iwl3945: do not delay hardware scan if it is a direct scan

iwl3945 <---> mac80211 <----> wpa_supplicant <---> NetworkManager

When a hardware scan is completed and another scan is requested in less
than two seconds, iwlwifi will not do the second scan and will pass the
error code -EAGAIN back to mac80211 where it quickly dies. The error
code is not passed along to the calling program wpa_supplicant. After a
timeout, wpa_supplicant will just give up but it will not know why the
scan failed. This is a weakness in the design.

I ran into this issue when I was trying to figure out why it takes more
an a minute for NetworkManager to connect after Networking has been
disabled and then re-enabled. I found a good deal of unnecessary work
being done because mac80211 requests authentication when the interface
is not configured, the ANY mode. I created an experimental passive
(NOTANY) mode for mac80211 to eliminate this case. Then NetworkManager
became so fast that I ran into the iwlwifi 2 second delay next scan
issue which we are discussing.

The patch resolves the problem by bypassing the delay if the scan request
is a direct scan. It should do less harm to the hardware.

Signed-off-by: Bill Moss <bmoss@CLEMSON.EDU>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

drivers/net/wireless/iwlwifi/iwl3945-base.c

@@ -7087,9 +7087,10 @@ static int iwl3945_mac_hw_scan(struct ieee80211_hw *hw, u8 *ssid, size_t len)
 		rc = -EAGAIN;
 		goto out_unlock;
 	}
-	/* if we just finished scan ask for delay */
-	if (priv->last_scan_jiffies && time_after(priv->last_scan_jiffies +
-				IWL_DELAY_NEXT_SCAN, jiffies)) {
+	/* if we just finished scan ask for delay for a broadcast scan */
+	if ((len == 0) && priv->last_scan_jiffies &&
+	    time_after(priv->last_scan_jiffies + IWL_DELAY_NEXT_SCAN,
+		       jiffies)) {
 		rc = -EAGAIN;
 		goto out_unlock;
 	}