2018-07-04 09:50:03

by Sebastian Gottschall

[permalink] [raw]
Subject: [PATCH v8] ath10k: fix crash in recent 3.5.3 9984 firmware due wrong handling of peer_bw_rxnss_override parameter

From: Sebastian Gottschall <[email protected]>

current handling of peer_bw_rxnss_override parameter is based on guessing the VHT160/8080 capability by rx rate. this is wrong and may lead
to a non initialized peer_bw_rxnss_override parameter which is required since VHT160 operation mode only supports 2x2 chainmasks in addition the original code
initialized the parameter with wrong masked values.
This patch uses the peer phymode and peer nss information for correct initialisation of the peer_bw_rxnss_override parameter.
if this peer information is not available, we initialize the parameter by minimum nss which is suggested by QCA as temporary workaround according
to the QCA sourcecodes.

Signed-off-by: Sebastian Gottschall <[email protected]>

v2: remove debug messages
v3: apply some cosmetics, update documentation
v4: fix compile warning and truncate nss to maximum of 2x2 since current chipsets only support 2x2 at vht160
v5: handle maximum nss for chipsets supportig vht160 with 1x1 only
v7: use more simple code variant and take care about hw/sw chainmask configuration
v8: fix some code style issues
---
drivers/net/wireless/ath/ath10k/mac.c | 54 +++++++++++++++++++--------
drivers/net/wireless/ath/ath10k/wmi.c | 7 +---
drivers/net/wireless/ath/ath10k/wmi.h | 20 +++++++++-
3 files changed, 58 insertions(+), 23 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 5be6386ede8f..20a54d3bab6e 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -2468,7 +2468,7 @@ static void ath10k_peer_assoc_h_vht(struct ath10k *ar,
const u16 *vht_mcs_mask;
u8 ampdu_factor;
u8 max_nss, vht_mcs;
- int i;
+ int i, nss160;

if (WARN_ON(ath10k_mac_vif_chan(vif, &def)))
return;
@@ -2528,23 +2528,45 @@ static void ath10k_peer_assoc_h_vht(struct ath10k *ar,
__le16_to_cpu(vht_cap->vht_mcs.tx_highest);
arg->peer_vht_rates.tx_mcs_set = ath10k_peer_assoc_h_vht_limit(
__le16_to_cpu(vht_cap->vht_mcs.tx_mcs_map), vht_mcs_mask);
+ arg->peer_bw_rxnss_override = 0;
+ nss160 = 1; /* 1x1 default config for VHT160 */

- ath10k_dbg(ar, ATH10K_DBG_MAC, "mac vht peer %pM max_mpdu %d flags 0x%x\n",
- sta->addr, arg->peer_max_mpdu, arg->peer_flags);
+ /* only local 4x4 configuration do support 2x2 for VHT160,
+ * everything else must use 1x1
+ */

- if (arg->peer_vht_rates.rx_max_rate &&
- (sta->vht_cap.cap & IEEE80211_VHT_CAP_SUPP_CHAN_WIDTH_MASK)) {
- switch (arg->peer_vht_rates.rx_max_rate) {
- case 1560:
- /* Must be 2x2 at 160Mhz is all it can do. */
- arg->peer_bw_rxnss_override = 2;
- break;
- case 780:
- /* Can only do 1x1 at 160Mhz (Long Guard Interval) */
- arg->peer_bw_rxnss_override = 1;
- break;
- }
+ if (ar->cfg_rx_chainmask == 15)
+ nss160 = arg->peer_num_spatial_streams <= 2 ? 1 : 2;
+
+ /* if peer provides 1x1 nss160 information using max rate
+ * vht information, we reduce local nss160 to 1x1.
+ * consider that it has been observed that some client
+ * devices provide zero here, no matter which transmission
+ * rate is possible. in that case the local nss configuration
+ * will be used at maxmimum configuration possible. (see above)
+ */
+
+ if (arg->peer_vht_rates.rx_max_rate == 780)
+ nss160 = 1;
+
+ /* in case if peer is connected with vht160 or vht80+80,
+ * we need to properly adjust rxnss parameters otherwise
+ * firmware will raise a assert
+ */
+ switch (arg->peer_phymode) {
+ case MODE_11AC_VHT80_80:
+ arg->peer_bw_rxnss_override = BW_NSS_FWCONF_80_80(nss160);
+ /* fall through */
+ case MODE_11AC_VHT160:
+ arg->peer_bw_rxnss_override |= BW_NSS_FWCONF_160(nss160);
+ break;
+ default:
+ break;
}
+
+ ath10k_dbg(ar, ATH10K_DBG_MAC, "mac vht peer %pM max_mpdu %d flags 0x%x peer_bw_rxnss_override 0x%x\n",
+ sta->addr, arg->peer_max_mpdu, arg->peer_flags,
+ arg->peer_bw_rxnss_override);
}

