Subject: [PATCH v2 10/11] ath10k: keep track of queue depth per txq

This will be necessary for later.

Signed-off-by: Michal Kazior <[email protected]>
---
drivers/net/wireless/ath/ath10k/core.h | 1 +
drivers/net/wireless/ath/ath10k/mac.c | 5 +++++
drivers/net/wireless/ath/ath10k/txrx.c | 7 +++++++
3 files changed, 13 insertions(+)

diff --git a/drivers/net/wireless/ath/ath10k/core.h b/drivers/net/wireless/ath/ath10k/core.h
index 89f789f3e5b4..926ecb2244a5 100644
--- a/drivers/net/wireless/ath/ath10k/core.h
+++ b/drivers/net/wireless/ath/ath10k/core.h
@@ -311,6 +311,7 @@ struct ath10k_peer {

struct ath10k_txq {
struct list_head list;
+ unsigned long num_fw_queued;
};

struct ath10k_sta {
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 8d02d53fdc2c..5bf614f1f75a 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -3669,6 +3669,7 @@ static int ath10k_mac_tx_push_txq(struct ieee80211_hw *hw,
const bool is_presp = false;
struct ath10k *ar = hw->priv;
struct ath10k_htt *htt = &ar->htt;
+ struct ath10k_txq *artxq = (void *)txq->drv_priv;
struct ieee80211_vif *vif = txq->vif;
struct ieee80211_sta *sta = txq->sta;
enum ath10k_hw_txrx_mode txmode;
@@ -3708,6 +3709,10 @@ static int ath10k_mac_tx_push_txq(struct ieee80211_hw *hw,
return ret;
}

+ spin_lock_bh(&ar->htt.tx_lock);
+ artxq->num_fw_queued++;
+ spin_unlock_bh(&ar->htt.tx_lock);
+
return 0;
}

diff --git a/drivers/net/wireless/ath/ath10k/txrx.c b/drivers/net/wireless/ath/ath10k/txrx.c
index 202e5192235b..ea4d3000c8c3 100644
--- a/drivers/net/wireless/ath/ath10k/txrx.c
+++ b/drivers/net/wireless/ath/ath10k/txrx.c
@@ -55,7 +55,9 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
struct ath10k *ar = htt->ar;
struct device *dev = ar->dev;
struct ieee80211_tx_info *info;
+ struct ieee80211_txq *txq;
struct ath10k_skb_cb *skb_cb;
+ struct ath10k_txq *artxq;
struct sk_buff *msdu;
bool limit_mgmt_desc = false;

@@ -80,11 +82,16 @@ void ath10k_txrx_tx_unref(struct ath10k_htt *htt,
}

skb_cb = ATH10K_SKB_CB(msdu);
+ txq = skb_cb->txq;
+ artxq = (void *)txq->drv_priv;

if (unlikely(skb_cb->flags & ATH10K_SKB_F_MGMT) &&
ar->hw_params.max_probe_resp_desc_thres)
limit_mgmt_desc = true;

+ if (txq)
+ artxq->num_fw_queued--;
+
ath10k_htt_tx_free_msdu_id(htt, tx_done->msdu_id);
ath10k_htt_tx_dec_pending(htt, limit_mgmt_desc);
if (htt->num_pending_tx == 0)
--
2.1.4

2016-03-01 10:32:08

by Michal Kazior

[permalink] [raw]

Subject: [PATCH v2 00/11] ath10k: implement push-pull tx model

This adds support for the new logic where host
tells firmware how many frames are queued for each
station/tid and then firmware asks host to submit
frames for given station/tid.

