xref: /wlan-driver/qcacld-3.0/core/dp/txrx/ol_tx_sched.c (revision 5113495b16420b49004c444715d2daae2066e7dc)

/*
 * Copyright (c) 2012-2020 The Linux Foundation. All rights reserved.
 * Copyright (c) 2022 Qualcomm Innovation Center, Inc. All rights reserved.
 *
 * Permission to use, copy, modify, and/or distribute this software for
 * any purpose with or without fee is hereby granted, provided that the
 * above copyright notice and this permission notice appear in all
 * copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL
 * WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE
 * AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL
 * DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR
 * PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
 * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
 * PERFORMANCE OF THIS SOFTWARE.
 */

#include <qdf_nbuf.h>         /* qdf_nbuf_t, etc. */
#include <htt.h>              /* HTT_TX_EXT_TID_MGMT */
#include <ol_htt_tx_api.h>    /* htt_tx_desc_tid */
#include <ol_txrx_api.h>      /* ol_txrx_vdev_handle */
#include <ol_txrx_ctrl_api.h> /* ol_txrx_sync */
#include <ol_txrx_internal.h> /* TXRX_ASSERT1 */
#include <ol_txrx_types.h>    /* pdev stats, etc. */
#include <ol_tx_desc.h>       /* ol_tx_desc */
#include <ol_tx_send.h>       /* ol_tx_send */
#include <ol_tx_sched.h>      /* OL_TX_SCHED, etc. */
#include <ol_tx_queue.h>
#include <ol_txrx.h>
#include <qdf_types.h>
#include <qdf_mem.h>         /* qdf_os_mem_alloc_consistent et al */
#include <cdp_txrx_handle.h>
#if defined(CONFIG_HL_SUPPORT)

#if defined(DEBUG_HL_LOGGING)
static void
ol_tx_sched_log(struct ol_txrx_pdev_t *pdev);

#else
static void
ol_tx_sched_log(struct ol_txrx_pdev_t *pdev)
{
}
#endif /* defined(DEBUG_HL_LOGGING) */

#if DEBUG_HTT_CREDIT
#define OL_TX_DISPATCH_LOG_CREDIT()                                           \
	do {								      \
		QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO_LOW,	\
			"TX %d bytes\n", qdf_nbuf_len(msdu));	\
		QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO_LOW,	\
			" <HTT> Decrease credit %d - 1 = %d, len:%d.\n",  \
			qdf_atomic_read(&pdev->target_tx_credit),	\
			qdf_atomic_read(&pdev->target_tx_credit) - 1,	\
			qdf_nbuf_len(msdu));				\
	} while (0)
#else
#define OL_TX_DISPATCH_LOG_CREDIT()
#endif

/*--- generic definitions used by the scheduler framework for all algs ---*/

struct ol_tx_sched_ctx {
	ol_tx_desc_list head;
	int frms;
};

typedef TAILQ_HEAD(ol_tx_frms_queue_list_s, ol_tx_frms_queue_t)
	ol_tx_frms_queue_list;

	/*--- scheduler algorithm selection ---*/

	/*--- scheduler options -----------------------------------------------
	 * 1. Round-robin scheduler:
	 *    Select the TID that is at the head of the list of active TIDs.
	 *    Select the head tx queue for this TID.
	 *    Move the tx queue to the back of the list of tx queues for
	 *    this TID.
	 *    Move the TID to the back of the list of active TIDs.
	 *    Send as many frames from the tx queue as credit allows.
	 * 2. Weighted-round-robin advanced scheduler:
	 *    Keep an ordered list of which TID gets selected next.
	 *    Use a weighted-round-robin scheme to determine when to promote
	 *    a TID within this list.
	 *    If a TID at the head of the list is inactive, leave it at the
	 *    head, but check the next TIDs.
	 *    If the credit available is less than the credit threshold for the
	 *    next active TID, don't send anything, and leave the TID at the
	 *    head of the list.
	 *    After a TID is selected, move it to the back of the list.
	 *    Select the head tx queue for this TID.
	 *    Move the tx queue to the back of the list of tx queues for this
	 *    TID.
	 *    Send no more frames than the limit specified for the TID.
	 */
#define OL_TX_SCHED_RR  1
#define OL_TX_SCHED_WRR_ADV 2

#ifndef OL_TX_SCHED
	/*#define OL_TX_SCHED OL_TX_SCHED_RR*/
#define OL_TX_SCHED OL_TX_SCHED_WRR_ADV /* default */
#endif


#if OL_TX_SCHED == OL_TX_SCHED_RR

#define ol_tx_sched_rr_t ol_tx_sched_t

#define OL_TX_SCHED_NUM_CATEGORIES (OL_TX_NUM_TIDS + OL_TX_VDEV_NUM_QUEUES)

#define ol_tx_sched_init                ol_tx_sched_init_rr
#define ol_tx_sched_select_init(pdev)   /* no-op */
#define ol_tx_sched_select_batch        ol_tx_sched_select_batch_rr
#define ol_tx_sched_txq_enqueue         ol_tx_sched_txq_enqueue_rr
#define ol_tx_sched_txq_deactivate      ol_tx_sched_txq_deactivate_rr
#define ol_tx_sched_category_tx_queues  ol_tx_sched_category_tx_queues_rr
#define ol_tx_sched_txq_discard         ol_tx_sched_txq_discard_rr
#define ol_tx_sched_category_info       ol_tx_sched_category_info_rr
#define ol_tx_sched_discard_select_category \
		ol_tx_sched_discard_select_category_rr

#elif OL_TX_SCHED == OL_TX_SCHED_WRR_ADV

#define ol_tx_sched_wrr_adv_t ol_tx_sched_t

#define OL_TX_SCHED_NUM_CATEGORIES OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES

#define ol_tx_sched_init                ol_tx_sched_init_wrr_adv
#define ol_tx_sched_select_init(pdev) \
		do { \
			qdf_spin_lock_bh(&pdev->tx_queue_spinlock); \
			ol_tx_sched_select_init_wrr_adv(pdev); \
			qdf_spin_unlock_bh(&pdev->tx_queue_spinlock); \
		} while (0)
#define ol_tx_sched_select_batch        ol_tx_sched_select_batch_wrr_adv
#define ol_tx_sched_txq_enqueue         ol_tx_sched_txq_enqueue_wrr_adv
#define ol_tx_sched_txq_deactivate      ol_tx_sched_txq_deactivate_wrr_adv
#define ol_tx_sched_category_tx_queues  ol_tx_sched_category_tx_queues_wrr_adv
#define ol_tx_sched_txq_discard         ol_tx_sched_txq_discard_wrr_adv
#define ol_tx_sched_category_info       ol_tx_sched_category_info_wrr_adv
#define ol_tx_sched_discard_select_category \
		ol_tx_sched_discard_select_category_wrr_adv

#else

#error Unknown OL TX SCHED specification

#endif /* OL_TX_SCHED */

	/*--- round-robin scheduler ----------------------------------------*/
#if OL_TX_SCHED == OL_TX_SCHED_RR

