Re: baseband PMD APIs

Re: baseband PMD APIs

[Thomas Monjalon]

To bbdev maintainers,

Do you think we can avoid having some PMD-specific configuration?
It looks blocking migration from a driver to another.
How do we manage that?


13/10/2022 11:07, Akhil Goyal:
> There are 6 bbdev PMDs, out of which 3 have pmd APIs - all from Intel
> These are the PMD APIs for baseband drivers.
> rte_acc_configure -> this is being used by 2 drivers(acc100 and acc200) merged in a single folder.
> rte_fpga_5gnr_fec_configure
> rte_fpga_lte_fec_configure
> 
> The fpga ones have almost same syntax.

RE: baseband PMD APIs

[Chautru, Nicolas]

Hi, 
There are a few options:
1) Keep as is, but not too scalable to keep on adding PMD APIs. Still there is no plan on adding more API, so no rush. 
2) Keep API internal only (but in that case only work with static lib I believe)
3) Take out off DPDK, but that is used by users notably OEMs and for test purpose (bbdev-test) and hence proven valuable to have everything in one place. 
4) Include under bbdev API, my concern is that this is not really meant to be used by something else than bbdev-test hence can be confusing to VRAN vendor by polluting the API. 

For the sake of discussion I have captured what 4) may look like here: https://patches.dpdk.org/project/dpdk/patch/20221014014205.38074-2-nicolas.chautru@intel.com/
Note that the actual configuration is just a void* since that configuration structure is device specific (we cannot standardize this really, nor would there be any value to do this here). 

Note that this is definitely not blocking migration from one driver to another Thomas. That companion function is purely to help bbdev-test.

Happy to discuss more, I don't believe we need a swift decision though. 

Thanks
Nic

> -----Original Message-----
> From: Thomas Monjalon <thomas@monjalon.net>
> Sent: Thursday, October 13, 2022 2:25 AM
> To: Chautru, Nicolas <nicolas.chautru@intel.com>; Gagandeep Singh
> <g.singh@nxp.com>; Hemant Agrawal <hemant.agrawal@nxp.com>
> Cc: Akhil Goyal <gakhil@marvell.com>; david.marchand@redhat.com;
> maxime.coquelin@redhat.com; Tom Rix <trix@redhat.com>; dev@dpdk.org
> Subject: Re: baseband PMD APIs
> 
> To bbdev maintainers,
> 
> Do you think we can avoid having some PMD-specific configuration?
> It looks blocking migration from a driver to another.
> How do we manage that?
> 
> 
> 13/10/2022 11:07, Akhil Goyal:
> > There are 6 bbdev PMDs, out of which 3 have pmd APIs - all from Intel
> > These are the PMD APIs for baseband drivers.
> > rte_acc_configure -> this is being used by 2 drivers(acc100 and acc200)
> merged in a single folder.
> > rte_fpga_5gnr_fec_configure
> > rte_fpga_lte_fec_configure
> >
> > The fpga ones have almost same syntax.
> 
> 
> 

Re: baseband PMD APIs

[Thomas Monjalon]

14/10/2022 03:46, Chautru, Nicolas:
> Hi, 
> There are a few options:
> 1) Keep as is, but not too scalable to keep on adding PMD APIs. Still there is no plan on adding more API, so no rush. 
> 2) Keep API internal only (but in that case only work with static lib I believe)
> 3) Take out off DPDK, but that is used by users notably OEMs and for test purpose (bbdev-test) and hence proven valuable to have everything in one place. 
> 4) Include under bbdev API, my concern is that this is not really meant to be used by something else than bbdev-test hence can be confusing to VRAN vendor by polluting the API. 
> 
> For the sake of discussion I have captured what 4) may look like here: https://patches.dpdk.org/project/dpdk/patch/20221014014205.38074-2-nicolas.chautru@intel.com/
> Note that the actual configuration is just a void* since that configuration structure is device specific (we cannot standardize this really, nor would there be any value to do this here). 
> 
> Note that this is definitely not blocking migration from one driver to another Thomas. That companion function is purely to help bbdev-test.

If it is only for testing, it is not an issue.
I just want to avoid having different API paths for major API functions.


> Happy to discuss more, I don't believe we need a swift decision though. 
> 
> Thanks
> Nic
> 
> > -----Original Message-----
> > From: Thomas Monjalon <thomas@monjalon.net>
> > Sent: Thursday, October 13, 2022 2:25 AM
> > To: Chautru, Nicolas <nicolas.chautru@intel.com>; Gagandeep Singh
> > <g.singh@nxp.com>; Hemant Agrawal <hemant.agrawal@nxp.com>
> > Cc: Akhil Goyal <gakhil@marvell.com>; david.marchand@redhat.com;
> > maxime.coquelin@redhat.com; Tom Rix <trix@redhat.com>; dev@dpdk.org
> > Subject: Re: baseband PMD APIs
> > 
> > To bbdev maintainers,
> > 
> > Do you think we can avoid having some PMD-specific configuration?
> > It looks blocking migration from a driver to another.
> > How do we manage that?
> > 
> > 
> > 13/10/2022 11:07, Akhil Goyal:
> > > There are 6 bbdev PMDs, out of which 3 have pmd APIs - all from Intel
> > > These are the PMD APIs for baseband drivers.
> > > rte_acc_configure -> this is being used by 2 drivers(acc100 and acc200)
> > merged in a single folder.
> > > rte_fpga_5gnr_fec_configure
> > > rte_fpga_lte_fec_configure
> > >
> > > The fpga ones have almost same syntax.
> > 
> > 
> > 
> 
>