Re: [dpdk-dev] [PATCH] eal: forcing IOVA as PA in ppc

[David Marchand <david.marchand@redhat.com>]

On Fri, Aug 2, 2019 at 10:09 AM Takeshi T Yoshimura <TYOS@jp.ibm.com> wrote:
>
> -----David Marchand <david.marchand@redhat.com> wrote: -----
>
> >To: Takeshi T Yoshimura <TYOS@jp.ibm.com>
> >From: David Marchand <david.marchand@redhat.com>
> >Date: 08/01/2019 06:47PM
> >Cc: dev <dev@dpdk.org>, David Christensen <drc@linux.vnet.ibm.com>
> >Subject: [EXTERNAL] Re: [dpdk-dev] [PATCH] eal: forcing IOVA as PA in
> >ppc
> >
> >On Thu, Aug 1, 2019 at 10:46 AM Takeshi T Yoshimura <TYOS@jp.ibm.com>
> >wrote:
> >>
> >> > - Are the (below) devices bound to vfio-pci or any other kmod at
> >this point of the init?
> >>
> >> Yes, vfio-pci is bound to 0030:01:00.0 in the log.
> >>
> >> > - Is the spdk_nvme driver loaded at this point?
> >>
> >> No, spdk just attempts to initialize dpdk at this point.
> >> After the initialization finishes, spdk probes the nvme driver.
> >
> >Do you mean that spdk registers its drivers after rte_eal_init() has
> >returned?
> >
> >Nothing should prevent from registering the drivers them before
> >calling rte_eal_init(), did you try this?
> >If the drivers are in the form of shared libraries (.so files), did
> >you try to load them using the -d eal option?
> >
> >
> >--
> >David Marchand
> >
> >
>
> SPDK apps have to call rte_eal_init() before registering drivers,
> since the driver allocates DMA memory.

Registering a driver before calling rte_eal_init is something that
works most of the time.
Not sure I get your point here.


> Do I have to change existing apps and spdk core to workaround this issue to use new versions of dpdk?

I understand this is a regression and I am looking at it.
If we are too short for 19.08, then we can go with your workaround,
but it needs a better commitlog.


-- 
David Marchand