slab corruption during packet flood on e100
Hi,
When routing a large stream of UDP packets through this machine, coming in on eth1 (e100), headed towards dummy0, I get a flood of slab corruption messages a la below. Box is a 2-way HT SMP machine.
Is this a known problem with the e100 driver perhaps?
thanks,
Lennert
Dec 31 11:42:05 phi kernel: Slab corruption: start=306ef74c, expend=306eff4b, problemat=306ef75e Dec 31 11:42:05 phi kernel: Last user: [<02279033>](kfree_skbmem+0x13/0x30) Dec 31 11:42:05 phi kernel: Data: *[snip]*28 A0 *[snip]* Dec 31 11:42:05 phi kernel: *[snip]* Dec 31 11:42:05 phi kernel: *[snip]*A5 Dec 31 11:42:05 phi kernel: Next: 71 F0 2C .33 90 27 02 ************************ Dec 31 11:42:05 phi kernel: slab error in check_poison_obj(): cache `size-2048': object was modified after freeing Dec 31 11:42:05 phi kernel: Call Trace: Dec 31 11:42:05 phi kernel: [<0214bee9>] check_poison_obj+0x109/0x1a0 Dec 31 11:42:05 phi kernel: [<0214dfb1>] __kmalloc+0x211/0x270 Dec 31 11:42:05 phi kernel: [<02278e07>] alloc_skb+0x47/0xf0 Dec 31 11:42:05 phi kernel: [<02278e07>] alloc_skb+0x47/0xf0 Dec 31 11:42:05 phi kernel: [<42b4671e>] e100_poll+0x1ae/0x6c0 [e100] Dec 31 11:42:05 phi kernel: [<0210e1e8>] do_nmi+0x38/0x60 Dec 31 11:42:05 phi kernel: [<0227dff5>] net_rx_action+0x85/0x140 Dec 31 11:42:05 phi kernel: [<0212cff7>] do_softirq+0xc7/0xd0 Dec 31 11:42:06 phi kernel: [<0210efaa>] do_IRQ+0x11a/0x160 Dec 31 11:42:06 phi kernel: [<02126c10>] remove_wait_queue+0x50/0x80 Dec 31 11:42:06 phi kernel: [<0217a2ee>] poll_freewait+0x2e/0x50 Dec 31 11:42:06 phi kernel: [<0217b0fc>] sys_poll+0x22c/0x260 Dec 31 11:42:06 phi kernel: [<0217a310>] __pollwait+0x0/0xd0 Dec 31 11:42:06 phi kernel:
Quoted 6 lines · Lennert Buytenhek
When routing a large stream of UDP packets through this machine, coming
in on eth1 (e100), headed towards dummy0, I get a flood of slab corruption messages a la below. Box is a 2-way HT SMP machine.Is this a known problem with the e100 driver perhaps?
The experimental net driver tree has what appears to be a big e100 rewrite in it. Can you test 2.6.1-rc1 sometime?
Quoted 8 lines · Andrew Morton
Quoted 5 lines
When routing a large stream of UDP packets through this machine, coming
in on eth1 (e100), headed towards dummy0, I get a flood of slab corruption messages a la below. Box is a 2-way HT SMP machine.Is this a known problem with the e100 driver perhaps?
The experimental net driver tree has what appears to be a big e100 rewrite in it. Can you test 2.6.1-rc1 sometime?
2.6.1-rc1 seems just fine. No packet drops, no overruns, and most importantly, no slab corruption warnings.
--L