ath10k: implement NAPI support

Add NAPI support for rx and tx completion. NAPI poll is scheduled
from interrupt handler. The design is as below

 - on interrupt
     - schedule napi and mask interrupts
 - on poll
   - process all pipes (no actual Tx/Rx)
   - process Rx within budget
   - if quota exceeds budget reschedule napi poll by returning budget
   - process Tx completions and update budget if necessary
   - process Tx fetch indications (pull-push)
   - push any other pending Tx (if possible)
   - before resched or napi completion replenish htt rx ring buffer
   - if work done < budget, complete napi poll and unmask interrupts

This change also get rid of two tasklets (intr_tq and txrx_compl_task).

Measured peak throughput with NAPI on IPQ4019 platform in controlled
environment. No noticeable reduction in throughput is seen and also
observed improvements in CPU usage. Approx. 15% CPU usage got reduced
in UDP uplink case.

DL: AP DUT Tx
UL: AP DUT Rx

IPQ4019 (avg. cpu usage %)

========
                TOT              +NAPI
              ===========      =============
TCP DL       644 Mbps (42%)    645 Mbps (36%)
TCP UL       673 Mbps (30%)    675 Mbps (26%)
UDP DL       682 Mbps (49%)    680 Mbps (49%)
UDP UL       720 Mbps (28%)    717 Mbps (11%)
Signed-off-by: Rajkumar Manoharan <rmanohar@qti.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>

drivers/net/wireless/ath/ath10k/ahb.c

@@ -462,13 +462,13 @@ static void ath10k_ahb_halt_chip(struct ath10k *ar)
 static irqreturn_t ath10k_ahb_interrupt_handler(int irq, void *arg)
 {
 	struct ath10k *ar = arg;
-	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
 
 	if (!ath10k_pci_irq_pending(ar))
 		return IRQ_NONE;
 
 	ath10k_pci_disable_and_clear_legacy_irq(ar);
-	tasklet_schedule(&ar_pci->intr_tq);
+	ath10k_pci_irq_msi_fw_mask(ar);
+	napi_schedule(&ar->napi);
 
 	return IRQ_HANDLED;
 }
@@ -717,6 +717,9 @@ static void ath10k_ahb_hif_stop(struct ath10k *ar)
 	synchronize_irq(ar_ahb->irq);
 
 	ath10k_pci_flush(ar);
+
+	napi_synchronize(&ar->napi);
+	napi_disable(&ar->napi);
 }
 
 static int ath10k_ahb_hif_power_up(struct ath10k *ar)
@@ -748,6 +751,7 @@ static int ath10k_ahb_hif_power_up(struct ath10k *ar)
 		ath10k_err(ar, "could not wake up target CPU: %d\n", ret);
 		goto err_ce_deinit;
 	}
+	napi_enable(&ar->napi);
 
 	return 0;
 
@@ -831,7 +835,7 @@ static int ath10k_ahb_probe(struct platform_device *pdev)
 		goto err_resource_deinit;
 	}
 
-	ath10k_pci_init_irq_tasklets(ar);
+	ath10k_pci_init_napi(ar);
 
 	ret = ath10k_ahb_request_irq_legacy(ar);
 	if (ret)

drivers/net/wireless/ath/ath10k/core.c

@@ -2322,6 +2322,8 @@ struct ath10k *ath10k_core_create(size_t priv_size, struct device *dev,
 	INIT_WORK(&ar->register_work, ath10k_core_register_work);
 	INIT_WORK(&ar->restart_work, ath10k_core_restart);
 
+	init_dummy_netdev(&ar->napi_dev);
+
 	ret = ath10k_debug_create(ar);
 	if (ret)
 		goto err_free_aux_wq;

drivers/net/wireless/ath/ath10k/core.h

@@ -65,6 +65,10 @@
 #define ATH10K_KEEPALIVE_MAX_IDLE 3895
 #define ATH10K_KEEPALIVE_MAX_UNRESPONSIVE 3900
 
+/* NAPI poll budget */
+#define ATH10K_NAPI_BUDGET      64
+#define ATH10K_NAPI_QUOTA_LIMIT 60
+
 struct ath10k;
 
 enum ath10k_bus {
@@ -954,6 +958,10 @@ struct ath10k {
 	struct ath10k_thermal thermal;
 	struct ath10k_wow wow;
 
