Difuse/difos
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 2

ramips: improve tx clean up and add fe_tx_ring struct

Browse source 
if there is any new tx need to clean up. do it in next napi poll.
collect tx related members to fe_tx_ring struct. for better
cache usage and more readable.

Signed-off-by: michael lee <igvtee@gmail.com>

SVN-Revision: 45895
This commit is contained in:
Felix Fietkau 2015-06-05 08:59:57 +00:00
parent 2a66c9886f
commit 4480fbce62
3 changed files with 82 additions and 69 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
target/linux/ramips/files/drivers/net/ethernet/ralink/ralink_ethtool.c
View file

@ -144,7 +144,7 @@ static int fe_set_ringparam(struct net_device *dev,
dev->netdev_ops->ndo_stop(dev); dev->netdev_ops->ndo_stop(dev);
priv->tx_ring_size = BIT(fls(ring->tx_pending) - 1); priv->tx_ring.tx_ring_size = BIT(fls(ring->tx_pending) - 1);
priv->rx_ring_size = BIT(fls(ring->rx_pending) - 1); priv->rx_ring_size = BIT(fls(ring->rx_pending) - 1);
dev->netdev_ops->ndo_open(dev); dev->netdev_ops->ndo_open(dev);
@ -160,7 +160,7 @@ static void fe_get_ringparam(struct net_device *dev,
ring->rx_max_pending = MAX_DMA_DESC; ring->rx_max_pending = MAX_DMA_DESC;
ring->tx_max_pending = MAX_DMA_DESC; ring->tx_max_pending = MAX_DMA_DESC;
ring->rx_pending = priv->rx_ring_size; ring->rx_pending = priv->rx_ring_size;
ring->tx_pending = priv->tx_ring_size; ring->tx_pending = priv->tx_ring.tx_ring_size;
} }
static void fe_get_strings(struct net_device *dev, u32 stringset, u8 *data) static void fe_get_strings(struct net_device *dev, u32 stringset, u8 *data)

124
target/linux/ramips/files/drivers/net/ethernet/ralink/ralink_soc_eth.c
View file

