[2.6.7, e1000] large MTU allocation failure...

[2.6.7, e1000] large MTU allocation failure...

[Daniel Blueman]

When I try to enable jumbo frames, and increase the MTU size to 9000 octets
[1], I see allocation failures in the kernel logs [2].

Is there another way of enabling jumbo frames and use a large MTU?

Kernel is stock 2.6.7, IA32.

--- [1]

# ifconfig eth0 mtu 9000

--- [2]

Intel(R) PRO/1000 Network Driver - version 5.2.52-k4
Copyright (c) 1999-2004 Intel Corporation.
e1000: eth0: e1000_probe: Intel(R) PRO/1000 Network
Connection
e1000: eth0: e1000_watchdog: NIC Link is Up 1000 Mbps
Full Duplex
ifconfig: page allocation failure. order:3, mode:0x20
[<c012ac54>] __alloc_pages+0x2b4/0x300
[<c012acbf>] __get_free_pages+0x1f/0x40
[<c012dc37>] kmem_getpages+0x17/0xb0
[<c012e6f6>] cache_grow+0x96/0x1f0
[<c012e9ae>] cache_alloc_refill+0x15e/0x230
[<c012ef07>] __kmalloc+0x67/0x70
[<c021d552>] alloc_skb+0x32/0xd0
[<c01fa035>] e1000_alloc_rx_buffers+0x55/0xf0
[<c01f75fd>] e1000_up+0x3d/0x90
[<c01f92c6>] e1000_change_mtu+0x76/0x100
[<c02228e5>] dev_set_mtu+0x65/0x70
[<c0222ebf>] dev_ioctl+0x1df/0x250
[<c02602e0>] udp_ioctl+0x0/0xc0
[<c02676f9>] inet_ioctl+0x89/0xa0
[<c021a722>] sock_ioctl+0xd2/0x220
[<c01514a9>] sys_ioctl+0xe9/0x240
[<c0103bf7>] syscall_call+0x7/0xb

-- 
Daniel J Blueman

Re: [2.6.7, e1000] large MTU allocation failure...

[Nick Piggin]

Daniel Blueman wrote:
> When I try to enable jumbo frames, and increase the MTU size to 9000 octets
> [1], I see allocation failures in the kernel logs [2].
> 
> Is there another way of enabling jumbo frames and use a large MTU?
> 
> Kernel is stock 2.6.7, IA32.
> 
> --- [1]
> 
> # ifconfig eth0 mtu 9000
> 
> --- [2]
> 
> Intel(R) PRO/1000 Network Driver - version 5.2.52-k4
> Copyright (c) 1999-2004 Intel Corporation.
> e1000: eth0: e1000_probe: Intel(R) PRO/1000 Network
> Connection
> e1000: eth0: e1000_watchdog: NIC Link is Up 1000 Mbps
> Full Duplex
> ifconfig: page allocation failure. order:3, mode:0x20

Order 3 GFP_ATOMIC allocation failures aren't really surprising.

I wonder if things can be improved in the network system. If not,
I have a possible workaround for the problem.