	/*--- definitions ---*/

	struct ol_tx_active_queues_in_tid_t {
		/* list_elem is used to queue up into up level queues*/
		TAILQ_ENTRY(ol_tx_active_queues_in_tid_t) list_elem;
		u_int32_t frms;
		u_int32_t bytes;
		ol_tx_frms_queue_list head;
		bool    active;
		int tid;
	};

	struct ol_tx_sched_rr_t {
		struct ol_tx_active_queues_in_tid_t
			tx_active_queues_in_tid_array[OL_TX_NUM_TIDS
						+ OL_TX_VDEV_NUM_QUEUES];
	TAILQ_HEAD(ol_tx_active_tids_s, ol_tx_active_queues_in_tid_t)
							tx_active_tids_list;
		u_int8_t discard_weights[OL_TX_NUM_TIDS
					+ OL_TX_VDEV_NUM_QUEUES];
	};

#define TX_SCH_MAX_CREDIT_FOR_THIS_TID(tidq) 16

/*--- functions ---*/

/*
 * The scheduler sync spinlock has been acquired outside this function,
 * so there is no need to worry about mutex within this function.
 */
static int
ol_tx_sched_select_batch_rr(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_sched_ctx *sctx,
	u_int32_t credit)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;
	struct ol_tx_frms_queue_t *next_tq;
	u_int16_t frames, used_credits = 0, tx_limit, tx_limit_flag = 0;
	int bytes;

	TX_SCHED_DEBUG_PRINT("Enter");

	if (TAILQ_EMPTY(&scheduler->tx_active_tids_list))
		return used_credits;

	txq_queue = TAILQ_FIRST(&scheduler->tx_active_tids_list);

	TAILQ_REMOVE(&scheduler->tx_active_tids_list, txq_queue, list_elem);
	txq_queue->active = false;

	next_tq = TAILQ_FIRST(&txq_queue->head);
	TAILQ_REMOVE(&txq_queue->head, next_tq, list_elem);

	credit = QDF_MIN(credit, TX_SCH_MAX_CREDIT_FOR_THIS_TID(next_tq));
	frames = next_tq->frms; /* download as many frames as credit allows */
	tx_limit = ol_tx_bad_peer_dequeue_check(next_tq,
					frames,
					&tx_limit_flag);
	frames = ol_tx_dequeue(
			pdev, next_tq, &sctx->head, tx_limit, &credit, &bytes);
	ol_tx_bad_peer_update_tx_limit(pdev, next_tq, frames, tx_limit_flag);

	used_credits = credit;
	txq_queue->frms -= frames;
	txq_queue->bytes -= bytes;

	if (next_tq->frms > 0) {
		TAILQ_INSERT_TAIL(&txq_queue->head, next_tq, list_elem);
		TAILQ_INSERT_TAIL(
				&scheduler->tx_active_tids_list,
						txq_queue, list_elem);
		txq_queue->active = true;
	} else if (!TAILQ_EMPTY(&txq_queue->head)) {
		/*
		 * This tx queue is empty, but there's another tx queue for the
		 * same TID that is not empty.
		 *Thus, the TID as a whole is active.
		 */
		TAILQ_INSERT_TAIL(
				&scheduler->tx_active_tids_list,
						txq_queue, list_elem);
		txq_queue->active = true;
	}
	sctx->frms += frames;

	TX_SCHED_DEBUG_PRINT("Leave");
	return used_credits;
}

static inline void
ol_tx_sched_txq_enqueue_rr(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int tid,
	int frms,
	int bytes)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;

	txq_queue = &scheduler->tx_active_queues_in_tid_array[tid];
	if (txq->flag != ol_tx_queue_active)
		TAILQ_INSERT_TAIL(&txq_queue->head, txq, list_elem);

	txq_queue->frms += frms;
	txq_queue->bytes += bytes;

	if (!txq_queue->active) {
		TAILQ_INSERT_TAIL(
				&scheduler->tx_active_tids_list,
				txq_queue, list_elem);
		txq_queue->active = true;
	}
}

static inline void
ol_tx_sched_txq_deactivate_rr(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int tid)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;

	txq_queue = &scheduler->tx_active_queues_in_tid_array[tid];
	txq_queue->frms -= txq->frms;
	txq_queue->bytes -= txq->bytes;

	TAILQ_REMOVE(&txq_queue->head, txq, list_elem);
	/*if (txq_queue->frms == 0 && txq_queue->active) {*/
	if (TAILQ_EMPTY(&txq_queue->head) && txq_queue->active) {
		TAILQ_REMOVE(&scheduler->tx_active_tids_list, txq_queue,
			     list_elem);
		txq_queue->active = false;
	}
}

ol_tx_frms_queue_list *
ol_tx_sched_category_tx_queues_rr(struct ol_txrx_pdev_t *pdev, int tid)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;

	txq_queue = &scheduler->tx_active_queues_in_tid_array[tid];
	return &txq_queue->head;
}

int
ol_tx_sched_discard_select_category_rr(struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_rr_t *scheduler;
	u_int8_t i, tid = 0;
	int max_score = 0;

	scheduler = pdev->tx_sched.scheduler;
	/*
	 * Choose which TID's tx frames to drop next based on two factors:
	 * 1.  Which TID has the most tx frames present
	 * 2.  The TID's priority (high-priority TIDs have a low discard_weight)
	 */
	for (i = 0; i < (OL_TX_NUM_TIDS + OL_TX_VDEV_NUM_QUEUES); i++) {
		int score;

		score =
			scheduler->tx_active_queues_in_tid_array[i].frms *
			scheduler->discard_weights[i];
		if (max_score == 0 || score > max_score) {
			max_score = score;
			tid = i;
		}
	}
	return tid;
}

void
ol_tx_sched_txq_discard_rr(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int tid, int frames, int bytes)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;

	txq_queue = &scheduler->tx_active_queues_in_tid_array[tid];

	if (0 == txq->frms)
		TAILQ_REMOVE(&txq_queue->head, txq, list_elem);

	txq_queue->frms -= frames;
	txq_queue->bytes -= bytes;
	if (txq_queue->active == true && txq_queue->frms == 0) {
		TAILQ_REMOVE(&scheduler->tx_active_tids_list, txq_queue,
			     list_elem);
		txq_queue->active = false;
	}
}

void
ol_tx_sched_category_info_rr(
	struct ol_txrx_pdev_t *pdev,
	int cat, int *active,
	int *frms, int *bytes)
{
	struct ol_tx_sched_rr_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_active_queues_in_tid_t *txq_queue;

	txq_queue = &scheduler->tx_active_queues_in_tid_array[cat];

