Files
android_kernel_samsung_sm86…/dp/wifi3.0/dp_rx_mon_status.c
Keyur Parekh 49cdc74b40 qcacmn: Modify SGI value in case of VHT SIG TLV
In VHT case, for some packets SGI value is 3, which
represent 3.2us value. Which is not applicable for VHT.
So if value 3 received from HW for SGI, than convert that
to default value, which is 0.4us.

Change-Id: I92716f8041f0db20b60dcd5e5d6681fa1bfa708f
2018-05-16 03:57:43 -07:00

857 lines
25 KiB
C

/*
* Copyright (c) 2017-2018 The Linux Foundation. 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 "dp_types.h"
#include "dp_rx.h"
#include "dp_peer.h"
#include "hal_rx.h"
#include "hal_api.h"
#include "qdf_trace.h"
#include "qdf_nbuf.h"
#include "hal_api_mon.h"
#include "dp_rx_mon.h"
#include "dp_internal.h"
#include "qdf_mem.h" /* qdf_mem_malloc,free */
/**
* dp_rx_populate_cdp_indication_ppdu() - Populate cdp rx indication structure
* @pdev: pdev ctx
* @ppdu_info: ppdu info structure from ppdu ring
* @ppdu_nbuf: qdf nbuf abstraction for linux skb
*
* Return: none
*/
#ifdef FEATURE_PERPKT_INFO
static inline void
dp_rx_populate_cdp_indication_ppdu(struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info,
qdf_nbuf_t ppdu_nbuf)
{
struct dp_peer *peer;
struct dp_soc *soc = pdev->soc;
struct dp_ast_entry *ast_entry;
struct cdp_rx_indication_ppdu *cdp_rx_ppdu;
uint32_t ast_index;
cdp_rx_ppdu = (struct cdp_rx_indication_ppdu *)ppdu_nbuf->data;
cdp_rx_ppdu->first_data_seq_ctrl =
ppdu_info->rx_status.first_data_seq_ctrl;
cdp_rx_ppdu->ppdu_id = ppdu_info->com_info.ppdu_id;
cdp_rx_ppdu->length = ppdu_info->rx_status.ppdu_len;
cdp_rx_ppdu->duration = ppdu_info->rx_status.duration;
cdp_rx_ppdu->u.bw = ppdu_info->rx_status.bw;
cdp_rx_ppdu->tcp_msdu_count = ppdu_info->rx_status.tcp_msdu_count;
cdp_rx_ppdu->udp_msdu_count = ppdu_info->rx_status.udp_msdu_count;
cdp_rx_ppdu->other_msdu_count = ppdu_info->rx_status.other_msdu_count;
cdp_rx_ppdu->u.nss = ppdu_info->rx_status.nss;
cdp_rx_ppdu->u.mcs = ppdu_info->rx_status.mcs;
if ((ppdu_info->rx_status.sgi == VHT_SGI_NYSM) &&
(ppdu_info->rx_status.preamble_type == HAL_RX_PKT_TYPE_11AC))
cdp_rx_ppdu->u.gi = CDP_SGI_0_4_US;
else
cdp_rx_ppdu->u.gi = ppdu_info->rx_status.sgi;
cdp_rx_ppdu->u.ldpc = ppdu_info->rx_status.ldpc;
cdp_rx_ppdu->u.preamble = ppdu_info->rx_status.preamble_type;
cdp_rx_ppdu->u.ppdu_type = ppdu_info->rx_status.reception_type;
cdp_rx_ppdu->rssi = ppdu_info->rx_status.rssi_comb;
cdp_rx_ppdu->timestamp = ppdu_info->rx_status.tsft;
cdp_rx_ppdu->channel = ppdu_info->rx_status.chan_num;
cdp_rx_ppdu->beamformed = ppdu_info->rx_status.beamformed;
cdp_rx_ppdu->num_msdu = (cdp_rx_ppdu->tcp_msdu_count +
cdp_rx_ppdu->udp_msdu_count +
cdp_rx_ppdu->other_msdu_count);
cdp_rx_ppdu->num_mpdu = ppdu_info->com_info.mpdu_cnt_fcs_ok;
if (ppdu_info->com_info.mpdu_cnt_fcs_ok > 1)
cdp_rx_ppdu->is_ampdu = 1;
else
cdp_rx_ppdu->is_ampdu = 0;
cdp_rx_ppdu->tid = ppdu_info->rx_status.tid;
