The allocation of 'unlink' can be moved before a spin_lock.
This slighly simplifies the error handling if the memory allocation fails,
aligns the code structure with what is done in 'vhci_tx_urb()' and reduces
potential lock contention.
Signed-off-by: Christophe JAILLET <[email protected]>
---
drivers/usb/usbip/vhci_hcd.c | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
index 65850e9c7190..b909a634260c 100644
--- a/drivers/usb/usbip/vhci_hcd.c
+++ b/drivers/usb/usbip/vhci_hcd.c
@@ -905,17 +905,16 @@ static int vhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
/* tcp connection is alive */
struct vhci_unlink *unlink;
- spin_lock(&vdev->priv_lock);
-
/* setup CMD_UNLINK pdu */
unlink = kzalloc(sizeof(struct vhci_unlink), GFP_ATOMIC);
if (!unlink) {
- spin_unlock(&vdev->priv_lock);
spin_unlock_irqrestore(&vhci->lock, flags);
usbip_event_add(&vdev->ud, VDEV_EVENT_ERROR_MALLOC);
return -ENOMEM;
}
+ spin_lock(&vdev->priv_lock);
+
unlink->seqnum = atomic_inc_return(&vhci_hcd->seqnum);
if (unlink->seqnum == 0xffff)
pr_info("seqnum max\n");
--
2.20.1
On 3/21/20 9:29 AM, Christophe JAILLET wrote:
> The allocation of 'unlink' can be moved before a spin_lock.
> This slighly simplifies the error handling if the memory allocation fails,
slightly (spelling nit)
> aligns the code structure with what is done in 'vhci_tx_urb()' and reduces
> potential lock contention.
>
Are you seeing any problems or is this a potential lock contention?
If you are seeing issues, please share the problem seen.
> Signed-off-by: Christophe JAILLET <[email protected]>
> ---
> drivers/usb/usbip/vhci_hcd.c | 5 ++---
> 1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> index 65850e9c7190..b909a634260c 100644
> --- a/drivers/usb/usbip/vhci_hcd.c
> +++ b/drivers/usb/usbip/vhci_hcd.c
> @@ -905,17 +905,16 @@ static int vhci_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
> /* tcp connection is alive */
> struct vhci_unlink *unlink;
>
> - spin_lock(&vdev->priv_lock);
> -
This change might simplify the error path, however it could
open a race window with the unlink activity during
vhci_shutdown_connection() when the connection is being taken
down. It would be safer to hold both locks as soon as the
connection check is done.
> /* setup CMD_UNLINK pdu */
> unlink = kzalloc(sizeof(struct vhci_unlink), GFP_ATOMIC);
> if (!unlink) {
> - spin_unlock(&vdev->priv_lock);
> spin_unlock_irqrestore(&vhci->lock, flags);
> usbip_event_add(&vdev->ud, VDEV_EVENT_ERROR_MALLOC);
> return -ENOMEM;
> }
>
> + spin_lock(&vdev->priv_lock);
> +
> unlink->seqnum = atomic_inc_return(&vhci_hcd->seqnum);
> if (unlink->seqnum == 0xffff)
> pr_info("seqnum max\n");
>
thanks,
-- Shuah
Le 23/03/2020 à 17:48, shuah a écrit :
> On 3/21/20 9:29 AM, Christophe JAILLET wrote:
>> The allocation of 'unlink' can be moved before a spin_lock.
>> This slighly simplifies the error handling if the memory allocation
>> fails,
>
> slightly (spelling nit)
>
>> aligns the code structure with what is done in 'vhci_tx_urb()' and
>> reduces
>> potential lock contention.
>>
>
> Are you seeing any problems or is this a potential lock contention?
> If you are seeing issues, please share the problem seen.
>
No, the issue is just theoretical.
>
>> Signed-off-by: Christophe JAILLET <[email protected]>
>> ---
>> drivers/usb/usbip/vhci_hcd.c | 5 ++---
>> 1 file changed, 2 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
>> index 65850e9c7190..b909a634260c 100644
>> --- a/drivers/usb/usbip/vhci_hcd.c
>> +++ b/drivers/usb/usbip/vhci_hcd.c
>> @@ -905,17 +905,16 @@ static int vhci_urb_dequeue(struct usb_hcd
>> *hcd, struct urb *urb, int status)
>> /* tcp connection is alive */
>> struct vhci_unlink *unlink;
>> - spin_lock(&vdev->priv_lock);
>> -
>
> This change might simplify the error path, however it could
> open a race window with the unlink activity during
> vhci_shutdown_connection() when the connection is being taken
> down. It would be safer to hold both locks as soon as the
> connection check is done.
My proposal was just a small clean-up (from my point of view at least).
If it can have some side effects, please, just consider it as a NACK.
CJ
>
>> /* setup CMD_UNLINK pdu */
>> unlink = kzalloc(sizeof(struct vhci_unlink), GFP_ATOMIC);
>> if (!unlink) {
>> - spin_unlock(&vdev->priv_lock);
>> spin_unlock_irqrestore(&vhci->lock, flags);
>> usbip_event_add(&vdev->ud, VDEV_EVENT_ERROR_MALLOC);
>> return -ENOMEM;
>> }
>> + spin_lock(&vdev->priv_lock);
>> +
>> unlink->seqnum = atomic_inc_return(&vhci_hcd->seqnum);
>> if (unlink->seqnum == 0xffff)
>> pr_info("seqnum max\n");
>>
>
> thanks,
> -- Shuah
>