	*active = txq_queue->active;
	*frms = txq_queue->frms;
	*bytes = txq_queue->bytes;
}

enum {
	ol_tx_sched_discard_weight_voice = 1,
	ol_tx_sched_discard_weight_video = 4,
	ol_tx_sched_discard_weight_ucast_default = 8,
	ol_tx_sched_discard_weight_mgmt_non_qos = 1, /* 0? */
	ol_tx_sched_discard_weight_mcast = 1, /* 0? also for probe & assoc */
};

void *
ol_tx_sched_init_rr(
	struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_rr_t *scheduler;
	int i;

	scheduler = qdf_mem_malloc(sizeof(struct ol_tx_sched_rr_t));
	if (!scheduler)
		return scheduler;

	for (i = 0; i < (OL_TX_NUM_TIDS + OL_TX_VDEV_NUM_QUEUES); i++) {
		scheduler->tx_active_queues_in_tid_array[i].tid = i;
		TAILQ_INIT(&scheduler->tx_active_queues_in_tid_array[i].head);
		scheduler->tx_active_queues_in_tid_array[i].active = 0;
		scheduler->tx_active_queues_in_tid_array[i].frms = 0;
		scheduler->tx_active_queues_in_tid_array[i].bytes = 0;
	}
	for (i = 0; i < OL_TX_NUM_TIDS; i++) {
		scheduler->tx_active_queues_in_tid_array[i].tid = i;
		if (i < OL_TX_NON_QOS_TID) {
			int ac = TXRX_TID_TO_WMM_AC(i);

			switch (ac) {
			case TXRX_WMM_AC_VO:
				scheduler->discard_weights[i] =
					ol_tx_sched_discard_weight_voice;
			case TXRX_WMM_AC_VI:
				scheduler->discard_weights[i] =
					ol_tx_sched_discard_weight_video;
			default:
				scheduler->discard_weights[i] =
				ol_tx_sched_discard_weight_ucast_default;
			};
		} else {
			scheduler->discard_weights[i] =
				ol_tx_sched_discard_weight_mgmt_non_qos;
		}
	}
	for (i = 0; i < OL_TX_VDEV_NUM_QUEUES; i++) {
		int j = i + OL_TX_NUM_TIDS;

		scheduler->tx_active_queues_in_tid_array[j].tid =
							OL_TX_NUM_TIDS - 1;
		scheduler->discard_weights[j] =
					ol_tx_sched_discard_weight_mcast;
	}
	TAILQ_INIT(&scheduler->tx_active_tids_list);

	return scheduler;
}

void
ol_txrx_set_wmm_param(struct cdp_soc_t *soc_hdl, uint8_t pdev_id,
		      struct ol_tx_wmm_param_t wmm_param)
{
	QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO_LOW,
		  "Dummy function when OL_TX_SCHED_RR is enabled\n");
}

/**
 * ol_tx_sched_stats_display() - tx sched stats display
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_stats_display(struct ol_txrx_pdev_t *pdev)
{
}

/**
 * ol_tx_sched_cur_state_display() - tx sched cur stat display
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_cur_state_display(struct ol_txrx_pdev_t *pdev)
{
}

/**
 * ol_tx_sched_cur_state_display() - reset tx sched stats
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_stats_clear(struct ol_txrx_pdev_t *pdev)
{
}

#endif /* OL_TX_SCHED == OL_TX_SCHED_RR */

/*--- advanced scheduler ----------------------------------------------------*/
#if OL_TX_SCHED == OL_TX_SCHED_WRR_ADV

/*--- definitions ---*/

struct ol_tx_sched_wrr_adv_category_info_t {
	struct {
		int wrr_skip_weight;
		u_int32_t credit_threshold;
		u_int16_t send_limit;
		int credit_reserve;
		int discard_weight;
	} specs;
	struct {
		int wrr_count;
		int frms;
		int bytes;
		ol_tx_frms_queue_list head;
		bool active;
	} state;
#ifdef DEBUG_HL_LOGGING
	struct {
		char *cat_name;
		unsigned int queued;
		unsigned int dispatched;
		unsigned int discard;
	} stat;
#endif
};

#define OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(cat, \
		wrr_skip_weight, \
		credit_threshold, \
		send_limit, \
		credit_reserve, \
		discard_weights) \
		enum { OL_TX_SCHED_WRR_ADV_ ## cat ## _WRR_SKIP_WEIGHT = \
			(wrr_skip_weight) }; \
		enum { OL_TX_SCHED_WRR_ADV_ ## cat ## _CREDIT_THRESHOLD = \
			(credit_threshold) }; \
		enum { OL_TX_SCHED_WRR_ADV_ ## cat ## _SEND_LIMIT = \
			(send_limit) }; \
		enum { OL_TX_SCHED_WRR_ADV_ ## cat ## _CREDIT_RESERVE = \
			(credit_reserve) }; \
		enum { OL_TX_SCHED_WRR_ADV_ ## cat ## _DISCARD_WEIGHT = \
			(discard_weights) };
/* Rome:
 * For high-volume traffic flows (VI, BE, BK), use a credit threshold
 * roughly equal to a large A-MPDU (occupying half the target memory
 * available for holding tx frames) to download AMPDU-sized batches
 * of traffic.
 * For high-priority, low-volume traffic flows (VO and mgmt), use no
 * credit threshold, to minimize download latency.
 */
/*                                            WRR           send
 *                                           skip  credit  limit credit disc
 *                                            wts  thresh (frms) reserv  wts
 */
#ifdef HIF_SDIO
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(VO,           1,     17,    24,     0,  1);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(VI,           3,     17,    16,     1,  4);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(BE,          10,     17,    16,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(BK,          12,      6,     6,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(NON_QOS_DATA,10,     17,    16,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(UCAST_MGMT,   1,      1,     4,     0,  1);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(MCAST_DATA,  10,     17,     4,     1,  4);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(MCAST_MGMT,   1,      1,     4,     0,  1);
#else
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(VO,           1,     16,    24,     0,  1);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(VI,           3,     16,    16,     1,  4);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(BE,          10,     12,    12,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(BK,          12,      6,     6,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(NON_QOS_DATA, 12,      6,     4,     1,  8);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(UCAST_MGMT,   1,      1,     4,     0,  1);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(MCAST_DATA,  10,     16,     4,     1,  4);
OL_TX_SCHED_WRR_ADV_CAT_CFG_SPEC(MCAST_MGMT,   1,      1,     4,     0,  1);
#endif

#ifdef DEBUG_HL_LOGGING

#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INIT(category, scheduler)               \
	do {                                                                 \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category]  \
		.stat.queued = 0;					\
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category]  \
		.stat.discard = 0;					\
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category]  \
		.stat.dispatched = 0;					\
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category]  \
		.stat.cat_name = #category;				\
	} while (0)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_QUEUED(category, frms)             \
	category->stat.queued += frms;
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISCARD(category, frms)           \
	category->stat.discard += frms;
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISPATCHED(category, frms)         \
	category->stat.dispatched += frms;
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_DUMP(scheduler)                        \
	ol_tx_sched_wrr_adv_cat_stat_dump(scheduler)
#define OL_TX_SCHED_WRR_ADV_CAT_CUR_STATE_DUMP(scheduler)                   \
	ol_tx_sched_wrr_adv_cat_cur_state_dump(scheduler)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_CLEAR(scheduler)                       \
	ol_tx_sched_wrr_adv_cat_stat_clear(scheduler)

#else   /* DEBUG_HL_LOGGING */

#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INIT(category, scheduler)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_QUEUED(category, frms)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISCARD(category, frms)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISPATCHED(category, frms)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_DUMP(scheduler)
#define OL_TX_SCHED_WRR_ADV_CAT_CUR_STATE_DUMP(scheduler)
#define OL_TX_SCHED_WRR_ADV_CAT_STAT_CLEAR(scheduler)

