[PATCH] bnx2x: clean up VF PCI resources on probe failure

[PATCH] bnx2x: clean up VF PCI resources on probe failure

[Haoxiang Li]

bnx2x_init_one() allocates VF PCI resources by calling
bnx2x_vf_pci_alloc(), but later error paths fall through
the common cleanup path without calling bnx2x_vf_pci_dealloc().

This can leak the VF PCI resources allocated for the VF
when probe fails after bnx2x_vf_pci_alloc() succeeds.

Add a VF-specific error label to call bnx2x_vf_pci_dealloc()
before falling through to the existing common cleanup path.

Fixes: 6411280ac94d ("bnx2x: Segregate SR-IOV code")
Cc: stable@vger.kernel.org
Signed-off-by: Haoxiang Li <lihaoxiang@isrc.iscas.ac.cn>
---
 drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
index da0f8c353e6a..acb0379d94b9 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
@@ -13932,13 +13932,13 @@ static int bnx2x_init_one(struct pci_dev *pdev,
 		dev_err(&bp->pdev->dev,
 			"Cannot map doorbell space, aborting\n");
 		rc = -ENOMEM;
-		goto init_one_freemem;
+		goto init_one_vf_pci_dealloc;
 	}
 
 	if (IS_VF(bp)) {
 		rc = bnx2x_vfpf_acquire(bp, tx_count, rx_count);
 		if (rc)
-			goto init_one_freemem;
+			goto init_one_vf_pci_dealloc;
 
 #ifdef CONFIG_BNX2X_SRIOV
 		/* VF with OLD Hypervisor or old PF do not support filtering */
@@ -13952,7 +13952,7 @@ static int bnx2x_init_one(struct pci_dev *pdev,
 	/* Enable SRIOV if capability found in configuration space */
 	rc = bnx2x_iov_init_one(bp, int_mode, BNX2X_MAX_NUM_OF_VFS);
 	if (rc)
-		goto init_one_freemem;
+		goto init_one_vf_pci_dealloc;
 
 	/* calc qm_cid_count */
 	bp->qm_cid_count = bnx2x_set_qm_cid_count(bp);
@@ -13971,7 +13971,7 @@ static int bnx2x_init_one(struct pci_dev *pdev,
 	rc = bnx2x_set_int_mode(bp);
 	if (rc) {
 		dev_err(&pdev->dev, "Cannot set interrupts\n");
-		goto init_one_freemem;
+		goto init_one_vf_pci_dealloc;
 	}
 	BNX2X_DEV_INFO("set interrupts successfully\n");
 
@@ -13979,7 +13979,7 @@ static int bnx2x_init_one(struct pci_dev *pdev,
 	rc = register_netdev(dev);
 	if (rc) {
 		dev_err(&pdev->dev, "Cannot register net device\n");
-		goto init_one_freemem;
+		goto init_one_vf_pci_dealloc;
 	}
 	BNX2X_DEV_INFO("device name after netdev register %s\n", dev->name);
 
@@ -14001,6 +14001,10 @@ static int bnx2x_init_one(struct pci_dev *pdev,
 
 	return 0;
 
+init_one_vf_pci_dealloc:
+	if (IS_VF(bp))
+		bnx2x_vf_pci_dealloc(bp);
+
 init_one_freemem:
 	bnx2x_free_mem_bp(bp);
 
-- 
2.25.1

Re: [PATCH] bnx2x: clean up VF PCI resources on probe failure

[Jakub Kicinski]

This is an AI-generated review of your patch. The human sending this
email says "Sashiko points out other problems with the error path
here, they should all be fixed in one series."
---
bnx2x: clean up VF PCI resources on probe failure

bnx2x_init_one() allocates VF PCI resources by calling bnx2x_vf_pci_alloc(),
but later error paths fall through the common cleanup path without calling
bnx2x_vf_pci_dealloc(). This commit adds a VF-specific error label to fix
the leak.

Fixes: 6411280ac94d ("bnx2x: Segregate SR-IOV code")
Cc: stable@vger.kernel.org
Signed-off-by: Haoxiang Li <lihaoxiang@isrc.iscas.ac.cn>

> diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
> index da0f8c353e6a7..acb0379d94b99 100644
> --- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
> +++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
[ ... ]
> @@ -13979,7 +13979,7 @@ static int bnx2x_init_one(struct pci_dev *pdev,
>  	rc = register_netdev(dev);
>  	if (rc) {
>  		dev_err(&pdev->dev, "Cannot register net device\n");
> -		goto init_one_freemem;
> +		goto init_one_vf_pci_dealloc;

[Severity: High]
This is a pre-existing issue, but if bnx2x_set_int_mode() succeeds earlier in
bnx2x_init_one(), MSI or MSI-X interrupts are configured. If register_netdev()
subsequently fails, the error path skips disabling these interrupts.

Does this leave the interrupts enabled and leak PCI IRQ vectors since there is
no call to bnx2x_disable_msi() on this error path?

>  	}
>  	BNX2X_DEV_INFO("device name after netdev register %s\n", dev->name);
>  
> @@ -14001,6 +14001,10 @@ static int bnx2x_init_one(struct pci_dev *pdev,
>  
>  	return 0;
>  
> +init_one_vf_pci_dealloc:
> +	if (IS_VF(bp))
> +		bnx2x_vf_pci_dealloc(bp);
> +

[Severity: High]
This isn't a bug introduced by this patch, but adding this mailbox cleanup here
seems to prevent releasing the acquired PF resources.

Earlier in bnx2x_init_one(), if IS_VF(bp) is true, we call bnx2x_vfpf_acquire()
to request resources from the PF. If a failure occurs later in the probe
function, the new error path calls bnx2x_vf_pci_dealloc() which frees the
VF-to-PF mailbox.

Because the mailbox memory is destroyed without first calling
bnx2x_vfpf_release(), does this result in a permanent leak of PF resources
assigned to the VF?

>  init_one_freemem:
>  	bnx2x_free_mem_bp(bp);
-- 
pw-bot: cr