qed: Correct MSI-x for storage

When qedr is enabled, qed would try dividing the msi-x vectors between
L2 and RoCE, starting with L2 and providing it with sufficient vectors
for its queues.

Problem is qed would also do that for storage partitions, and as those
don't need queues it would lead qed to award those partitions with 0
msi-x vectors, causing them to believe theye're using INTa and
preventing them from operating.

Fixes: 51ff172 ("qed: Add support for RoCE hw init")
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/qlogic/qed/qed_main.c

@@ -745,7 +745,8 @@ static int qed_slowpath_setup_int(struct qed_dev *cdev,
 	cdev->int_params.fp_msix_cnt = cdev->int_params.out.num_vectors -
 				       cdev->num_hwfns;
 
-	if (!IS_ENABLED(CONFIG_QED_RDMA))
+	if (!IS_ENABLED(CONFIG_QED_RDMA) ||
+	    QED_LEADING_HWFN(cdev)->hw_info.personality != QED_PCI_ETH_ROCE)
 		return 0;
 
 	for_each_hwfn(cdev, i)