#endif  /* DEBUG_HL_LOGGING */

#define OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(category, scheduler) \
	do { \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category] \
		.specs.wrr_skip_weight = \
		OL_TX_SCHED_WRR_ADV_ ## category ## _WRR_SKIP_WEIGHT; \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category] \
		.specs.credit_threshold = \
		OL_TX_SCHED_WRR_ADV_ ## category ## _CREDIT_THRESHOLD; \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category] \
		.specs.send_limit = \
		OL_TX_SCHED_WRR_ADV_ ## category ## _SEND_LIMIT; \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category] \
		.specs.credit_reserve = \
		OL_TX_SCHED_WRR_ADV_ ## category ## _CREDIT_RESERVE; \
		scheduler->categories[OL_TX_SCHED_WRR_ADV_CAT_ ## category] \
		.specs.discard_weight = \
		OL_TX_SCHED_WRR_ADV_ ## category ## _DISCARD_WEIGHT; \
		OL_TX_SCHED_WRR_ADV_CAT_STAT_INIT(category, scheduler); \
	} while (0)

struct ol_tx_sched_wrr_adv_t {
	int order[OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES];
	int index;
	struct ol_tx_sched_wrr_adv_category_info_t
		categories[OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES];
};

#define OL_TX_AIFS_DEFAULT_VO   2
#define OL_TX_AIFS_DEFAULT_VI   2
#define OL_TX_AIFS_DEFAULT_BE   3
#define OL_TX_AIFS_DEFAULT_BK   7
#define OL_TX_CW_MIN_DEFAULT_VO   3
#define OL_TX_CW_MIN_DEFAULT_VI   7
#define OL_TX_CW_MIN_DEFAULT_BE   15
#define OL_TX_CW_MIN_DEFAULT_BK   15

/*--- functions ---*/

#ifdef DEBUG_HL_LOGGING
static void ol_tx_sched_wrr_adv_cat_stat_dump(
	struct ol_tx_sched_wrr_adv_t *scheduler)
{
	int i;

	txrx_nofl_info("Scheduler Stats:");
	txrx_nofl_info("====category(CRR,CRT,WSW): Queued  Discard  Dequeued  frms  wrr===");
	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; ++i) {
		txrx_nofl_info("%12s(%2d, %2d, %2d):  %6d  %7d  %8d  %4d  %3d",
			       scheduler->categories[i].stat.cat_name,
			       scheduler->categories[i].specs.credit_reserve,
			       scheduler->categories[i].specs.
					credit_threshold,
			       scheduler->categories[i].
					specs.wrr_skip_weight,
			       scheduler->categories[i].stat.queued,
			       scheduler->categories[i].stat.discard,
			       scheduler->categories[i].stat.dispatched,
			       scheduler->categories[i].state.frms,
			       scheduler->categories[i].state.wrr_count);
	}
}

static void ol_tx_sched_wrr_adv_cat_cur_state_dump(
	struct ol_tx_sched_wrr_adv_t *scheduler)
{
	int i;

	txrx_nofl_info("Scheduler State Snapshot:");
	txrx_nofl_info("====category(CRR,CRT,WSW): IS_Active  Pend_Frames  Pend_bytes  wrr===");
	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; ++i) {
		txrx_nofl_info("%12s(%2d, %2d, %2d):  %9d  %11d  %10d  %3d",
			       scheduler->categories[i].stat.cat_name,
			       scheduler->categories[i].specs.credit_reserve,
			       scheduler->categories[i].specs.
					credit_threshold,
			       scheduler->categories[i].specs.
					wrr_skip_weight,
			       scheduler->categories[i].state.active,
			       scheduler->categories[i].state.frms,
			       scheduler->categories[i].state.bytes,
			       scheduler->categories[i].state.wrr_count);
	}
}

static void ol_tx_sched_wrr_adv_cat_stat_clear(
	struct ol_tx_sched_wrr_adv_t *scheduler)
{
	int i;

	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; ++i) {
		scheduler->categories[i].stat.queued = 0;
		scheduler->categories[i].stat.discard = 0;
		scheduler->categories[i].stat.dispatched = 0;
	}
}

#endif

static void
ol_tx_sched_select_init_wrr_adv(struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	/* start selection from the front of the ordered list */
	scheduler->index = 0;
}

static void
ol_tx_sched_wrr_adv_rotate_order_list_tail(
		struct ol_tx_sched_wrr_adv_t *scheduler, int idx)
{
	int value;
	/* remember the value of the specified element */
	value = scheduler->order[idx];
	/* shift all further elements up one space */
	for (; idx < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES-1; idx++)
		scheduler->order[idx] = scheduler->order[idx + 1];

	/* put the specified element at the end */
	scheduler->order[idx] = value;
}

static void
ol_tx_sched_wrr_adv_credit_sanity_check(struct ol_txrx_pdev_t *pdev,
					u_int32_t credit)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	int i;
	int okay = 1;

	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; i++) {
		if (scheduler->categories[i].specs.credit_threshold > credit) {
			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_ERROR,
				  "*** Config error: credit (%d) not enough to support category %d threshold (%d)\n",
				  credit, i,
				  scheduler->categories[i].specs.
						credit_threshold);
			okay = 0;
		}
	}
	qdf_assert(okay);
}

/*
 * The scheduler sync spinlock has been acquired outside this function,
 * so there is no need to worry about mutex within this function.
 */
static int
ol_tx_sched_select_batch_wrr_adv(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_sched_ctx *sctx,
	u_int32_t credit)
{
	static int first = 1;
	int category_index = 0;
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_frms_queue_t *txq, *first_txq = NULL;
	int index;
	struct ol_tx_sched_wrr_adv_category_info_t *category = NULL;
	int frames, bytes, used_credits = 0, tx_limit;
	u_int16_t tx_limit_flag;
	u32 credit_rem = credit;

	/*
	 * Just for good measure, do a sanity check that the initial credit
	 * is enough to cover every category's credit threshold.
	 */
	if (first) {
		first = 0;
		ol_tx_sched_wrr_adv_credit_sanity_check(pdev, credit);
	}

	/* choose the traffic category from the ordered list */
	index = scheduler->index;
	while (index < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES) {
		category_index = scheduler->order[index];
		category = &scheduler->categories[category_index];
		if (!category->state.active) {
			/* move on to the next category */
			index++;
			continue;
		}
		if (++category->state.wrr_count <
					category->specs.wrr_skip_weight) {
			/* skip this category (move it to the back) */
			ol_tx_sched_wrr_adv_rotate_order_list_tail(scheduler,
								   index);
			/*
			 * try again (iterate) on the new element
			 * that was moved up
			 */
			continue;
		}
		/* found the first active category whose WRR turn is present */
		break;
	}
	if (index >= OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES) {
		/* no categories are active */
		return 0;
	}

	/* is there enough credit for the selected category? */
	if (credit < category->specs.credit_threshold) {
		/*
		 * Can't send yet - wait until more credit becomes available.
		 * In the meantime, restore the WRR counter (since we didn't
		 * service this category after all).
		 */
		category->state.wrr_count = category->state.wrr_count - 1;
		return 0;
	}
	/* enough credit is available - go ahead and send some frames */
	/*
	 * This category was serviced - reset the WRR counter, and move this
	 * category to the back of the order list.
	 */
	category->state.wrr_count = 0;
	ol_tx_sched_wrr_adv_rotate_order_list_tail(scheduler, index);
	/*
	 * With this category moved to the back, if there's still any credit
	 * left, set up the next invocation of this function to start from
	 * where this one left off, by looking at the category that just got
	 * shifted forward into the position the service category was
	 * occupying.
	 */
	scheduler->index = index;

