net: stmmac: Fix NAPI poll in TX path when in multi-queue

Commit 8fce333170 introduced the concept of NAPI per-channel and
independent cleaning of TX path.

This is currently breaking performance in some cases. The scenario
happens when all packets are being received in Queue 0 but the TX is
performed in Queue != 0.

Fix this by using different NAPI instances per each TX and RX queue, as
suggested by Florian.

Changes from v2:
	- Only force restart transmission if there are pending packets
Changes from v1:
	- Pass entire ring size to TX clean path (Florian)

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Jose Abreu 2019-02-19 10:38:47 +01:00 committed by David S. Miller
parent d0e698d57a
commit 4ccb45857c
2 changed files with 70 additions and 54 deletions

View File

@ -79,11 +79,10 @@ struct stmmac_rx_queue {
}; };
struct stmmac_channel { struct stmmac_channel {
struct napi_struct napi ____cacheline_aligned_in_smp; struct napi_struct rx_napi ____cacheline_aligned_in_smp;
struct napi_struct tx_napi ____cacheline_aligned_in_smp;
struct stmmac_priv *priv_data; struct stmmac_priv *priv_data;
u32 index; u32 index;
int has_rx;
int has_tx;
}; };
struct stmmac_tc_entry { struct stmmac_tc_entry {

