[PATCH] net/failsafe: fix interrupt handle leak

[PATCH] net/failsafe: fix interrupt handle leak

[David Marchand]

A intr_handle is being allocated as a hack to get a (proxy) eventfd from
the Linux interrupt implementation.
But this handle is never freed.

Remove this convoluted hack and create an eventfd in Linux case.

Fixes: d61138d4f0e2 ("drivers: remove direct access to interrupt handle")
Cc: stable@dpdk.org

Signed-off-by: David Marchand <david.marchand@redhat.com>
---
 drivers/net/failsafe/failsafe_ops.c | 32 ++++++++++-------------------
 1 file changed, 11 insertions(+), 21 deletions(-)

diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
index 55e21d635c..2c23d0e70a 100644
--- a/drivers/net/failsafe/failsafe_ops.c
+++ b/drivers/net/failsafe/failsafe_ops.c
@@ -6,6 +6,9 @@
 #include <stdbool.h>
 #include <stdint.h>
 #include <unistd.h>
+#ifdef RTE_EXEC_ENV_LINUX
+#include <sys/eventfd.h>
+#endif
 
 #include <rte_debug.h>
 #include <rte_atomic.h>
@@ -387,28 +390,11 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
 		const struct rte_eth_rxconf *rx_conf,
 		struct rte_mempool *mb_pool)
 {
-	/*
-	 * FIXME: Add a proper interface in rte_eal_interrupts for
-	 * allocating eventfd as an interrupt vector.
-	 * For the time being, fake as if we are using MSIX interrupts,
-	 * this will cause rte_intr_efd_enable to allocate an eventfd for us.
-	 */
-	struct rte_intr_handle *intr_handle;
 	struct sub_device *sdev;
 	struct rxq *rxq;
 	uint8_t i;
 	int ret;
 
-	intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
-	if (intr_handle == NULL)
-		return -ENOMEM;
-
-	if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX))
-		return -rte_errno;
-
-	if (rte_intr_efds_index_set(intr_handle, 0, -1))
-		return -rte_errno;
-
 	fs_lock(dev, 0);
 	if (rx_conf->rx_deferred_start) {
 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
@@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
 	rxq->info.nb_desc = nb_rx_desc;
 	rxq->priv = PRIV(dev);
 	rxq->sdev = PRIV(dev)->subs;
-	ret = rte_intr_efd_enable(intr_handle, 1);
-	if (ret < 0) {
+#ifdef RTE_EXEC_ENV_LINUX
+	rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
+	if (rxq->event_fd < 0) {
+		ERROR("Failed to create an eventfd: %s", strerror(errno));
 		fs_unlock(dev, 0);
-		return ret;
+		return -errno;
 	}
-	rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
+#else
+	rxq->event_fd = -1;
+#endif
 	dev->data->rx_queues[rx_queue_id] = rxq;
 	FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 		ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
-- 
2.23.0

Re: [PATCH] net/failsafe: fix interrupt handle leak

[Ferruh Yigit]

On 3/24/2022 3:09 PM, David Marchand wrote:
> A intr_handle is being allocated as a hack to get a (proxy) eventfd from
> the Linux interrupt implementation.
> But this handle is never freed.
> 
> Remove this convoluted hack and create an eventfd in Linux case.
> 
> Fixes: d61138d4f0e2 ("drivers: remove direct access to interrupt handle")
> Cc: stable@dpdk.org
> 
> Signed-off-by: David Marchand <david.marchand@redhat.com>
> ---
>   drivers/net/failsafe/failsafe_ops.c | 32 ++++++++++-------------------
>   1 file changed, 11 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
> index 55e21d635c..2c23d0e70a 100644
> --- a/drivers/net/failsafe/failsafe_ops.c
> +++ b/drivers/net/failsafe/failsafe_ops.c
> @@ -6,6 +6,9 @@
>   #include <stdbool.h>
>   #include <stdint.h>
>   #include <unistd.h>
> +#ifdef RTE_EXEC_ENV_LINUX
> +#include <sys/eventfd.h>
> +#endif
>   
>   #include <rte_debug.h>
>   #include <rte_atomic.h>
> @@ -387,28 +390,11 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
>   		const struct rte_eth_rxconf *rx_conf,
>   		struct rte_mempool *mb_pool)
>   {
> -	/*
> -	 * FIXME: Add a proper interface in rte_eal_interrupts for
> -	 * allocating eventfd as an interrupt vector.
> -	 * For the time being, fake as if we are using MSIX interrupts,
> -	 * this will cause rte_intr_efd_enable to allocate an eventfd for us.
> -	 */
> -	struct rte_intr_handle *intr_handle;
>   	struct sub_device *sdev;
>   	struct rxq *rxq;
>   	uint8_t i;
>   	int ret;
>   
> -	intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
> -	if (intr_handle == NULL)
> -		return -ENOMEM;
> -
> -	if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX))
> -		return -rte_errno;
> -
> -	if (rte_intr_efds_index_set(intr_handle, 0, -1))
> -		return -rte_errno;
> -
>   	fs_lock(dev, 0);
>   	if (rx_conf->rx_deferred_start) {
>   		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
> @@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
>   	rxq->info.nb_desc = nb_rx_desc;
>   	rxq->priv = PRIV(dev);
>   	rxq->sdev = PRIV(dev)->subs;
> -	ret = rte_intr_efd_enable(intr_handle, 1);
> -	if (ret < 0) {
> +#ifdef RTE_EXEC_ENV_LINUX
> +	rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
> +	if (rxq->event_fd < 0) {
> +		ERROR("Failed to create an eventfd: %s", strerror(errno));
>   		fs_unlock(dev, 0);
> -		return ret;
> +		return -errno;
>   	}
> -	rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
> +#else
> +	rxq->event_fd = -1;
> +#endif