	/*
	 * Take the tx queue from the head of the category list.
	 */
	txq = TAILQ_FIRST(&category->state.head);

	while (txq) {
		TAILQ_REMOVE(&category->state.head, txq, list_elem);
		credit = ol_tx_txq_group_credit_limit(pdev, txq, credit);
		if (credit > category->specs.credit_reserve) {
			credit -= category->specs.credit_reserve;
			tx_limit = ol_tx_bad_peer_dequeue_check(txq,
					category->specs.send_limit,
					&tx_limit_flag);
			frames = ol_tx_dequeue(
					pdev, txq, &sctx->head,
					tx_limit, &credit, &bytes);
			ol_tx_bad_peer_update_tx_limit(pdev, txq,
						       frames,
						       tx_limit_flag);

			OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISPATCHED(category,
								    frames);
			/* Update used global credits */
			used_credits = credit;
			credit =
			ol_tx_txq_update_borrowed_group_credits(pdev, txq,
								credit);
			category->state.frms -= frames;
			category->state.bytes -= bytes;
			if (txq->frms > 0) {
				TAILQ_INSERT_TAIL(&category->state.head,
						  txq, list_elem);
			} else {
				if (category->state.frms == 0)
					category->state.active = 0;
			}
			sctx->frms += frames;
			ol_tx_txq_group_credit_update(pdev, txq, -credit, 0);
			break;
		} else {
			/*
			 * Current txq belongs to a group which does not have
			 * enough credits,
			 * Iterate over to next txq and see if we can download
			 * packets from that queue.
			 */
			if (ol_tx_if_iterate_next_txq(first_txq, txq)) {
				credit = credit_rem;
				if (!first_txq)
					first_txq = txq;

				TAILQ_INSERT_TAIL(&category->state.head,
						  txq, list_elem);

				txq = TAILQ_FIRST(&category->state.head);
			} else {
				TAILQ_INSERT_HEAD(&category->state.head, txq,
					  list_elem);
				break;
			}
		}
	} /* while(txq) */

	return used_credits;
}

static inline void
ol_tx_sched_txq_enqueue_wrr_adv(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int tid,
	int frms,
	int bytes)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_sched_wrr_adv_category_info_t *category;

	category = &scheduler->categories[pdev->tid_to_ac[tid]];
	category->state.frms += frms;
	category->state.bytes += bytes;
	OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_QUEUED(category, frms);
	if (txq->flag != ol_tx_queue_active) {
		TAILQ_INSERT_TAIL(&category->state.head, txq, list_elem);
		category->state.active = 1; /* may have already been active */
	}
}

static inline void
ol_tx_sched_txq_deactivate_wrr_adv(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int tid)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_sched_wrr_adv_category_info_t *category;

	category = &scheduler->categories[pdev->tid_to_ac[tid]];
	category->state.frms -= txq->frms;
	category->state.bytes -= txq->bytes;

	TAILQ_REMOVE(&category->state.head, txq, list_elem);

	if (category->state.frms == 0 && category->state.active)
		category->state.active = 0;
}

static ol_tx_frms_queue_list *
ol_tx_sched_category_tx_queues_wrr_adv(struct ol_txrx_pdev_t *pdev, int cat)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_sched_wrr_adv_category_info_t *category;

	category = &scheduler->categories[cat];
	return &category->state.head;
}

static int
ol_tx_sched_discard_select_category_wrr_adv(struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_wrr_adv_t *scheduler;
	u_int8_t i, cat = 0;
	int max_score = 0;

	scheduler = pdev->tx_sched.scheduler;
	/*
	 * Choose which category's tx frames to drop next based on two factors:
	 * 1.  Which category has the most tx frames present
	 * 2.  The category's priority (high-priority categories have a low
	 *     discard_weight)
	 */
	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; i++) {
		int score;

		score =
			scheduler->categories[i].state.frms *
			scheduler->categories[i].specs.discard_weight;
		if (max_score == 0 || score > max_score) {
			max_score = score;
			cat = i;
		}
	}
	return cat;
}

static void
ol_tx_sched_txq_discard_wrr_adv(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_frms_queue_t *txq,
	int cat, int frames, int bytes)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_sched_wrr_adv_category_info_t *category;

	category = &scheduler->categories[cat];

	if (0 == txq->frms)
		TAILQ_REMOVE(&category->state.head, txq, list_elem);


	category->state.frms -= frames;
	category->state.bytes -= bytes;
	OL_TX_SCHED_WRR_ADV_CAT_STAT_INC_DISCARD(category, frames);
	if (category->state.frms == 0)
		category->state.active = 0;
}

static void
ol_tx_sched_category_info_wrr_adv(
	struct ol_txrx_pdev_t *pdev,
	int cat, int *active,
	int *frms, int *bytes)
{
	struct ol_tx_sched_wrr_adv_t *scheduler = pdev->tx_sched.scheduler;
	struct ol_tx_sched_wrr_adv_category_info_t *category;

	category = &scheduler->categories[cat];
	*active = category->state.active;
	*frms = category->state.frms;
	*bytes = category->state.bytes;
}

/**
 * ol_tx_sched_wrr_param_update() - update the WRR TX sched params
 * @pdev: Pointer to PDEV structure.
 * @scheduler: Pointer to tx scheduler.
 *
 * Update the WRR TX schedule parameters for each category if it is
 * specified in the ini file by user.
 *
 * Return: none
 */
static void ol_tx_sched_wrr_param_update(struct ol_txrx_pdev_t *pdev,
					 struct ol_tx_sched_wrr_adv_t *
					 scheduler)
{
	int i;
	static const char * const tx_sched_wrr_name[4] = {
		"BE",
		"BK",
		"VI",
		"VO"
	};

	if (!scheduler)
		return;

	QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO,
		"%s: Tuning the TX scheduler wrr parameters by ini file:",
		__func__);

	QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO,
		"         skip credit limit credit disc");

	for (i = OL_TX_SCHED_WRR_ADV_CAT_BE;
		i <= OL_TX_SCHED_WRR_ADV_CAT_VO; i++) {
		if (ol_cfg_get_wrr_skip_weight(pdev->ctrl_pdev, i)) {
			scheduler->categories[i].specs.wrr_skip_weight =
				ol_cfg_get_wrr_skip_weight(pdev->ctrl_pdev, i);
			scheduler->categories[i].specs.credit_threshold =
				ol_cfg_get_credit_threshold(pdev->ctrl_pdev, i);
			scheduler->categories[i].specs.send_limit =
				ol_cfg_get_send_limit(pdev->ctrl_pdev, i);
			scheduler->categories[i].specs.credit_reserve =
				ol_cfg_get_credit_reserve(pdev->ctrl_pdev, i);
			scheduler->categories[i].specs.discard_weight =
				ol_cfg_get_discard_weight(pdev->ctrl_pdev, i);

			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO,
				"%s-update: %d,  %d,    %d,   %d,    %d",
				tx_sched_wrr_name[i],
				scheduler->categories[i].specs.wrr_skip_weight,
				scheduler->categories[i].specs.credit_threshold,
				scheduler->categories[i].specs.send_limit,
				scheduler->categories[i].specs.credit_reserve,
				scheduler->categories[i].specs.discard_weight);
		} else {
			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO,
				"%s-orig: %d,  %d,    %d,   %d,    %d",
				tx_sched_wrr_name[i],
				scheduler->categories[i].specs.wrr_skip_weight,
				scheduler->categories[i].specs.credit_threshold,
				scheduler->categories[i].specs.send_limit,
				scheduler->categories[i].specs.credit_reserve,
				scheduler->categories[i].specs.discard_weight);
		}
	}
}

static void *
ol_tx_sched_init_wrr_adv(
		struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_wrr_adv_t *scheduler;
	int i;

	scheduler = qdf_mem_malloc(
			sizeof(struct ol_tx_sched_wrr_adv_t));
	if (!scheduler)
		return scheduler;

