Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 01b54b14 authored by Jing Huang's avatar Jing Huang Committed by David S. Miller
Browse files

bna: tx rx cleanup fix



This patch removes busy wait in tx/rx cleanup. bnad_cb_tx_cleanup() and
bnad_cb_rx_cleanup() functions are called from irq context, and currently
they do busy wait for the in-flight transmit or the currently executing napi
polling routine to complete. To fix the issue, we create a workqueue to defer
tx & rx cleanup processing, an in the tx rx cleanup handler, we will
wait respective in flight processing to complete, before freeing the buffers.

Signed-off-by: default avatarJing Huang <huangj@brocade.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent f96c1d24
Loading
Loading
Loading
Loading
+121 −80
Original line number Original line Diff line number Diff line
@@ -80,8 +80,6 @@ do { \
	(sizeof(struct bnad_skb_unmap) * ((_depth) - 1));	\
	(sizeof(struct bnad_skb_unmap) * ((_depth) - 1));	\
} while (0)
} while (0)


#define BNAD_TXRX_SYNC_MDELAY	250	/* 250 msecs */

static void
static void
bnad_add_to_list(struct bnad *bnad)
bnad_add_to_list(struct bnad *bnad)
{
{
@@ -141,7 +139,8 @@ bnad_pci_unmap_skb(struct device *pdev, struct bnad_skb_unmap *array,


	for (j = 0; j < frag; j++) {
	for (j = 0; j < frag; j++) {
		dma_unmap_page(pdev, dma_unmap_addr(&array[index], dma_addr),
		dma_unmap_page(pdev, dma_unmap_addr(&array[index], dma_addr),
			  skb_frag_size(&skb_shinfo(skb)->frags[j]), DMA_TO_DEVICE);
			  skb_frag_size(&skb_shinfo(skb)->frags[j]),
						DMA_TO_DEVICE);
		dma_unmap_addr_set(&array[index], dma_addr, 0);
		dma_unmap_addr_set(&array[index], dma_addr, 0);
		BNA_QE_INDX_ADD(index, 1, depth);
		BNA_QE_INDX_ADD(index, 1, depth);
	}
	}
@@ -453,12 +452,8 @@ bnad_poll_cq(struct bnad *bnad, struct bna_ccb *ccb, int budget)
	struct bna_pkt_rate *pkt_rt = &ccb->pkt_rate;
	struct bna_pkt_rate *pkt_rt = &ccb->pkt_rate;
	struct bnad_rx_ctrl *rx_ctrl = (struct bnad_rx_ctrl *)(ccb->ctrl);
	struct bnad_rx_ctrl *rx_ctrl = (struct bnad_rx_ctrl *)(ccb->ctrl);


	set_bit(BNAD_FP_IN_RX_PATH, &rx_ctrl->flags);
	if (!test_bit(BNAD_RXQ_STARTED, &ccb->rcb[0]->flags))

	if (!test_bit(BNAD_RXQ_STARTED, &ccb->rcb[0]->flags)) {
		clear_bit(BNAD_FP_IN_RX_PATH, &rx_ctrl->flags);
		return 0;
		return 0;
	}


	prefetch(bnad->netdev);
	prefetch(bnad->netdev);
	BNA_CQ_QPGE_PTR_GET(ccb->producer_index, ccb->sw_qpt, cmpl,
	BNA_CQ_QPGE_PTR_GET(ccb->producer_index, ccb->sw_qpt, cmpl,
@@ -533,9 +528,8 @@ bnad_poll_cq(struct bnad *bnad, struct bna_ccb *ccb, int budget)


		if (skb->ip_summed == CHECKSUM_UNNECESSARY)
		if (skb->ip_summed == CHECKSUM_UNNECESSARY)
			napi_gro_receive(&rx_ctrl->napi, skb);
			napi_gro_receive(&rx_ctrl->napi, skb);
		else {
		else
			netif_receive_skb(skb);
			netif_receive_skb(skb);
		}


