return false;
 }
 
+static void iwl_napi_add(struct iwl_op_mode *op_mode,
+                        struct napi_struct *napi,
+                        struct net_device *napi_dev,
+                        int (*poll)(struct napi_struct *, int),
+                        int weight)
+{
+       struct iwl_priv *priv = IWL_OP_MODE_GET_DVM(op_mode);
+
+       ieee80211_napi_add(priv->hw, napi, napi_dev, poll, weight);
+}
+
 static const struct iwl_op_mode_ops iwl_dvm_ops = {
        .start = iwl_op_mode_dvm_start,
        .stop = iwl_op_mode_dvm_stop,
        .cmd_queue_full = iwl_cmd_queue_full,
        .nic_config = iwl_nic_config,
        .wimax_active = iwl_wimax_active,
+       .napi_add = iwl_napi_add,
 };
 
 /*****************************************************************************
 
 #ifndef __iwl_op_mode_h__
 #define __iwl_op_mode_h__
 
+#include <linux/netdevice.h>
 #include <linux/debugfs.h>
 
 struct iwl_op_mode;
  * @stop: stop the op_mode. Must free all the memory allocated.
  *     May sleep
  * @rx: Rx notification to the op_mode. rxb is the Rx buffer itself. Cmd is the
- *     HCMD this Rx responds to.
- *     This callback may sleep, it is called from a threaded IRQ handler.
+ *     HCMD this Rx responds to. Can't sleep.
+ * @napi_add: NAPI initialisation. The transport is fully responsible for NAPI,
+ *     but the higher layers need to know about it (in particular mac80211 to
+ *     to able to call the right NAPI RX functions); this function is needed
+ *     to eventually call netif_napi_add() with higher layer involvement.
  * @queue_full: notifies that a HW queue is full.
  *     Must be atomic and called with BH disabled.
  * @queue_not_full: notifies that a HW queue is not full any more.
        void (*stop)(struct iwl_op_mode *op_mode);
        int (*rx)(struct iwl_op_mode *op_mode, struct iwl_rx_cmd_buffer *rxb,
                  struct iwl_device_cmd *cmd);
+       void (*napi_add)(struct iwl_op_mode *op_mode,
+                        struct napi_struct *napi,
+                        struct net_device *napi_dev,
+                        int (*poll)(struct napi_struct *, int),
+                        int weight);
        void (*queue_full)(struct iwl_op_mode *op_mode, int queue);
        void (*queue_not_full)(struct iwl_op_mode *op_mode, int queue);
        bool (*hw_rf_kill)(struct iwl_op_mode *op_mode, bool state);
                                  struct iwl_rx_cmd_buffer *rxb,
                                  struct iwl_device_cmd *cmd)
 {
-       might_sleep();
        return op_mode->ops->rx(op_mode, rxb, cmd);
 }
 
        return op_mode->ops->exit_d0i3(op_mode);
 }
 
+static inline void iwl_op_mode_napi_add(struct iwl_op_mode *op_mode,
+                                       struct napi_struct *napi,
+                                       struct net_device *napi_dev,
+                                       int (*poll)(struct napi_struct *, int),
+                                       int weight)
+{
+       if (!op_mode->ops->napi_add)
+               return;
+       op_mode->ops->napi_add(op_mode, napi, napi_dev, poll, weight);
+}
+
 #endif /* __iwl_op_mode_h__ */
 
        return ret;
 }
 
+static void iwl_mvm_napi_add(struct iwl_op_mode *op_mode,
+                            struct napi_struct *napi,
+                            struct net_device *napi_dev,
+                            int (*poll)(struct napi_struct *, int),
+                            int weight)
+{
+       struct iwl_mvm *mvm = IWL_OP_MODE_GET_MVM(op_mode);
+
+       ieee80211_napi_add(mvm->hw, napi, napi_dev, poll, weight);
+}
+
 static const struct iwl_op_mode_ops iwl_mvm_ops = {
        .start = iwl_op_mode_mvm_start,
        .stop = iwl_op_mode_mvm_stop,
        .nic_config = iwl_mvm_nic_config,
        .enter_d0i3 = iwl_mvm_enter_d0i3,
        .exit_d0i3 = iwl_mvm_exit_d0i3,
+       .napi_add = iwl_mvm_napi_add,
 };
 
 
        memcpy(IEEE80211_SKB_RXCB(skb), stats, sizeof(*stats));
 