	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(VO, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(VI, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(BE, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(BK, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(NON_QOS_DATA, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(UCAST_MGMT, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(MCAST_DATA, scheduler);
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(MCAST_MGMT, scheduler);

	ol_tx_sched_wrr_param_update(pdev, scheduler);

	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; i++) {
		scheduler->categories[i].state.active = 0;
		scheduler->categories[i].state.frms = 0;
		/*scheduler->categories[i].state.bytes = 0;*/
		TAILQ_INIT(&scheduler->categories[i].state.head);
		/*
		 * init categories to not be skipped before
		 * their initial selection
		 */
		scheduler->categories[i].state.wrr_count =
			scheduler->categories[i].specs.wrr_skip_weight - 1;
	}

	/*
	 * Init the order array - the initial ordering doesn't matter, as the
	 * order array will get reshuffled as data arrives.
	 */
	for (i = 0; i < OL_TX_SCHED_WRR_ADV_NUM_CATEGORIES; i++)
		scheduler->order[i] = i;

	return scheduler;
}


/* WMM parameters are suppposed to be passed when associate with AP.
 * According to AIFS+CWMin, the function maps each queue to one of four default
 * settings of the scheduler, ie. VO, VI, BE, or BK.
 */
void
ol_txrx_set_wmm_param(struct cdp_soc_t *soc_hdl, uint8_t pdev_id,
		      struct ol_tx_wmm_param_t wmm_param)
{
	struct ol_txrx_soc_t *soc = cdp_soc_t_to_ol_txrx_soc_t(soc_hdl);
	ol_txrx_pdev_handle data_pdev =
				ol_txrx_get_pdev_from_pdev_id(soc, pdev_id);
	struct ol_tx_sched_wrr_adv_t def_cfg;
	struct ol_tx_sched_wrr_adv_t *scheduler =
					data_pdev->tx_sched.scheduler;
	u_int32_t i, ac_selected;
	u_int32_t  weight[QCA_WLAN_AC_ALL], default_edca[QCA_WLAN_AC_ALL];

	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(VO, (&def_cfg));
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(VI, (&def_cfg));
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(BE, (&def_cfg));
	OL_TX_SCHED_WRR_ADV_CAT_CFG_STORE(BK, (&def_cfg));

	/* default_eca = AIFS + CWMin */
	default_edca[OL_TX_SCHED_WRR_ADV_CAT_VO] =
		OL_TX_AIFS_DEFAULT_VO + OL_TX_CW_MIN_DEFAULT_VO;
	default_edca[OL_TX_SCHED_WRR_ADV_CAT_VI] =
		OL_TX_AIFS_DEFAULT_VI + OL_TX_CW_MIN_DEFAULT_VI;
	default_edca[OL_TX_SCHED_WRR_ADV_CAT_BE] =
		OL_TX_AIFS_DEFAULT_BE + OL_TX_CW_MIN_DEFAULT_BE;
	default_edca[OL_TX_SCHED_WRR_ADV_CAT_BK] =
		OL_TX_AIFS_DEFAULT_BK + OL_TX_CW_MIN_DEFAULT_BK;

	weight[OL_TX_SCHED_WRR_ADV_CAT_VO] =
		wmm_param.ac[QCA_WLAN_AC_VO].aifs +
				wmm_param.ac[QCA_WLAN_AC_VO].cwmin;
	weight[OL_TX_SCHED_WRR_ADV_CAT_VI] =
		wmm_param.ac[QCA_WLAN_AC_VI].aifs +
				wmm_param.ac[QCA_WLAN_AC_VI].cwmin;
	weight[OL_TX_SCHED_WRR_ADV_CAT_BK] =
		wmm_param.ac[QCA_WLAN_AC_BK].aifs +
				wmm_param.ac[QCA_WLAN_AC_BK].cwmin;
	weight[OL_TX_SCHED_WRR_ADV_CAT_BE] =
		wmm_param.ac[QCA_WLAN_AC_BE].aifs +
				wmm_param.ac[QCA_WLAN_AC_BE].cwmin;

	for (i = 0; i < QCA_WLAN_AC_ALL; i++) {
		if (default_edca[OL_TX_SCHED_WRR_ADV_CAT_VO] >= weight[i])
			ac_selected = OL_TX_SCHED_WRR_ADV_CAT_VO;
		else if (default_edca[OL_TX_SCHED_WRR_ADV_CAT_VI] >= weight[i])
			ac_selected = OL_TX_SCHED_WRR_ADV_CAT_VI;
		else if (default_edca[OL_TX_SCHED_WRR_ADV_CAT_BE] >= weight[i])
			ac_selected = OL_TX_SCHED_WRR_ADV_CAT_BE;
		else
			ac_selected = OL_TX_SCHED_WRR_ADV_CAT_BK;


		scheduler->categories[i].specs.wrr_skip_weight =
			def_cfg.categories[ac_selected].specs.wrr_skip_weight;
		scheduler->categories[i].specs.credit_threshold =
			def_cfg.categories[ac_selected].specs.credit_threshold;
		scheduler->categories[i].specs.send_limit =
			def_cfg.categories[ac_selected].specs.send_limit;
		scheduler->categories[i].specs.credit_reserve =
			def_cfg.categories[ac_selected].specs.credit_reserve;
		scheduler->categories[i].specs.discard_weight =
			def_cfg.categories[ac_selected].specs.discard_weight;
	}
}

/**
 * ol_tx_sched_stats_display() - tx sched stats display
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_stats_display(struct ol_txrx_pdev_t *pdev)
{
	OL_TX_SCHED_WRR_ADV_CAT_STAT_DUMP(pdev->tx_sched.scheduler);
}

/**
 * ol_tx_sched_cur_state_display() - tx sched cur stat display
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_cur_state_display(struct ol_txrx_pdev_t *pdev)
{
	OL_TX_SCHED_WRR_ADV_CAT_CUR_STATE_DUMP(pdev->tx_sched.scheduler);
}

/**
 * ol_tx_sched_cur_state_display() - reset tx sched stats
 * @pdev: Pointer to the PDEV structure.
 *
 * Return: none.
 */
void ol_tx_sched_stats_clear(struct ol_txrx_pdev_t *pdev)
{
	OL_TX_SCHED_WRR_ADV_CAT_STAT_CLEAR(pdev->tx_sched.scheduler);
}

#endif /* OL_TX_SCHED == OL_TX_SCHED_WRR_ADV */

/*--- congestion control discard --------------------------------------------*/

static struct ol_tx_frms_queue_t *
ol_tx_sched_discard_select_txq(
		struct ol_txrx_pdev_t *pdev,
		ol_tx_frms_queue_list *tx_queues)
{
	struct ol_tx_frms_queue_t *txq;
	struct ol_tx_frms_queue_t *selected_txq = NULL;
	int max_frms = 0;