static void ath10k_peer_assoc_h_qos(struct ath10k *ar,
@@ -2696,9 +2718,9 @@ static int ath10k_peer_assoc_prepare(struct ath10k *ar,
ath10k_peer_assoc_h_crypto(ar, vif, sta, arg);
ath10k_peer_assoc_h_rates(ar, vif, sta, arg);
ath10k_peer_assoc_h_ht(ar, vif, sta, arg);
+ ath10k_peer_assoc_h_phymode(ar, vif, sta, arg);
ath10k_peer_assoc_h_vht(ar, vif, sta, arg);
ath10k_peer_assoc_h_qos(ar, vif, sta, arg);
- ath10k_peer_assoc_h_phymode(ar, vif, sta, arg);

return 0;
}
diff --git a/drivers/net/wireless/ath/ath10k/wmi.c b/drivers/net/wireless/ath/ath10k/wmi.c
index 2c36256a441d..3797dca317ff 100644
--- a/drivers/net/wireless/ath/ath10k/wmi.c
+++ b/drivers/net/wireless/ath/ath10k/wmi.c
@@ -7211,12 +7211,7 @@ ath10k_wmi_peer_assoc_fill_10_4(struct ath10k *ar, void *buf,
struct wmi_10_4_peer_assoc_complete_cmd *cmd = buf;

ath10k_wmi_peer_assoc_fill_10_2(ar, buf, arg);
- if (arg->peer_bw_rxnss_override)
- cmd->peer_bw_rxnss_override =
- __cpu_to_le32((arg->peer_bw_rxnss_override - 1) |
- BIT(PEER_BW_RXNSS_OVERRIDE_OFFSET));
- else
- cmd->peer_bw_rxnss_override = 0;
+ cmd->peer_bw_rxnss_override = __cpu_to_le32(arg->peer_bw_rxnss_override);
}

static int
diff --git a/drivers/net/wireless/ath/ath10k/wmi.h b/drivers/net/wireless/ath/ath10k/wmi.h
index 46ae19bb2c92..92b9263ff4f6 100644
--- a/drivers/net/wireless/ath/ath10k/wmi.h
+++ b/drivers/net/wireless/ath/ath10k/wmi.h
@@ -6380,7 +6380,25 @@ struct wmi_10_2_peer_assoc_complete_cmd {
__le32 info0; /* WMI_PEER_ASSOC_INFO0_ */
} __packed;

-#define PEER_BW_RXNSS_OVERRIDE_OFFSET 31
+#define BW_NSS_FWCONF_MAP_ENABLE BIT(31)
+#define BW_NSS_FWCONF_MAP_160MHZ_S (0)
+#define BW_NSS_FWCONF_MAP_160MHZ_M (0x00000007)
+#define BW_NSS_FWCONF_MAP_80_80MHZ_S (3)
+#define BW_NSS_FWCONF_MAP_80_80MHZ_M (0x00000038)
+#define BW_NSS_FWCONF_MAP_M (0x0000003F)
+
+#define GET_BW_NSS_FWCONF_160(x) ((((x) & BW_NSS_FWCONF_MAP_160MHZ_M) \
+ >> BW_NSS_FWCONF_MAP_160MHZ_S) + 1)
+#define GET_BW_NSS_FWCONF_80_80(x) ((((x) & BW_NSS_FWCONF_MAP_80_80MHZ_M) \
+ >> BW_NSS_FWCONF_MAP_80_80MHZ_S) + 1)
+
+/* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
+#define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
+ (((x - 1) << BW_NSS_FWCONF_MAP_160MHZ_S) \
+ & BW_NSS_FWCONF_MAP_160MHZ_M))
+#define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
+ (((x - 1) << BW_NSS_FWCONF_MAP_80_80MHZ_S) \
+ & BW_NSS_FWCONF_MAP_80_80MHZ_M))

struct wmi_10_4_peer_assoc_complete_cmd {
struct wmi_10_2_peer_assoc_complete_cmd cmd;
--
2.17.1