-       ieee80211_rx_ni(mvm->hw, skb);
+       ieee80211_rx(mvm->hw, skb);
 }
 
 static void iwl_mvm_calc_rssi(struct iwl_mvm *mvm,
 
                info->status.status_driver_data[0] =
                                (void *)(uintptr_t)tx_resp->reduced_tpc;
 
-               ieee80211_tx_status_ni(mvm->hw, skb);
+               ieee80211_tx_status(mvm->hw, skb);
        }
 
        if (txq_id >= mvm->first_agg_queue) {
 
        while (!skb_queue_empty(&reclaimed_skbs)) {
                skb = __skb_dequeue(&reclaimed_skbs);
-               ieee80211_tx_status_ni(mvm->hw, skb);
+               ieee80211_tx_status(mvm->hw, skb);
        }
 
        return 0;
 
        struct iwl_trans *trans;
        struct iwl_drv *drv;
 
+       struct net_device napi_dev;
+       struct napi_struct napi;
+
        /* INT ICT Table */
        __le32 *ict_tbl;
        dma_addr_t ict_tbl_dma;
 
        /* Reuse the page if possible. For notification packets and
         * SKBs that fail to Rx correctly, add them back into the
         * rx_free list for reuse later. */
-       spin_lock(&rxq->lock);
        if (rxb->page != NULL) {
                rxb->page_dma =
                        dma_map_page(trans->dev, rxb->page, 0,
                }
        } else
                list_add_tail(&rxb->list, &rxq->rx_used);
-       spin_unlock(&rxq->lock);
 }
 
 /*
        u32 count = 8;
        int total_empty;
 
+restart:
+       spin_lock(&rxq->lock);
        /* uCode's read index (stored in shared DRAM) indicates the last Rx
         * buffer that the driver may process (last buffer filled by ucode). */
        r = le16_to_cpu(ACCESS_ONCE(rxq->rb_stts->closed_rb_num)) & 0x0FFF;
                        count++;
                        if (count >= 8) {
                                rxq->read = i;
+                               spin_unlock(&rxq->lock);
                                iwl_pcie_rx_replenish_now(trans);
                                count = 0;
+                               goto restart;
                        }
                }
        }
 
        /* Backtrack one entry */
        rxq->read = i;
+       spin_unlock(&rxq->lock);
+
        if (fill_rx)
                iwl_pcie_rx_replenish_now(trans);
        else
                iwl_pcie_rxq_restock(trans);
+
+       if (trans_pcie->napi.poll)
+               napi_gro_flush(&trans_pcie->napi, false);
 }
 
 /*
                iwl_write8(trans, CSR_INT_PERIODIC_REG,
                            CSR_INT_PERIODIC_DIS);
 
-               iwl_pcie_rx_handle(trans);
-
                /*
                 * Enable periodic interrupt in 8 msec only if we received
                 * real RX interrupt (instead of just periodic int), to catch
                                   CSR_INT_PERIODIC_ENA);
 
                isr_stats->rx++;
+
+               local_bh_disable();
+               iwl_pcie_rx_handle(trans);
+               local_bh_enable();
        }
 
        /* This "Tx" DMA channel is used only for loading uCode */
 
        iwl_trans_pcie_write32(trans, HBUS_TARG_PRPH_WDAT, val);
 }
 
+static int iwl_pcie_dummy_napi_poll(struct napi_struct *napi, int budget)
+{
+       WARN_ON(1);
+       return 0;
+}
+
 static void iwl_trans_pcie_configure(struct iwl_trans *trans,
                                     const struct iwl_trans_config *trans_cfg)
 {
 
        trans_pcie->command_names = trans_cfg->command_names;
        trans_pcie->bc_table_dword = trans_cfg->bc_table_dword;
+
+       /* Initialize NAPI here - it should be before registering to mac80211
+        * in the opmode but after the HW struct is allocated.
+        * As this function may be called again in some corner cases don't
+        * do anything if NAPI was already initialized.
+        */
+       if (!trans_pcie->napi.poll && trans->op_mode->ops->napi_add) {
+               init_dummy_netdev(&trans_pcie->napi_dev);
+               iwl_op_mode_napi_add(trans->op_mode, &trans_pcie->napi,
+                                    &trans_pcie->napi_dev,
+                                    iwl_pcie_dummy_napi_poll, 64);
+       }
 }
 
 void iwl_trans_pcie_free(struct iwl_trans *trans)
        pci_disable_device(trans_pcie->pci_dev);
        kmem_cache_destroy(trans->dev_cmd_pool);
 
+       if (trans_pcie->napi.poll)
+               netif_napi_del(&trans_pcie->napi);
+
        kfree(trans);
 }