	/* return the tx queue with the most frames */
	TAILQ_FOREACH(txq, tx_queues, list_elem) {
		if (txq->frms > max_frms) {
			max_frms = txq->frms;
			selected_txq = txq;
		}
	}
	return selected_txq;
}

u_int16_t
ol_tx_sched_discard_select(
		struct ol_txrx_pdev_t *pdev,
		u_int16_t frms,
		ol_tx_desc_list *tx_descs,
		bool force)
{
	int cat;
	struct ol_tx_frms_queue_t *txq;
	int bytes;
	u_int32_t credit;
	struct ol_tx_sched_notify_ctx_t notify_ctx;

	/*
	 * first decide what category of traffic (e.g. TID or AC)
	 * to discard next
	 */
	cat = ol_tx_sched_discard_select_category(pdev);

	/* then decide which peer within this category to discard from next */
	txq = ol_tx_sched_discard_select_txq(
			pdev, ol_tx_sched_category_tx_queues(pdev, cat));
	if (!txq)
		/* No More pending Tx Packets in Tx Queue. Exit Discard loop */
		return 0;


	if (force == false) {
		/*
		 * Now decide how many frames to discard from this peer-TID.
		 * Don't discard more frames than the caller has specified.
		 * Don't discard more than a fixed quantum of frames at a time.
		 * Don't discard more than 50% of the queue's frames at a time,
		 * but if there's only 1 frame left, go ahead and discard it.
		 */
#define OL_TX_DISCARD_QUANTUM 10
		if (OL_TX_DISCARD_QUANTUM < frms)
			frms = OL_TX_DISCARD_QUANTUM;


		if (txq->frms > 1 && frms >= (txq->frms >> 1))
			frms = txq->frms >> 1;
	}

	/*
	 * Discard from the head of the queue, because:
	 * 1.  Front-dropping gives applications like TCP that include ARQ
	 *     an early notification of congestion.
	 * 2.  For time-sensitive applications like RTP, the newest frames are
	 *     most relevant.
	 */
	credit = 10000; /* no credit limit */
	frms = ol_tx_dequeue(pdev, txq, tx_descs, frms, &credit, &bytes);

	notify_ctx.event = OL_TX_DISCARD_FRAMES;
	notify_ctx.frames = frms;
	notify_ctx.bytes = bytes;
	notify_ctx.txq = txq;
	notify_ctx.info.ext_tid = cat;
	ol_tx_sched_notify(pdev, &notify_ctx);

	TX_SCHED_DEBUG_PRINT("Tx Drop : %d", frms);
	return frms;
}

/*--- scheduler framework ---------------------------------------------------*/

/*
 * The scheduler mutex spinlock has been acquired outside this function,
 * so there is need to take locks inside this function.
 */
void
ol_tx_sched_notify(
		struct ol_txrx_pdev_t *pdev,
		struct ol_tx_sched_notify_ctx_t *ctx)
{
	struct ol_tx_frms_queue_t *txq = ctx->txq;
	int tid;

	if (!pdev->tx_sched.scheduler)
		return;

	switch (ctx->event) {
	case OL_TX_ENQUEUE_FRAME:
		tid = ctx->info.tx_msdu_info->htt.info.ext_tid;
		ol_tx_sched_txq_enqueue(pdev, txq, tid, 1, ctx->bytes);
		break;
	case OL_TX_DELETE_QUEUE:
		tid = ctx->info.ext_tid;
		if (txq->flag == ol_tx_queue_active)
			ol_tx_sched_txq_deactivate(pdev, txq, tid);

		break;
	case OL_TX_PAUSE_QUEUE:
		tid = ctx->info.ext_tid;
		if (txq->flag == ol_tx_queue_active)
			ol_tx_sched_txq_deactivate(pdev, txq, tid);

		break;
	case OL_TX_UNPAUSE_QUEUE:
		tid = ctx->info.ext_tid;
		if (txq->frms != 0)
			ol_tx_sched_txq_enqueue(pdev, txq, tid,
						txq->frms, txq->bytes);

		break;
	case OL_TX_DISCARD_FRAMES:
		/* not necessarily TID, could be category */
		tid = ctx->info.ext_tid;
		ol_tx_sched_txq_discard(pdev, txq, tid,
					ctx->frames, ctx->bytes);
		break;
	default:
		QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_ERROR,
			  "Error: unknown sched notification (%d)\n",
			  ctx->event);
		qdf_assert(0);
		break;
	}
}

#define OL_TX_MSDU_ID_STORAGE_ERR(ptr) (!ptr)

static void
ol_tx_sched_dispatch(
	struct ol_txrx_pdev_t *pdev,
	struct ol_tx_sched_ctx *sctx)
{
	qdf_nbuf_t msdu, prev = NULL, head_msdu = NULL;
	struct ol_tx_desc_t *tx_desc;
	u_int16_t *msdu_id_storage;
	u_int16_t msdu_id;
	int num_msdus = 0;

	TX_SCHED_DEBUG_PRINT("Enter");
	while (sctx->frms) {
		tx_desc = TAILQ_FIRST(&sctx->head);
		if (!tx_desc) {
			/* TODO: find its reason */
			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_ERROR,
				  "%s: err, no enough tx_desc from stx->head.\n",
				  __func__);
			break;
		}
		msdu = tx_desc->netbuf;
		TAILQ_REMOVE(&sctx->head, tx_desc, tx_desc_list_elem);
		if (!head_msdu)
			head_msdu = msdu;

		if (prev)
			qdf_nbuf_set_next(prev, msdu);

		prev = msdu;

#ifndef ATH_11AC_TXCOMPACT
		/*
		 * When the tx frame is downloaded to the target, there are two
		 * outstanding references:
		 * 1.  The host download SW (HTT, HTC, HIF)
		 *     This reference is cleared by the ol_tx_send_done callback
		 *     functions.
		 * 2.  The target FW
		 *     This reference is cleared by the ol_tx_completion_handler
		 *     function.
		 * It is extremely probable that the download completion is
		 * processed before the tx completion message.  However, under
		 * exceptional conditions the tx completion may be processed
		 *first. Thus, rather that assuming that reference (1) is
		 *done before reference (2),
		 * explicit reference tracking is needed.
		 * Double-increment the ref count to account for both references
		 * described above.
		 */
		qdf_atomic_init(&tx_desc->ref_cnt);
		qdf_atomic_inc(&tx_desc->ref_cnt);
		qdf_atomic_inc(&tx_desc->ref_cnt);
#endif

		/*Store the MSDU Id for each MSDU*/
		/* store MSDU ID */
		msdu_id = ol_tx_desc_id(pdev, tx_desc);
		msdu_id_storage = ol_tx_msdu_id_storage(msdu);
		if (OL_TX_MSDU_ID_STORAGE_ERR(msdu_id_storage)) {
			/*
			 * Send the prior frames as a batch,
			 *then send this as a single,
			 * then resume handling the remaining frames.
			 */
			if (head_msdu)
				ol_tx_send_batch(pdev, head_msdu, num_msdus);

			prev = NULL;
			head_msdu = prev;
			num_msdus = 0;

			if (htt_tx_send_std(pdev->htt_pdev, msdu, msdu_id)) {
				ol_tx_target_credit_incr(pdev, msdu);
				ol_tx_desc_frame_free_nonstd(pdev, tx_desc,
							     1 /* error */);
			}
		} else {
			*msdu_id_storage = msdu_id;
			num_msdus++;
		}
		sctx->frms--;
	}

	/*Send Batch Of Frames*/
	if (head_msdu)
		ol_tx_send_batch(pdev, head_msdu, num_msdus);
	TX_SCHED_DEBUG_PRINT("Leave");
}

