[PATCH] staging: brcm80211: bugfix for softmac crash on multi cpu configurations

[PATCH] staging: brcm80211: bugfix for softmac crash on multi cpu configurations

[Roland Vossen]

Solved a locking issue that resulted in driver crashes with the 43224 and 43225
chips. The problem has been reported on several fora. Root cause was two fold:
hardware was being manipulated by two unsynchronized threads, and a scan
operation could interfere with an ongoing dynamic calibration process. Fix was
to invoke a lock on wl_ops_config() operation and to set internal flags when a
scan operation is started and stopped.

Please add this to the staging-linus branch.

Reviewed-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Roland Vossen <rvossen@broadcom.com>
---
 drivers/staging/brcm80211/sys/wl_mac80211.c  |   12 ++++++++++--
 drivers/staging/brcm80211/sys/wlc_mac80211.c |   13 +++++++++++++
 drivers/staging/brcm80211/sys/wlc_pub.h      |    2 ++
 3 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/brcm80211/sys/wl_mac80211.c b/drivers/staging/brcm80211/sys/wl_mac80211.c
index f123588..cd8392b 100644
--- a/drivers/staging/brcm80211/sys/wl_mac80211.c
+++ b/drivers/staging/brcm80211/sys/wl_mac80211.c
@@ -263,9 +263,7 @@ ieee_set_channel(struct ieee80211_hw *hw, struct ieee80211_channel *chan,
 	switch (type) {
 	case NL80211_CHAN_HT20:
 	case NL80211_CHAN_NO_HT:
-		WL_LOCK(wl);
 		err = wlc_set(wl->wlc, WLC_SET_CHANNEL, chan->hw_value);
-		WL_UNLOCK(wl);
 		break;
 	case NL80211_CHAN_HT40MINUS:
 	case NL80211_CHAN_HT40PLUS:
@@ -285,6 +283,7 @@ static int wl_ops_config(struct ieee80211_hw *hw, u32 changed)
 	int err = 0;
 	int new_int;
 
+	WL_LOCK(wl);
 	if (changed & IEEE80211_CONF_CHANGE_LISTEN_INTERVAL) {
 		WL_NONE("%s: Setting listen interval to %d\n",
 			__func__, conf->listen_interval);
@@ -341,6 +340,7 @@ static int wl_ops_config(struct ieee80211_hw *hw, u32 changed)
 	}
 
  config_out:
+	WL_UNLOCK(wl);
 	return err;
 }
 
@@ -459,13 +459,21 @@ wl_ops_set_tim(struct ieee80211_hw *hw, struct ieee80211_sta *sta, bool set)
 
 static void wl_ops_sw_scan_start(struct ieee80211_hw *hw)
 {
+	struct wl_info *wl = hw->priv;
 	WL_NONE("Scan Start\n");
+	WL_LOCK(wl);
+	wlc_scan_start(wl->wlc);
+	WL_UNLOCK(wl);
 	return;
 }
 
 static void wl_ops_sw_scan_complete(struct ieee80211_hw *hw)
 {
+	struct wl_info *wl = hw->priv;
 	WL_NONE("Scan Complete\n");
+	WL_LOCK(wl);
+	wlc_scan_stop(wl->wlc);
+	WL_UNLOCK(wl);
 	return;
 }
 
diff --git a/drivers/staging/brcm80211/sys/wlc_mac80211.c b/drivers/staging/brcm80211/sys/wlc_mac80211.c
index a130386..e37e805 100644
--- a/drivers/staging/brcm80211/sys/wlc_mac80211.c
+++ b/drivers/staging/brcm80211/sys/wlc_mac80211.c
@@ -8461,3 +8461,16 @@ static void wlc_txq_free(struct wlc_info *wlc, struct osl_info *osh,
 
 	kfree(qi);
 }
+
+/*
+ * Flag 'scan in progress' to withold dynamic phy calibration
+ */
+void wlc_scan_start(struct wlc_info *wlc)
+{
+	wlc_phy_hold_upd(wlc->band->pi, PHY_HOLD_FOR_SCAN, true);
+}
+
+void wlc_scan_stop(struct wlc_info *wlc)
+{
+	wlc_phy_hold_upd(wlc->band->pi, PHY_HOLD_FOR_SCAN, false);
+}
diff --git a/drivers/staging/brcm80211/sys/wlc_pub.h b/drivers/staging/brcm80211/sys/wlc_pub.h
index 146a690..aff4130 100644
--- a/drivers/staging/brcm80211/sys/wlc_pub.h
+++ b/drivers/staging/brcm80211/sys/wlc_pub.h
@@ -570,6 +570,8 @@ extern void wlc_enable_mac(struct wlc_info *wlc);
 extern u16 wlc_rate_shm_offset(struct wlc_info *wlc, u8 rate);
 extern u32 wlc_get_rspec_history(struct wlc_bsscfg *cfg);
 extern u32 wlc_get_current_highest_rate(struct wlc_bsscfg *cfg);
+extern void wlc_scan_start(struct wlc_info *wlc);
+extern void wlc_scan_stop(struct wlc_info *wlc);
 
 static inline int wlc_iovar_getuint(struct wlc_info *wlc, const char *name,
 				    uint *arg)
-- 
1.7.1

Re: [PATCH] staging: brcm80211: bugfix for softmac crash on multi cpu configurations

[Greg KH]

On Tue, Jan 25, 2011 at 11:51:56AM +0100, Roland Vossen wrote:
> Solved a locking issue that resulted in driver crashes with the 43224 and 43225
> chips. The problem has been reported on several fora. Root cause was two fold:
> hardware was being manipulated by two unsynchronized threads, and a scan
> operation could interfere with an ongoing dynamic calibration process. Fix was
> to invoke a lock on wl_ops_config() operation and to set internal flags when a
> scan operation is started and stopped.
> 
> Please add this to the staging-linus branch.

Should this also go to the .37 stable kernel tree?

thanks,

greg k-h

RE: [PATCH] staging: brcm80211: bugfix for softmac crash on multi cpu configurations

[Roland Vossen]

>Should this also go to the .37 stable kernel tree?

Yes it should. Should I send you a patch for the .37 stable kernel tree ?

Thanks, Roland.

Re: [PATCH] staging: brcm80211: bugfix for softmac crash on multi cpu configurations

[Greg KH]

On Tue, Jan 25, 2011 at 03:24:38AM -0800, Roland Vossen wrote:
> >Should this also go to the .37 stable kernel tree?
> 
> Yes it should. Should I send you a patch for the .37 stable kernel tree ?

Not at the moment, I'll just mark it for the stable tree, and if it
doesn't apply there properly, I'll email you asking for an updated
version.

For more details about the stable kernel process, see the file,
Documentation/stable_kernel_rules.txt

thanks,

greg k-h