@ -56,7 +56,7 @@
#define TX_DMA_DESP2_DEF (TX_DMA_LS0 | TX_DMA_DONE) #define TX_DMA_DESP2_DEF (TX_DMA_LS0 | TX_DMA_DONE)
#define TX_DMA_DESP4_DEF (TX_DMA_QN(3) | TX_DMA_PN(1)) #define TX_DMA_DESP4_DEF (TX_DMA_QN(3) | TX_DMA_PN(1))
#define NEXT_TX_DESP_IDX(X) (((X) + 1) & (priv->tx_ring_size - 1)) #define NEXT_TX_DESP_IDX(X) (((X) + 1) & (ring->tx_ring_size - 1))
#define NEXT_RX_DESP_IDX(X) (((X) + 1) & (priv->rx_ring_size - 1)) #define NEXT_RX_DESP_IDX(X) (((X) + 1) & (priv->rx_ring_size - 1))
#define SYSC_REG_RSTCTRL 0x34 #define SYSC_REG_RSTCTRL 0x34
@ -310,51 +310,56 @@ static void fe_txd_unmap(struct device *dev, struct fe_tx_buf *tx_buf)
static void fe_clean_tx(struct fe_priv *priv) static void fe_clean_tx(struct fe_priv *priv)
{ {
int i; int i;
struct device *dev = &priv->netdev->dev;
struct fe_tx_ring *ring = &priv->tx_ring;
if (priv->tx_buf) { if (ring->tx_buf) {
for (i = 0; i < priv->tx_ring_size; i++) for (i = 0; i < ring->tx_ring_size; i++)
fe_txd_unmap(&priv->netdev->dev, &priv->tx_buf[i]); fe_txd_unmap(dev, &ring->tx_buf[i]);
kfree(priv->tx_buf); kfree(ring->tx_buf);
priv->tx_buf = NULL; ring->tx_buf = NULL;
} }
if (priv->tx_dma) { if (ring->tx_dma) {
dma_free_coherent(&priv->netdev->dev, dma_free_coherent(dev,
priv->tx_ring_size * sizeof(*priv->tx_dma), ring->tx_ring_size * sizeof(*ring->tx_dma),
priv->tx_dma, ring->tx_dma,
priv->tx_phys); ring->tx_phys);
priv->tx_dma = NULL; ring->tx_dma = NULL;
} }
netdev_reset_queue(priv->netdev);
} }
static int fe_alloc_tx(struct fe_priv *priv) static int fe_alloc_tx(struct fe_priv *priv)
{ {
int i; int i;
struct fe_tx_ring *ring = &priv->tx_ring;
priv->tx_free_idx = 0; ring->tx_free_idx = 0;
priv->tx_buf = kcalloc(priv->tx_ring_size, sizeof(*priv->tx_buf), ring->tx_buf = kcalloc(ring->tx_ring_size, sizeof(*ring->tx_buf),
GFP_KERNEL); GFP_KERNEL);
if (!priv->tx_buf) if (!ring->tx_buf)
goto no_tx_mem; goto no_tx_mem;
priv->tx_dma = dma_alloc_coherent(&priv->netdev->dev, ring->tx_dma = dma_alloc_coherent(&priv->netdev->dev,
priv->tx_ring_size * sizeof(*priv->tx_dma), ring->tx_ring_size * sizeof(*ring->tx_dma),
&priv->tx_phys, &ring->tx_phys,
GFP_ATOMIC | __GFP_ZERO); GFP_ATOMIC | __GFP_ZERO);
if (!priv->tx_dma) if (!ring->tx_dma)
goto no_tx_mem; goto no_tx_mem;
for (i = 0; i < priv->tx_ring_size; i++) { for (i = 0; i < ring->tx_ring_size; i++) {
if (priv->soc->tx_dma) { if (priv->soc->tx_dma) {
priv->soc->tx_dma(&priv->tx_dma[i]); priv->soc->tx_dma(&ring->tx_dma[i]);
} }
priv->tx_dma[i].txd2 = TX_DMA_DESP2_DEF; ring->tx_dma[i].txd2 = TX_DMA_DESP2_DEF;
} }
wmb(); wmb();
fe_reg_w32(priv->tx_phys, FE_REG_TX_BASE_PTR0); fe_reg_w32(ring->tx_phys, FE_REG_TX_BASE_PTR0);
fe_reg_w32(priv->tx_ring_size, FE_REG_TX_MAX_CNT0); fe_reg_w32(ring->tx_ring_size, FE_REG_TX_MAX_CNT0);
fe_reg_w32(0, FE_REG_TX_CTX_IDX0); fe_reg_w32(0, FE_REG_TX_CTX_IDX0);
fe_reg_w32(FE_PST_DTX_IDX0, FE_REG_PDMA_RST_CFG); fe_reg_w32(FE_PST_DTX_IDX0, FE_REG_PDMA_RST_CFG);
@ -383,8 +388,6 @@ static void fe_free_dma(struct fe_priv *priv)
{ {
fe_clean_tx(priv); fe_clean_tx(priv);
fe_clean_rx(priv); fe_clean_rx(priv);
netdev_reset_queue(priv->netdev);
} }
void fe_stats_update(struct fe_priv *priv) void fe_stats_update(struct fe_priv *priv)
@ -523,7 +526,7 @@ static int fe_vlan_rx_kill_vid(struct net_device *dev,
} }
static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev, static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
int idx, int tx_num) int idx, int tx_num, struct fe_tx_ring *ring)
{ {
struct fe_priv *priv = netdev_priv(dev); struct fe_priv *priv = netdev_priv(dev);
struct skb_frag_struct *frag; struct skb_frag_struct *frag;
@ -534,7 +537,7 @@ static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
u32 def_txd4; u32 def_txd4;
int i, j, k, frag_size, frag_map_size, offset; int i, j, k, frag_size, frag_map_size, offset;
tx_buf = &priv->tx_buf[idx]; tx_buf = &ring->tx_buf[idx];
memset(tx_buf, 0, sizeof(*tx_buf)); memset(tx_buf, 0, sizeof(*tx_buf));
memset(&txd, 0, sizeof(txd)); memset(&txd, 0, sizeof(txd));
nr_frags = skb_shinfo(skb)->nr_frags; nr_frags = skb_shinfo(skb)->nr_frags;
@ -606,7 +609,7 @@ static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
txd.txd2 = TX_DMA_PLEN0(frag_map_size); txd.txd2 = TX_DMA_PLEN0(frag_map_size);
txd.txd4 = def_txd4; txd.txd4 = def_txd4;
tx_buf = &priv->tx_buf[j]; tx_buf = &ring->tx_buf[j];
