Re: [dpdk-dev] [RFC PATCH v3 3/3] drivers/net: use unmask API in interrupt handlers

[Nithin Kumar Dabilpuram <ndabilpuram@marvell.com>]

On Wed, Jul 17, 2019 at 06:01:47AM +0000, Hyong Youb Kim (hyonkim) wrote:
> > -----Original Message-----
> > From: Nithin Dabilpuram <ndabilpuram@marvell.com>
> > Sent: Wednesday, July 17, 2019 1:44 AM
> [...]
> > Subject: [RFC PATCH v3 3/3] drivers/net: use unmask API in interrupt
> > handlers
> > 
> > Replace rte_intr_enable() with rte_intr_unmask() API
> > for unmasking in interrupt handlers and rx_queue_intr_enable()
> > in callbacks of PMD's whose original intent was to unmask interrupts
> > after handling is completed if device is backed by UIO, IGB_UIO or
> > VFIO(with INTx).
> > 
> > Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
> > ---
> > 
> > v3:
> > * Change all PMD's that use rte_intr_enable() in
> >   rx_queue_intr_enable() or in irq handler to use
> >   new unmask api.
> > 
> >  drivers/net/atlantic/atl_ethdev.c    |  2 +-
> >  drivers/net/avp/avp_ethdev.c         |  2 +-
> >  drivers/net/axgbe/axgbe_ethdev.c     |  4 ++--
> >  drivers/net/bnx2x/bnx2x_ethdev.c     |  2 +-
> >  drivers/net/e1000/em_ethdev.c        |  4 ++--
> >  drivers/net/e1000/igb_ethdev.c       |  6 +++---
> >  drivers/net/fm10k/fm10k_ethdev.c     |  6 +++---
> >  drivers/net/i40e/i40e_ethdev.c       |  2 +-
> >  drivers/net/iavf/iavf_ethdev.c       |  2 +-
> >  drivers/net/ice/ice_ethdev.c         |  4 ++--
> >  drivers/net/ixgbe/ixgbe_ethdev.c     |  6 +++---
> >  drivers/net/nfp/nfp_net.c            |  2 +-
> >  drivers/net/qede/qede_ethdev.c       |  8 ++++----
> >  drivers/net/sfc/sfc_intr.c           |  4 ++--
> >  drivers/net/virtio/virtio_ethdev.c   | 16 +++++++++++++++-
> >  drivers/net/vmxnet3/vmxnet3_ethdev.c |  2 +-
> >  16 files changed, 43 insertions(+), 29 deletions(-)
> > 
> [...]
> > diff --git a/drivers/net/nfp/nfp_net.c b/drivers/net/nfp/nfp_net.c
> > index 1a7aa17..5484d31 100644
> > --- a/drivers/net/nfp/nfp_net.c
> > +++ b/drivers/net/nfp/nfp_net.c
> > @@ -1412,7 +1412,7 @@ nfp_net_irq_unmask(struct rte_eth_dev *dev)
> >  	if (hw->ctrl & NFP_NET_CFG_CTRL_MSIXAUTO) {
> >  		/* If MSI-X auto-masking is used, clear the entry */
> >  		rte_wmb();
> > -		rte_intr_enable(&pci_dev->intr_handle);
> > +		rte_intr_unmask(&pci_dev->intr_handle);
> 
> I mentioned this part earlier. The comment suggests the driver really
> needs to clear Mask bit in MSI-X Table, if MSI-X is used?
> rte_intr_unmask() + vfio-pci does not do that. The maintainers should
> say something..
> 

Ok, if this comment meant that rte_intr_enable() is a must even with
VFIO-MSIX, then I can leave it as it is with rte_intr_unmask().
If it for MSIX interrupts via IGB_UIO, then rte_intr_unmask() will work.


> [...]
> > diff --git a/drivers/net/qede/qede_ethdev.c
> > b/drivers/net/qede/qede_ethdev.c
> > index 82363e6..6648139 100644
> > --- a/drivers/net/qede/qede_ethdev.c
> > +++ b/drivers/net/qede/qede_ethdev.c
> > @@ -248,8 +248,8 @@ qede_interrupt_handler_intx(void *param)
> >  	if (status & 0x1) {
> >  		qede_interrupt_action(ECORE_LEADING_HWFN(edev));
> > 
> > -		if (rte_intr_enable(eth_dev->intr_handle))
> > -			DP_ERR(edev, "rte_intr_enable failed\n");
> > +		if (rte_intr_unmask(eth_dev->intr_handle))
> > +			DP_ERR(edev, "rte_intr_unmask failed\n");
> >  	}
> >  }
> > 
> > @@ -261,8 +261,8 @@ qede_interrupt_handler(void *param)
> >  	struct ecore_dev *edev = &qdev->edev;
> > 
> >  	qede_interrupt_action(ECORE_LEADING_HWFN(edev));
> > -	if (rte_intr_enable(eth_dev->intr_handle))
> > -		DP_ERR(edev, "rte_intr_enable failed\n");
> > +	if (rte_intr_unmask(eth_dev->intr_handle))
> > +		DP_ERR(edev, "rte_intr_unmask failed\n");
> >  }
> 
> I thought you were going to remove rte_intr_enable from this handler
> altogether.
> 

Here again, if it was done for case of IGB_UIO with MSIx, then
I think it is needed. If maintainer confirms that it is not needed in this
i.e qede_interrupt_handler(), then I can remove this unmask.


> [...]
> 
> Thanks.
> -Hyong
>