[PATCH] config/arm: make SoC-generic more generic

[Ruifeng Wang]

> -----Original Message-----
> From: fengchengwen <fengchengwen at huawei.com>
> Sent: Tuesday, November 15, 2022 3:55 PM
> To: Jerin Jacob <jerinjacobk at gmail.com>; Richardson, Bruce <bruce.richardson at intel.com>;
> Konstantin Ananyev <konstantin.v.ananyev at yandex.ru>
> Cc: thomas at monjalon.net; dev at dpdk.org; Ruifeng Wang <Ruifeng.Wang at arm.com>
> Subject: Re: [PATCH] config/arm: make SoC-generic more generic
> 
> Hi Jerin,
> 
> On 2022/11/15 15:08, Jerin Jacob wrote:
> > On Tue, Nov 15, 2022 at 7:46 AM Chengwen Feng <fengchengwen at huawei.com> wrote:
> >>
> >> The non-generic soc has 1280 max-lcores and 32 max-numas, but the
> >> implementer-generic still has 256 max-lcores and 4 max-numas, which
> >> may result in restrictions on the use of binaries compiled by
> >> soc-generic in these systems.
> >>
> >> This patch changes the soc-generic max-lcores to 1280 and max-numas
> >> to
> >> 32 to make it more generic.
> >
> > I think, it is not specific to ARM architecture issue. Even x86 config
> > set as below.
> > dpdk_conf.set('RTE_MAX_LCORE', 128)
> > dpdk_conf.set('RTE_MAX_NUMA_NODES', 32)
> >
> > I think, it makes sense to keep generic as "most commonly used max"
> > per application, expectation cases can have separate config.
> >
> > Also, this is for SINGLE application, even if you have 1280 cores a
> > single application most likely won't use that.
> > So in that way this patch is not needed.
> 
> If we keep it, then dpdk application could not run with lcore >= 256
> 
> I have a test for it:
> 1. set RTE_MAX_LCORE to 64
> 2. run testpmd with lcore 80-81: ./build/native/app/dpdk-testpmd -a 0000:bd:00.0 --file-
> prefix=feng -l 80-81 -- -i
>    it will display error:
>      EAL: Detected CPU lcores: 64
>      EAL: Detected NUMA nodes: 3
>      EAL: lcore 80 >= RTE_MAX_LCORE (64)
>      EAL: lcore 81 >= RTE_MAX_LCORE (64)
>      EAL: To use high physical core ids, please use --lcores to map them to lcore ids
> below RTE_MAX_LCORE, e.g. --lcores 0 at 80,1 at 81

Actually, EAL provides suggestions here. Physical cores above RTE_MAX_LCORE can be mapped and used.

>      EAL: invalid core list syntax
> 
> >
> >>
> >> Also, because single numa is a special case of multiple numas, mark
> >> soc-generic's numa flag as true.
> >>
> >> Signed-off-by: Chengwen Feng <fengchengwen at huawei.com>
> >> ---
> >>  config/arm/meson.build | 7 ++++---
> >>  1 file changed, 4 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/config/arm/meson.build b/config/arm/meson.build index
> >> 6442ec9596..97b91cf609 100644
> >> --- a/config/arm/meson.build
> >> +++ b/config/arm/meson.build
> >> @@ -32,8 +32,8 @@ implementer_generic = {
> >>      'flags': [
> >>          ['RTE_MACHINE', '"armv8a"'],
> >>          ['RTE_USE_C11_MEM_MODEL', true],
> >> -        ['RTE_MAX_LCORE', 256],
> >> -        ['RTE_MAX_NUMA_NODES', 4]
> >> +        ['RTE_MAX_LCORE', 1280],
> >> +        ['RTE_MAX_NUMA_NODES', 32]
> >>      ],
> >>      'part_number_config': {
> >>          'generic': {
> >> @@ -271,7 +271,8 @@ implementers = {
> >>  soc_generic = {
> >>      'description': 'Generic un-optimized build for armv8 aarch64 exec mode',
> >>      'implementer': 'generic',
> >> -    'part_number': 'generic'
> >> +    'part_number': 'generic',
> >> +    'numa': true
> >>  }
> >>
> >>  soc_generic_aarch32 = {
> >> --
> >> 2.17.1
> >>
> > .
> >