[dpdk-dev] [PATCH] net/ixgbe: fix VFIO interrupt mapping in PF

[dpdk-dev] [PATCH] net/ixgbe: fix VFIO interrupt mapping in PF

[Wei Dai]

When a PF port is bound to VFIO-PIC, only miscellaneous interrupt
is mapped to VFIO vector 0 in ixgbe_dev_init( ).
In ixgbe_dev_start(), if previous VFIO interrupt mapping set in
ixgbe_dev_init( ) is not cleard, it will fail when calling
rte_intr_enable( ) tries to map Rx queue interrupt to other VFIO
vectors. This patch clears the VFIO interrupt mappings before
setting both miscellaneous and Rx queue interrupt mappings again
to avoid failure.

Fixes: 0a45657a6794 ("pci: rework interrupt handling")
Cc: stable@dpdk.org

Signed-off-by: Wei Dai <wei.dai@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 9ca5cbc..9030d3d 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -2692,6 +2692,15 @@ ixgbe_dev_start(struct rte_eth_dev *dev)
 				     " no intr multiplex");
 	}
 
+	/* When a PF port is bound to VFIO-PCI only miscellaneous interrupt
+	*  is mapped to VFIO vector 0 in ixgbe_dev_init( ).
+	*  If previous VFIO interrupt mapping set in ixgbe_dev_init( ) is
+	*  not cleared, it will fail when following rte_intr_enable( ) tries
+	*  to map Rx queue interrupt to other VFIO vectors.
+	*  So clear uio/vfio intr/evevnfd first to avoid failure.
+	*/
+	rte_intr_disable(intr_handle);
+
 	/* check if rxq interrupt is enabled */
 	if (dev->data->dev_conf.intr_conf.rxq != 0 &&
 	    rte_intr_dp_is_en(intr_handle))
-- 
2.7.5

[dpdk-dev] [PATCH] net/ixgbe: fix VFIO interrupt mapping in PF

[Wei Dai]

When a PF port is bound to VFIO-PIC, only miscellaneous interrupt
is mapped to VFIO vector 0 in ixgbe_dev_init( ).
In ixgbe_dev_start(), if previous VFIO interrupt mapping set in
ixgbe_dev_init( ) is not cleard, it will fail when calling
rte_intr_enable( ) tries to map Rx queue interrupt to other VFIO
vectors. This patch clears the VFIO interrupt mappings before
setting both miscellaneous and Rx queue interrupt mappings again
to avoid failure.

Fixes: 0a45657a6794 ("pci: rework interrupt handling")
Cc: stable@dpdk.org

Signed-off-by: Wei Dai <wei.dai@intel.com>
---
 drivers/net/ixgbe/ixgbe_ethdev.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
index 9ca5cbc..5292694 100644
--- a/drivers/net/ixgbe/ixgbe_ethdev.c
+++ b/drivers/net/ixgbe/ixgbe_ethdev.c
@@ -2692,6 +2692,15 @@ ixgbe_dev_start(struct rte_eth_dev *dev)
 				     " no intr multiplex");
 	}
 
+	/* When a PF port is bound to VFIO-PCI only miscellaneous interrupt
+	 * is mapped to VFIO vector 0 in ixgbe_dev_init( ).
+	 * If previous VFIO interrupt mapping set in ixgbe_dev_init( ) is
+	 * not cleared, it will fail when following rte_intr_enable( ) tries
+	 * to map Rx queue interrupt to other VFIO vectors.
+	 * So clear uio/vfio intr/evevnfd first to avoid failure.
+	 */
+	rte_intr_disable(intr_handle);
+
 	/* check if rxq interrupt is enabled */
 	if (dev->data->dev_conf.intr_conf.rxq != 0 &&
 	    rte_intr_dp_is_en(intr_handle))
-- 
2.7.5

Re: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF

[Wenzhuo Lu]

Hi Wei,

> +	rte_intr_disable(intr_handle);
TBH, confused by this patch. I see the intr already disabled at the begining
of this function, ixgbe_dev_start, why we need to disable it again?

Re: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF

[Yang, Qiming]

+1 why we disable interrupt in dev_start twice?

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wenzhuo Lu
> Sent: Tuesday, January 9, 2018 4:24 PM
> To: Dai, Wei <wei.dai@intel.com>
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF
> 
> Hi Wei,
> 
> > +	rte_intr_disable(intr_handle);
> TBH, confused by this patch. I see the intr already disabled at the begining of this
> function, ixgbe_dev_start, why we need to disable it again?

Re: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF

[Dai, Wei]

I add this code line to align same style and same action taken in ixgbe VF and i40e VF.
In ixgbe VF or i40e VF,  without this line, the initialization of Rx queue interrupt mapping to VFIO-PCI vectors would fail.
I have tested it in ixgbe PF port with example/l3fwd-power and 
found the interrupt can be initialized successfully and Rx queue can be waked up by incoming packets
no matter whether this line exists or not.
So I decide to reject it now.


> -----Original Message-----
> From: Yang, Qiming
> Sent: Wednesday, January 10, 2018 2:29 PM
> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Dai, Wei <wei.dai@intel.com>
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF
> 
> +1 why we disable interrupt in dev_start twice?
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Wenzhuo Lu
> > Sent: Tuesday, January 9, 2018 4:24 PM
> > To: Dai, Wei <wei.dai@intel.com>
> > Cc: dev@dpdk.org
> > Subject: Re: [dpdk-dev] net/ixgbe: fix VFIO interrupt mapping in PF
> >
> > Hi Wei,
> >
> > > +	rte_intr_disable(intr_handle);
> > TBH, confused by this patch. I see the intr already disabled at the
> > begining of this function, ixgbe_dev_start, why we need to disable it
> again?