memset(tx_buf, 0, sizeof(*tx_buf)); memset(tx_buf, 0, sizeof(*tx_buf));
tx_buf->flags |= FE_TX_FLAGS_PAGE0; tx_buf->flags |= FE_TX_FLAGS_PAGE0;
@ -623,7 +626,7 @@ static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
if (!((i == (nr_frags -1)) && if (!((i == (nr_frags -1)) &&
(frag_map_size == frag_size))) { (frag_map_size == frag_size))) {
fe_set_txd(&txd, &priv->tx_dma[j]); fe_set_txd(&txd, &ring->tx_dma[j]);
memset(&txd, 0, sizeof(txd)); memset(&txd, 0, sizeof(txd));
} }
} }
@ -638,7 +641,7 @@ static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
txd.txd2 |= TX_DMA_LS1; txd.txd2 |= TX_DMA_LS1;
else else
txd.txd2 |= TX_DMA_LS0; txd.txd2 |= TX_DMA_LS0;
fe_set_txd(&txd, &priv->tx_dma[j]); fe_set_txd(&txd, &ring->tx_dma[j]);
/* store skb to cleanup */ /* store skb to cleanup */
tx_buf->skb = skb; tx_buf->skb = skb;
@ -655,8 +658,8 @@ static int fe_tx_map_dma(struct sk_buff *skb, struct net_device *dev,
err_dma: err_dma:
j = idx; j = idx;
for (i = 0; i < tx_num; i++) { for (i = 0; i < tx_num; i++) {
ptxd = &priv->tx_dma[j]; ptxd = &ring->tx_dma[j];
tx_buf = &priv->tx_buf[j]; tx_buf = &ring->tx_buf[j];
/* unmap dma */ /* unmap dma */
fe_txd_unmap(&dev->dev, tx_buf); fe_txd_unmap(&dev->dev, tx_buf);
@ -700,10 +703,10 @@ static inline int fe_skb_padto(struct sk_buff *skb, struct fe_priv *priv) {
return ret; return ret;
} }
static inline u32 fe_empty_txd(struct fe_priv *priv, u32 tx_fill_idx) static inline u32 fe_empty_txd(struct fe_tx_ring *ring, u32 tx_fill_idx)
{ {
return (u32)(priv->tx_ring_size - ((tx_fill_idx - priv->tx_free_idx) & return (u32)(ring->tx_ring_size - ((tx_fill_idx - ring->tx_free_idx) &
(priv->tx_ring_size - 1))); (ring->tx_ring_size - 1)));
} }
static inline int fe_cal_txd_req(struct sk_buff *skb) static inline int fe_cal_txd_req(struct sk_buff *skb)
@ -727,6 +730,7 @@ static inline int fe_cal_txd_req(struct sk_buff *skb)
static int fe_start_xmit(struct sk_buff *skb, struct net_device *dev) static int fe_start_xmit(struct sk_buff *skb, struct net_device *dev)
{ {
struct fe_priv *priv = netdev_priv(dev); struct fe_priv *priv = netdev_priv(dev);
struct fe_tx_ring *ring = &priv->tx_ring;
struct net_device_stats *stats = &dev->stats; struct net_device_stats *stats = &dev->stats;
u32 tx; u32 tx;
int tx_num; int tx_num;
@ -739,7 +743,7 @@ static int fe_start_xmit(struct sk_buff *skb, struct net_device *dev)
tx_num = fe_cal_txd_req(skb); tx_num = fe_cal_txd_req(skb);
tx = fe_reg_r32(FE_REG_TX_CTX_IDX0); tx = fe_reg_r32(FE_REG_TX_CTX_IDX0);
if (unlikely(fe_empty_txd(priv, tx) <= tx_num)) if (unlikely(fe_empty_txd(ring, tx) <= tx_num))
{ {
netif_stop_queue(dev); netif_stop_queue(dev);
netif_err(priv, tx_queued,dev, netif_err(priv, tx_queued,dev,
@ -747,7 +751,7 @@ static int fe_start_xmit(struct sk_buff *skb, struct net_device *dev)
return NETDEV_TX_BUSY; return NETDEV_TX_BUSY;
} }
if (fe_tx_map_dma(skb, dev, tx, tx_num) < 0) { if (fe_tx_map_dma(skb, dev, tx, tx_num, ring) < 0) {
stats->tx_dropped++; stats->tx_dropped++;
} else { } else {
stats->tx_packets++; stats->tx_packets++;
@ -868,7 +872,8 @@ release_desc:
return done; return done;
} }
static int fe_poll_tx(struct fe_priv *priv, int budget, u32 tx_intr) static int fe_poll_tx(struct fe_priv *priv, int budget, u32 tx_intr,
int *tx_again)
{ {
struct net_device *netdev = priv->netdev; struct net_device *netdev = priv->netdev;
struct device *dev = &netdev->dev; struct device *dev = &netdev->dev;
@ -877,13 +882,13 @@ static int fe_poll_tx(struct fe_priv *priv, int budget, u32 tx_intr)
struct fe_tx_buf *tx_buf; struct fe_tx_buf *tx_buf;
int done = 0; int done = 0;
u32 idx, hwidx; u32 idx, hwidx;
struct fe_tx_ring *ring = &priv->tx_ring;
idx = ring->tx_free_idx;
hwidx = fe_reg_r32(FE_REG_TX_DTX_IDX0); hwidx = fe_reg_r32(FE_REG_TX_DTX_IDX0);
idx = priv->tx_free_idx;
txpoll_again:
while ((idx != hwidx) && budget) { while ((idx != hwidx) && budget) {
tx_buf = &priv->tx_buf[idx]; tx_buf = &ring->tx_buf[idx];
skb = tx_buf->skb; skb = tx_buf->skb;
if (!skb) if (!skb)
@ -897,23 +902,25 @@ txpoll_again:
fe_txd_unmap(dev, tx_buf); fe_txd_unmap(dev, tx_buf);
idx = NEXT_TX_DESP_IDX(idx); idx = NEXT_TX_DESP_IDX(idx);
} }
priv->tx_free_idx = idx; ring->tx_free_idx = idx;
if (budget) { if (idx == hwidx) {
fe_reg_w32(tx_intr, FE_REG_FE_INT_STATUS); /* read hw index again make sure no new tx packet */
