spinlock semaphore question 2.6.32 mac80211
 

Hi,
I am trying to cut in a function in the mac80211 layer in the 2.6.32 kernel and am stuck I was hoping someone can help.

The issue is that I need to call msleep() or schedule() or down() inside multiple layers of existing spinlock and rcu_lock critical sections, when I attempt this I get the "scheduling while atomic" or "scheduling from idle thread" error.

This is using 2.6.32 non-SMP. I have tried both preempt and not preempt as kernel options.

I have read around that if you are non-SMP the spin locks shouldn't do anything and I found an option to let the RCU's be preemptable.

Is there an obvious way to tackle this problem?

I am trying to cut in this fcn :
void ieee80211_aes_ccm_encrypt(struct crypto_cipher *tfm, u8 *scratch,
u8 *data, size_t data_len,
u8 *cdata, u8 *mic)
in

net/mac80211/aes_ccm.c

with a blocking version, without rewriting the kernel

Any help or 411 on spinlocks,interrupts and rcu_locks in the networking layer will help a ton
Thanks

Why do you need to block there? You're running the encryption elsewhere and you need to wait for the result?

Exactly, The chip (freescale) has a dedicated encryption co-processor and I have to wait for the result

Why don't you acquire the those locks before this function executes?
Or unlock those locks?
Or just use tasklets or workqueues?

archieval thanks for the info,
I am reading up on workqueues and tasklets, I think the workqueue is what I am looking for, (which permits sleeping).

I still cant think of a clever way force a function running in a interrupt context to wait for one in a user context.
You can always split the function up into a top and bottom half, I was just trying to be lazier than that.