[dpdk-stable] [EXT] [PATCH] net/qede: remove interrupt reconfigure in handler

[David Marchand]

On Mon, Jul 1, 2019 at 11:15 AM Shahed Shaikh <shshaikh at marvell.com> wrote:

>
>
> >From: David Marchand <david.marchand at redhat.com>
> >Sent: Wednesday, June 26, 2019 1:07 PM
> >To: Rasesh Mody <rmody at marvell.com>
> >Cc: dev at dpdk.org; stable at dpdk.org; Shahed Shaikh <shshaikh at marvell.com>
> >Subject: Re: [EXT] [PATCH] net/qede: remove interrupt reconfigure in
> handler
>
> >On Wed, Jun 26, 2019 at 12:50 AM Rasesh Mody <mailto:rmody at marvell.com>
> wrote:
> >>From: David Marchand <mailto:david.marchand at redhat.com>
> >>Sent: Tuesday, June 25, 2019 6:39 AM
> >>
> >>----------------------------------------------------------------------
> >>rte_intr_enable/rte_intr_disable configure the interrupt context on the
> >>kernel side (either uio or vfio).
> >>In VFIO case, calling it from the interrupt handlers triggers an unneeded
> >>interrupt handlers reconfiguration.
> >>During this reconfiguration window, the device can trigger interrupts
> which
> >>are left unserviced.
> >>
> >>Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=1654824
> >>Fixes: 245aec289338 ("net/qede: fix legacy interrupt mode")
> >>Fixes: 2ea6f76aff40 ("qede: add core driver")
> >>Cc: mailto:stable at dpdk.org
> >>
> >>Signed-off-by: David Marchand <mailto:david.marchand at redhat.com>
> >>---
> >Change looks good, thanks.
> >Acked-by: Rasesh Mody <mailto:rmody at marvell.com>
>
> >Something still bothers me about the meaning of rte_intr_enable()...
> >Let me write a mail to a little more people :-)
>
> >For now, let's put this patch on hold.
>
> Another question I have is, is it required to re-enable interrupt by PMD
> at the end of interrupt handling by calling rte_intr_enable()?
> Does DPDK core / vfio/uio module take care of re-enabling the interrupt
> after the interrupt is handled?
>

This is exactly why I wanted to put this on hold.
This patch is just wrong, auto NAK for me :-).

I am currently reading the VFIO api and how UIO behaves to try and come
with a common fix on the DPDK infrastructure side.


-- 
David Marchand