The patch count is a bit high but I tried
splitting the patches as much as possible to keep
them short and easy to review. Hopefully it's not
going to be a huge headache.

v2:
- squashes some patches
- reworked pending frame scheduling

Michal Kazior (11):
ath10k: refactor tx code
ath10k: unify txpath decision
ath10k: refactor tx pending management
ath10k: maintain peer_id for each sta and vif
ath10k: add fast peer_map lookup
ath10k: add new htt message generation/parsing logic
ath10k: implement wake_tx_queue
ath10k: implement updating shared htt txq state
ath10k: store txq in skb_cb
ath10k: keep track of queue depth per txq
ath10k: implement push-pull tx

drivers/net/wireless/ath/ath10k/core.c | 2 +
drivers/net/wireless/ath/ath10k/core.h | 16 +
drivers/net/wireless/ath/ath10k/htt.h | 21 +-
drivers/net/wireless/ath/ath10k/htt_rx.c | 298 ++++++++++++++++-
drivers/net/wireless/ath/ath10k/htt_tx.c | 261 +++++++++++----
drivers/net/wireless/ath/ath10k/mac.c | 534 +++++++++++++++++++++++++++----
drivers/net/wireless/ath/ath10k/mac.h | 6 +
drivers/net/wireless/ath/ath10k/txrx.c | 11 +-
8 files changed, 1019 insertions(+), 130 deletions(-)

--
2.1.4

2016-03-01 10:32:21

by Michal Kazior

[permalink] [raw]

Subject: [PATCH v2 08/11] ath10k: implement updating shared htt txq state

Firmware 10.4.3 onwards can support a pull-push Tx
model where it shares a Tx queue state with the
host.

The host updates the DMA region it pointed to
during HTT setup whenever number of software
queued from (on host) changes. Based on this
information firmware issues fetch requests to the
host telling the host how many frames from a list
of given stations/tids should be submitted to the
firmware.

The code won't be called because not all
appropriate HTT events are processed yet.

Signed-off-by: Michal Kazior <[email protected]>
---
drivers/net/wireless/ath/ath10k/htt.h | 3 +
drivers/net/wireless/ath/ath10k/htt_tx.c | 104 +++++++++++++++++++++++++++++++
drivers/net/wireless/ath/ath10k/mac.c | 3 +
3 files changed, 110 insertions(+)

diff --git a/drivers/net/wireless/ath/ath10k/htt.h b/drivers/net/wireless/ath/ath10k/htt.h
index 65dcd22f31df..b1e40f44e76b 100644
--- a/drivers/net/wireless/ath/ath10k/htt.h
+++ b/drivers/net/wireless/ath/ath10k/htt.h
@@ -1667,6 +1667,7 @@ struct ath10k_htt {
} txbuf;

struct {
+ bool enabled;
struct htt_q_state *vaddr;
dma_addr_t paddr;
u16 num_peers;
@@ -1758,6 +1759,8 @@ int ath10k_htt_tx_fetch_resp(struct ath10k *ar,
struct htt_tx_fetch_record *records,
size_t num_records);

+void ath10k_htt_tx_txq_update(struct ieee80211_hw *hw,
+ struct ieee80211_txq *txq);
void ath10k_htt_tx_dec_pending(struct ath10k_htt *htt,
bool is_mgmt);
int ath10k_htt_tx_inc_pending(struct ath10k_htt *htt,
diff --git a/drivers/net/wireless/ath/ath10k/htt_tx.c b/drivers/net/wireless/ath/ath10k/htt_tx.c
index 225f0561b3fd..6643be8692b5 100644
--- a/drivers/net/wireless/ath/ath10k/htt_tx.c
+++ b/drivers/net/wireless/ath/ath10k/htt_tx.c
@@ -22,6 +22,110 @@
#include "txrx.h"
#include "debug.h"