How this impacts the BSD? I don't know if driver used on BSD but 
technically it looks supported.

@Gaetan, any objection to the change?

Re: [PATCH] net/failsafe: fix interrupt handle leak

[David Marchand]

Hello Ferruh,

On Fri, Apr 29, 2022 at 3:56 PM Ferruh Yigit <ferruh.yigit@xilinx.com> wrote:
> > @@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
> >       rxq->info.nb_desc = nb_rx_desc;
> >       rxq->priv = PRIV(dev);
> >       rxq->sdev = PRIV(dev)->subs;
> > -     ret = rte_intr_efd_enable(intr_handle, 1);
> > -     if (ret < 0) {
> > +#ifdef RTE_EXEC_ENV_LINUX
> > +     rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
> > +     if (rxq->event_fd < 0) {
> > +             ERROR("Failed to create an eventfd: %s", strerror(errno));
> >               fs_unlock(dev, 0);
> > -             return ret;
> > +             return -errno;
> >       }
> > -     rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
> > +#else
> > +     rxq->event_fd = -1;
> > +#endif
>
> How this impacts the BSD? I don't know if driver used on BSD but
> technically it looks supported.

Atm, the driver calls rte_intr_efd_enable() which on Linux triggers an
eventfd creation and a intr_handle->nb_intr++, but does nothing on
FreeBSD.
I did not test it with FreeBSD but I expect the previous call to
rte_intr_efds_index_get(intr_handle, 0); return -1 (because of a
earlier call to rte_intr_efds_index_set(intr_handle, 0, -1)).


-- 
David Marchand

Re: [PATCH] net/failsafe: fix interrupt handle leak

[David Marchand]

On Fri, Apr 29, 2022 at 4:25 PM David Marchand
<david.marchand@redhat.com> wrote:
> eventfd creation and a intr_handle->nb_intr++, but does nothing on

Scratch the part about nb_intr++ :-)
I did not re-read before sending.


-- 
David Marchand

Re: [PATCH] net/failsafe: fix interrupt handle leak

[David Marchand]

On Fri, Apr 29, 2022 at 3:56 PM Ferruh Yigit <ferruh.yigit@xilinx.com> wrote:
>
> On 3/24/2022 3:09 PM, David Marchand wrote:
> > A intr_handle is being allocated as a hack to get a (proxy) eventfd from
> > the Linux interrupt implementation.
> > But this handle is never freed.
> >
> > Remove this convoluted hack and create an eventfd in Linux case.
> >
> > Fixes: d61138d4f0e2 ("drivers: remove direct access to interrupt handle")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: David Marchand <david.marchand@redhat.com>
> > ---
> >   drivers/net/failsafe/failsafe_ops.c | 32 ++++++++++-------------------
> >   1 file changed, 11 insertions(+), 21 deletions(-)
> >
> > diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
> > index 55e21d635c..2c23d0e70a 100644
> > --- a/drivers/net/failsafe/failsafe_ops.c
> > +++ b/drivers/net/failsafe/failsafe_ops.c
> > @@ -6,6 +6,9 @@
> >   #include <stdbool.h>
> >   #include <stdint.h>
> >   #include <unistd.h>
> > +#ifdef RTE_EXEC_ENV_LINUX
> > +#include <sys/eventfd.h>
> > +#endif
> >
> >   #include <rte_debug.h>
> >   #include <rte_atomic.h>
> > @@ -387,28 +390,11 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
> >               const struct rte_eth_rxconf *rx_conf,
> >               struct rte_mempool *mb_pool)
> >   {
> > -     /*
> > -      * FIXME: Add a proper interface in rte_eal_interrupts for
> > -      * allocating eventfd as an interrupt vector.
> > -      * For the time being, fake as if we are using MSIX interrupts,
> > -      * this will cause rte_intr_efd_enable to allocate an eventfd for us.
> > -      */
> > -     struct rte_intr_handle *intr_handle;
> >       struct sub_device *sdev;
> >       struct rxq *rxq;
> >       uint8_t i;
> >       int ret;
> >
> > -     intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
> > -     if (intr_handle == NULL)
> > -             return -ENOMEM;
> > -
> > -     if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX))
> > -             return -rte_errno;
> > -
> > -     if (rte_intr_efds_index_set(intr_handle, 0, -1))
> > -             return -rte_errno;
> > -
> >       fs_lock(dev, 0);
> >       if (rx_conf->rx_deferred_start) {
> >               FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
> > @@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
> >       rxq->info.nb_desc = nb_rx_desc;
> >       rxq->priv = PRIV(dev);
> >       rxq->sdev = PRIV(dev)->subs;
> > -     ret = rte_intr_efd_enable(intr_handle, 1);
> > -     if (ret < 0) {
> > +#ifdef RTE_EXEC_ENV_LINUX
> > +     rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
> > +     if (rxq->event_fd < 0) {
> > +             ERROR("Failed to create an eventfd: %s", strerror(errno));
> >               fs_unlock(dev, 0);
> > -             return ret;
> > +             return -errno;
> >       }
> > -     rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
> > +#else
> > +     rxq->event_fd = -1;
> > +#endif
>
> How this impacts the BSD? I don't know if driver used on BSD but
> technically it looks supported.
>
> @Gaetan, any objection to the change?

