[PATCH] ibmveth fix buffer replenishing

This patch removes the allocation of RX skb's  buffers from a workqueue
to be called directly at RX processing time.  This change was suggested
by Dave Miller when the driver was starving the RX buffers and
deadlocking under heavy traffic:

> Allocating RX SKBs via tasklet is, IMHO, the worst way to
> do it.  It is no surprise that there are starvation cases.
>
> If tasklets or work queues get delayed in any way, you lose,
> and it's very easy for a card to catch up with the driver RX'ing
> packets very fast, no matter how aggressive you make the
> replenishing.  By the time you detect that you need to be
> "more aggressive" it is already too late.
> The only pseudo-reliable way is to allocate at RX processing time.
>

Signed-off-by: Santiago Leon <santil@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>

2 files changed, 8 insertions, 44 deletions

diff --git a/drivers/net/ibmveth.c b/drivers/net/ibmveth.c
index d985b804a762..aea1598b2253 100644
--- a/drivers/net/ibmveth.c
+++ b/drivers/net/ibmveth.c
@@ -96,7 +96,6 @@ static void ibmveth_proc_unregister_driver(void);
 static void ibmveth_proc_register_adapter(struct ibmveth_adapter *adapter);
 static void ibmveth_proc_unregister_adapter(struct ibmveth_adapter *adapter);
 static irqreturn_t ibmveth_interrupt(int irq, void *dev_instance, struct pt_regs *regs);
-static inline void ibmveth_schedule_replenishing(struct ibmveth_adapter*);
 static inline void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter);
 
 #ifdef CONFIG_PROC_FS
@@ -257,29 +256,7 @@ static void ibmveth_replenish_buffer_pool(struct ibmveth_adapter *adapter, struc
         atomic_add(buffers_added, &(pool->available));
 }
 
-/* check if replenishing is needed.  */
-static inline int ibmveth_is_replenishing_needed(struct ibmveth_adapter *adapter)
-{
-        int i;
-
-        for(i = 0; i < IbmVethNumBufferPools; i++)
-                if(adapter->rx_buff_pool[i].active &&
-                  (atomic_read(&adapter->rx_buff_pool[i].available) <
-                   adapter->rx_buff_pool[i].threshold))
-                        return 1;
-        return 0;
-}
-
-/* kick the replenish tasklet if we need replenishing and it isn't already running */
-static inline void ibmveth_schedule_replenishing(struct ibmveth_adapter *adapter)
-{
-        if(ibmveth_is_replenishing_needed(adapter) &&
-           (atomic_dec_if_positive(&adapter->not_replenishing) == 0)) {
-                schedule_work(&adapter->replenish_task);
-        }
-}
-
-/* replenish tasklet routine */
+/* replenish routine */
 static void ibmveth_replenish_task(struct ibmveth_adapter *adapter) 
 {
         int i;
@@ -292,10 +269,6 @@ static void ibmveth_replenish_task(struct ibmveth_adapter *adapter)
                                                      &adapter->rx_buff_pool[i]);
 
         adapter->rx_no_buffer = *(u64*)(((char*)adapter->buffer_list_addr) + 4096 - 8);
-
-        atomic_inc(&adapter->not_replenishing);
-
-        ibmveth_schedule_replenishing(adapter);
 }
 
 /* empty and free ana buffer pool - also used to do cleanup in error paths */
@@ -563,10 +536,10 @@ static int ibmveth_open(struct net_device *netdev)
                 return rc;
         }
 
-        netif_start_queue(netdev);
+        ibmveth_debug_printk("initial replenish cycle\n");
+        ibmveth_replenish_task(adapter);
 
-        ibmveth_debug_printk("scheduling initial replenish cycle\n");
-        ibmveth_schedule_replenishing(adapter);
+        netif_start_queue(netdev);
 
         ibmveth_debug_printk("open complete\n");
 
@@ -584,9 +557,6 @@ static int ibmveth_close(struct net_device *netdev)
 
         free_irq(netdev->irq, netdev);
 
-        cancel_delayed_work(&adapter->replenish_task);
-        flush_scheduled_work();
-
         do {
                 lpar_rc = h_free_logical_lan(adapter->vdev->unit_address);
         } while (H_isLongBusy(lpar_rc) || (lpar_rc == H_Busy));
@@ -795,7 +765,7 @@ static int ibmveth_poll(struct net_device *netdev, int *budget)
                 }
         } while(more_work && (frames_processed < max_frames_to_process));
 
-        ibmveth_schedule_replenishing(adapter);
+        ibmveth_replenish_task(adapter);
 
         if(more_work) {
                 /* more work to do - return that we are not done yet */
@@ -931,8 +901,10 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
 
         }
 
+        /* kick the interrupt handler so that the new buffer pools get
+           replenished or deallocated */
+        ibmveth_interrupt(dev->irq, dev, NULL);
 
-        ibmveth_schedule_replenishing(adapter);
         dev->mtu = new_mtu;
         return 0;       
 }
@@ -1017,14 +989,10 @@ static int __devinit ibmveth_probe(struct vio_dev *dev, const struct vio_device_
 
         ibmveth_debug_printk("adapter @ 0x%p\n", adapter);
 
-        INIT_WORK(&adapter->replenish_task, (void*)ibmveth_replenish_task, (void*)adapter);
-
         adapter->buffer_list_dma = DMA_ERROR_CODE;
         adapter->filter_list_dma = DMA_ERROR_CODE;
         adapter->rx_queue.queue_dma = DMA_ERROR_CODE;
 
-        atomic_set(&adapter->not_replenishing, 1);
-
         ibmveth_debug_printk("registering netdev...\n");
 
         rc = register_netdev(netdev);
diff --git a/drivers/net/ibmveth.h b/drivers/net/ibmveth.h
index a5d27a9cdf1f..a3ea02937d63 100644
--- a/drivers/net/ibmveth.h
+++ b/drivers/net/ibmveth.h
@@ -118,10 +118,6 @@ struct ibmveth_adapter {
     dma_addr_t filter_list_dma;
     struct ibmveth_buff_pool rx_buff_pool[IbmVethNumBufferPools];
     struct ibmveth_rx_q rx_queue;
-    atomic_t not_replenishing;
-
-    /* helper tasks */
-    struct work_struct replenish_task;
 
     /* adapter specific stats */
     u64 replenish_task_cycles;