Re: [PATCH] wpan: use GFP_ATOMIC instead of GFP_KERNEL

Hi,

thanks for your reply.

On Tue, Feb 05, 2013 at 08:09:57AM -0800, Eric Dumazet wrote:

On Tue, 2013-02-05 at 16:34 +0100, Alexander Aring wrote:

quoted

I got this message at runtime:

[   48.847347] BUG: scheduling while atomic: swapper/0/0/0x40000100
[   48.853870] Modules linked in: autofs4
[   48.857891] [<c0013db8>] (unwind_backtrace+0x0/0x11c) from [<c04cfb84>] (__schedule_bug+0x48/0x5c)
[   48.867342] [<c04cfb84>] (__schedule_bug+0x48/0x5c) from [<c04e0f4c>] (__schedule+0x68/0x5bc)
[   48.876357] [<c04e0f4c>] (__schedule+0x68/0x5bc) from [<c0063f0c>] (__cond_resched+0x24/0x34)
[   48.885368] [<c0063f0c>] (__cond_resched+0x24/0x34) from [<c04e152c>] (_cond_resched+0x34/0x44)
[   48.894559] [<c04e152c>] (_cond_resched+0x34/0x44) from [<c00e9b28>] (kmem_cache_alloc_trace+0x2c/0x158)
[   48.904578] [<c00e9b28>] (kmem_cache_alloc_trace+0x2c/0x158) from [<c04baa3c>] (mac802154_header_create+0x44/0x2)
[   48.915594] [<c04baa3c>] (mac802154_header_create+0x44/0x24c) from [<c04b8e50>] (lowpan_header_create.part.3+0x7)
[   48.927157] [<c04b8e50>] (lowpan_header_create.part.3+0x734/0x79c) from [<c03ebc04>] (neigh_connected_output+0x9)
[   48.938542] [<c03ebc04>] (neigh_connected_output+0x98/0xd0) from [<c0467d5c>] (ip6_finish_output2+0x374/0x45c)
[   48.949102] [<c0467d5c>] (ip6_finish_output2+0x374/0x45c) from [<c0479bcc>] (ndisc_send_skb+0x174/0x22c)
[   48.959101] [<c0479bcc>] (ndisc_send_skb+0x174/0x22c) from [<c047b138>] (ndisc_send_rs+0x3c/0x44)
[   48.968473] [<c047b138>] (ndisc_send_rs+0x3c/0x44) from [<c046e3f0>] (addrconf_rs_timer+0xbc/0x16c)
[   48.978032] [<c046e3f0>] (addrconf_rs_timer+0xbc/0x16c) from [<c00451fc>] (call_timer_fn+0x74/0x134)
[   48.987675] [<c00451fc>] (call_timer_fn+0x74/0x134) from [<c0046984>] (run_timer_softirq+0x258/0x2d4)
[   48.997417] [<c0046984>] (run_timer_softirq+0x258/0x2d4) from [<c003fb84>] (__do_softirq+0x118/0x248)
[   49.007156] [<c003fb84>] (__do_softirq+0x118/0x248) from [<c003ff64>] (irq_exit+0x44/0x84)
[   49.015890] [<c003ff64>] (irq_exit+0x44/0x84) from [<c000e780>] (handle_IRQ+0x7c/0xb8)
[   49.024249] [<c000e780>] (handle_IRQ+0x7c/0xb8) from [<c000857c>] (omap3_intc_handle_irq+0x54/0x68)
[   49.033801] [<c000857c>] (omap3_intc_handle_irq+0x54/0x68) from [<c04e2900>] (__irq_svc+0x40/0x50)
[   49.043239] Exception stack(0xc074bf68 to 0xc074bfb0)
[   49.048576] bf60:                   ffffffed 00000000 002f8000 00000000 c074a000 c07b6608
[   49.057199] bf80: c04ec10c c0757618 80004059 413fc082 00000000 00000000 00000000 c074bfb0
[   49.065821] bfa0: c000e904 c000e908 60000013 ffffffff
[   49.071162] [<c04e2900>] (__irq_svc+0x40/0x50) from [<c000e908>] (default_idle+0x24/0x2c)
[   49.079789] [<c000e908>] (default_idle+0x24/0x2c) from [<c000ea7c>] (cpu_idle+0x9c/0xf0)
[   49.088328] [<c000ea7c>] (cpu_idle+0x9c/0xf0) from [<c06ff7dc>] (start_kernel+0x258/0x2a4)

It seems that mac802154_header_create is calling from a irq context.
So kzalloc need GFP_ATOMIC flag.

This patch will fix it.

Signed-off-by: Alexander Aring <alex.aring@gmail.com>
---
 net/mac802154/wpan.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/mac802154/wpan.c b/net/mac802154/wpan.c
index 98c867b..0d9e110 100644
--- a/net/mac802154/wpan.c
+++ b/net/mac802154/wpan.c

@@ -143,7 +143,7 @@ static int mac802154_header_create(struct sk_buff *skb,
 	if (!daddr)
 		return -EINVAL;
 
-	head = kzalloc(MAC802154_FRAME_HARD_HEADER_LEN, GFP_KERNEL);
+	head = kzalloc(MAC802154_FRAME_HARD_HEADER_LEN, GFP_ATOMIC);
 	if (head == NULL)
 		return -ENOMEM;
 

Please remove this allocation, its absolutely not needed.

I don't know how I can do that. head is a temp buffer.
At the end of this function stands:

memcpy(skb_push(skb, pos), head, pos)

and 'pos' is determined at runtime. skb_push will add data to the
start of buffer, so head buffer will be at the start of skb. I don't
know where I should manipulate the skb buffer at beginning of this function.

Regards
Alex

40 bytes on the stack are fine.

Thanks