cdp_rx_ppdu->lsig_a = ppdu_info->rx_status.rate;
ast_index = ppdu_info->rx_status.ast_index;
if (ast_index > (WLAN_UMAC_PSOC_MAX_PEERS * 2)) {
cdp_rx_ppdu->peer_id = HTT_INVALID_PEER;
return;
}
ast_entry = soc->ast_table[ast_index];
if (!ast_entry) {
cdp_rx_ppdu->peer_id = HTT_INVALID_PEER;
return;
}
peer = ast_entry->peer;
if (!peer || peer->peer_ids[0] == HTT_INVALID_PEER) {
cdp_rx_ppdu->peer_id = HTT_INVALID_PEER;
return;
}
qdf_mem_copy(cdp_rx_ppdu->mac_addr,
peer->mac_addr.raw, DP_MAC_ADDR_LEN);
cdp_rx_ppdu->peer_id = peer->peer_ids[0];
cdp_rx_ppdu->vdev_id = peer->vdev->vdev_id;
}
#else
static inline void
dp_rx_populate_cdp_indication_ppdu(struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info,
qdf_nbuf_t ppdu_nbuf)
{
}
#endif
/**
* dp_rx_stats_update() - Update per-peer statistics
* @soc: Datapath SOC handle
* @peer: Datapath peer handle
* @ppdu: PPDU Descriptor
*
* Return: None
*/
#ifdef FEATURE_PERPKT_INFO
static void dp_rx_stats_update(struct dp_soc *soc, struct dp_peer *peer,
struct cdp_rx_indication_ppdu *ppdu)
{
struct dp_pdev *pdev = NULL;
uint8_t mcs, preamble, ac = 0;
uint16_t num_msdu;
mcs = ppdu->u.mcs;
preamble = ppdu->u.preamble;
num_msdu = ppdu->num_msdu;
if (!peer)
return;
pdev = peer->vdev->pdev;
dp_mark_peer_inact(peer, false);
if (soc->process_rx_status)
return;
DP_STATS_UPD(peer, rx.rssi, ppdu->rssi);
if ((preamble == DOT11_A) || (preamble == DOT11_B))
ppdu->u.nss = 1;
if (ppdu->u.nss)
DP_STATS_INC(peer, rx.nss[ppdu->u.nss - 1], num_msdu);
DP_STATS_INC(peer, rx.sgi_count[ppdu->u.gi], num_msdu);
DP_STATS_INC(peer, rx.bw[ppdu->u.bw], num_msdu);
DP_STATS_INC(peer, rx.reception_type[ppdu->u.ppdu_type], num_msdu);
DP_STATS_INCC(peer, rx.ampdu_cnt, num_msdu, ppdu->is_ampdu);
DP_STATS_INCC(peer, rx.non_ampdu_cnt, num_msdu, !(ppdu->is_ampdu));
DP_STATS_UPD(peer, rx.rx_rate, mcs);
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[MAX_MCS - 1], num_msdu,
((mcs >= MAX_MCS_11A) && (preamble == DOT11_A)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[mcs], num_msdu,
((mcs < MAX_MCS_11A) && (preamble == DOT11_A)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[MAX_MCS - 1], num_msdu,
((mcs >= MAX_MCS_11B) && (preamble == DOT11_B)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[mcs], num_msdu,
((mcs < MAX_MCS_11B) && (preamble == DOT11_B)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[MAX_MCS - 1], num_msdu,
((mcs >= MAX_MCS_11A) && (preamble == DOT11_N)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[mcs], num_msdu,
((mcs < MAX_MCS_11A) && (preamble == DOT11_N)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[MAX_MCS - 1], num_msdu,
((mcs >= MAX_MCS_11AC) && (preamble == DOT11_AC)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[mcs], num_msdu,
((mcs < MAX_MCS_11AC) && (preamble == DOT11_AC)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[MAX_MCS - 1], num_msdu,
