Thread (42 messages) flat view 42 messages, 4 authors, 2025-06-27

Re: [PATCH v2 02/17] psp: base PSP device support

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date: 2025-06-27 16:10:34

Jakub Kicinski wrote:
On Thu, 26 Jun 2025 13:20:45 -0400 Willem de Bruijn wrote:
quoted
quoted
quoted
IDPF does support multiple "vports" (num_alloc_vports), and with that
struct net_device, from a single BDF.  
Upstream? If yes then I'm very bad at reviewing code :D  
I then don't think I want to focus your attention on this, but..

See use num_alloc_ports in idpf_init_task. Which keeps requeueing
itself until num_default_vports is reached. Which is a variable
received from the device in VIRTCHNL2_OP_GET_CAPS.
That's not too bad, one of the older drivers had a sysfs interface
for creating the sub-interfaces IIRC :/

We should be able to share one psp_dev if the implementation shares PSP
between blocks.
Great.
I was trying to write the code so that it'd be possible
to attach the psp_dev to veth / netkit
IIRC the main_netdev was supposed to be special in terms of
permissions, the admin in the netns where main_netdev sits is
the admin of the device (for rotations and config).
But I was planning to add a secondary list of "attached devices"
which have access to non-privileged operations. 
That makes sense. I like how this approach enables support for such
devices without having to explicitly pipe through net_device_ops
(what legacy Google PSP implementation had to do for each of ipvlan,
bonding, etc).
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help