2018-07-04 10:48:05

by Sebastian Gottschall

[permalink] [raw]
Subject: Re: [PATCH v8] ath10k: fix crash in recent 3.5.3 9984 firmware due wrong handling of peer_bw_rxnss_override parameter



Am 04.07.2018 um 12:40 schrieb Sven Eckelmann:
> On Mittwoch, 4. Juli 2018 11:29:53 CEST [email protected] wrote:
>> +/* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
>> +#define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
>> + (((x - 1) << BW_NSS_FWCONF_MAP_160MHZ_S) \
>> + & BW_NSS_FWCONF_MAP_160MHZ_M))
>> +#define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
>> + (((x - 1) << BW_NSS_FWCONF_MAP_80_80MHZ_S) \
>> + & BW_NSS_FWCONF_MAP_80_80MHZ_M))
> Please check the shift and mask macros in core.h
>
>
> #define BW_NSS_FWCONF_MAP_160MHZ_LSB 0
> #define BW_NSS_FWCONF_MAP_160MHZ_MASK 0x00000007
> #define BW_NSS_FWCONF_MAP_80_80MHZ_LSB 3
> #define BW_NSS_FWCONF_MAP_80_80MHZ_MASK 0x00000038
> #define BW_NSS_FWCONF_MAP_MASK 0x0000003F
>
> #define GET_BW_NSS_FWCONF_160(x) (MS((x), BW_NSS_FWCONF_MAP_160MHZ) + 1)
> #define GET_BW_NSS_FWCONF_80_80(x) (MS((x), BW_NSS_FWCONF_MAP_80_80MHZ_M) + 1)
>
> /* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
> #define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
> (SM((x) - 1, BW_NSS_FWCONF_MAP_160MHZ))
> #define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
> (SM((x) - 1, BW_NSS_FWCONF_MAP_80_80MHZ))
>
> (these snippets were not yet tested and most likely need further adjustments)
>
>
> Kind regards,
> Sven
can you explain your request? the macros are unchanged and just
reformated for maximum line length restriction. the values have been
taken from QCA's wmi header

Sebastian

2018-07-04 10:40:42

by Sven Eckelmann

[permalink] [raw]
Subject: Re: [PATCH v8] ath10k: fix crash in recent 3.5.3 9984 firmware due wrong handling of peer_bw_rxnss_override parameter

On Mittwoch, 4. Juli 2018 11:29:53 CEST [email protected] wrote:
> +/* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
> +#define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
> + (((x - 1) << BW_NSS_FWCONF_MAP_160MHZ_S) \
> + & BW_NSS_FWCONF_MAP_160MHZ_M))
> +#define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
> + (((x - 1) << BW_NSS_FWCONF_MAP_80_80MHZ_S) \
> + & BW_NSS_FWCONF_MAP_80_80MHZ_M))

Please check the shift and mask macros in core.h


#define BW_NSS_FWCONF_MAP_160MHZ_LSB 0
#define BW_NSS_FWCONF_MAP_160MHZ_MASK 0x00000007
#define BW_NSS_FWCONF_MAP_80_80MHZ_LSB 3
#define BW_NSS_FWCONF_MAP_80_80MHZ_MASK 0x00000038
#define BW_NSS_FWCONF_MAP_MASK 0x0000003F

#define GET_BW_NSS_FWCONF_160(x) (MS((x), BW_NSS_FWCONF_MAP_160MHZ) + 1)
#define GET_BW_NSS_FWCONF_80_80(x) (MS((x), BW_NSS_FWCONF_MAP_80_80MHZ_M) + 1)

/* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
#define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
(SM((x) - 1, BW_NSS_FWCONF_MAP_160MHZ))
#define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
(SM((x) - 1, BW_NSS_FWCONF_MAP_80_80MHZ))

(these snippets were not yet tested and most likely need further adjustments)


Kind regards,
Sven


Attachments:
signature.asc (833.00 B)
This is a digitally signed message part.