There was no feedback for months, can we get this merged?
Thanks.


-- 
David Marchand

Re: [PATCH] net/failsafe: fix interrupt handle leak

[Ferruh Yigit]

On 8/29/2022 11:23 AM, David Marchand wrote:

> 
> On Fri, Apr 29, 2022 at 3:56 PM Ferruh Yigit <ferruh.yigit@xilinx.com> wrote:
>>
>> On 3/24/2022 3:09 PM, David Marchand wrote:
>>> A intr_handle is being allocated as a hack to get a (proxy) eventfd from
>>> the Linux interrupt implementation.
>>> But this handle is never freed.
>>>
>>> Remove this convoluted hack and create an eventfd in Linux case.
>>>
>>> Fixes: d61138d4f0e2 ("drivers: remove direct access to interrupt handle")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: David Marchand <david.marchand@redhat.com>
>>> ---
>>>    drivers/net/failsafe/failsafe_ops.c | 32 ++++++++++-------------------
>>>    1 file changed, 11 insertions(+), 21 deletions(-)
>>>
>>> diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
>>> index 55e21d635c..2c23d0e70a 100644
>>> --- a/drivers/net/failsafe/failsafe_ops.c
>>> +++ b/drivers/net/failsafe/failsafe_ops.c
>>> @@ -6,6 +6,9 @@
>>>    #include <stdbool.h>
>>>    #include <stdint.h>
>>>    #include <unistd.h>
>>> +#ifdef RTE_EXEC_ENV_LINUX
>>> +#include <sys/eventfd.h>
>>> +#endif
>>>
>>>    #include <rte_debug.h>
>>>    #include <rte_atomic.h>
>>> @@ -387,28 +390,11 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
>>>                const struct rte_eth_rxconf *rx_conf,
>>>                struct rte_mempool *mb_pool)
>>>    {
>>> -     /*
>>> -      * FIXME: Add a proper interface in rte_eal_interrupts for
>>> -      * allocating eventfd as an interrupt vector.
>>> -      * For the time being, fake as if we are using MSIX interrupts,
>>> -      * this will cause rte_intr_efd_enable to allocate an eventfd for us.
>>> -      */
>>> -     struct rte_intr_handle *intr_handle;
>>>        struct sub_device *sdev;
>>>        struct rxq *rxq;
>>>        uint8_t i;
>>>        int ret;
>>>
>>> -     intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
>>> -     if (intr_handle == NULL)
>>> -             return -ENOMEM;
>>> -
>>> -     if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX))
>>> -             return -rte_errno;
>>> -
>>> -     if (rte_intr_efds_index_set(intr_handle, 0, -1))
>>> -             return -rte_errno;
>>> -
>>>        fs_lock(dev, 0);
>>>        if (rx_conf->rx_deferred_start) {
>>>                FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
>>> @@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev,
>>>        rxq->info.nb_desc = nb_rx_desc;
>>>        rxq->priv = PRIV(dev);
>>>        rxq->sdev = PRIV(dev)->subs;
>>> -     ret = rte_intr_efd_enable(intr_handle, 1);
>>> -     if (ret < 0) {
>>> +#ifdef RTE_EXEC_ENV_LINUX
>>> +     rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
>>> +     if (rxq->event_fd < 0) {
>>> +             ERROR("Failed to create an eventfd: %s", strerror(errno));
>>>                fs_unlock(dev, 0);
>>> -             return ret;
>>> +             return -errno;
>>>        }
>>> -     rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0);
>>> +#else
>>> +     rxq->event_fd = -1;
>>> +#endif
>>
>> How this impacts the BSD? I don't know if driver used on BSD but
>> technically it looks supported.
>>
>> @Gaetan, any objection to the change?
> 
> There was no feedback for months, can we get this merged?
> Thanks.
> 

There is no comment from maintainer, but patch is out for a while, so 
agree to proceed:

Acked-by: Ferruh Yigit <ferruh.yigit@xilinx.com>

Applied to dpdk-next-net/main, thanks.