On 24 March 2016 at 13:23, Mohammed Shafi Shajakhan
<mohammed@xxxxxxxxxxxxxx> wrote:
> On Thu, Mar 24, 2016 at 08:49:12AM +0100, Michal Kazior wrote:
>> On 24 March 2016 at 08:19, Mohammed Shafi Shajakhan
>> <mohammed@xxxxxxxxxxxxxx> wrote:
>> > Hi Michal,
>> >
>> > On Wed, Mar 16, 2016 at 11:17:57AM +0100, Michal Kazior wrote:
>> >> The rate control is offloaded by firmware so it's
>> >> challanging to provide expected throughput value
>> >> for given station.
>> >>
>> >> This approach is naive as it reports last tx rate
>> >> used for given station as provided by firmware
>> >> stat event.
>> >>
>> >> This should be sufficient for airtime estimation
>> >> used for fq-codel-in-mac80211 tx scheduling
>> >> purposes now.
>> >>
>> >> This patch uses a very hacky way to get the stats.
>> >> This is sufficient for proof-of-concept but must
>> >> be cleaned up properly eventually.
>> >>
>> >> Signed-off-by: Michal Kazior <michal.kazior@xxxxxxxxx>
>> >> ---
>> >> drivers/net/wireless/ath/ath10k/core.h | 5 +++
>> >> drivers/net/wireless/ath/ath10k/debug.c | 61 +++++++++++++++++++++++++++++----
>> >> drivers/net/wireless/ath/ath10k/mac.c | 26 ++++++++------
>> >> drivers/net/wireless/ath/ath10k/wmi.h | 2 +-
>> >> 4 files changed, 76 insertions(+), 18 deletions(-)
>> >>
>> >> diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
>> >> index 23ba03fb7a5f..3f76669d44cf 100644
>> >> --- a/drivers/net/wireless/ath/ath10k/core.h
>> >> +++ b/drivers/net/wireless/ath/ath10k/core.h
>> >> @@ -331,6 +331,9 @@ struct ath10k_sta {
>> >> /* protected by conf_mutex */
>> >> bool aggr_mode;
>> >> u64 rx_duration;
>> >> +
>> >> + u32 tx_rate_kbps;
>> >> + u32 rx_rate_kbps;
>> >> #endif
>> >> };
>> >>
>> >> @@ -372,6 +375,8 @@ struct ath10k_vif {
>> >> s8 def_wep_key_idx;
>> >>
>> >> u16 tx_seq_no;
>> >> + u32 tx_rate_kbps;
>> >> + u32 rx_rate_kbps;
>> >>
>> >> union {
>> >> struct {
>> >> diff --git a/drivers/net/wireless/ath/ath10k/debug.c b/drivers/net/wireless/ath/ath10k/debug.c
>> >> index 076d29b53ddf..cc7ebf04ae00 100644
>> >> --- a/drivers/net/wireless/ath/ath10k/debug.c
>> >> +++ b/drivers/net/wireless/ath/ath10k/debug.c
>> >> @@ -316,6 +316,58 @@ static void ath10k_debug_fw_stats_reset(struct ath10k *ar)
>> >> spin_unlock_bh(&ar->data_lock);
>> >> }
>> >>
>> >> +static void ath10k_mac_update_txrx_rate_iter(void *data,
>> >> + u8 *mac,
>> >> + struct ieee80211_vif *vif)
>> >> +{
>> >> + struct ath10k_fw_stats_peer *peer = data;
>> >> + struct ath10k_vif *arvif;
>> >> +
>> >> + if (memcmp(vif->addr, peer->peer_macaddr, ETH_ALEN))
>> >> + return;
>> >> +
>> >> + arvif = (void *)vif->drv_priv;
>> >> + arvif->tx_rate_kbps = peer->peer_tx_rate;
>> >> + arvif->rx_rate_kbps = peer->peer_rx_rate;
>> >> +}
>> >> +
>> >> +static void ath10k_mac_update_txrx_rate(struct ath10k *ar,
>> >> + struct ath10k_fw_stats *stats)
>> >> +{
>> >> + struct ieee80211_hw *hw = ar->hw;
>> >> + struct ath10k_fw_stats_peer *peer;
>> >> + struct ath10k_sta *arsta;
>> >> + struct ieee80211_sta *sta;
>> >> + const u8 *localaddr = NULL;
>> >> +
>> >> + rcu_read_lock();
>> >> +
>> >> + list_for_each_entry(peer, &stats->peers, list) {
>> >> + /* This doesn't account for multiple STA connected on different
>> >> + * vifs. Unfortunately there's no way to derive that from the available
>> >> + * information.
