2024-02-08 19:54:45

by Bitterblue Smith

[permalink] [raw]
Subject: [PATCH] wifi: rtlwifi: rtl8192cu: Fix TX aggregation

rtl8192cu is checking rtl_mac.tids when deciding if it should enable
aggregation. This is wrong because rtl_mac.tids is not initialised
anywhere. Check rtl_sta_info.tids instead, which is initialised.

Also, when enabling aggregation also enable RTS. The vendor driver does
this, my router does this. It seems like the thing to do.

Also also, it seems right to set the AMPDU density only when enabling
aggregation.

Also also also, delete the unused member rtl_mac.tids and the unused
macros RTL_AGG_ON and RTL_AGG_OFF.

Naturally, with working AMPDU the download/upload speeds are better.
Before: 59/32 Mbps.
After: 68/46 Mbps.

Signed-off-by: Bitterblue Smith <[email protected]>
---
.../wireless/realtek/rtlwifi/rtl8192cu/trx.c | 27 ++++++++++---------
.../wireless/realtek/rtlwifi/rtl8192cu/trx.h | 2 --
drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 ---
3 files changed, 15 insertions(+), 17 deletions(-)

diff --git a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
index e5c81c1c63c0..cbbd1dab8af0 100644
--- a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
+++ b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
@@ -475,8 +475,9 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
struct rtl_priv *rtlpriv = rtl_priv(hw);
struct rtl_mac *mac = rtl_mac(rtl_priv(hw));
struct rtl_ps_ctl *ppsc = rtl_psc(rtl_priv(hw));
- u8 *qc = ieee80211_get_qos_ctl(hdr);
- u8 tid = qc[0] & IEEE80211_QOS_CTL_TID_MASK;
+ struct rtl_sta_info *sta_entry;
+ u8 agg_state = RTL_AGG_STOP;
+ u8 ampdu_density = 0;
u16 seq_number;
__le16 fc = hdr->frame_control;
u8 rate_flag = info->control.rates[0].flags;
@@ -498,10 +499,20 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
set_tx_desc_tx_rate(txdesc, tcb_desc->hw_rate);
if (tcb_desc->use_shortgi || tcb_desc->use_shortpreamble)
set_tx_desc_data_shortgi(txdesc, 1);
- if (mac->tids[tid].agg.agg_state == RTL_AGG_ON &&
- info->flags & IEEE80211_TX_CTL_AMPDU) {
+
+ if (sta) {
+ sta_entry = (struct rtl_sta_info *)sta->drv_priv;
+ agg_state = sta_entry->tids[ieee80211_get_tid(hdr)].agg.agg_state;
+ ampdu_density = sta->deflink.ht_cap.ampdu_density;
+ }
+
+ if (agg_state == RTL_AGG_OPERATIONAL &&
+ info->flags & IEEE80211_TX_CTL_AMPDU) {
set_tx_desc_agg_enable(txdesc, 1);
set_tx_desc_max_agg_num(txdesc, 0x14);
+ set_tx_desc_ampdu_density(txdesc, ampdu_density);
+ tcb_desc->rts_enable = 1;
+ tcb_desc->rts_rate = DESC_RATE24M;
} else {
set_tx_desc_agg_break(txdesc, 1);
}
@@ -536,14 +547,6 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
set_tx_desc_data_bw(txdesc, 0);
set_tx_desc_data_sc(txdesc, 0);
}
- rcu_read_lock();
- sta = ieee80211_find_sta(mac->vif, mac->bssid);
- if (sta) {
- u8 ampdu_density = sta->deflink.ht_cap.ampdu_density;
-
- set_tx_desc_ampdu_density(txdesc, ampdu_density);
- }
- rcu_read_unlock();
if (info->control.hw_key) {
struct ieee80211_key_conf *keyconf = info->control.hw_key;

diff --git a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.h b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.h
index 5f81cab205cc..caeebe4480ff 100644
--- a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.h
+++ b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.h
@@ -11,8 +11,6 @@
#define RTL92C_SIZE_MAX_RX_BUFFER 15360 /* 8192 */
#define RX_DRV_INFO_SIZE_UNIT 8