hwidx = fe_reg_r32(FE_REG_TX_DTX_IDX0); hwidx = fe_reg_r32(FE_REG_TX_DTX_IDX0);
if (idx != hwidx) if (idx == hwidx)
goto txpoll_again; fe_reg_w32(tx_intr, FE_REG_FE_INT_STATUS);
} else
*tx_again = 1;
if (!done) } else
return 0; *tx_again = 1;
if (done) {
netdev_completed_queue(netdev, done, bytes_compl); netdev_completed_queue(netdev, done, bytes_compl);
if (unlikely(netif_queue_stopped(netdev) && if (unlikely(netif_queue_stopped(netdev) &&
netif_carrier_ok(netdev))) { netif_carrier_ok(netdev))) {
netif_wake_queue(netdev); netif_wake_queue(netdev);
} }
}
return done; return done;
} }
@ -922,7 +929,7 @@ static int fe_poll(struct napi_struct *napi, int budget)
{ {
struct fe_priv *priv = container_of(napi, struct fe_priv, rx_napi); struct fe_priv *priv = container_of(napi, struct fe_priv, rx_napi);
struct fe_hw_stats *hwstat = priv->hw_stats; struct fe_hw_stats *hwstat = priv->hw_stats;
int tx_done, rx_done; int tx_done, rx_done, tx_again;
u32 status, fe_status, status_reg, mask; u32 status, fe_status, status_reg, mask;
u32 tx_intr, rx_intr, status_intr; u32 tx_intr, rx_intr, status_intr;
@ -930,7 +937,7 @@ static int fe_poll(struct napi_struct *napi, int budget)
tx_intr = priv->soc->tx_int; tx_intr = priv->soc->tx_int;
rx_intr = priv->soc->rx_int; rx_intr = priv->soc->rx_int;
status_intr = priv->soc->status_int; status_intr = priv->soc->status_int;
tx_done = rx_done = 0; tx_done = rx_done = tx_again = 0;
if (fe_reg_table[FE_REG_FE_INT_STATUS2]) { if (fe_reg_table[FE_REG_FE_INT_STATUS2]) {
fe_status = fe_reg_r32(FE_REG_FE_INT_STATUS2); fe_status = fe_reg_r32(FE_REG_FE_INT_STATUS2);
@ -939,7 +946,7 @@ static int fe_poll(struct napi_struct *napi, int budget)
status_reg = FE_REG_FE_INT_STATUS; status_reg = FE_REG_FE_INT_STATUS;
if (status & tx_intr) if (status & tx_intr)
tx_done = fe_poll_tx(priv, budget, tx_intr); tx_done = fe_poll_tx(priv, budget, tx_intr, &tx_again);
if (status & rx_intr) if (status & rx_intr)
rx_done = fe_poll_rx(napi, budget, priv, rx_intr); rx_done = fe_poll_rx(napi, budget, priv, rx_intr);
@ -959,7 +966,7 @@ static int fe_poll(struct napi_struct *napi, int budget)
tx_done, rx_done, status, mask); tx_done, rx_done, status, mask);
} }
if ((tx_done < budget) && (rx_done < budget)) { if (!tx_again && (rx_done < budget)) {
status = fe_reg_r32(FE_REG_FE_INT_STATUS); status = fe_reg_r32(FE_REG_FE_INT_STATUS);
if (status & (tx_intr | rx_intr )) if (status & (tx_intr | rx_intr ))
goto poll_again; goto poll_again;
@ -975,6 +982,7 @@ poll_again:
static void fe_tx_timeout(struct net_device *dev) static void fe_tx_timeout(struct net_device *dev)
{ {
struct fe_priv *priv = netdev_priv(dev); struct fe_priv *priv = netdev_priv(dev);
struct fe_tx_ring *ring = &priv->tx_ring;
priv->netdev->stats.tx_errors++; priv->netdev->stats.tx_errors++;
netif_err(priv, tx_err, dev, netif_err(priv, tx_err, dev,
@ -987,7 +995,7 @@ static void fe_tx_timeout(struct net_device *dev)
fe_reg_r32(FE_REG_TX_MAX_CNT0), fe_reg_r32(FE_REG_TX_MAX_CNT0),
fe_reg_r32(FE_REG_TX_CTX_IDX0), fe_reg_r32(FE_REG_TX_CTX_IDX0),
fe_reg_r32(FE_REG_TX_DTX_IDX0), fe_reg_r32(FE_REG_TX_DTX_IDX0),
priv->tx_free_idx ring->tx_free_idx
); );
netif_info(priv, drv, dev, "rx_ring=%d, " \ netif_info(priv, drv, dev, "rx_ring=%d, " \
"base=%08x, max=%u, calc=%u, drx=%u\n", 0, "base=%08x, max=%u, calc=%u, drx=%u\n", 0,
@ -1480,13 +1488,13 @@ static int fe_probe(struct platform_device *pdev)
priv->msg_enable = netif_msg_init(fe_msg_level, FE_DEFAULT_MSG_ENABLE); priv->msg_enable = netif_msg_init(fe_msg_level, FE_DEFAULT_MSG_ENABLE);
priv->frag_size = fe_max_frag_size(ETH_DATA_LEN); priv->frag_size = fe_max_frag_size(ETH_DATA_LEN);
priv->rx_buf_size = fe_max_buf_size(priv->frag_size); priv->rx_buf_size = fe_max_buf_size(priv->frag_size);
priv->tx_ring_size = priv->rx_ring_size = NUM_DMA_DESC; priv->tx_ring.tx_ring_size = priv->rx_ring_size = NUM_DMA_DESC;
INIT_WORK(&priv->pending_work, fe_pending_work); INIT_WORK(&priv->pending_work, fe_pending_work);
napi_weight = 32; napi_weight = 32;
if (priv->flags & FE_FLAG_NAPI_WEIGHT) { if (priv->flags & FE_FLAG_NAPI_WEIGHT) {
napi_weight *= 4; napi_weight *= 4;
priv->tx_ring_size *= 4; priv->tx_ring.tx_ring_size *= 4;
priv->rx_ring_size *= 4; priv->rx_ring_size *= 4;
} }
netif_napi_add(netdev, &priv->rx_napi, fe_poll, napi_weight); netif_napi_add(netdev, &priv->rx_napi, fe_poll, napi_weight);

