leandrof/linux
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
39aead8373
linux/drivers/net/wireless/mediatek/mt76/mt7615/dma.c
Felix Fietkau 781eef5b34 mt76: convert from tx tasklet to tx worker thread 
This improves performance by allowing the scheduler to move the tx scheduling
work to idle CPUs. Since tx scheduling work is very latency sensitive and
kept short via AQL, sched_set_fifo_low is used to keep worker priority above
normal tasks

Signed-off-by: Felix Fietkau <nbd@nbd.name>
2020-09-24 18:10:18 +02:00

302 lines
7.7 KiB
C
Raw Blame History

// SPDX-License-Identifier: ISC
/* Copyright (C) 2019 MediaTek Inc.
*
* Author: Ryder Lee <ryder.lee@mediatek.com>
* Roy Luo <royluo@google.com>
* Lorenzo Bianconi <lorenzo@kernel.org>
* Felix Fietkau <nbd@nbd.name>
*/
#include "mt7615.h"
#include "../dma.h"
#include "mac.h"
static int
mt7615_init_tx_queue(struct mt7615_dev *dev, int qid, int idx, int n_desc)
{
struct mt76_queue *hwq;
int err;
hwq = devm_kzalloc(dev->mt76.dev, sizeof(*hwq), GFP_KERNEL);
if (!hwq)
return -ENOMEM;
err = mt76_queue_alloc(dev, hwq, idx, n_desc, 0, MT_TX_RING_BASE);
if (err < 0)
return err;
dev->mt76.q_tx[qid] = hwq;
return 0;
}
static int
mt7622_init_tx_queues_multi(struct mt7615_dev *dev)
{
static const u8 wmm_queue_map[] = {
[IEEE80211_AC_BK] = MT7622_TXQ_AC0,
[IEEE80211_AC_BE] = MT7622_TXQ_AC1,
[IEEE80211_AC_VI] = MT7622_TXQ_AC2,
[IEEE80211_AC_VO] = MT7622_TXQ_AC3,
};
int ret;
int i;
for (i = 0; i < ARRAY_SIZE(wmm_queue_map); i++) {
ret = mt7615_init_tx_queue(dev, i, wmm_queue_map[i],
MT7615_TX_RING_SIZE / 2);
if (ret)
return ret;
}
ret = mt7615_init_tx_queue(dev, MT_TXQ_PSD,
MT7622_TXQ_MGMT, MT7615_TX_MGMT_RING_SIZE);
if (ret)
return ret;
ret = mt7615_init_tx_queue(dev, MT_TXQ_MCU,
MT7622_TXQ_MCU, MT7615_TX_MCU_RING_SIZE);
return ret;
}
static int
mt7615_init_tx_queues(struct mt7615_dev *dev)
{
int ret, i;
ret = mt7615_init_tx_queue(dev, MT_TXQ_FWDL,
MT7615_TXQ_FWDL,
MT7615_TX_FWDL_RING_SIZE);
if (ret)
return ret;
if (!is_mt7615(&dev->mt76))
return mt7622_init_tx_queues_multi(dev);
ret = mt7615_init_tx_queue(dev, 0, 0, MT7615_TX_RING_SIZE);
if (ret)
return ret;
for (i = 1; i < MT_TXQ_MCU; i++)
dev->mt76.q_tx[i] = dev->mt76.q_tx[0];
ret = mt7615_init_tx_queue(dev, MT_TXQ_MCU, MT7615_TXQ_MCU,
MT7615_TX_MCU_RING_SIZE);
return 0;
}
static int mt7615_poll_tx(struct napi_struct *napi, int budget)
{
struct mt7615_dev *dev;
dev = container_of(napi, struct mt7615_dev, mt76.tx_napi);
mt76_queue_tx_cleanup(dev, MT_TXQ_MCU, false);
if (napi_complete_done(napi, 0))
mt7615_irq_enable(dev, mt7615_tx_mcu_int_mask(dev));
return 0;
}
int mt7615_wait_pdma_busy(struct mt7615_dev *dev)
{
struct mt76_dev *mdev = &dev->mt76;
if (!is_mt7663(mdev)) {
u32 mask = MT_PDMA_TX_BUSY | MT_PDMA_RX_BUSY;
u32 reg = mt7615_reg_map(dev, MT_PDMA_BUSY);
