[dpdk-dev] [PATCH 2/7] net/mlx4: fix Rx interrupts with multiple ports

[Adrien Mazarguil]

On Fri, Jun 16, 2017 at 02:07:54PM +0100, Ferruh Yigit wrote:
> On 6/14/2017 12:49 PM, Adrien Mazarguil wrote:
> > Several Ethernet device structures are allocated on top of a common PCI
> > device for mlx4 adapters with multiple ports. These inherit a common
> > interrupt handle from their parent PCI device, which prevents Rx interrupts
> > from working properly on all ports as their configuration is overwritten.
> > 
> > Use a local interrupt handle to address this issue.
> 
> Hi Adrien,
> 
> I am not clear why local copy required, and main concern from my point
> of view is if this is a common problem for all PMDs and should be
> addressed in higher level?

This issue only affects PMDs that handle multiple Ethernet ports through a
single PCI device. Such PMDs (like mlx4) identify themselves as PCI drivers
that manually have to register multiple rte_eth_dev instances through
rte_eth_dev_allocate(), which they then have to initialize.

> The variable is already per eth_dev, but this patch moves it the private
> data. What overwrites it within eth_dev?

Calling rte_eth_copy_pci_info() makes the rte_eth_dev structure inherit the
default interrupt handle of the underlying PCI device. By "inherit", I mean
eth_dev->intr_handle points to it, in that sense it's not per eth_dev but
per PCI device.

mlx4 Rx interrupts are associated with a given Verbs context, and each port
has its own Verbs context, so they cannot be shared, while other PMDs using
other methods for catching interrupts may be perfectly fine with a single
vector associated with the PCI device. It depends on the PMD, for instance
there is no such problem with mlx5 as exactly one PCI device is associated
with a given port.

This patch merely allocates a specific interrupt handle associated with the
eth_dev itself and makes the eth_dev handle point to that instead of the
default PCI handle. This "local" handle is initialized using the PCI handle
as a template before modifying the pointer. It's completely safe.

> > Fixes: 9f05a4b81809 ("net/mlx4: support user space Rx interrupt event")
> > 
> > Signed-off-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com>
> > Acked-by: Moti Haimovsky <motih at mellanox.com>
> > ---
> >  drivers/net/mlx4/mlx4.c | 9 +++++++++
> >  drivers/net/mlx4/mlx4.h | 1 +
> >  2 files changed, 10 insertions(+)
> > 
> > diff --git a/drivers/net/mlx4/mlx4.c b/drivers/net/mlx4/mlx4.c
> > index 178562e..2b4722f 100644
> > --- a/drivers/net/mlx4/mlx4.c
> > +++ b/drivers/net/mlx4/mlx4.c
> > @@ -6207,6 +6207,15 @@ mlx4_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
> >  
> >  		eth_dev->device->driver = &mlx4_driver.driver;
> >  
> > +		/*
> > +		 * Copy and override interrupt handle to prevent it from
> > +		 * being shared between all ethdev instances of a given PCI
> > +		 * device. This is required to properly handle Rx interrupts
> > +		 * on all ports.
> > +		 */
> > +		priv->intr_handle_dev = *eth_dev->intr_handle;
> > +		eth_dev->intr_handle = &priv->intr_handle_dev;
> > +
> >  		priv->dev = eth_dev;
> >  		eth_dev->dev_ops = &mlx4_dev_ops;
> >  
> > diff --git a/drivers/net/mlx4/mlx4.h b/drivers/net/mlx4/mlx4.h
> > index c46fc23..b74fbf8 100644
> > --- a/drivers/net/mlx4/mlx4.h
> > +++ b/drivers/net/mlx4/mlx4.h
> > @@ -345,6 +345,7 @@ struct priv {
> >  	unsigned int txqs_n; /* TX queues array size. */
> >  	struct rxq *(*rxqs)[]; /* RX queues. */
> >  	struct txq *(*txqs)[]; /* TX queues. */
> > +	struct rte_intr_handle intr_handle_dev; /* Device interrupt handler. */
> >  	struct rte_intr_handle intr_handle; /* Interrupt handler. */
> >  	struct rte_flow_drop *flow_drop_queue; /* Flow drop queue. */
> >  	LIST_HEAD(mlx4_flows, rte_flow) flows;
> > 
> 

-- 
Adrien Mazarguil
6WIND