Marcus Better | 15 Oct 13:37 2007
Picon

Re: [PATCH v2] mac80211: fix set_channel regression

Johannes Berg wrote:
> On Mon, 2007-10-15 at 09:04 +0200, Marcus Better wrote:
>> The new patch doesn't seem to solve the problem either. I get the same
>> association failure as before.

Update: on my second recompile (after adding the printks) it started
working. I'm not sure why it failed the first time.

However there is a huge performance regression. "tptest" shows a TCP
throughput of under 300 kbit for the driver version 2.0.9 (current
wireless-2.6), compared to 7 Mbit for version 2.0.7 (with kernel
2.6.23-rc3). The UDP speed was also affected but not as much, down from
around 5-10 Mbit to 1 Mbit.

> Can you add back your printk patch and check what type 'channel' was and
> maybe print channel->channel?

With the printks below I get the following:

Oct 15 13:14:43 better kernel: selected mode 2
Oct 15 13:14:43 better kernel: selected channel 8

Thanks,

Marcus

--- a/net/mac80211/ieee80211_ioctl.c
+++ b/net/mac80211/ieee80211_ioctl.c
 <at>  <at>  -314,6 +314,8  <at>  <at>  int ieee80211_set_channel(struct ieee80211_local
*local, int
        }

        if (set) {
+               printk(KERN_DEBUG "selected mode %d\n",
local->oper_hw_mode->mod
+               printk(KERN_DEBUG "selected channel %d\n",
local->oper_channel->
                if (local->sta_scanning)
                        ret = 0;
                else


Gmane