2018-07-04 11:05:19

by Sebastian Gottschall

[permalink] [raw]
Subject: Re: [PATCH v8] ath10k: fix crash in recent 3.5.3 9984 firmware due wrong handling of peer_bw_rxnss_override parameter

the values within the patch are named

#define BW_NSS_FWCONF_MAP_ENABLE???????????? BIT(31)
#define BW_NSS_FWCONF_MAP_160MHZ_S?????????? (0)
#define BW_NSS_FWCONF_MAP_160MHZ_M?????????? (0x00000007)
#define BW_NSS_FWCONF_MAP_80_80MHZ_S???????? (3)
#define BW_NSS_FWCONF_MAP_80_80MHZ_M???????? (0x00000038)
#define BW_NSS_FWCONF_MAP_M????????????????? (0x0000003F)


so your comment doesnt seem to belong to the patch. maybe the openwrt
guys renamed some values in the patch. my one is based on my original
sources


Am 04.07.2018 um 12:48 schrieb Sebastian Gottschall:
>
>
> Am 04.07.2018 um 12:40 schrieb Sven Eckelmann:
>> On Mittwoch, 4. Juli 2018 11:29:53 CEST [email protected] wrote:
>>> +/* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
>>> +#define BW_NSS_FWCONF_160(x) (BW_NSS_FWCONF_MAP_ENABLE | \
>>> +?????????????????????????????????????? (((x - 1) <<
>>> BW_NSS_FWCONF_MAP_160MHZ_S) \
>>> +?????????????????????????????????????? & BW_NSS_FWCONF_MAP_160MHZ_M))
>>> +#define BW_NSS_FWCONF_80_80(x) (BW_NSS_FWCONF_MAP_ENABLE | \
>>> +?????????????????????????????????????? (((x - 1) <<
>>> BW_NSS_FWCONF_MAP_80_80MHZ_S) \
>>> +?????????????????????????????????????? &
>>> BW_NSS_FWCONF_MAP_80_80MHZ_M))
>> Please check the shift and mask macros in core.h
>>
>>
>> #define BW_NSS_FWCONF_MAP_160MHZ_LSB?????????? 0
>> #define BW_NSS_FWCONF_MAP_160MHZ_MASK????????? 0x00000007
>> #define BW_NSS_FWCONF_MAP_80_80MHZ_LSB???????? 3
>> #define BW_NSS_FWCONF_MAP_80_80MHZ_MASK??????? 0x00000038
>> #define BW_NSS_FWCONF_MAP_MASK???????????????? 0x0000003F
>>
>> #define GET_BW_NSS_FWCONF_160(x)???? (MS((x),
>> BW_NSS_FWCONF_MAP_160MHZ) + 1)
>> #define GET_BW_NSS_FWCONF_80_80(x)?? (MS((x),
>> BW_NSS_FWCONF_MAP_80_80MHZ_M) + 1)
>>
>> /* Values defined to set 160 MHz Bandwidth NSS Mapping into FW*/
>> #define BW_NSS_FWCONF_160(x)???????? (BW_NSS_FWCONF_MAP_ENABLE | \
>> ?????????????????????????????????????? (SM((x) - 1,
>> BW_NSS_FWCONF_MAP_160MHZ))
>> #define BW_NSS_FWCONF_80_80(x)?????? (BW_NSS_FWCONF_MAP_ENABLE | \
>> ?????????????????????????????????????? (SM((x) - 1,
>> BW_NSS_FWCONF_MAP_80_80MHZ))
>>
>> (these snippets were not yet tested and most likely need further
>> adjustments)
>>
>>
>> Kind regards,
>> ????Sven
> can you explain your request? the macros are unchanged and just
> reformated for maximum line length restriction. the values have been
> taken from QCA's wmi header
>
> Sebastian
>
>

2018-07-04 11:12:52

by Sven Eckelmann

[permalink] [raw]
Subject: Re: [PATCH v8] ath10k: fix crash in recent 3.5.3 9984 firmware due wrong handling of peer_bw_rxnss_override parameter

On Mittwoch, 4. Juli 2018 12:48:06 CEST Sebastian Gottschall wrote:
[...]
> can you explain your request? the macros are unchanged and just
> reformated for maximum line length restriction. the values have been
> taken from QCA's wmi header

You use long shift and mask statements in your patch. But you didn't use the
macros which are there to shorten them. So please check the macros SM and MS
from core.h and adjust your code accordingly. The code which I've pasted does
this (with an unintentionally added a copy and paste error).

Kind regards,
Sven


Attachments:
signature.asc (833.00 B)
This is a digitally signed message part.