if (!mt76_poll_msec(dev, reg, mask, 0, 1000)) {
dev_err(mdev->dev, "PDMA engine busy\n");
return -EIO;
}
return 0;
}
if (!mt76_poll_msec(dev, MT_PDMA_BUSY_STATUS,
MT_PDMA_TX_IDX_BUSY, 0, 1000)) {
dev_err(mdev->dev, "PDMA engine tx busy\n");
return -EIO;
}
if (!mt76_poll_msec(dev, MT_PSE_PG_INFO,
MT_PSE_SRC_CNT, 0, 1000)) {
dev_err(mdev->dev, "PSE engine busy\n");
return -EIO;
}
if (!mt76_poll_msec(dev, MT_PDMA_BUSY_STATUS,
MT_PDMA_BUSY_IDX, 0, 1000)) {
dev_err(mdev->dev, "PDMA engine busy\n");
return -EIO;
}
return 0;
}
static void mt7622_dma_sched_init(struct mt7615_dev *dev)
{
u32 reg = mt7615_reg_map(dev, MT_DMASHDL_BASE);
int i;
mt76_rmw(dev, reg + MT_DMASHDL_PKT_MAX_SIZE,
MT_DMASHDL_PKT_MAX_SIZE_PLE | MT_DMASHDL_PKT_MAX_SIZE_PSE,
FIELD_PREP(MT_DMASHDL_PKT_MAX_SIZE_PLE, 1) |
FIELD_PREP(MT_DMASHDL_PKT_MAX_SIZE_PSE, 8));
for (i = 0; i <= 5; i++)
mt76_wr(dev, reg + MT_DMASHDL_GROUP_QUOTA(i),
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MIN, 0x10) |
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MAX, 0x800));
mt76_wr(dev, reg + MT_DMASHDL_Q_MAP(0), 0x42104210);
mt76_wr(dev, reg + MT_DMASHDL_Q_MAP(1), 0x42104210);
mt76_wr(dev, reg + MT_DMASHDL_Q_MAP(2), 0x5);
mt76_wr(dev, reg + MT_DMASHDL_Q_MAP(3), 0);
mt76_wr(dev, reg + MT_DMASHDL_SCHED_SET0, 0x6012345f);
mt76_wr(dev, reg + MT_DMASHDL_SCHED_SET1, 0xedcba987);
}
static void mt7663_dma_sched_init(struct mt7615_dev *dev)
{
int i;
mt76_rmw(dev, MT_DMA_SHDL(MT_DMASHDL_PKT_MAX_SIZE),
MT_DMASHDL_PKT_MAX_SIZE_PLE | MT_DMASHDL_PKT_MAX_SIZE_PSE,
FIELD_PREP(MT_DMASHDL_PKT_MAX_SIZE_PLE, 1) |
FIELD_PREP(MT_DMASHDL_PKT_MAX_SIZE_PSE, 8));
/* enable refill control group 0, 1, 2, 4, 5 */
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_REFILL), 0xffc80000);
/* enable group 0, 1, 2, 4, 5, 15 */
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_OPTIONAL), 0x70068037);
/* each group min quota must larger then PLE_PKT_MAX_SIZE_NUM */
for (i = 0; i < 5; i++)
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_GROUP_QUOTA(i)),
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MIN, 0x40) |
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MAX, 0x800));
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_GROUP_QUOTA(5)),
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MIN, 0x40) |
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MAX, 0x40));
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_GROUP_QUOTA(15)),
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MIN, 0x20) |
FIELD_PREP(MT_DMASHDL_GROUP_QUOTA_MAX, 0x20));
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_Q_MAP(0)), 0x42104210);
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_Q_MAP(1)), 0x42104210);
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_Q_MAP(2)), 0x00050005);