View File

@ -155,7 +155,10 @@ static void stmmac_disable_all_queues(struct stmmac_priv *priv)
for (queue = 0; queue < maxq; queue++) { for (queue = 0; queue < maxq; queue++) {
struct stmmac_channel *ch = &priv->channel[queue]; struct stmmac_channel *ch = &priv->channel[queue];
napi_disable(&ch->napi); if (queue < rx_queues_cnt)
napi_disable(&ch->rx_napi);
if (queue < tx_queues_cnt)
napi_disable(&ch->tx_napi);
} }
} }
@ -173,7 +176,10 @@ static void stmmac_enable_all_queues(struct stmmac_priv *priv)
for (queue = 0; queue < maxq; queue++) { for (queue = 0; queue < maxq; queue++) {
struct stmmac_channel *ch = &priv->channel[queue]; struct stmmac_channel *ch = &priv->channel[queue];
napi_enable(&ch->napi); if (queue < rx_queues_cnt)
napi_enable(&ch->rx_napi);
if (queue < tx_queues_cnt)
napi_enable(&ch->tx_napi);
} }
} }
@ -1955,6 +1961,10 @@ static int stmmac_tx_clean(struct stmmac_priv *priv, int budget, u32 queue)
mod_timer(&priv->eee_ctrl_timer, STMMAC_LPI_T(eee_timer)); mod_timer(&priv->eee_ctrl_timer, STMMAC_LPI_T(eee_timer));
} }
/* We still have pending packets, let's call for a new scheduling */
if (tx_q->dirty_tx != tx_q->cur_tx)
mod_timer(&tx_q->txtimer, STMMAC_COAL_TIMER(10));
__netif_tx_unlock_bh(netdev_get_tx_queue(priv->dev, queue)); __netif_tx_unlock_bh(netdev_get_tx_queue(priv->dev, queue));
return count; return count;
@ -2045,23 +2055,15 @@ static int stmmac_napi_check(struct stmmac_priv *priv, u32 chan)
int status = stmmac_dma_interrupt_status(priv, priv->ioaddr, int status = stmmac_dma_interrupt_status(priv, priv->ioaddr,
&priv->xstats, chan); &priv->xstats, chan);
struct stmmac_channel *ch = &priv->channel[chan]; struct stmmac_channel *ch = &priv->channel[chan];
bool needs_work = false;
if ((status & handle_rx) && ch->has_rx) { if ((status & handle_rx) && (chan < priv->plat->rx_queues_to_use)) {
needs_work = true;
} else {
status &= ~handle_rx;
}
if ((status & handle_tx) && ch->has_tx) {
needs_work = true;
} else {
status &= ~handle_tx;
}
if (needs_work && napi_schedule_prep(&ch->napi)) {
stmmac_disable_dma_irq(priv, priv->ioaddr, chan); stmmac_disable_dma_irq(priv, priv->ioaddr, chan);
__napi_schedule(&ch->napi); napi_schedule_irqoff(&ch->rx_napi);
}
if ((status & handle_tx) && (chan < priv->plat->tx_queues_to_use)) {
stmmac_disable_dma_irq(priv, priv->ioaddr, chan);
napi_schedule_irqoff(&ch->tx_napi);
} }
return status; return status;
@ -2257,8 +2259,14 @@ static void stmmac_tx_timer(struct timer_list *t)
ch = &priv->channel[tx_q->queue_index]; ch = &priv->channel[tx_q->queue_index];
if (likely(napi_schedule_prep(&ch->napi))) /*
__napi_schedule(&ch->napi); * If NAPI is already running we can miss some events. Let's rearm
* the timer and try again.
*/
if (likely(napi_schedule_prep(&ch->tx_napi)))
__napi_schedule(&ch->tx_napi);
else
mod_timer(&tx_q->txtimer, STMMAC_COAL_TIMER(10));
} }
/** /**
@ -3514,7 +3522,7 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
else else
skb->ip_summed = CHECKSUM_UNNECESSARY; skb->ip_summed = CHECKSUM_UNNECESSARY;
napi_gro_receive(&ch->napi, skb); napi_gro_receive(&ch->rx_napi, skb);
priv->dev->stats.rx_packets++; priv->dev->stats.rx_packets++;
priv->dev->stats.rx_bytes += frame_len; priv->dev->stats.rx_bytes += frame_len;
@ -3529,40 +3537,45 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
return count; return count;
} }
/** static int stmmac_napi_poll_rx(struct napi_struct *napi, int budget)
* stmmac_poll - stmmac poll method (NAPI)
* @napi : pointer to the napi structure.
* @budget : maximum number of packets that the current CPU can receive from
* all interfaces.
* Description :
* To look at the incoming frames and clear the tx resources.
*/
static int stmmac_napi_poll(struct napi_struct *napi, int budget)
{ {
struct stmmac_channel *ch = struct stmmac_channel *ch =
container_of(napi, struct stmmac_channel, napi); container_of(napi, struct stmmac_channel, rx_napi);
struct stmmac_priv *priv = ch->priv_data; struct stmmac_priv *priv = ch->priv_data;
int work_done, rx_done = 0, tx_done = 0;
u32 chan = ch->index; u32 chan = ch->index;
int work_done;
priv->xstats.napi_poll++; priv->xstats.napi_poll++;
if (ch->has_tx) work_done = stmmac_rx(priv, budget, chan);
tx_done = stmmac_tx_clean(priv, budget, chan); if (work_done < budget && napi_complete_done(napi, work_done))
if (ch->has_rx) stmmac_enable_dma_irq(priv, priv->ioaddr, chan);
rx_done = stmmac_rx(priv, budget, chan); return work_done;
}
work_done = max(rx_done, tx_done); static int stmmac_napi_poll_tx(struct napi_struct *napi, int budget)
{
struct stmmac_channel *ch =
container_of(napi, struct stmmac_channel, tx_napi);
struct stmmac_priv *priv = ch->priv_data;
struct stmmac_tx_queue *tx_q;
u32 chan = ch->index;
int work_done;
priv->xstats.napi_poll++;
work_done = stmmac_tx_clean(priv, DMA_TX_SIZE, chan);
work_done = min(work_done, budget); work_done = min(work_done, budget);
if (work_done < budget && napi_complete_done(napi, work_done)) { if (work_done < budget && napi_complete_done(napi, work_done))
int stat;
stmmac_enable_dma_irq(priv, priv->ioaddr, chan); stmmac_enable_dma_irq(priv, priv->ioaddr, chan);
stat = stmmac_dma_interrupt_status(priv, priv->ioaddr,
&priv->xstats, chan); /* Force transmission restart */
if (stat && napi_reschedule(napi)) tx_q = &priv->tx_queue[chan];
stmmac_disable_dma_irq(priv, priv->ioaddr, chan); if (tx_q->cur_tx != tx_q->dirty_tx) {
stmmac_enable_dma_transmission(priv, priv->ioaddr);
stmmac_set_tx_tail_ptr(priv, priv->ioaddr, tx_q->tx_tail_addr,
chan);
} }
return work_done; return work_done;
@ -4342,13 +4355,14 @@ int stmmac_dvr_probe(struct device *device,
ch->priv_data = priv; ch->priv_data = priv;
ch->index = queue; ch->index = queue;
if (queue < priv->plat->rx_queues_to_use) if (queue < priv->plat->rx_queues_to_use) {
ch->has_rx = true; netif_napi_add(ndev, &ch->rx_napi, stmmac_napi_poll_rx,
if (queue < priv->plat->tx_queues_to_use) NAPI_POLL_WEIGHT);
ch->has_tx = true; }
if (queue < priv->plat->tx_queues_to_use) {
netif_napi_add(ndev, &ch->napi, stmmac_napi_poll, netif_napi_add(ndev, &ch->tx_napi, stmmac_napi_poll_tx,
NAPI_POLL_WEIGHT); NAPI_POLL_WEIGHT);
}
} }
mutex_init(&priv->lock); mutex_init(&priv->lock);
@ -4404,7 +4418,10 @@ error_mdio_register:
for (queue = 0; queue < maxq; queue++) { for (queue = 0; queue < maxq; queue++) {
struct stmmac_channel *ch = &priv->channel[queue]; struct stmmac_channel *ch = &priv->channel[queue];
netif_napi_del(&ch->napi); if (queue < priv->plat->rx_queues_to_use)
netif_napi_del(&ch->rx_napi);
if (queue < priv->plat->tx_queues_to_use)
netif_napi_del(&ch->tx_napi);
} }
error_hw_init: error_hw_init:
destroy_workqueue(priv->wq); destroy_workqueue(priv->wq);