues ]