[PATCH] libertas_sdio lockup fix

[Jes Sorensen]

On 04/01/10 09:06, Dan Williams wrote:
> We're trying to kill lbs_prepare_and_send_command() completely though
> and move most commands to separate handlers.  Any chance you could port
> the patch to do that instead?  Basically create a function modeled after
> cmd.c::lbs_get_tx_power()

Hi Dan,

Thanks for the feedback. The problem I had was that I know zilch about
the libertas chips, so I was guessing my way through based on the old
patch that I found.

I'll try and setup a build environment and see if I can reproduce it as
the last test I did was a full rpm build. My test box is a 1.1GHz with
16GB SSD so it's not really up for kernel builds.

> int lbs_get_log (struct lbs_private *priv, u32 *retries, u32 *code, u32 *misc)
> {
> 	struct cmd_ds_802_11_get_log cmd;
> 	int ret;
>
> 	lbs_deb_enter(LBS_DEB_CMD);
>
> 	memset(&cmd, 0, sizeof(cmd));
> 	cmd.hdr.size = cpu_to_le16(sizeof(cmd));
>
> 	ret = lbs_cmd_with_response(priv, CMD_802_11_GET_LOG,&cmd);
> 	if (ret == 0) {
> 		*retries = get_unaligned_le32(cmd.retry);
> 		*code = get_unaligned_le32(cmd.code);
> 		*misc = get_unaligned_le32(cmd.misc);
> 	}
>
> 	lbs_deb_leave(LBS_DEB_CMD);
> 	return ret;
> }
>
> Then at the bottom of the function call lbs_get_log() and use the
> returned u32 values to populate priv->wstats.  But really, it looks like
> the only functional changes here are:

This part looks easy enough.

> 1) the cmd size passed to the 8686 is larger with this patch, since you
> haven't removed the header from struct cmd_ds_802_11_get_log; it could
> be that we got the size slightly wrong or something when converting it
> to a "direct" command?
>     - pad the structure in host.h by another u32 at the bottom and see if
> this helps

The size in my patch was pure guessware to be honest. The original patch
used sizeof(struct cmd_ds_802_11_get_log) + S_DS_GEN and I didn't find
S_DS_GEN anywhere.

> 2) calling GET_LOG after RSSI.  Not sure why that would make a
> difference, but it could
>     - do the "direct" command thing above but move the lbs_get_log() call
> after the RSSI call

I think I can figure this out.

> 3) not calling GET_LOG through the blocking command path during the WEXT
> stats requests
>     - call GET_LOG async and just ignore updating the wstats values to
> see if the blocking nature of "direct" commands is indeed the issue
> since the wext stats stuff gets called in different places with
> different locking semantics

I am not sure if I follow you here. Is there an easy async path for this
or do I need to come up with something?

> Any chance you could try these three options and see if they also fix
> the problem?  I'd like to figure out why this is happing instead of just
> apply what looks like a partial revert first.

I am happy to help as much as I can. Main problem is that the libertas
codebase is pretty unknown to me.

Cheers,
Jes