((mcs >= (MAX_MCS - 1)) && (preamble == DOT11_AX)));
DP_STATS_INCC(peer,
rx.pkt_type[preamble].mcs_count[mcs], num_msdu,
((mcs < (MAX_MCS - 1)) && (preamble == DOT11_AX)));
/*
* If invalid TID, it could be a non-qos frame, hence do not update
* any AC counters
*/
ac = TID_TO_WME_AC(ppdu->tid);
if (ppdu->tid != HAL_TID_INVALID)
DP_STATS_INC(peer, rx.wme_ac_type[ac], num_msdu);
if (soc->cdp_soc.ol_ops->update_dp_stats) {
soc->cdp_soc.ol_ops->update_dp_stats(pdev->osif_pdev,
&peer->stats, ppdu->peer_id,
UPDATE_PEER_STATS);
}
}
#endif
/**
* dp_rx_handle_mcopy_mode() - Allocate and deliver first MSDU payload
* @soc: core txrx main context
* @pdev: pdev strcuture
* @ppdu_info: structure for rx ppdu ring
*
* Return: QDF_STATUS_SUCCESS - If nbuf to be freed by caller
* QDF_STATUS_E_ALREADY - If nbuf not to be freed by caller
*/
#ifdef FEATURE_PERPKT_INFO
static inline QDF_STATUS
dp_rx_handle_mcopy_mode(struct dp_soc *soc, struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info, qdf_nbuf_t nbuf)
{
uint8_t size = 0;
if (ppdu_info->msdu_info.first_msdu_payload == NULL)
return QDF_STATUS_SUCCESS;
if (pdev->m_copy_id.rx_ppdu_id == ppdu_info->com_info.ppdu_id)
return QDF_STATUS_SUCCESS;
pdev->m_copy_id.rx_ppdu_id = ppdu_info->com_info.ppdu_id;
/* Include 2 bytes of reserved space appended to the msdu payload */
size = (ppdu_info->msdu_info.first_msdu_payload -
qdf_nbuf_data(nbuf)) + 2;
ppdu_info->msdu_info.first_msdu_payload = NULL;
if (qdf_nbuf_pull_head(nbuf, size) == NULL)
return QDF_STATUS_SUCCESS;
/* only retain RX MSDU payload in the skb */
qdf_nbuf_trim_tail(nbuf, qdf_nbuf_len(nbuf) -
ppdu_info->msdu_info.payload_len);
dp_wdi_event_handler(WDI_EVENT_RX_DATA, soc,
nbuf, HTT_INVALID_PEER, WDI_NO_VAL, pdev->pdev_id);
return QDF_STATUS_E_ALREADY;
}
#else
static inline QDF_STATUS
dp_rx_handle_mcopy_mode(struct dp_soc *soc, struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info, qdf_nbuf_t nbuf)
{
return QDF_STATUS_SUCCESS;
}
#endif
/**
* dp_rx_handle_ppdu_stats() - Allocate and deliver ppdu stats to cdp layer
* @soc: core txrx main context
* @pdev: pdev strcuture
* @ppdu_info: structure for rx ppdu ring
*
* Return: none
*/
#ifdef FEATURE_PERPKT_INFO
static inline void
dp_rx_handle_ppdu_stats(struct dp_soc *soc, struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info)
{
qdf_nbuf_t ppdu_nbuf;
struct dp_peer *peer;
struct cdp_rx_indication_ppdu *cdp_rx_ppdu;
/*
* Do not allocate if fcs error,
* ast idx invalid / fctl invalid
*/
if (ppdu_info->com_info.mpdu_cnt_fcs_ok == 0)
return;
if (!pdev->mcopy_mode) {
if (!ppdu_info->rx_status.frame_control_info_valid)
return;
if (ppdu_info->rx_status.ast_index == HAL_AST_IDX_INVALID)
return;
}
ppdu_nbuf = qdf_nbuf_alloc(soc->osdev,
sizeof(struct hal_rx_ppdu_info), 0, 0, FALSE);
if (ppdu_nbuf) {
dp_rx_populate_cdp_indication_ppdu(pdev, ppdu_info, ppdu_nbuf);
qdf_nbuf_put_tail(ppdu_nbuf,
sizeof(struct cdp_rx_indication_ppdu));