next:
next:
		cmpl->valid = 0;
		cmpl->valid = 0;
@@ -839,20 +833,9 @@ bnad_cb_tcb_destroy(struct bnad *bnad, struct bna_tcb *tcb)
{
{
	struct bnad_tx_info *tx_info =
	struct bnad_tx_info *tx_info =
			(struct bnad_tx_info *)tcb->txq->tx->priv;
			(struct bnad_tx_info *)tcb->txq->tx->priv;
	struct bnad_unmap_q *unmap_q = tcb->unmap_q;

	while (test_and_set_bit(BNAD_TXQ_FREE_SENT, &tcb->flags))
		cpu_relax();

	bnad_free_all_txbufs(bnad, tcb);

	unmap_q->producer_index = 0;
	unmap_q->consumer_index = 0;

	smp_mb__before_clear_bit();
	clear_bit(BNAD_TXQ_FREE_SENT, &tcb->flags);


	tx_info->tcb[tcb->id] = NULL;
	tx_info->tcb[tcb->id] = NULL;
	tcb->priv = NULL;
}
}


static void
static void
@@ -865,12 +848,6 @@ bnad_cb_rcb_setup(struct bnad *bnad, struct bna_rcb *rcb)
	unmap_q->q_depth = BNAD_RX_UNMAPQ_DEPTH;
	unmap_q->q_depth = BNAD_RX_UNMAPQ_DEPTH;
}
}


static void
bnad_cb_rcb_destroy(struct bnad *bnad, struct bna_rcb *rcb)
{
	bnad_free_all_rxbufs(bnad, rcb);
}

static void
static void
bnad_cb_ccb_setup(struct bnad *bnad, struct bna_ccb *ccb)
bnad_cb_ccb_setup(struct bnad *bnad, struct bna_ccb *ccb)
{
{
@@ -916,7 +893,6 @@ bnad_cb_tx_resume(struct bnad *bnad, struct bna_tx *tx)
{
{
	struct bnad_tx_info *tx_info = (struct bnad_tx_info *)tx->priv;
	struct bnad_tx_info *tx_info = (struct bnad_tx_info *)tx->priv;
	struct bna_tcb *tcb;
	struct bna_tcb *tcb;
	struct bnad_unmap_q *unmap_q;
	u32 txq_id;
	u32 txq_id;
	int i;
	int i;


@@ -926,23 +902,9 @@ bnad_cb_tx_resume(struct bnad *bnad, struct bna_tx *tx)
			continue;
			continue;
		txq_id = tcb->id;
		txq_id = tcb->id;


		unmap_q = tcb->unmap_q;
		BUG_ON(test_bit(BNAD_TXQ_TX_STARTED, &tcb->flags));

		if (test_bit(BNAD_TXQ_TX_STARTED, &tcb->flags))
			continue;

		while (test_and_set_bit(BNAD_TXQ_FREE_SENT, &tcb->flags))
			cpu_relax();

		bnad_free_all_txbufs(bnad, tcb);

		unmap_q->producer_index = 0;
		unmap_q->consumer_index = 0;

		smp_mb__before_clear_bit();
		clear_bit(BNAD_TXQ_FREE_SENT, &tcb->flags);

		set_bit(BNAD_TXQ_TX_STARTED, &tcb->flags);
		set_bit(BNAD_TXQ_TX_STARTED, &tcb->flags);
		BUG_ON(*(tcb->hw_consumer_index) != 0);


		if (netif_carrier_ok(bnad->netdev)) {
		if (netif_carrier_ok(bnad->netdev)) {
			printk(KERN_INFO "bna: %s %d TXQ_STARTED\n",
			printk(KERN_INFO "bna: %s %d TXQ_STARTED\n",
@@ -963,6 +925,54 @@ bnad_cb_tx_resume(struct bnad *bnad, struct bna_tx *tx)
	}
	}
}
}


