Re: [PATCH v2] mac80211: resume properly, add suspend/resume test

[Johannes Berg <johannes@sipsolutions.net>]

[-- Attachment #1: Type: text/plain, Size: 1128 bytes --]

On Fri, 2009-03-13 at 14:51 -0400, Bob Copeland wrote:

> > [  289.753866]
> > [  289.753869] =================================
> > [  289.754164] [ INFO: inconsistent lock state ]
> > [  289.754372] 2.6.29-rc7-wl-17155-g2f7ab92-dirty #29
> > [  289.754598] ---------------------------------
> > [  289.754805] inconsistent {in-softirq-W} -> {softirq-on-W} usage.
> 
> [softirq-on-W stack looks like:
> 
>   sta_notify
>     ath_node_detach
>       ath_tx_node_cleanup
>         spin_lock()
> 
> vs the ath9k_tx_tasklet]
> 
> > Looks like we need to disable BHs for the sta functions or something?

> Yup... or pm can take sta_lock with spin_lock_irqsave() like the other
> callers of sta_notify.

Yes, we probably should do that in mac80211 for consistency, I'll cook
up a patch tomorrow or so.

> Feel free to add my Acked-by on both the PM patches.  I'd been using
> CONFIG_PM_DEBUG to simulate suspend/resume but since you have to
> do 'devices' anyway to get the desired effect, the debug file here
> is nicer.

PM_DEBUG works too, this just seemed a little more self-contained.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]