#ifdef QCA_TX_PADDING_CREDIT_SUPPORT
static void replenish_tx_pad_credit(struct ol_txrx_pdev_t *pdev)
{
	int replenish_credit = 0, avail_targ_tx_credit = 0;
	int cur_tx_pad_credit = 0, grp_credit = 0, i = 0;
	qdf_atomic_t *tx_grp_credit = NULL;

	cur_tx_pad_credit = qdf_atomic_read(&pdev->pad_reserve_tx_credit);
	if (cur_tx_pad_credit < MIN_TX_PAD_CREDIT_THRESH) {
		replenish_credit = MAX_TX_PAD_CREDIT_THRESH - cur_tx_pad_credit;
		avail_targ_tx_credit = qdf_atomic_read(&pdev->target_tx_credit);
		replenish_credit = (replenish_credit < avail_targ_tx_credit) ?
				   replenish_credit : avail_targ_tx_credit;
		if (replenish_credit < 0) {
			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_FATAL,
				  "Tx Pad Credits = %d Target Tx Credits = %d",
				  cur_tx_pad_credit,
				  avail_targ_tx_credit);
			qdf_assert(0);
		}
		qdf_atomic_add(replenish_credit, &pdev->pad_reserve_tx_credit);
		qdf_atomic_add(-replenish_credit, &pdev->target_tx_credit);

		while (replenish_credit > 0) {
			for (i = 0; i < OL_TX_MAX_TXQ_GROUPS; i++) {
				tx_grp_credit = &pdev->txq_grps[i].credit;
				grp_credit = qdf_atomic_read(tx_grp_credit);
				if (grp_credit) {
					qdf_atomic_add(-1, tx_grp_credit);
					replenish_credit--;
				}
				if (!replenish_credit)
					break;
			}
		}
	}
}
#else
static void replenish_tx_pad_credit(struct ol_txrx_pdev_t *pdev)
{
}
#endif

void
ol_tx_sched(struct ol_txrx_pdev_t *pdev)
{
	struct ol_tx_sched_ctx sctx;
	u_int32_t credit;

	TX_SCHED_DEBUG_PRINT("Enter");
	qdf_spin_lock_bh(&pdev->tx_queue_spinlock);
	if (pdev->tx_sched.tx_sched_status != ol_tx_scheduler_idle) {
		qdf_spin_unlock_bh(&pdev->tx_queue_spinlock);
		return;
	}
	pdev->tx_sched.tx_sched_status = ol_tx_scheduler_running;

	ol_tx_sched_log(pdev);
	/*
	 *adf_os_print("BEFORE tx sched:\n");
	 *ol_tx_queues_display(pdev);
	 */
	replenish_tx_pad_credit(pdev);
	qdf_spin_unlock_bh(&pdev->tx_queue_spinlock);

	TAILQ_INIT(&sctx.head);
	sctx.frms = 0;

	ol_tx_sched_select_init(pdev);
	while (qdf_atomic_read(&pdev->target_tx_credit) > 0) {
		int num_credits;

		qdf_spin_lock_bh(&pdev->tx_queue_spinlock);
		replenish_tx_pad_credit(pdev);
		credit = qdf_atomic_read(&pdev->target_tx_credit);
		num_credits = ol_tx_sched_select_batch(pdev, &sctx, credit);
		if (num_credits > 0) {
#if DEBUG_HTT_CREDIT
			QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_INFO,
				  " <HTT> Decrease credit %d - %d = %d.\n",
				  qdf_atomic_read(&pdev->target_tx_credit),
				  num_credits,
				  qdf_atomic_read(&pdev->target_tx_credit) -
				  num_credits);
#endif
			DPTRACE(qdf_dp_trace_credit_record(QDF_TX_SCHED,
				QDF_CREDIT_DEC, num_credits,
				qdf_atomic_read(&pdev->target_tx_credit) -
				num_credits,
				qdf_atomic_read(&pdev->txq_grps[0].credit),
				qdf_atomic_read(&pdev->txq_grps[1].credit)));

			qdf_atomic_add(-num_credits, &pdev->target_tx_credit);
		}
		qdf_spin_unlock_bh(&pdev->tx_queue_spinlock);

		if (num_credits == 0)
			break;
	}
	ol_tx_sched_dispatch(pdev, &sctx);

	qdf_spin_lock_bh(&pdev->tx_queue_spinlock);
	/*
	 *adf_os_print("AFTER tx sched:\n");
	 *ol_tx_queues_display(pdev);
	 */

	pdev->tx_sched.tx_sched_status = ol_tx_scheduler_idle;
	qdf_spin_unlock_bh(&pdev->tx_queue_spinlock);
	TX_SCHED_DEBUG_PRINT("Leave");
}

void *
ol_tx_sched_attach(
	struct ol_txrx_pdev_t *pdev)
{
	pdev->tx_sched.tx_sched_status = ol_tx_scheduler_idle;
	return ol_tx_sched_init(pdev);
}

void
ol_tx_sched_detach(
	struct ol_txrx_pdev_t *pdev)
{
	if (pdev->tx_sched.scheduler) {
		qdf_mem_free(pdev->tx_sched.scheduler);
		pdev->tx_sched.scheduler = NULL;
	}
}

/*--- debug functions -------------------------------------------------------*/

#if defined(DEBUG_HL_LOGGING)

static void
ol_tx_sched_log(struct ol_txrx_pdev_t *pdev)
{
	u_int8_t  *buf;
	u_int32_t *active_bitmap;
	int i, j, num_cats_active;
	int active, frms, bytes;
	int credit;

	/* don't bother recording state if credit is zero */
	credit = qdf_atomic_read(&pdev->target_tx_credit);
	if (credit == 0)
		return;


	/*
	 * See how many TIDs are active, so queue state can be stored only
	 * for those TIDs.
	 * Do an initial iteration through all categories to see if any
	 * are active.  Doing an extra iteration is inefficient, but
	 * efficiency is not a dominant concern when logging is enabled.
	 */
	num_cats_active = 0;
	for (i = 0; i < OL_TX_SCHED_NUM_CATEGORIES; i++) {
		ol_tx_sched_category_info(pdev, i, &active, &frms, &bytes);
		if (active)
			num_cats_active++;
	}
	/* don't bother recording state if there are no active queues */
	if (num_cats_active == 0)
		return;


	ol_tx_queue_log_sched(pdev, credit, &num_cats_active,
			      &active_bitmap, &buf);

	if (num_cats_active == 0)
		return;

	*active_bitmap = 0;
	for (i = 0, j = 0;
			i < OL_TX_SCHED_NUM_CATEGORIES && j < num_cats_active;
			i++) {
		u_int8_t *p;

		ol_tx_sched_category_info(pdev, i, &active, &frms, &bytes);
		if (!active)
			continue;

		p = &buf[j*6];
		p[0]   = (frms >> 0) & 0xff;
		p[1] = (frms >> 8) & 0xff;

		p[2] = (bytes >> 0) & 0xff;
		p[3] = (bytes >> 8) & 0xff;
		p[4] = (bytes >> 16) & 0xff;
		p[5] = (bytes >> 24) & 0xff;
		j++;
		*active_bitmap |= 1 << i;
	}
}

#endif /* defined(DEBUG_HL_LOGGING) */

#endif /* defined(CONFIG_HL_SUPPORT) */