/*
 * Free all TxQs buffers and then notify TX_E_CLEANUP_DONE to Tx fsm.
 */
static void
bnad_tx_cleanup(struct delayed_work *work)
{
	struct bnad_tx_info *tx_info =
		container_of(work, struct bnad_tx_info, tx_cleanup_work);
	struct bnad *bnad = NULL;
	struct bnad_unmap_q *unmap_q;
	struct bna_tcb *tcb;
	unsigned long flags;
	uint32_t i, pending = 0;

	for (i = 0; i < BNAD_MAX_TXQ_PER_TX; i++) {
		tcb = tx_info->tcb[i];
		if (!tcb)
			continue;

		bnad = tcb->bnad;

		if (test_and_set_bit(BNAD_TXQ_FREE_SENT, &tcb->flags)) {
			pending++;
			continue;
		}

		bnad_free_all_txbufs(bnad, tcb);

		unmap_q = tcb->unmap_q;
		unmap_q->producer_index = 0;
		unmap_q->consumer_index = 0;

		smp_mb__before_clear_bit();
		clear_bit(BNAD_TXQ_FREE_SENT, &tcb->flags);
	}

	if (pending) {
		queue_delayed_work(bnad->work_q, &tx_info->tx_cleanup_work,
			msecs_to_jiffies(1));
		return;
	}

	spin_lock_irqsave(&bnad->bna_lock, flags);
	bna_tx_cleanup_complete(tx_info->tx);
	spin_unlock_irqrestore(&bnad->bna_lock, flags);
}


static void
static void
bnad_cb_tx_cleanup(struct bnad *bnad, struct bna_tx *tx)
bnad_cb_tx_cleanup(struct bnad *bnad, struct bna_tx *tx)
{
{
@@ -976,8 +986,7 @@ bnad_cb_tx_cleanup(struct bnad *bnad, struct bna_tx *tx)
			continue;
			continue;
	}
	}


	mdelay(BNAD_TXRX_SYNC_MDELAY);
	queue_delayed_work(bnad->work_q, &tx_info->tx_cleanup_work, 0);
	bna_tx_cleanup_complete(tx);
}
}


static void
static void
@@ -1001,6 +1010,44 @@ bnad_cb_rx_stall(struct bnad *bnad, struct bna_rx *rx)
	}
	}
}
}


/*
 * Free all RxQs buffers and then notify RX_E_CLEANUP_DONE to Rx fsm.
 */
static void
bnad_rx_cleanup(void *work)
{
	struct bnad_rx_info *rx_info =
		container_of(work, struct bnad_rx_info, rx_cleanup_work);
	struct bnad_rx_ctrl *rx_ctrl;
	struct bnad *bnad = NULL;
	unsigned long flags;
	uint32_t i;

	for (i = 0; i < BNAD_MAX_RXP_PER_RX; i++) {
		rx_ctrl = &rx_info->rx_ctrl[i];

		if (!rx_ctrl->ccb)
			continue;

		bnad = rx_ctrl->ccb->bnad;

		/*
		 * Wait till the poll handler has exited
		 * and nothing can be scheduled anymore
		 */
		napi_disable(&rx_ctrl->napi);

		bnad_cq_cmpl_init(bnad, rx_ctrl->ccb);
		bnad_free_all_rxbufs(bnad, rx_ctrl->ccb->rcb[0]);
		if (rx_ctrl->ccb->rcb[1])
			bnad_free_all_rxbufs(bnad, rx_ctrl->ccb->rcb[1]);
	}

	spin_lock_irqsave(&bnad->bna_lock, flags);
	bna_rx_cleanup_complete(rx_info->rx);
	spin_unlock_irqrestore(&bnad->bna_lock, flags);
}

static void
static void
bnad_cb_rx_cleanup(struct bnad *bnad, struct bna_rx *rx)
bnad_cb_rx_cleanup(struct bnad *bnad, struct bna_rx *rx)
{
{
@@ -1009,8 +1056,6 @@ bnad_cb_rx_cleanup(struct bnad *bnad, struct bna_rx *rx)
	struct bnad_rx_ctrl *rx_ctrl;
	struct bnad_rx_ctrl *rx_ctrl;
	int i;
	int i;


	mdelay(BNAD_TXRX_SYNC_MDELAY);

