> -----Original Message----- > From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Zhang, Qi Z > Sent: Tuesday, March 21, 2017 7:34 PM > To: Thomas Monjalon <thomas.monja...@6wind.com> > Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yi...@intel.com> > Subject: Re: [dpdk-dev] [PATCH v2] net/i40e: fix wrong handle when enable > interrupt > > Hi Thomas: > > > > -----Original Message----- > > From: Zhang, Qi Z > > Sent: Tuesday, March 21, 2017 8:10 PM > > To: Wu, Jingjing <jingjing...@intel.com>; Zhang, Helin > > <helin.zh...@intel.com> > > Cc: dev@dpdk.org; Zhang, Qi Z <qi.z.zh...@intel.com>; sta...@dpdk.org > > Subject: [PATCH v2] net/i40e: fix wrong handle when enable interrupt > > > > In i40e_dev_interrupt_handler, when call rte_intr_enable, the > > intr_handle is the copy when we registered. > > According to interrupt handle framework, if the requirement of > > intr_handle is changed, we need to unregister then register a new > > copy. This happens on i40e driver when bind to vfio-pci, the > > rte_intr_efd_enable function will modify the max_intr according the > > queue number, so a new copy of intr_handle need to be registered. > > Without this fix, we saw lw3fwd-power does not work due to wrong > > interrupt count in vfio_irq_set when set vfio interrupt. > > > > Fixes: 2ce7a1ed09fc ("net/i40e: localize mapping of ethdev to PCI > > device") > > > > Cc: sta...@dpdk.org > > > > Signed-off-by: Qi Zhang <qi.z.zh...@intel.com> > > --- > > v2: > > - follow current design, when intr_handle is modified, unregister > > the old one and register the new one. > > - there should be a patch set to fix on other devices. > > > > drivers/net/i40e/i40e_ethdev.c | 8 ++++++++ > > 1 file changed, 8 insertions(+) > > > > diff --git a/drivers/net/i40e/i40e_ethdev.c > > b/drivers/net/i40e/i40e_ethdev.c index 9c76baa..e7bbea5 100644 > > --- a/drivers/net/i40e/i40e_ethdev.c > > +++ b/drivers/net/i40e/i40e_ethdev.c > > @@ -1886,6 +1886,14 @@ i40e_dev_start(struct rte_eth_dev *dev) > > ret = rte_intr_efd_enable(intr_handle, intr_vector); > > if (ret) > > return ret; > > + /** > > + * intr_handle may be modified in rte_intr_efd_enable > > + * so unregster the old one and register the new one. > > + */ > > + rte_intr_callback_unregister(intr_handle, > > + i40e_dev_interrupt_handler, dev); > > + rte_intr_callback_register(intr_handle, > > + i40e_dev_interrupt_handler, dev); > To me, this fix looks a little bit weird. > May I know why we need to register an copy of intr_handle ? > What do you think about my previous clean up patch. > http://dpdk.org/dev/patchwork/patch/21529/ > Thanks > Qi.
It seems [dpdk-dev,v2,1/3] vfio: keep interrupt source read only http://www.dpdk.org/dev/patchwork/patch/21528/ is already be acked by Anatoly. I think we need to support that change, but not this workaround.