cdp_rx_ppdu = (struct cdp_rx_indication_ppdu *)ppdu_nbuf->data;
peer = dp_peer_find_by_id(soc, cdp_rx_ppdu->peer_id);
if (peer && cdp_rx_ppdu->peer_id != HTT_INVALID_PEER) {
dp_rx_stats_update(soc, peer, cdp_rx_ppdu);
dp_wdi_event_handler(WDI_EVENT_RX_PPDU_DESC, soc,
ppdu_nbuf, cdp_rx_ppdu->peer_id,
WDI_NO_VAL, pdev->pdev_id);
} else if (pdev->mcopy_mode) {
dp_wdi_event_handler(WDI_EVENT_RX_PPDU_DESC, soc,
ppdu_nbuf, HTT_INVALID_PEER,
WDI_NO_VAL, pdev->pdev_id);
} else {
qdf_nbuf_free(ppdu_nbuf);
}
}
}
#else
static inline void
dp_rx_handle_ppdu_stats(struct dp_soc *soc, struct dp_pdev *pdev,
struct hal_rx_ppdu_info *ppdu_info)
{
}
#endif
/**
* dp_rx_mon_status_process_tlv() - Process status TLV in status
* buffer on Rx status Queue posted by status SRNG processing.
* @soc: core txrx main context
* @mac_id: mac_id which is one of 3 mac_ids _ring
*
* Return: none
*/
static inline void
dp_rx_mon_status_process_tlv(struct dp_soc *soc, uint32_t mac_id,
uint32_t quota)
{
struct dp_pdev *pdev = dp_get_pdev_for_mac_id(soc, mac_id);
struct hal_rx_ppdu_info *ppdu_info;
qdf_nbuf_t status_nbuf;
uint8_t *rx_tlv;
uint8_t *rx_tlv_start;
uint32_t tlv_status = HAL_TLV_STATUS_BUF_DONE;
QDF_STATUS m_copy_status = QDF_STATUS_SUCCESS;
struct cdp_pdev_mon_stats *rx_mon_stats;
ppdu_info = &pdev->ppdu_info;
rx_mon_stats = &pdev->rx_mon_stats;
if (pdev->mon_ppdu_status != DP_PPDU_STATUS_START)
return;
while (!qdf_nbuf_is_queue_empty(&pdev->rx_status_q)) {
status_nbuf = qdf_nbuf_queue_remove(&pdev->rx_status_q);
rx_tlv = qdf_nbuf_data(status_nbuf);
rx_tlv_start = rx_tlv;
#ifndef REMOVE_PKT_LOG
#if defined(WDI_EVENT_ENABLE)
dp_wdi_event_handler(WDI_EVENT_RX_DESC, soc,
status_nbuf, HTT_INVALID_PEER, WDI_NO_VAL, mac_id);
#endif
#endif
if ((pdev->monitor_vdev != NULL) || (pdev->enhanced_stats_en) ||
pdev->mcopy_mode) {
do {
tlv_status = hal_rx_status_get_tlv_info(rx_tlv,
ppdu_info);
dp_rx_mon_update_dbg_ppdu_stats(ppdu_info,
rx_mon_stats);
rx_tlv = hal_rx_status_get_next_tlv(rx_tlv);
if ((rx_tlv - rx_tlv_start) >= RX_BUFFER_SIZE)
break;
} while (tlv_status == HAL_TLV_STATUS_PPDU_NOT_DONE);
}
if (pdev->mcopy_mode) {
m_copy_status = dp_rx_handle_mcopy_mode(soc,
pdev, ppdu_info, status_nbuf);
if (m_copy_status == QDF_STATUS_SUCCESS)
qdf_nbuf_free(status_nbuf);
} else {
qdf_nbuf_free(status_nbuf);
}
if (tlv_status == HAL_TLV_STATUS_PPDU_DONE) {
rx_mon_stats->status_ppdu_done++;
if (pdev->enhanced_stats_en ||
pdev->mcopy_mode)
dp_rx_handle_ppdu_stats(soc, pdev, ppdu_info);
pdev->mon_ppdu_status = DP_PPDU_STATUS_DONE;
dp_rx_mon_dest_process(soc, mac_id, quota);
pdev->mon_ppdu_status = DP_PPDU_STATUS_START;
pdev->ppdu_info.com_info.last_ppdu_id =
pdev->ppdu_info.com_info.ppdu_id;
}
}
return;
}
/*
* dp_rx_mon_status_srng_process() - Process monitor status ring
* post the status ring buffer to Rx status Queue for later
* processing when status ring is filled with status TLV.