	for (i = 0; i < BNAD_MAX_RXP_PER_RX; i++) {
	for (i = 0; i < BNAD_MAX_RXP_PER_RX; i++) {
		rx_ctrl = &rx_info->rx_ctrl[i];
		rx_ctrl = &rx_info->rx_ctrl[i];
		ccb = rx_ctrl->ccb;
		ccb = rx_ctrl->ccb;
@@ -1021,12 +1066,9 @@ bnad_cb_rx_cleanup(struct bnad *bnad, struct bna_rx *rx)


		if (ccb->rcb[1])
		if (ccb->rcb[1])
			clear_bit(BNAD_RXQ_STARTED, &ccb->rcb[1]->flags);
			clear_bit(BNAD_RXQ_STARTED, &ccb->rcb[1]->flags);

		while (test_bit(BNAD_FP_IN_RX_PATH, &rx_ctrl->flags))
			cpu_relax();
	}
	}


	bna_rx_cleanup_complete(rx);
	queue_work(bnad->work_q, &rx_info->rx_cleanup_work);
}
}


static void
static void
@@ -1046,13 +1088,12 @@ bnad_cb_rx_post(struct bnad *bnad, struct bna_rx *rx)
		if (!ccb)
		if (!ccb)
			continue;
			continue;


		bnad_cq_cmpl_init(bnad, ccb);
		napi_enable(&rx_ctrl->napi);


		for (j = 0; j < BNAD_MAX_RXQ_PER_RXP; j++) {
		for (j = 0; j < BNAD_MAX_RXQ_PER_RXP; j++) {
			rcb = ccb->rcb[j];
			rcb = ccb->rcb[j];
			if (!rcb)
			if (!rcb)
				continue;
				continue;
			bnad_free_all_rxbufs(bnad, rcb);


			set_bit(BNAD_RXQ_STARTED, &rcb->flags);
			set_bit(BNAD_RXQ_STARTED, &rcb->flags);
			set_bit(BNAD_RXQ_POST_OK, &rcb->flags);
			set_bit(BNAD_RXQ_POST_OK, &rcb->flags);
@@ -1704,7 +1745,7 @@ bnad_napi_poll_rx(struct napi_struct *napi, int budget)


#define BNAD_NAPI_POLL_QUOTA		64
#define BNAD_NAPI_POLL_QUOTA		64
static void
static void
bnad_napi_init(struct bnad *bnad, u32 rx_id)
bnad_napi_add(struct bnad *bnad, u32 rx_id)
{
{
	struct bnad_rx_ctrl *rx_ctrl;
	struct bnad_rx_ctrl *rx_ctrl;
	int i;
	int i;
@@ -1718,30 +1759,14 @@ bnad_napi_init(struct bnad *bnad, u32 rx_id)
}
}


static void
static void
bnad_napi_enable(struct bnad *bnad, u32 rx_id)
bnad_napi_delete(struct bnad *bnad, u32 rx_id)
{
	struct bnad_rx_ctrl *rx_ctrl;
	int i;

	/* Initialize & enable NAPI */
	for (i = 0; i <	bnad->num_rxp_per_rx; i++) {
		rx_ctrl = &bnad->rx_info[rx_id].rx_ctrl[i];

		napi_enable(&rx_ctrl->napi);
	}
}

static void
bnad_napi_disable(struct bnad *bnad, u32 rx_id)
{
{
	int i;
	int i;


	/* First disable and then clean up */
	/* First disable and then clean up */
	for (i = 0; i < bnad->num_rxp_per_rx; i++) {
	for (i = 0; i < bnad->num_rxp_per_rx; i++)
		napi_disable(&bnad->rx_info[rx_id].rx_ctrl[i].napi);
		netif_napi_del(&bnad->rx_info[rx_id].rx_ctrl[i].napi);
		netif_napi_del(&bnad->rx_info[rx_id].rx_ctrl[i].napi);
}
}
}


/* Should be held with conf_lock held */
/* Should be held with conf_lock held */
void
void
@@ -1832,6 +1857,9 @@ bnad_setup_tx(struct bnad *bnad, u32 tx_id)
		goto err_return;
		goto err_return;
	tx_info->tx = tx;
	tx_info->tx = tx;