+	/* NAPI */
+	struct net_device napi_dev;
+	struct napi_struct napi;
+
 	/* must be last */
 	u8 drv_priv[0] __aligned(sizeof(void *));
 };

drivers/net/wireless/ath/ath10k/htt.h

@@ -1665,7 +1665,6 @@ struct ath10k_htt {
 
 	/* This is used to group tx/rx completions separately and process them
 	 * in batches to reduce cache stalls */
-	struct tasklet_struct txrx_compl_task;
 	struct sk_buff_head rx_compl_q;
 	struct sk_buff_head rx_in_ord_compl_q;
 	struct sk_buff_head tx_fetch_ind_q;
@@ -1798,5 +1797,6 @@ int ath10k_htt_tx(struct ath10k_htt *htt,
 		  struct sk_buff *msdu);
 void ath10k_htt_rx_pktlog_completion_handler(struct ath10k *ar,
 					     struct sk_buff *skb);
+int ath10k_htt_txrx_compl_task(struct ath10k *ar, int budget);
 
 #endif

drivers/net/wireless/ath/ath10k/htt_tx.c

@@ -390,8 +390,6 @@ void ath10k_htt_tx_free(struct ath10k_htt *htt)
 {
 	int size;
 
-	tasklet_kill(&htt->txrx_compl_task);
-
 	idr_for_each(&htt->pending_tx, ath10k_htt_tx_clean_up_pending, htt->ar);
 	idr_destroy(&htt->pending_tx);
 

drivers/net/wireless/ath/ath10k/pci.c

@@ -1506,12 +1506,10 @@ void ath10k_pci_hif_send_complete_check(struct ath10k *ar, u8 pipe,
 	ath10k_ce_per_engine_service(ar, pipe);
 }
 
-void ath10k_pci_kill_tasklet(struct ath10k *ar)
+static void ath10k_pci_rx_retry_sync(struct ath10k *ar)
 {
 	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
 
-	tasklet_kill(&ar_pci->intr_tq);
-
 	del_timer_sync(&ar_pci->rx_post_retry);
 }
 
@@ -1570,7 +1568,7 @@ void ath10k_pci_hif_get_default_pipe(struct ath10k *ar,
 						 ul_pipe, dl_pipe);
 }
 
-static void ath10k_pci_irq_msi_fw_mask(struct ath10k *ar)
+void ath10k_pci_irq_msi_fw_mask(struct ath10k *ar)
 {
 	u32 val;
 
@@ -1753,7 +1751,7 @@ void ath10k_pci_ce_deinit(struct ath10k *ar)
 
 void ath10k_pci_flush(struct ath10k *ar)
 {
-	ath10k_pci_kill_tasklet(ar);
+	ath10k_pci_rx_retry_sync(ar);
 	ath10k_pci_buffer_cleanup(ar);
 }
 
@@ -1780,6 +1778,8 @@ static void ath10k_pci_hif_stop(struct ath10k *ar)
 	ath10k_pci_irq_disable(ar);
 	ath10k_pci_irq_sync(ar);
 	ath10k_pci_flush(ar);
+	napi_synchronize(&ar->napi);
+	napi_disable(&ar->napi);
 
 	spin_lock_irqsave(&ar_pci->ps_lock, flags);
 	WARN_ON(ar_pci->ps_wake_refcount > 0);
@@ -2533,6 +2533,7 @@ static int ath10k_pci_hif_power_up(struct ath10k *ar)
 		ath10k_err(ar, "could not wake up target CPU: %d\n", ret);
 		goto err_ce;
 	}
+	napi_enable(&ar->napi);
 
 	return 0;
 
@@ -2772,35 +2773,53 @@ static irqreturn_t ath10k_pci_interrupt_handler(int irq, void *arg)
 		return IRQ_NONE;
 	}
 
-	if (ar_pci->oper_irq_mode == ATH10K_PCI_IRQ_LEGACY) {
-		if (!ath10k_pci_irq_pending(ar))
-			return IRQ_NONE;
-
-		ath10k_pci_disable_and_clear_legacy_irq(ar);
-	}
+	if ((ar_pci->oper_irq_mode == ATH10K_PCI_IRQ_LEGACY) &&
+	    !ath10k_pci_irq_pending(ar))
+		return IRQ_NONE;
 
-	tasklet_schedule(&ar_pci->intr_tq);
+	ath10k_pci_disable_and_clear_legacy_irq(ar);
+	ath10k_pci_irq_msi_fw_mask(ar);
+	napi_schedule(&ar->napi);
 
 	return IRQ_HANDLED;
 }
 
-static void ath10k_pci_tasklet(unsigned long data)
+static int ath10k_pci_napi_poll(struct napi_struct *ctx, int budget)
 {
-	struct ath10k *ar = (struct ath10k *)data;
-	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k *ar = container_of(ctx, struct ath10k, napi);
+	int done = 0;
 
 	if (ath10k_pci_has_fw_crashed(ar)) {
-		ath10k_pci_irq_disable(ar);
 		ath10k_pci_fw_crashed_clear(ar);
 		ath10k_pci_fw_crashed_dump(ar);
-		return;
+		napi_complete(ctx);
+		return done;
 	}
 
 	ath10k_ce_per_engine_service_any(ar);
 
