Re: [PATCH v5.18] ath9k: Properly clear TX status area before reporting to mac80211

["Toke Høiland-Jørgensen" <toke@toke.dk>]

Peter Seiderer <ps.report@gmx.net> writes:

> Hello Toke,
>
> On Wed, 30 Mar 2022 18:44:09 +0200, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
>
>> The ath9k driver was not properly clearing the status area in the
>> ieee80211_tx_info struct before reporting TX status to mac80211. Instead,
>> it was manually filling in fields, which meant that fields introduced later
>> were left as-is.
>> 
>> Conveniently, mac80211 actually provides a helper to zero out the status
>> area, so use that to make sure we zero everything.
>> 
>> The last commit touching the driver function writing the status information
>> seems to have actually been fixing an issue that was also caused by the
>> area being uninitialised; but it only added clearing of a single field
>> instead of the whole struct. That is now redundant, though, so revert that
>> commit and use it as a convenient Fixes tag.
>> 
>> Fixes: cc591d77aba1 ("ath9k: Make sure to zero status.tx_time before reporting TX status")
>> Reported-by: Bagas Sanjaya <bagasdotme@gmail.com>
>> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
>> ---
>>  drivers/net/wireless/ath/ath9k/xmit.c | 5 ++---
>>  1 file changed, 2 insertions(+), 3 deletions(-)
>> 
>> diff --git a/drivers/net/wireless/ath/ath9k/xmit.c b/drivers/net/wireless/ath/ath9k/xmit.c
>> index d0caf1de2bde..cbcf96ac303e 100644
>> --- a/drivers/net/wireless/ath/ath9k/xmit.c
>> +++ b/drivers/net/wireless/ath/ath9k/xmit.c
>> @@ -2553,6 +2553,8 @@ static void ath_tx_rc_status(struct ath_softc *sc, struct ath_buf *bf,
>>  	struct ath_hw *ah = sc->sc_ah;
>>  	u8 i, tx_rateindex;
>>  
>> +	ieee80211_tx_info_clear_status(tx_info);
>> +
>
> As this also clears the status.rates[].count, see include/net/mac80211.h:
>
> 1195 static inline void
> 1196 ieee80211_tx_info_clear_status(struct ieee80211_tx_info *info)
> 1197 {
> 1198         int i;
> 1199         
> 1200         BUILD_BUG_ON(offsetof(struct ieee80211_tx_info, status.rates) !=
> 1201                      offsetof(struct ieee80211_tx_info, control.rates));
> 1202         BUILD_BUG_ON(offsetof(struct ieee80211_tx_info, status.rates) !=
> 1203                      offsetof(struct ieee80211_tx_info, driver_rates));
> 1204         BUILD_BUG_ON(offsetof(struct ieee80211_tx_info, status.rates) != 8)     ;       
> 1205         /* clear the rate counts */
> 1206         for (i = 0; i < IEEE80211_TX_MAX_RATES; i++)
> 1207                 info->status.rates[i].count = 0;
> 1208         memset_after(&info->status, 0, rates);
> 1209 }
>
> I would have expected some lines added to restore the count (for the
> rates with index < tx_rateindex), e.g. as done in
> drivers/net/wireless/ath/ath5k/base.c:
>
> 1731         ieee80211_tx_info_clear_status(info);
> 1732 
> 1733         for (i = 0; i < ts->ts_final_idx; i++) {
> 1734                 struct ieee80211_tx_rate *r =
> 1735                         &info->status.rates[i];
> 1736 
> 1737                 r->count = tries[i];
> 1738         }
>
> In drivers/net/wireless/ath/ath9k/xmit.c this is only done/changed for
> the tx_rateindex index (which is often zero in case the first suggested rate
> succeeds, but in noisy environment is sometimes > 0)...

Ah, you're right! I looked at that code, and somehow parsed that as "OK,
it's setting all the rate counts, we're fine". But obviously that's not
what that code is doing, so we'll lose some information now. Bugger :(

I'll send a follow-up, thanks for flagging this!

-Toke