RE: [PATCH 3/9] rtw88: Move rtw_update_sta_info() out of rtw_ra_mask_info_update_iter()

[Pkshih <pkshih@xxxxxxxxxxx>]

> -----Original Message-----
> From: Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx>
> Sent: Wednesday, December 29, 2021 5:15 AM
> To: linux-wireless@xxxxxxxxxxxxxxx
> Cc: tony0620emma@xxxxxxxxx; kvalo@xxxxxxxxxxxxxx; johannes@xxxxxxxxxxxxxxxx; netdev@xxxxxxxxxxxxxxx;
> linux-kernel@xxxxxxxxxxxxxxx; Neo Jou <neojou@xxxxxxxxx>; Jernej Skrabec <jernej.skrabec@xxxxxxxxx>;
> Pkshih <pkshih@xxxxxxxxxxx>; Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx>
> Subject: [PATCH 3/9] rtw88: Move rtw_update_sta_info() out of rtw_ra_mask_info_update_iter()
> 
> rtw_update_sta_info() internally access some registers while being
> called unter an atomic lock acquired by rtw_iterate_vifs_atomic(). Move
> rtw_update_sta_info() call out of (rtw_ra_mask_info_update_iter) in
> preparation for SDIO support where register access may sleep.
> 
> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx>
> ---
> v1 -> v2:
> - this patch is new in v2
> - keep rtw_iterate_vifs_atomic() to prevent deadlocks as Johannes
>   suggested. Keep track of all relevant stations inside
>   rtw_ra_mask_info_update_iter() and the iter-data and then call
>   rtw_update_sta_info() while held under rtwdev->mutex instead
> 
>  drivers/net/wireless/realtek/rtw88/mac80211.c | 14 +++++++++++++-
>  1 file changed, 13 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/wireless/realtek/rtw88/mac80211.c
> b/drivers/net/wireless/realtek/rtw88/mac80211.c
> index ae7d97de5fdf..3bd12354a8a1 100644
> --- a/drivers/net/wireless/realtek/rtw88/mac80211.c
> +++ b/drivers/net/wireless/realtek/rtw88/mac80211.c

[...]

> @@ -699,11 +702,20 @@ static void rtw_ra_mask_info_update(struct rtw_dev *rtwdev,
>  				    const struct cfg80211_bitrate_mask *mask)
>  {
>  	struct rtw_iter_bitrate_mask_data br_data;
> +	unsigned int i;
> +
> +	mutex_lock(&rtwdev->mutex);

I think this lock is used to protect br_data.si[i], right?

And, I prefer to move mutex lock to caller, like:

@@ -734,7 +734,9 @@ static int rtw_ops_set_bitrate_mask(struct ieee80211_hw *hw,
 {
        struct rtw_dev *rtwdev = hw->priv;

+       mutex_lock(&rtwdev->mutex);
        rtw_ra_mask_info_update(rtwdev, vif, mask);
+       mutex_unlock(&rtwdev->mutex);

        return 0;
 }

> 
>  	br_data.rtwdev = rtwdev;
>  	br_data.vif = vif;
>  	br_data.mask = mask;
> +	br_data.num_si = 0;
>  	rtw_iterate_stas_atomic(rtwdev, rtw_ra_mask_info_update_iter, &br_data);
> +
> +	for (i = 0; i < br_data.num_si; i++)
> +		rtw_update_sta_info(rtwdev, br_data.si[i]);
> +
> +	mutex_unlock(&rtwdev->mutex);
>  }
> 

--
Ping-Ke