-	/* Re-enable legacy irq that was disabled in the irq handler */
-	if (ar_pci->oper_irq_mode == ATH10K_PCI_IRQ_LEGACY)
+	done = ath10k_htt_txrx_compl_task(ar, budget);
+
+	if (done < budget) {
+		napi_complete(ctx);
+		/* In case of MSI, it is possible that interrupts are received
+		 * while NAPI poll is inprogress. So pending interrupts that are
+		 * received after processing all copy engine pipes by NAPI poll
+		 * will not be handled again. This is causing failure to
+		 * complete boot sequence in x86 platform. So before enabling
+		 * interrupts safer to check for pending interrupts for
+		 * immediate servicing.
+		 */
+		if (CE_INTERRUPT_SUMMARY(ar)) {
+			napi_reschedule(ctx);
+			goto out;
+		}
 		ath10k_pci_enable_legacy_irq(ar);
+		ath10k_pci_irq_msi_fw_unmask(ar);
+	}
+
+out:
+	return done;
 }
 
 static int ath10k_pci_request_irq_msi(struct ath10k *ar)
@@ -2858,11 +2877,10 @@ static void ath10k_pci_free_irq(struct ath10k *ar)
 	free_irq(ar_pci->pdev->irq, ar);
 }
 
-void ath10k_pci_init_irq_tasklets(struct ath10k *ar)
+void ath10k_pci_init_napi(struct ath10k *ar)
 {
-	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
-
-	tasklet_init(&ar_pci->intr_tq, ath10k_pci_tasklet, (unsigned long)ar);
+	netif_napi_add(&ar->napi_dev, &ar->napi, ath10k_pci_napi_poll,
+		       ATH10K_NAPI_BUDGET);
 }
 
 static int ath10k_pci_init_irq(struct ath10k *ar)
@@ -2870,7 +2888,7 @@ static int ath10k_pci_init_irq(struct ath10k *ar)
 	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
 	int ret;
 
-	ath10k_pci_init_irq_tasklets(ar);
+	ath10k_pci_init_napi(ar);
 
 	if (ath10k_pci_irq_mode != ATH10K_PCI_IRQ_AUTO)
 		ath10k_info(ar, "limiting irq mode to: %d\n",
@@ -3131,7 +3149,8 @@ int ath10k_pci_setup_resource(struct ath10k *ar)
 
 void ath10k_pci_release_resource(struct ath10k *ar)
 {
-	ath10k_pci_kill_tasklet(ar);
+	ath10k_pci_rx_retry_sync(ar);
+	netif_napi_del(&ar->napi);
 	ath10k_pci_ce_deinit(ar);
 	ath10k_pci_free_pipes(ar);
 }
@@ -3298,7 +3317,7 @@ static int ath10k_pci_probe(struct pci_dev *pdev,
 
 err_free_irq:
 	ath10k_pci_free_irq(ar);
-	ath10k_pci_kill_tasklet(ar);
+	ath10k_pci_rx_retry_sync(ar);
 
 err_deinit_irq:
 	ath10k_pci_deinit_irq(ar);

drivers/net/wireless/ath/ath10k/pci.h

@@ -177,8 +177,6 @@ struct ath10k_pci {
 	/* Operating interrupt mode */
 	enum ath10k_pci_irq_mode oper_irq_mode;
 
-	struct tasklet_struct intr_tq;
-
 	struct ath10k_pci_pipe pipe_info[CE_COUNT_MAX];
 
 	/* Copy Engine used for Diagnostic Accesses */
@@ -294,8 +292,7 @@ void ath10k_pci_free_pipes(struct ath10k *ar);
 void ath10k_pci_free_pipes(struct ath10k *ar);
 void ath10k_pci_rx_replenish_retry(unsigned long ptr);
 void ath10k_pci_ce_deinit(struct ath10k *ar);
-void ath10k_pci_init_irq_tasklets(struct ath10k *ar);
-void ath10k_pci_kill_tasklet(struct ath10k *ar);
+void ath10k_pci_init_napi(struct ath10k *ar);
 int ath10k_pci_init_pipes(struct ath10k *ar);
 int ath10k_pci_init_config(struct ath10k *ar);
 void ath10k_pci_rx_post(struct ath10k *ar);
@@ -303,6 +300,7 @@ void ath10k_pci_flush(struct ath10k *ar);
 void ath10k_pci_enable_legacy_irq(struct ath10k *ar);
 bool ath10k_pci_irq_pending(struct ath10k *ar);
 void ath10k_pci_disable_and_clear_legacy_irq(struct ath10k *ar);
+void ath10k_pci_irq_msi_fw_mask(struct ath10k *ar);
 int ath10k_pci_wait_for_target_init(struct ath10k *ar);
 int ath10k_pci_setup_resource(struct ath10k *ar);
 void ath10k_pci_release_resource(struct ath10k *ar);