On Thu, Oct 03, 2013 at 10:14:33AM +0300, Eli Cohen wrote:

On Wed, Oct 02, 2013 at 12:49:06PM +0200, Alexander Gordeev wrote:

quoted

 
+	err = pci_msix_table_size(dev->pdev);
+	if (err < 0)
+		return err;
+
 	nvec = dev->caps.num_ports * num_online_cpus() + MLX5_EQ_VEC_COMP_BASE;
 	nvec = min_t(int, nvec, num_eqs);
+	nvec = min_t(int, nvec, err);
 	if (nvec <= MLX5_EQ_VEC_COMP_BASE)
 		return -ENOSPC;

Making sure we don't request more vectors then the device's is capable
of -- looks good.

quoted

@@ -131,20 +136,15 @@ static int mlx5_enable_msix(struct mlx5_core_dev *dev)
 	for (i = 0; i < nvec; i++)
 		table->msix_arr[i].entry = i;
 
-retry:
-	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
 	err = pci_enable_msix(dev->pdev, table->msix_arr, nvec);
-	if (err <= 0) {
+	if (err) {
+		kfree(table->msix_arr);
 		return err;
-	} else if (err > MLX5_EQ_VEC_COMP_BASE) {
-		nvec = err;
-		goto retry;
 	}
 

According to latest sources, pci_enable_msix() may still fail so why
do you want to remove this code?

pci_enable_msix() may fail, but it can not return a positive number.

We first calculate how many MSI-Xs we need, adjust to what we can get,
check if that is enough and only then go for it.

quoted

-	mlx5_core_dbg(dev, "received %d MSI vectors out of %d requested\n", err, nvec);
-	kfree(table->msix_arr);
+	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
 
-	return -ENOSPC;
+	return 0;
 }
 
 static void mlx5_disable_msix(struct mlx5_core_dev *dev)

-- 
Regards,
Alexander Gordeev
agordeev@redhat.com