vmxnet3: Fix tx queue race condition with XDP

If XDP traffic runs on a CPU which is greater than or equal to
the number of the Tx queues of the NIC, then vmxnet3_xdp_get_tq()
always picks up queue 0 for transmission as it uses reciprocal scale
instead of simple modulo operation.

vmxnet3_xdp_xmit() and vmxnet3_xdp_xmit_frame() use the above
returned queue without any locking which can lead to race conditions
when multiple XDP xmits run in parallel on different CPU's.

This patch uses a simple module scheme when the current CPU equals or
exceeds the number of Tx queues on the NIC. It also adds locking in
vmxnet3_xdp_xmit() and vmxnet3_xdp_xmit_frame() functions.

Fixes: 54f00cc ("vmxnet3: Add XDP support.")
Signed-off-by: Sankararaman Jayaraman <[email protected]>
Signed-off-by: Ronak Doshi <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: Sankararaman Jayaraman

drivers/net/vmxnet3/vmxnet3_xdp.c

@@ -28,7 +28,7 @@ vmxnet3_xdp_get_tq(struct vmxnet3_adapter *adapter)
 	if (likely(cpu < tq_number))
 		tq = &adapter->tx_queue[cpu];
 	else
-		tq = &adapter->tx_queue[reciprocal_scale(cpu, tq_number)];
+		tq = &adapter->tx_queue[cpu % tq_number];
 
 	return tq;
 }
@@ -124,6 +124,7 @@ vmxnet3_xdp_xmit_frame(struct vmxnet3_adapter *adapter,
 	u32 buf_size;
 	u32 dw2;
 
+	spin_lock_irq(&tq->tx_lock);
 	dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 	dw2 |= xdpf->len;
 	ctx.sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
@@ -134,6 +135,7 @@ vmxnet3_xdp_xmit_frame(struct vmxnet3_adapter *adapter,
 
 	if (vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) == 0) {
 		tq->stats.tx_ring_full++;
+		spin_unlock_irq(&tq->tx_lock);
 		return -ENOSPC;
 	}
 
@@ -142,8 +144,10 @@ vmxnet3_xdp_xmit_frame(struct vmxnet3_adapter *adapter,
 		tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
 					       xdpf->data, buf_size,
 					       DMA_TO_DEVICE);
-		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
+		if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr)) {
+			spin_unlock_irq(&tq->tx_lock);
 			return -EFAULT;
+		}
 		tbi->map_type |= VMXNET3_MAP_SINGLE;
 	} else { /* XDP buffer from page pool */
 		page = virt_to_page(xdpf->data);
@@ -182,6 +186,7 @@ vmxnet3_xdp_xmit_frame(struct vmxnet3_adapter *adapter,
 	dma_wmb();
 	gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
 						  VMXNET3_TXD_GEN);
+	spin_unlock_irq(&tq->tx_lock);
 
 	/* No need to handle the case when tx_num_deferred doesn't reach
 	 * threshold. Backend driver at hypervisor side will poll and reset
@@ -225,6 +230,7 @@ vmxnet3_xdp_xmit(struct net_device *dev,
 {
 	struct vmxnet3_adapter *adapter = netdev_priv(dev);
 	struct vmxnet3_tx_queue *tq;
+	struct netdev_queue *nq;
 	int i;
 
 	if (unlikely(test_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state)))
@@ -236,13 +242,17 @@ vmxnet3_xdp_xmit(struct net_device *dev,
 	if (tq->stopped)
 		return -ENETDOWN;
 
+	nq = netdev_get_tx_queue(adapter->netdev, tq->qid);
+
+	__netif_tx_lock(nq, smp_processor_id());
 	for (i = 0; i < n; i++) {
 		if (vmxnet3_xdp_xmit_frame(adapter, frames[i], tq, true)) {
 			tq->stats.xdp_xmit_err++;
 			break;
 		}
 	}
 	tq->stats.xdp_xmit += i;
+	__netif_tx_unlock(nq);
 
 	return i;
 }