* Allocate a new buffer to status ring if the filled buffer
* is posted.
*
* @soc: core txrx main context
* @mac_id: mac_id which is one of 3 mac_ids
* @quota: No. of ring entry that can be serviced in one shot.
* Return: uint32_t: No. of ring entry that is processed.
*/
static inline uint32_t
dp_rx_mon_status_srng_process(struct dp_soc *soc, uint32_t mac_id,
uint32_t quota)
{
struct dp_pdev *pdev = dp_get_pdev_for_mac_id(soc, mac_id);
void *hal_soc;
void *mon_status_srng;
void *rxdma_mon_status_ring_entry;
QDF_STATUS status;
uint32_t work_done = 0;
int mac_for_pdev = dp_get_mac_id_for_mac(soc, mac_id);
mon_status_srng = pdev->rxdma_mon_status_ring[mac_for_pdev].hal_srng;
qdf_assert(mon_status_srng);
if (!mon_status_srng || !hal_srng_initialized(mon_status_srng)) {
QDF_TRACE(QDF_MODULE_ID_TXRX, QDF_TRACE_LEVEL_ERROR,
"%s %d : HAL Monitor Status Ring Init Failed -- %pK\n",
__func__, __LINE__, mon_status_srng);
return work_done;
}
hal_soc = soc->hal_soc;
qdf_assert(hal_soc);
if (qdf_unlikely(hal_srng_access_start(hal_soc, mon_status_srng)))
goto done;
/* mon_status_ring_desc => WBM_BUFFER_RING STRUCT =>
* BUFFER_ADDR_INFO STRUCT
*/
while (qdf_likely((rxdma_mon_status_ring_entry =
hal_srng_src_peek(hal_soc, mon_status_srng))
&& quota--)) {
uint32_t rx_buf_cookie;
qdf_nbuf_t status_nbuf;
struct dp_rx_desc *rx_desc;
uint8_t *status_buf;
qdf_dma_addr_t paddr;
uint64_t buf_addr;
buf_addr =
(HAL_RX_BUFFER_ADDR_31_0_GET(
rxdma_mon_status_ring_entry) |
((uint64_t)(HAL_RX_BUFFER_ADDR_39_32_GET(
rxdma_mon_status_ring_entry)) << 32));
if (qdf_likely(buf_addr)) {
rx_buf_cookie =
HAL_RX_BUF_COOKIE_GET(
rxdma_mon_status_ring_entry);
rx_desc = dp_rx_cookie_2_va_mon_status(soc,
rx_buf_cookie);
qdf_assert(rx_desc);
status_nbuf = rx_desc->nbuf;
qdf_nbuf_sync_for_cpu(soc->osdev, status_nbuf,
QDF_DMA_FROM_DEVICE);
status_buf = qdf_nbuf_data(status_nbuf);
status = hal_get_rx_status_done(status_buf);
if (status != QDF_STATUS_SUCCESS) {
uint32_t hp, tp;
hal_api_get_tphp(hal_soc, mon_status_srng,
&tp, &hp);
QDF_TRACE(QDF_MODULE_ID_DP,
QDF_TRACE_LEVEL_ERROR,
"[%s][%d] status not done - hp:%u, tp:%u",
__func__, __LINE__, hp, tp);
/* WAR for missing status: Skip status entry */
hal_srng_src_get_next(hal_soc, mon_status_srng);
continue;
}
qdf_nbuf_set_pktlen(status_nbuf, RX_BUFFER_SIZE);
qdf_nbuf_unmap_single(soc->osdev, status_nbuf,
QDF_DMA_FROM_DEVICE);
/* Put the status_nbuf to queue */
qdf_nbuf_queue_add(&pdev->rx_status_q, status_nbuf);
} else {
union dp_rx_desc_list_elem_t *desc_list = NULL;
union dp_rx_desc_list_elem_t *tail = NULL;
struct rx_desc_pool *rx_desc_pool;
uint32_t num_alloc_desc;
rx_desc_pool = &soc->rx_desc_status[mac_id];
num_alloc_desc = dp_rx_get_free_desc_list(soc, mac_id,
rx_desc_pool,
1,
&desc_list,
&tail);
rx_desc = &desc_list->rx_desc;
}
status_nbuf = dp_rx_nbuf_prepare(soc, pdev);
/*
* qdf_nbuf alloc or map failed,
* free the dp rx desc to free list,
* fill in NULL dma address at current HP entry,
* keep HP in mon_status_ring unchanged,
* wait next time dp_rx_mon_status_srng_process
* to fill in buffer at current HP.
