Re: [PATCH v2] drivers/bus: set device NUMA node to unknown by default

[Thomas Monjalon <thomas@monjalon.net>]

05/10/2022 11:04, Olivier Matz:
> Hi David,
> 
> On Wed, Oct 05, 2022 at 10:52:49AM +0200, David Marchand wrote:
> > On Tue, Oct 4, 2022 at 4:59 PM Olivier Matz <olivier.matz@6wind.com> wrote:
> > >
> > > The dev->device.numa_node field is set by each bus driver for
> > > every device it manages to indicate on which NUMA node this device lies.
> > >
> > > When this information is unknown, the assigned value is not consistent
> > > across the bus drivers.
> > >
> > > Set the default value to SOCKET_ID_ANY (-1) by all bus drivers
> > > when the NUMA information is unavailable. This change impacts
> > > rte_eth_dev_socket_id() in the same manner.
> > >
> > > Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
> > > ---
> > >
> > > v2
> > > * use SOCKET_ID_ANY instead of -1 in drivers/dma/idxd (David)
> > > * document the behavior change of rte_eth_dev_socket_id()
> > > * fix few examples where rte_eth_dev_socket_id() was expected to
> > >   return 0 on unknown socket
> > 
> > Cc: ethdev maintainers.
> > 
> > [snip]
> > 
> > > diff --git a/doc/guides/rel_notes/release_22_11.rst b/doc/guides/rel_notes/release_22_11.rst
> > > index 53fe21453c..d52f823694 100644
> > > --- a/doc/guides/rel_notes/release_22_11.rst
> > > +++ b/doc/guides/rel_notes/release_22_11.rst
> > > @@ -317,6 +317,12 @@ ABI Changes
> > >  * eventdev: Added ``weight`` and ``affinity`` fields
> > >    to ``rte_event_queue_conf`` structure.
> > >
> > > +* bus: Changed the device numa node to -1 when NUMA information is unavailable.
> > > +  The ``dev->device.numa_node`` field is set by each bus driver for
> > > +  every device it manages to indicate on which NUMA node this device lies.
> > > +  When this information is unknown, the assigned value was not consistent
> > > +  across the bus drivers. This similarly impacts ``rte_eth_dev_socket_id()``.
> > > +
> > >
> > >  Known Issues
> > >  ------------
> > 
> > [snip]
> > 
> > > diff --git a/lib/ethdev/rte_ethdev.h b/lib/ethdev/rte_ethdev.h
> > > index a21f58b9cd..dd8d25d6d4 100644
> > > --- a/lib/ethdev/rte_ethdev.h
> > > +++ b/lib/ethdev/rte_ethdev.h
> > > @@ -2445,8 +2445,8 @@ int rte_eth_hairpin_unbind(uint16_t tx_port, uint16_t rx_port);
> > >   *   The port identifier of the Ethernet device
> > >   * @return
> > >   *   The NUMA socket ID to which the Ethernet device is connected or
> > > - *   a default of zero if the socket could not be determined.
> > > - *   -1 is returned is the port_id value is out of range.
> > > + *   a default of -1 (SOCKET_ID_ANY) if the socket could not be determined.
> > > + *   -1 is also returned if the port_id is invalid.
> > >   */
> > >  int rte_eth_dev_socket_id(uint16_t port_id);
> > 
> > It would be better to distinguish the two cases, using rte_errno.
> > Something like:
> > 
> > diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c
> > index 2821770e2d..1baf302804 100644
> > --- a/lib/ethdev/rte_ethdev.c
> > +++ b/lib/ethdev/rte_ethdev.c
> > @@ -562,8 +562,16 @@ rte_eth_dev_owner_get(const uint16_t port_id,
> > struct rte_eth_dev_owner *owner)
> >  int
> >  rte_eth_dev_socket_id(uint16_t port_id)
> >  {
> > -       RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
> > -       return rte_eth_devices[port_id].data->numa_node;
> > +       int socket_id = SOCKET_ID_ANY;
> > +
> > +       if (!rte_eth_dev_is_valid_port(port_id))
> > +               rte_errno = EINVAL;
> > +       } else {
> > +               socket_id = rte_eth_devices[port_id].data->numa_node;
> > +               if (socket_id == SOCKET_ID_ANY)
> > +                       rte_errno = 0;
> > +       }
> > +       return socket_id;
> >  }
> > 
> >  void *
> > diff --git a/lib/ethdev/rte_ethdev.h b/lib/ethdev/rte_ethdev.h
> > index dd8d25d6d4..03456b2dbb 100644
> > --- a/lib/ethdev/rte_ethdev.h
> > +++ b/lib/ethdev/rte_ethdev.h
> > @@ -2444,9 +2444,11 @@ int rte_eth_hairpin_unbind(uint16_t tx_port,
> > uint16_t rx_port);
> >   * @param port_id
> >   *   The port identifier of the Ethernet device
> >   * @return
> > - *   The NUMA socket ID to which the Ethernet device is connected or
> > - *   a default of -1 (SOCKET_ID_ANY) if the socket could not be determined.
> > - *   -1 is also returned if the port_id is invalid.
> > + *   - The NUMA socket ID which the Ethernet device is connected to.
> > + *   - -1 (which translates to SOCKET_ID_ANY) if the socket could not be
> > + *     determined. rte_errno is then set to:
> > + *     - EINVAL is the port_id is invalid,
> > + *     - 0 is the socket could not be determined,
> >   */
> >  int rte_eth_dev_socket_id(uint16_t port_id);
> > 
> > WDYT?
> 
> As discussed off-list, it is indeed better.

+1

Note that the decision to take in case of EINVAL
requires some work for each call to rte_eth_dev_socket_id().
I suppose we can leave it as a future exercise.