Re: ath10k: antenna bitmask support?

[Sebastian Gottschall <s.gottschall@dd-wrt.com>]

Am 07.08.2018 um 04:08 schrieb Brian Norris:

> Hi all,
>
> I'm looking at ancient changes like this:
>
> commit 5572a95b4b5768187652a346356e39e7542ca6e0
> Author: Ben Greear <greearb@candelatech.com>
> Date:   Mon Nov 24 16:22:10 2014 +0200
>
>      ath10k: apply chainmask settings to vdev on creation
>
> since I see that some firmwares seem to crash a lot if you apply certain
> chainmask settings (e.g., 0x2).
>
> Is it really expected that you can set gaps in the chainmask or not? I
> ask because I've been trying to use this for doing some antenna
> configuration verification (e.g., disable all but one antenna and see
> what happens), and this works as expected on APs I have running IPQ8064
> (QCA988X?), but it crashes the firmware on IPQ4019.
>
> I similarly wonder about this opinionated statement in
> ath10k_check_chain_mask():
>
>          /* It is not clear that allowing gaps in chainmask
>           * is helpful.  Probably it will not do what user
>           * is hoping for, so warn in that case.
>           */
>
> It seems like we should reject unexpected values (-EINVAL) if they're
> really not supported. But then, I've found differences depending on the
> chipset it would seem, so I guess I'd have to figure out which chipsets
> (or firmwares?) support which features...
>
> On a related note, if we *do* support "gaps" in these bitmasks, wouldn't
> that mean we're handling 0x6, 0xa, etc., incorrectly in
> get_nss_from_chainmask()? My prototype patch (untested so far) would be:
>
> diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
> index 836e0a47b94a..3b557d3dc036 100644
> --- a/drivers/net/wireless/ath/ath10k/mac.c
> +++ b/drivers/net/wireless/ath/ath10k/mac.c
> @@ -19,6 +19,7 @@
>   #include "mac.h"
>   
>   #include <net/mac80211.h>
> +#include <linux/bitops.h>
>   #include <linux/etherdevice.h>
>   #include <linux/acpi.h>
>   
> @@ -4905,13 +4906,7 @@ static int ath10k_config(struct ieee80211_hw *hw, u32 changed)
>   
>   static u32 get_nss_from_chainmask(u16 chain_mask)
>   {
> -	if ((chain_mask & 0xf) == 0xf)
> -		return 4;
> -	else if ((chain_mask & 0x7) == 0x7)
> -		return 3;
> -	else if ((chain_mask & 0x3) == 0x3)
> -		return 2;
> -	return 1;
> +	return max_t(u32, hweight16(chain_mask & 0xf), 1);
>   }
>   
>   static int ath10k_mac_set_txbf_conf(struct ath10k_vif *arvif)
>
> _______________________________________________
> ath10k mailing list
> ath10k@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/ath10k
from what i know, gaps will not work. so valid masks are only like 0x1, 
0x3, 0x7, 0xf. so same behaviour as for ath9k


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k