+static u8 ath10k_htt_tx_txq_calc_size(size_t count)
+{
+ int exp;
+ int factor;
+
+ exp = 0;
+ factor = count >> 7;
+
+ while (factor >= 64 && exp < 4) {
+ factor >>= 3;
+ exp++;
+ }
+
+ if (exp == 4)
+ return 0xff;
+
+ if (count > 0)
+ factor = max(1, factor);
+
+ return SM(exp, HTT_TX_Q_STATE_ENTRY_EXP) |
+ SM(factor, HTT_TX_Q_STATE_ENTRY_FACTOR);
+}
+
+static void __ath10k_htt_tx_txq_recalc(struct ieee80211_hw *hw,
+ struct ieee80211_txq *txq)
+{
+ struct ath10k *ar = hw->priv;
+ struct ath10k_sta *arsta = (void *)txq->sta->drv_priv;
+ struct ath10k_vif *arvif = (void *)txq->vif->drv_priv;
+ unsigned long frame_cnt;
+ unsigned long byte_cnt;
+ int idx;
+ u32 bit;
+ u16 peer_id;
+ u8 tid;
+ u8 count;
+
+ lockdep_assert_held(&ar->htt.tx_lock);
+
+ if (!ar->htt.tx_q_state.enabled)
+ return;
+
+ if (txq->sta)
+ peer_id = arsta->peer_id;
+ else
+ peer_id = arvif->peer_id;
+
+ tid = txq->tid;
+ bit = BIT(peer_id % 32);
+ idx = peer_id / 32;
+
+ ieee80211_txq_get_depth(txq, &frame_cnt, &byte_cnt);
+ count = ath10k_htt_tx_txq_calc_size(byte_cnt);
+
+ if (unlikely(peer_id >= ar->htt.tx_q_state.num_peers) ||
+ unlikely(tid >= ar->htt.tx_q_state.num_tids)) {
+ ath10k_warn(ar, "refusing to update txq for peer_id %hu tid %hhu due to out of bounds\n",
+ peer_id, tid);
+ return;
+ }
+
+ ar->htt.tx_q_state.vaddr->count[tid][peer_id] = count;
+ ar->htt.tx_q_state.vaddr->map[tid][idx] &= ~bit;
+ ar->htt.tx_q_state.vaddr->map[tid][idx] |= count ? bit : 0;
+
+ ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update peer_id %hu tid %hhu count %hhu\n",
+ peer_id, tid, count);
+}
+
+static void __ath10k_htt_tx_txq_sync(struct ath10k *ar)
+{
+ u32 seq;
+ size_t size;
+
+ lockdep_assert_held(&ar->htt.tx_lock);
+
+ if (!ar->htt.tx_q_state.enabled)
+ return;
+
+ seq = le32_to_cpu(ar->htt.tx_q_state.vaddr->seq);
+ seq++;
+ ar->htt.tx_q_state.vaddr->seq = cpu_to_le32(seq);
+
+ ath10k_dbg(ar, ATH10K_DBG_HTT, "htt tx txq state update commit seq %u\n",
+ seq);
+
+ size = sizeof(*ar->htt.tx_q_state.vaddr);
+ dma_sync_single_for_device(ar->dev,
+ ar->htt.tx_q_state.paddr,
+ size,
+ DMA_TO_DEVICE);
+}
+
+void ath10k_htt_tx_txq_update(struct ieee80211_hw *hw,
+ struct ieee80211_txq *txq)
+{
+ struct ath10k *ar = hw->priv;
+
+ spin_lock_bh(&ar->htt.tx_lock);
+ __ath10k_htt_tx_txq_recalc(hw, txq);
+ __ath10k_htt_tx_txq_sync(ar);
+ spin_unlock_bh(&ar->htt.tx_lock);
+}
+
void ath10k_htt_tx_dec_pending(struct ath10k_htt *htt,
bool is_mgmt)
{
diff --git a/drivers/net/wireless/ath/ath10k/mac.c b/drivers/net/wireless/ath/ath10k/mac.c
index 74a42e3465e4..900c64b65b43 100644
--- a/drivers/net/wireless/ath/ath10k/mac.c
+++ b/drivers/net/wireless/ath/ath10k/mac.c
@@ -3725,6 +3725,7 @@ void ath10k_mac_tx_push_pending(struct ath10k *ar)
}

list_del_init(&artxq->list);
+ ath10k_htt_tx_txq_update(hw, txq);

if (artxq == last || (ret < 0 && ret != -ENOENT)) {
if (ret != -ENOENT)
@@ -3967,6 +3968,8 @@ static void ath10k_mac_op_wake_tx_queue(struct ieee80211_hw *hw,

tasklet_schedule(&ar->htt.txrx_compl_task);
}
+
+ ath10k_htt_tx_txq_update(hw, txq);
}

/* Must not be called with conf_mutex held as workers can use that also. */
--
2.1.4