-#define RTL_AGG_ON 1
-
enum usb_rx_agg_mode {
USB_RX_AGG_DISABLE,
USB_RX_AGG_DMA,
diff --git a/drivers/net/wireless/realtek/rtlwifi/wifi.h b/drivers/net/wireless/realtek/rtlwifi/wifi.h
index 3821f6e31447..b65d54115eac 100644
--- a/drivers/net/wireless/realtek/rtlwifi/wifi.h
+++ b/drivers/net/wireless/realtek/rtlwifi/wifi.h
@@ -1397,8 +1397,6 @@ struct rtl_phy {
#define RTL_AGG_PROGRESS 1
#define RTL_AGG_START 2
#define RTL_AGG_OPERATIONAL 3
-#define RTL_AGG_OFF 0
-#define RTL_AGG_ON 1
#define RTL_RX_AGG_START 1
#define RTL_RX_AGG_STOP 0
#define RTL_AGG_EMPTYING_HW_QUEUE_ADDBA 2
@@ -1473,7 +1471,6 @@ struct rtl_mac {
enum nl80211_iftype opmode;

/*Probe Beacon management */
- struct rtl_tid_data tids[MAX_TID_COUNT];
enum rtl_link_state link_state;

int n_channels;
--
2.43.0


2024-02-12 12:50:29

by Ping-Ke Shih

[permalink] [raw]
Subject: Re: [PATCH] wifi: rtlwifi: rtl8192cu: Fix TX aggregation

On Thu, 2024-02-08 at 21:54 +0200, Bitterblue Smith wrote:
>
> rtl8192cu is checking rtl_mac.tids when deciding if it should enable
> aggregation. This is wrong because rtl_mac.tids is not initialised
> anywhere. Check rtl_sta_info.tids instead, which is initialised.
>
> Also, when enabling aggregation also enable RTS. The vendor driver does
> this, my router does this. It seems like the thing to do.
>
> Also also, it seems right to set the AMPDU density only when enabling
> aggregation.
>
> Also also also, delete the unused member rtl_mac.tids and the unused
> macros RTL_AGG_ON and RTL_AGG_OFF.
>
> Naturally, with working AMPDU the download/upload speeds are better.
> Before: 59/32 Mbps.
> After: 68/46 Mbps.
>
> Signed-off-by: Bitterblue Smith <[email protected]>
> ---
> .../wireless/realtek/rtlwifi/rtl8192cu/trx.c | 27 ++++++++++---------
> .../wireless/realtek/rtlwifi/rtl8192cu/trx.h | 2 --
> drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 ---
> 3 files changed, 15 insertions(+), 17 deletions(-)
>
> diff --git a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
> b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
> index e5c81c1c63c0..cbbd1dab8af0 100644
> --- a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
> +++ b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
> @@ -475,8 +475,9 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
> struct rtl_priv *rtlpriv = rtl_priv(hw);
> struct rtl_mac *mac = rtl_mac(rtl_priv(hw));
> struct rtl_ps_ctl *ppsc = rtl_psc(rtl_priv(hw));
> - u8 *qc = ieee80211_get_qos_ctl(hdr);
> - u8 tid = qc[0] & IEEE80211_QOS_CTL_TID_MASK;
> + struct rtl_sta_info *sta_entry;
> + u8 agg_state = RTL_AGG_STOP;
> + u8 ampdu_density = 0;
> u16 seq_number;
> __le16 fc = hdr->frame_control;
> u8 rate_flag = info->control.rates[0].flags;
> @@ -498,10 +499,20 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
> set_tx_desc_tx_rate(txdesc, tcb_desc->hw_rate);
> if (tcb_desc->use_shortgi || tcb_desc->use_shortpreamble)
> set_tx_desc_data_shortgi(txdesc, 1);
> - if (mac->tids[tid].agg.agg_state == RTL_AGG_ON &&
> - info->flags & IEEE80211_TX_CTL_AMPDU) {
> +
> + if (sta) {
> + sta_entry = (struct rtl_sta_info *)sta->drv_priv;

nit: It would be better to be separated statement of ieee80211_get_tid(hdr)

tid = ieee80211_get_tid(hdr);
agg_state = sta_entry->tids[tid].agg.agg_state;