*/
if (qdf_unlikely(status_nbuf == NULL)) {
union dp_rx_desc_list_elem_t *desc_list = NULL;
union dp_rx_desc_list_elem_t *tail = NULL;
struct rx_desc_pool *rx_desc_pool;
rx_desc_pool = &soc->rx_desc_status[mac_id];
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"%s: fail to allocate or map qdf_nbuf",
__func__);
dp_rx_add_to_free_desc_list(&desc_list,
&tail, rx_desc);
dp_rx_add_desc_list_to_free_list(soc, &desc_list,
&tail, mac_id, rx_desc_pool);
hal_rxdma_buff_addr_info_set(
rxdma_mon_status_ring_entry,
0, 0, HAL_RX_BUF_RBM_SW3_BM);
work_done++;
break;
}
paddr = qdf_nbuf_get_frag_paddr(status_nbuf, 0);
rx_desc->nbuf = status_nbuf;
rx_desc->in_use = 1;
hal_rxdma_buff_addr_info_set(rxdma_mon_status_ring_entry,
paddr, rx_desc->cookie, HAL_RX_BUF_RBM_SW3_BM);
hal_srng_src_get_next(hal_soc, mon_status_srng);
work_done++;
}
done:
hal_srng_access_end(hal_soc, mon_status_srng);
return work_done;
}
/*
* dp_rx_mon_status_process() - Process monitor status ring and
* TLV in status ring.
*
* @soc: core txrx main context
* @mac_id: mac_id which is one of 3 mac_ids
* @quota: No. of ring entry that can be serviced in one shot.
* Return: uint32_t: No. of ring entry that is processed.
*/
static inline uint32_t
dp_rx_mon_status_process(struct dp_soc *soc, uint32_t mac_id, uint32_t quota) {
uint32_t work_done;
work_done = dp_rx_mon_status_srng_process(soc, mac_id, quota);
quota -= work_done;
dp_rx_mon_status_process_tlv(soc, mac_id, quota);
return work_done;
}
/**
* dp_mon_process() - Main monitor mode processing roution.
* This call monitor status ring process then monitor
* destination ring process.
* Called from the bottom half (tasklet/NET_RX_SOFTIRQ)
* @soc: core txrx main context
* @mac_id: mac_id which is one of 3 mac_ids
* @quota: No. of status ring entry that can be serviced in one shot.
* Return: uint32_t: No. of ring entry that is processed.
*/
uint32_t
dp_mon_process(struct dp_soc *soc, uint32_t mac_id, uint32_t quota) {
return dp_rx_mon_status_process(soc, mac_id, quota);
}
/**
* dp_rx_pdev_mon_detach() - detach dp rx for status ring
* @pdev: core txrx pdev context
* @mac_id: mac_id/pdev_id correspondinggly for MCL and WIN
*
* This function will detach DP RX status ring from
* main device context. will free DP Rx resources for
* status ring
*
* Return: QDF_STATUS_SUCCESS: success
* QDF_STATUS_E_RESOURCES: Error return
*/
QDF_STATUS
dp_rx_pdev_mon_status_detach(struct dp_pdev *pdev, int mac_id)
{
struct dp_soc *soc = pdev->soc;
struct rx_desc_pool *rx_desc_pool;
rx_desc_pool = &soc->rx_desc_status[mac_id];
if (rx_desc_pool->pool_size != 0)
dp_rx_desc_pool_free(soc, mac_id, rx_desc_pool);
return QDF_STATUS_SUCCESS;
}
/*
* dp_rx_buffers_replenish() - replenish monitor status ring with
* rx nbufs called during dp rx
* monitor status ring initialization
*
* @soc: core txrx main context
* @mac_id: mac_id which is one of 3 mac_ids
* @dp_rxdma_srng: dp monitor status circular ring
* @rx_desc_pool; Pointer to Rx descriptor pool
* @num_req_buffers: number of buffer to be replenished
* @desc_list: list of descs if called from dp rx monitor status
* process or NULL during dp rx initialization or
* out of buffer interrupt
* @tail: tail of descs list
* @owner: who owns the nbuf (host, NSS etc...)