>> >> + */
>> >> + sta = ieee80211_find_sta_by_ifaddr(hw,
>> >> + peer->peer_macaddr,
>> >> + localaddr);
>> >> + if (!sta) {
>> >> + /* This tries to update multicast rates */
>> >> + ieee80211_iterate_active_interfaces_atomic(
>> >> + hw,
>> >> + IEEE80211_IFACE_ITER_NORMAL,
>> >> + ath10k_mac_update_txrx_rate_iter,
>> >> + peer);
>> >> + continue;
>> >> + }
>> >> +
>> >> + arsta = (void *)sta->drv_priv;
>> >> + arsta->tx_rate_kbps = peer->peer_tx_rate;
>> >> + arsta->rx_rate_kbps = peer->peer_rx_rate;
>> >> + }
>> >> +
>> >> + rcu_read_unlock();
>> >> +}
>> >> +
>> >> void ath10k_debug_fw_stats_process(struct ath10k *ar, struct sk_buff *skb)
>> >> {
>> >> struct ath10k_fw_stats stats = {};
>> >> @@ -335,6 +387,8 @@ void ath10k_debug_fw_stats_process(struct ath10k *ar, struct sk_buff *skb)
>> >> goto free;
>> >> }
>> >>
>> >> + ath10k_mac_update_txrx_rate(ar, &stats);
>> >> +
>> >> /* Stat data may exceed htc-wmi buffer limit. In such case firmware
>> >> * splits the stats data and delivers it in a ping-pong fashion of
>> >> * request cmd-update event.
>> >> @@ -351,13 +405,6 @@ void ath10k_debug_fw_stats_process(struct ath10k *ar, struct sk_buff *skb)
>> >> if (peer_stats_svc)
>> >> ath10k_sta_update_rx_duration(ar, &stats.peers);
>> >>
>> >> - if (ar->debug.fw_stats_done) {
>> >> - if (!peer_stats_svc)
>> >> - ath10k_warn(ar, "received unsolicited stats update event\n");
>> >> -
>> >> - goto free;
>> >> - }
>> >> -
>> >
>> > [shafi] As you had suggested previously, should we completely clean up this ping
>> > - pong response approach for f/w stats, (or) this should be retained to support
>> > backward compatibility and also for supporting ping - pong response when user
>> > cats for fw-stats (via debugfs) (i did see in the commit message this needs to
>> > be cleaned up)
>>
>> I think it makes sense to remove the ping-pong logic and rely on
>> periodic updates alone, including fw_stats and ethstats handling.
>>
>>
>> >> - if (test_bit(WMI_SERVICE_PEER_STATS, ar->wmi.svc_map)) {
>> >> - param = ar->wmi.pdev_param->peer_stats_update_period;
>> >> - ret = ath10k_wmi_pdev_set_param(ar, param,
>> >> - PEER_DEFAULT_STATS_UPDATE_PERIOD);
>> >> - if (ret) {
>> >> - ath10k_warn(ar,
>> >> - "failed to set peer stats period : %d\n",
>> >> - ret);
>> >> - goto err_core_stop;
>> >> - }
>> >> + param = ar->wmi.pdev_param->peer_stats_update_period;
>> >> + ret = ath10k_wmi_pdev_set_param(ar, param,
>> >> + PEER_DEFAULT_STATS_UPDATE_PERIOD);
>> >> + if (ret) {
>> >> + ath10k_warn(ar,
>> >> + "failed to set peer stats period : %d\n",
>> >> + ret);
>> >> + goto err_core_stop;
>> >> }
>> >
>> > [shafi] If i am correct this change requires 'PEER_STATS' to be enabled by
>> > default.
>>
>> No, it does not. Periodic stats have been available since forever.
>
> [shafi] Michal, sorry i was talking about enabling WMI_PEER_STATS feature for
> 10.2, and we have a patch pushed recently to reduce the number of peers if
> 'WMI_PEER_STATS' feature is enabled(avoiding f/w crash due to memory
> constraints) . But this patch requires the feature to be
> turned ON always (with periodic stats update as well for evey 100ms). Please
> correct me if my understanding is wrong.
Periodic stats and extended stats are two separate things.
WMI_PEER_STATS is a feature which prompts firmware to gather more
statistics and then report them to host via stat update event and
includes, e.g. rx_duration. Due to how rx_duration was designed in
firmware it needs to be combined with reading out the stats often to
make it usable. Periodic stats were used instead of explicit
ping-pong.
Periodic stats is just one of two ways (the other being ping-pong)
asking firmware for stat update event. It has been in firmware since
forever.
Michał
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