15
target/linux/ramips/files/drivers/net/ethernet/ralink/ralink_soc_eth.h
View file

@ -454,6 +454,15 @@ struct fe_tx_buf
DEFINE_DMA_UNMAP_LEN(dma_len1); DEFINE_DMA_UNMAP_LEN(dma_len1);
}; };
struct fe_tx_ring
{
struct fe_tx_dma *tx_dma;
struct fe_tx_buf *tx_buf;
dma_addr_t tx_phys;
u16 tx_ring_size;
u16 tx_free_idx;
};
struct fe_priv struct fe_priv
{ {
spinlock_t page_lock; spinlock_t page_lock;
@ -473,10 +482,7 @@ struct fe_priv
dma_addr_t rx_phys; dma_addr_t rx_phys;
struct napi_struct rx_napi; struct napi_struct rx_napi;
struct fe_tx_dma *tx_dma; struct fe_tx_ring tx_ring;
struct fe_tx_buf *tx_buf;
dma_addr_t tx_phys;
unsigned int tx_free_idx;
struct fe_phy *phy; struct fe_phy *phy;
struct mii_bus *mii_bus; struct mii_bus *mii_bus;
@ -489,7 +495,6 @@ struct fe_priv
unsigned long vlan_map; unsigned long vlan_map;
struct work_struct pending_work; struct work_struct pending_work;
DECLARE_BITMAP(pending_flags, FE_FLAG_MAX); DECLARE_BITMAP(pending_flags, FE_FLAG_MAX);
u16 tx_ring_size;
u16 rx_ring_size; u16 rx_ring_size;
}; };