* Return: return success or failure
*/
static inline
QDF_STATUS dp_rx_mon_status_buffers_replenish(struct dp_soc *dp_soc,
uint32_t mac_id,
struct dp_srng *dp_rxdma_srng,
struct rx_desc_pool *rx_desc_pool,
uint32_t num_req_buffers,
union dp_rx_desc_list_elem_t **desc_list,
union dp_rx_desc_list_elem_t **tail,
uint8_t owner)
{
uint32_t num_alloc_desc;
uint16_t num_desc_to_free = 0;
uint32_t num_entries_avail;
uint32_t count = 0;
int sync_hw_ptr = 1;
qdf_dma_addr_t paddr;
qdf_nbuf_t rx_netbuf;
void *rxdma_ring_entry;
union dp_rx_desc_list_elem_t *next;
void *rxdma_srng;
struct dp_pdev *dp_pdev = dp_get_pdev_for_mac_id(dp_soc, mac_id);
rxdma_srng = dp_rxdma_srng->hal_srng;
qdf_assert(rxdma_srng);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"[%s][%d] requested %d buffers for replenish\n",
__func__, __LINE__, num_req_buffers);
/*
* if desc_list is NULL, allocate the descs from freelist
*/
if (!(*desc_list)) {
num_alloc_desc = dp_rx_get_free_desc_list(dp_soc, mac_id,
rx_desc_pool,
num_req_buffers,
desc_list,
tail);
if (!num_alloc_desc) {
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"[%s][%d] no free rx_descs in freelist\n",
__func__, __LINE__);
return QDF_STATUS_E_NOMEM;
}
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"[%s][%d] %d rx desc allocated\n", __func__, __LINE__,
num_alloc_desc);
num_req_buffers = num_alloc_desc;
}
hal_srng_access_start(dp_soc->hal_soc, rxdma_srng);
num_entries_avail = hal_srng_src_num_avail(dp_soc->hal_soc,
rxdma_srng, sync_hw_ptr);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"[%s][%d] no of availble entries in rxdma ring: %d\n",
__func__, __LINE__, num_entries_avail);
if (num_entries_avail < num_req_buffers) {
num_desc_to_free = num_req_buffers - num_entries_avail;
num_req_buffers = num_entries_avail;
}
while (count < num_req_buffers) {
rx_netbuf = dp_rx_nbuf_prepare(dp_soc, dp_pdev);
/*
* qdf_nbuf alloc or map failed,
* keep HP in mon_status_ring unchanged,
* wait dp_rx_mon_status_srng_process
* to fill in buffer at current HP.