> + agg_state = sta_entry->tids[ieee80211_get_tid(hdr)].agg.agg_state;
> + ampdu_density = sta->deflink.ht_cap.ampdu_density;
> + }
> +
> + if (agg_state == RTL_AGG_OPERATIONAL &&
> + info->flags & IEEE80211_TX_CTL_AMPDU) {
> set_tx_desc_agg_enable(txdesc, 1);
> set_tx_desc_max_agg_num(txdesc, 0x14);

Have you tried larger number of this?
As IEEE80211_MAX_AMPDU_BUF_HT is 0x40, set 0x3f here to get higher TX
throughput if USB speed can afford.

You can reference an equation of rtw89:

ampdu_num = (u8)((rtwsta->ampdu_params[tid].agg_num ?
rtwsta->ampdu_params[tid].agg_num :
4 << sta->deflink.ht_cap.ampdu_factor) - 1);

This is new for rtlwifi, so we can have this later.


[...]



2024-02-13 14:32:46

by Bitterblue Smith

[permalink] [raw]
Subject: Re: [PATCH] wifi: rtlwifi: rtl8192cu: Fix TX aggregation

On 12/02/2024 14:50, Ping-Ke Shih wrote:
> On Thu, 2024-02-08 at 21:54 +0200, Bitterblue Smith wrote:
>>
>> rtl8192cu is checking rtl_mac.tids when deciding if it should enable
>> aggregation. This is wrong because rtl_mac.tids is not initialised
>> anywhere. Check rtl_sta_info.tids instead, which is initialised.
>>
>> Also, when enabling aggregation also enable RTS. The vendor driver does
>> this, my router does this. It seems like the thing to do.
>>
>> Also also, it seems right to set the AMPDU density only when enabling
>> aggregation.
>>
>> Also also also, delete the unused member rtl_mac.tids and the unused
>> macros RTL_AGG_ON and RTL_AGG_OFF.
>>
>> Naturally, with working AMPDU the download/upload speeds are better.
>> Before: 59/32 Mbps.
>> After: 68/46 Mbps.
>>
>> Signed-off-by: Bitterblue Smith <[email protected]>
>> ---
>> .../wireless/realtek/rtlwifi/rtl8192cu/trx.c | 27 ++++++++++---------
>> .../wireless/realtek/rtlwifi/rtl8192cu/trx.h | 2 --
>> drivers/net/wireless/realtek/rtlwifi/wifi.h | 3 ---
>> 3 files changed, 15 insertions(+), 17 deletions(-)
>>
>> diff --git a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
>> b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
>> index e5c81c1c63c0..cbbd1dab8af0 100644
>> --- a/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
>> +++ b/drivers/net/wireless/realtek/rtlwifi/rtl8192cu/trx.c
>> @@ -475,8 +475,9 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
>> struct rtl_priv *rtlpriv = rtl_priv(hw);
>> struct rtl_mac *mac = rtl_mac(rtl_priv(hw));
>> struct rtl_ps_ctl *ppsc = rtl_psc(rtl_priv(hw));
>> - u8 *qc = ieee80211_get_qos_ctl(hdr);
>> - u8 tid = qc[0] & IEEE80211_QOS_CTL_TID_MASK;
>> + struct rtl_sta_info *sta_entry;
>> + u8 agg_state = RTL_AGG_STOP;
>> + u8 ampdu_density = 0;
>> u16 seq_number;
>> __le16 fc = hdr->frame_control;
>> u8 rate_flag = info->control.rates[0].flags;
>> @@ -498,10 +499,20 @@ void rtl92cu_tx_fill_desc(struct ieee80211_hw *hw,
>> set_tx_desc_tx_rate(txdesc, tcb_desc->hw_rate);
>> if (tcb_desc->use_shortgi || tcb_desc->use_shortpreamble)
>> set_tx_desc_data_shortgi(txdesc, 1);
>> - if (mac->tids[tid].agg.agg_state == RTL_AGG_ON &&
>> - info->flags & IEEE80211_TX_CTL_AMPDU) {
>> +
>> + if (sta) {
>> + sta_entry = (struct rtl_sta_info *)sta->drv_priv;
>
> nit: It would be better to be separated statement of ieee80211_get_tid(hdr)
>
> tid = ieee80211_get_tid(hdr);
> agg_state = sta_entry->tids[tid].agg.agg_state;
>

