Re: [PATCH] common/cnxk: allow enabling IOVA field in mbuf

[Bruce Richardson <bruce.richardson@intel.com>]

On Thu, Oct 24, 2024 at 04:40:40PM +0530, Jerin Jacob wrote:
> On Mon, Oct 14, 2024 at 4:37 PM Shijith Thotton <sthotton@marvell.com> wrote:
> >
> > Value of RTE_IOVA_IN_MBUF was always disabled on cnxk platforms, as IOVA
> > in the mbuf is not required. This change modifies that behavior,
> > allowing RTE_IOVA_IN_MBUF to be enabled if the build option
> > -Denable_iova_as_pa=true is explicitly specified.
> >
> > Signed-off-by: Shijith Thotton <sthotton@marvell.com>
> > ---
> >
> > diff --git a/config/arm/meson.build b/config/arm/meson.build
> > index 012935d5d7..ca54524376 100644
> > --- a/config/arm/meson.build
> > +++ b/config/arm/meson.build
> > @@ -439,10 +439,7 @@ soc_cn9k = {
> >      'description': 'Marvell OCTEON 9',
> >      'implementer': '0x43',
> >      'part_number': '0xb2',
> > -    'numa': false,
> > -    'flags': [
> > -        ['RTE_IOVA_IN_MBUF', 0]
> > -    ]
> > +    'numa': false
> >  }
> >
> >  soc_cn10k = {
> > @@ -451,8 +448,7 @@ soc_cn10k = {
> >      'flags': [
> >          ['RTE_MAX_LCORE', 24],
> >          ['RTE_MAX_NUMA_NODES', 1],
> > -        ['RTE_MEMPOOL_ALIGN', 128],
> > -        ['RTE_IOVA_IN_MBUF', 0]
> > +        ['RTE_MEMPOOL_ALIGN', 128]
> >      ],
> >      'part_number': '0xd49',
> >      'extra_march_features': ['crypto'],
> > diff --git a/drivers/common/cnxk/meson.build b/drivers/common/cnxk/meson.build
> > index dc2ddf1f20..bba780e750 100644
> > --- a/drivers/common/cnxk/meson.build
> > +++ b/drivers/common/cnxk/meson.build
> > @@ -108,4 +108,13 @@ deps += ['bus_pci', 'net', 'telemetry']
> >
> >  require_iova_in_mbuf = false
> >
> > +cnxk_socs = ['cn9k', 'cn10k', 'cn20k']
> > +
> > +# Enable RTE_IOVA_IN_MBUF only if enable_iova_as_pa is set explicitly, else disable it
> > +if meson.version().version_compare('>=1.1.0')
> > +    if '-Denable_iova_as_pa' not in meson.build_options() and soc_type in cnxk_socs
> > +        dpdk_conf.set10('RTE_IOVA_IN_MBUF', false)
> > +    endif
> > +endif
> 
> Since this is added in driver/common/cnxk, it will be late to decide.
> For example,
> 
> Following PMDs will have mis match:
> common - cpt, dpaax, idpf, ionic
> bus - cdx, dpaa, fslmc, ifpga, uacce
> 
> I think, this check needs to move up in the chain. @Richardson, Bruce
> Any thoughts on cleanly adding this kind of check in top-level meson
> objects?
> 
Can you explain what you mean by a mismatch?

Can I assume that the common/cnxk is processed before any of these other
drivers? If so, then whatever values or variables set by that meson.build
file can be queried by all the others.

/Bruce