*/
if (qdf_unlikely(rx_netbuf == NULL)) {
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"%s: qdf_nbuf allocate or map fail, count %d",
__func__, count);
break;
}
paddr = qdf_nbuf_get_frag_paddr(rx_netbuf, 0);
next = (*desc_list)->next;
rxdma_ring_entry = hal_srng_src_get_next(dp_soc->hal_soc,
rxdma_srng);
if (qdf_unlikely(rxdma_ring_entry == NULL)) {
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"[%s][%d] rxdma_ring_entry is NULL, count - %d\n",
__func__, __LINE__, count);
qdf_nbuf_unmap_single(dp_soc->osdev, rx_netbuf,
QDF_DMA_BIDIRECTIONAL);
qdf_nbuf_free(rx_netbuf);
break;
}
(*desc_list)->rx_desc.nbuf = rx_netbuf;
(*desc_list)->rx_desc.in_use = 1;
count++;
hal_rxdma_buff_addr_info_set(rxdma_ring_entry, paddr,
(*desc_list)->rx_desc.cookie, owner);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"[%s][%d] rx_desc=%pK, cookie=%d, nbuf=%pK, \
paddr=%pK\n",
__func__, __LINE__, &(*desc_list)->rx_desc,
(*desc_list)->rx_desc.cookie, rx_netbuf,
(void *)paddr);
*desc_list = next;
}
hal_srng_access_end(dp_soc->hal_soc, rxdma_srng);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"successfully replenished %d buffers\n", num_req_buffers);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"%d rx desc added back to free list\n", num_desc_to_free);
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_DEBUG,
"[%s][%d] desc_list=%pK, tail=%pK rx_desc=%pK, cookie=%d\n",
__func__, __LINE__, desc_list, tail, &(*desc_list)->rx_desc,
(*desc_list)->rx_desc.cookie);
/*
* add any available free desc back to the free list
*/
if (*desc_list) {
dp_rx_add_desc_list_to_free_list(dp_soc, desc_list, tail,
mac_id, rx_desc_pool);
}
return QDF_STATUS_SUCCESS;
}
/**
* dp_rx_pdev_mon_status_attach() - attach DP RX monitor status ring
* @pdev: core txrx pdev context
*
* This function will attach a DP RX monitor status ring into pDEV
* and replenish monitor status ring with buffer.
*
* Return: QDF_STATUS_SUCCESS: success
* QDF_STATUS_E_RESOURCES: Error return
*/
QDF_STATUS
dp_rx_pdev_mon_status_attach(struct dp_pdev *pdev, int ring_id) {
struct dp_soc *soc = pdev->soc;
union dp_rx_desc_list_elem_t *desc_list = NULL;
union dp_rx_desc_list_elem_t *tail = NULL;
struct dp_srng *rxdma_srng;
uint32_t rxdma_entries;
struct rx_desc_pool *rx_desc_pool;
QDF_STATUS status;
int mac_for_pdev = dp_get_mac_id_for_mac(soc, ring_id);
rxdma_srng = &pdev->rxdma_mon_status_ring[mac_for_pdev];
rxdma_entries = rxdma_srng->alloc_size/hal_srng_get_entrysize(
soc->hal_soc, RXDMA_MONITOR_STATUS);
rx_desc_pool = &soc->rx_desc_status[ring_id];
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_INFO_LOW,
"%s: Mon RX Status Pool[%d] allocation size=%d\n",
__func__, ring_id, rxdma_entries);
status = dp_rx_desc_pool_alloc(soc, ring_id, rxdma_entries+1,
rx_desc_pool);
if (!QDF_IS_STATUS_SUCCESS(status)) {
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"%s: dp_rx_desc_pool_alloc() failed \n", __func__);
return status;
}
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_INFO_LOW,
"%s: Mon RX Status Buffers Replenish ring_id=%d\n",
__func__, ring_id);
status = dp_rx_mon_status_buffers_replenish(soc, ring_id, rxdma_srng,
rx_desc_pool, rxdma_entries, &desc_list, &tail,
HAL_RX_BUF_RBM_SW3_BM);
if (!QDF_IS_STATUS_SUCCESS(status)) {
QDF_TRACE(QDF_MODULE_ID_DP, QDF_TRACE_LEVEL_ERROR,
"%s: dp_rx_buffers_replenish() failed \n", __func__);
return status;
}
qdf_nbuf_queue_init(&pdev->rx_status_q);
pdev->mon_ppdu_status = DP_PPDU_STATUS_START;
pdev->ppdu_info.com_info.last_ppdu_id = 0;
qdf_mem_zero(&(pdev->ppdu_info.rx_status),
sizeof(pdev->ppdu_info.rx_status));
qdf_mem_zero(&pdev->rx_mon_stats,
sizeof(pdev->rx_mon_stats));
dp_rx_mon_init_dbg_ppdu_stats(&pdev->ppdu_info,
&pdev->rx_mon_stats);
return QDF_STATUS_SUCCESS;
}