	INIT_DELAYED_WORK(&tx_info->tx_cleanup_work,
			(work_func_t)bnad_tx_cleanup);

	/* Register ISR for the Tx object */
	/* Register ISR for the Tx object */
	if (intr_info->intr_type == BNA_INTR_T_MSIX) {
	if (intr_info->intr_type == BNA_INTR_T_MSIX) {
		err = bnad_tx_msix_register(bnad, tx_info,
		err = bnad_tx_msix_register(bnad, tx_info,
@@ -1928,7 +1956,7 @@ bnad_cleanup_rx(struct bnad *bnad, u32 rx_id)
	if (rx_info->rx_ctrl[0].ccb->intr_type == BNA_INTR_T_MSIX)
	if (rx_info->rx_ctrl[0].ccb->intr_type == BNA_INTR_T_MSIX)
		bnad_rx_msix_unregister(bnad, rx_info, rx_config->num_paths);
		bnad_rx_msix_unregister(bnad, rx_info, rx_config->num_paths);


	bnad_napi_disable(bnad, rx_id);
	bnad_napi_delete(bnad, rx_id);


	spin_lock_irqsave(&bnad->bna_lock, flags);
	spin_lock_irqsave(&bnad->bna_lock, flags);
	bna_rx_destroy(rx_info->rx);
	bna_rx_destroy(rx_info->rx);
@@ -1952,7 +1980,7 @@ bnad_setup_rx(struct bnad *bnad, u32 rx_id)
	struct bna_rx_config *rx_config = &bnad->rx_config[rx_id];
	struct bna_rx_config *rx_config = &bnad->rx_config[rx_id];
	static const struct bna_rx_event_cbfn rx_cbfn = {
	static const struct bna_rx_event_cbfn rx_cbfn = {
		.rcb_setup_cbfn = bnad_cb_rcb_setup,
		.rcb_setup_cbfn = bnad_cb_rcb_setup,
		.rcb_destroy_cbfn = bnad_cb_rcb_destroy,
		.rcb_destroy_cbfn = NULL,
		.ccb_setup_cbfn = bnad_cb_ccb_setup,
		.ccb_setup_cbfn = bnad_cb_ccb_setup,
		.ccb_destroy_cbfn = bnad_cb_ccb_destroy,
		.ccb_destroy_cbfn = bnad_cb_ccb_destroy,
		.rx_stall_cbfn = bnad_cb_rx_stall,
		.rx_stall_cbfn = bnad_cb_rx_stall,
@@ -1998,11 +2026,14 @@ bnad_setup_rx(struct bnad *bnad, u32 rx_id)
	rx_info->rx = rx;
	rx_info->rx = rx;
	spin_unlock_irqrestore(&bnad->bna_lock, flags);
	spin_unlock_irqrestore(&bnad->bna_lock, flags);


	INIT_WORK(&rx_info->rx_cleanup_work,
			(work_func_t)(bnad_rx_cleanup));

	/*
	/*
	 * Init NAPI, so that state is set to NAPI_STATE_SCHED,
	 * Init NAPI, so that state is set to NAPI_STATE_SCHED,
	 * so that IRQ handler cannot schedule NAPI at this point.
	 * so that IRQ handler cannot schedule NAPI at this point.
	 */
	 */
	bnad_napi_init(bnad, rx_id);
	bnad_napi_add(bnad, rx_id);


	/* Register ISR for the Rx object */
	/* Register ISR for the Rx object */
	if (intr_info->intr_type == BNA_INTR_T_MSIX) {
	if (intr_info->intr_type == BNA_INTR_T_MSIX) {
@@ -2028,9 +2059,6 @@ bnad_setup_rx(struct bnad *bnad, u32 rx_id)
	bna_rx_enable(rx);
	bna_rx_enable(rx);
	spin_unlock_irqrestore(&bnad->bna_lock, flags);
	spin_unlock_irqrestore(&bnad->bna_lock, flags);