All right.

> >> + agg_state = sta_entry->tids[ieee80211_get_tid(hdr)].agg.agg_state;
>> + ampdu_density = sta->deflink.ht_cap.ampdu_density;
>> + }
>> +
>> + if (agg_state == RTL_AGG_OPERATIONAL &&
>> + info->flags & IEEE80211_TX_CTL_AMPDU) {
>> set_tx_desc_agg_enable(txdesc, 1);
>> set_tx_desc_max_agg_num(txdesc, 0x14);
>
> Have you tried larger number of this?
> As IEEE80211_MAX_AMPDU_BUF_HT is 0x40, set 0x3f here to get higher TX
> throughput if USB speed can afford.
>

With this patch, the upload speed is 46 Mbps. But that's just
because my device is detected as 1T2R and it's not allowed to use
rates above MCS7. When I fix that problem the upload speed is 82 Mbps
in the same conditions, about 3 meters from the router.

If I then move the device just 60 cm away, the upload speed is 92
Mbps.

I tried 0x1f now (this field is only 5 bits). The upload speed doesn't
get better. Maybe it gets a bit worse.

> You can reference an equation of rtw89:
>
> ampdu_num = (u8)((rtwsta->ampdu_params[tid].agg_num ?
> rtwsta->ampdu_params[tid].agg_num :
> 4 << sta->deflink.ht_cap.ampdu_factor) - 1);
>
> This is new for rtlwifi, so we can have this later.
>
>
> [...]
>
>
>


2024-02-15 07:25:40

by Ping-Ke Shih

[permalink] [raw]
Subject: RE: [PATCH] wifi: rtlwifi: rtl8192cu: Fix TX aggregation



> -----Original Message-----
> From: Bitterblue Smith <[email protected]>
> Sent: Tuesday, February 13, 2024 10:33 PM
> To: Ping-Ke Shih <[email protected]>; [email protected]
> Cc: [email protected]
> Subject: Re: [PATCH] wifi: rtlwifi: rtl8192cu: Fix TX aggregation
>
> On 12/02/2024 14:50, Ping-Ke Shih wrote:
> > On Thu, 2024-02-08 at 21:54 +0200, Bitterblue Smith wrote:
> >> set_tx_desc_max_agg_num(txdesc, 0x14);
> >
> > Have you tried larger number of this?
> > As IEEE80211_MAX_AMPDU_BUF_HT is 0x40, set 0x3f here to get higher TX
> > throughput if USB speed can afford.
> >
>
> With this patch, the upload speed is 46 Mbps. But that's just
> because my device is detected as 1T2R and it's not allowed to use
> rates above MCS7. When I fix that problem the upload speed is 82 Mbps
> in the same conditions, about 3 meters from the router.
>
> If I then move the device just 60 cm away, the upload speed is 92
> Mbps.
>
> I tried 0x1f now (this field is only 5 bits). The upload speed doesn't
> get better. Maybe it gets a bit worse.

It could have more interference with larger aggregation number. Let's keep
it as it was. Thanks for the try.

Ping-Ke