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_Q_MAP(3)), 0);
/* ALTX0 and ALTX1 QID mapping to group 5 */
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_SCHED_SET0), 0x6012345f);
mt76_wr(dev, MT_DMA_SHDL(MT_DMASHDL_SCHED_SET1), 0xedcba987);
}
int mt7615_dma_init(struct mt7615_dev *dev)
{
int rx_ring_size = MT7615_RX_RING_SIZE;
int rx_buf_size = MT_RX_BUF_SIZE;
int ret;
/* Increase buffer size to receive large VHT MPDUs */
if (dev->mt76.cap.has_5ghz)
rx_buf_size *= 2;
mt76_dma_attach(&dev->mt76);
mt76_wr(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_WRITEBACK_DONE |
MT_WPDMA_GLO_CFG_FIFO_LITTLE_ENDIAN |
MT_WPDMA_GLO_CFG_OMIT_TX_INFO);
mt76_rmw_field(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_BT_SIZE_BIT0, 0x1);
mt76_rmw_field(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_BT_SIZE_BIT21, 0x1);
mt76_rmw_field(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_DMA_BURST_SIZE, 0x3);
mt76_rmw_field(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_MULTI_DMA_EN, 0x3);
if (is_mt7615(&dev->mt76)) {
mt76_set(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_FIRST_TOKEN_ONLY);
mt76_wr(dev, MT_WPDMA_GLO_CFG1, 0x1);
mt76_wr(dev, MT_WPDMA_TX_PRE_CFG, 0xf0000);
mt76_wr(dev, MT_WPDMA_RX_PRE_CFG, 0xf7f0000);
mt76_wr(dev, MT_WPDMA_ABT_CFG, 0x4000026);
mt76_wr(dev, MT_WPDMA_ABT_CFG1, 0x18811881);
mt76_set(dev, 0x7158, BIT(16));
mt76_clear(dev, 0x7000, BIT(23));
}
mt76_wr(dev, MT_WPDMA_RST_IDX, ~0);
ret = mt7615_init_tx_queues(dev);
if (ret)
return ret;
/* init rx queues */
ret = mt76_queue_alloc(dev, &dev->mt76.q_rx[MT_RXQ_MCU], 1,
MT7615_RX_MCU_RING_SIZE, rx_buf_size,
MT_RX_RING_BASE);
if (ret)
return ret;
if (!is_mt7615(&dev->mt76))
rx_ring_size /= 2;
ret = mt76_queue_alloc(dev, &dev->mt76.q_rx[MT_RXQ_MAIN], 0,
rx_ring_size, rx_buf_size, MT_RX_RING_BASE);
if (ret)
return ret;
mt76_wr(dev, MT_DELAY_INT_CFG, 0);
ret = mt76_init_queues(dev);
if (ret < 0)
return ret;
netif_tx_napi_add(&dev->mt76.napi_dev, &dev->mt76.tx_napi,
mt7615_poll_tx, NAPI_POLL_WEIGHT);
napi_enable(&dev->mt76.tx_napi);
mt76_poll(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_DMA_BUSY |
MT_WPDMA_GLO_CFG_RX_DMA_BUSY, 0, 1000);
/* start dma engine */
mt76_set(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_DMA_EN |
MT_WPDMA_GLO_CFG_RX_DMA_EN);
/* enable interrupts for TX/RX rings */
mt7615_irq_enable(dev, MT_INT_RX_DONE_ALL | mt7615_tx_mcu_int_mask(dev) |
MT_INT_MCU_CMD);
if (is_mt7622(&dev->mt76))
mt7622_dma_sched_init(dev);
if (is_mt7663(&dev->mt76))
mt7663_dma_sched_init(dev);
return 0;
}
void mt7615_dma_cleanup(struct mt7615_dev *dev)
{
mt76_clear(dev, MT_WPDMA_GLO_CFG,
MT_WPDMA_GLO_CFG_TX_DMA_EN |
MT_WPDMA_GLO_CFG_RX_DMA_EN);
mt76_set(dev, MT_WPDMA_GLO_CFG, MT_WPDMA_GLO_CFG_SW_RESET);
mt76_dma_cleanup(&dev->mt76);
}
Reference in New Issue View Git Blame Copy Permalink