	/* Enable scheduling of NAPI */
	bnad_napi_enable(bnad, rx_id);

	return 0;
	return 0;


err_return:
err_return:
@@ -3129,6 +3157,7 @@ bnad_netdev_init(struct bnad *bnad, bool using_dac)
 * 2. Setup netdev pointer in pci_dev
 * 2. Setup netdev pointer in pci_dev
 * 3. Initialze Tx free tasklet
 * 3. Initialze Tx free tasklet
 * 4. Initialize no. of TxQ & CQs & MSIX vectors
 * 4. Initialize no. of TxQ & CQs & MSIX vectors
 * 5. Initialize work queue.
 */
 */
static int
static int
bnad_init(struct bnad *bnad,
bnad_init(struct bnad *bnad,
@@ -3174,6 +3203,12 @@ bnad_init(struct bnad *bnad,
	tasklet_init(&bnad->tx_free_tasklet, bnad_tx_free_tasklet,
	tasklet_init(&bnad->tx_free_tasklet, bnad_tx_free_tasklet,
		     (unsigned long)bnad);
		     (unsigned long)bnad);


	sprintf(bnad->wq_name, "%s_wq_%d", BNAD_NAME, bnad->id);
	bnad->work_q = create_singlethread_workqueue(bnad->wq_name);

	if (!bnad->work_q)
		return -ENOMEM;

	return 0;
	return 0;
}
}


@@ -3185,6 +3220,12 @@ bnad_init(struct bnad *bnad,
static void
static void
bnad_uninit(struct bnad *bnad)
bnad_uninit(struct bnad *bnad)
{
{
	if (bnad->work_q) {
		flush_workqueue(bnad->work_q);
		destroy_workqueue(bnad->work_q);
		bnad->work_q = NULL;
	}

	if (bnad->bar0)
	if (bnad->bar0)
		iounmap(bnad->bar0);
		iounmap(bnad->bar0);
	pci_set_drvdata(bnad->pcidev, NULL);
	pci_set_drvdata(bnad->pcidev, NULL);
+4 −0
Original line number Original line Diff line number Diff line
@@ -210,6 +210,7 @@ struct bnad_tx_info {
	struct bna_tx *tx; /* 1:1 between tx_info & tx */
	struct bna_tx *tx; /* 1:1 between tx_info & tx */
	struct bna_tcb *tcb[BNAD_MAX_TXQ_PER_TX];
	struct bna_tcb *tcb[BNAD_MAX_TXQ_PER_TX];
	u32 tx_id;
	u32 tx_id;
	struct delayed_work tx_cleanup_work;
} ____cacheline_aligned;
} ____cacheline_aligned;


struct bnad_rx_info {
struct bnad_rx_info {
@@ -217,6 +218,7 @@ struct bnad_rx_info {


	struct bnad_rx_ctrl rx_ctrl[BNAD_MAX_RXP_PER_RX];
	struct bnad_rx_ctrl rx_ctrl[BNAD_MAX_RXP_PER_RX];
	u32 rx_id;
	u32 rx_id;
	struct work_struct rx_cleanup_work;
} ____cacheline_aligned;
} ____cacheline_aligned;


/* Unmap queues for Tx / Rx cleanup */
/* Unmap queues for Tx / Rx cleanup */
@@ -319,6 +321,7 @@ struct bnad {
	mac_t			perm_addr;
	mac_t			perm_addr;


	struct tasklet_struct	tx_free_tasklet;
	struct tasklet_struct	tx_free_tasklet;
	struct workqueue_struct *work_q;


	/* Statistics */
	/* Statistics */
	struct bnad_stats stats;
	struct bnad_stats stats;
@@ -328,6 +331,7 @@ struct bnad {
	char			adapter_name[BNAD_NAME_LEN];
	char			adapter_name[BNAD_NAME_LEN];
	char			port_name[BNAD_NAME_LEN];
	char			port_name[BNAD_NAME_LEN];
	char			mbox_irq_name[BNAD_NAME_LEN];
	char			mbox_irq_name[BNAD_NAME_LEN];
	char			wq_name[BNAD_NAME_LEN];


	/* debugfs specific data */
	/* debugfs specific data */
	char	*regdata;
	char	*regdata;