[PATCH net-next v18 07/14] memory-provider: dmabuf devmem memory provider

Jakub Kicinski kuba at kernel.org
Tue Aug 13 15:26:15 UTC 2024


On Tue, 13 Aug 2024 16:11:15 +0100 Pavel Begunkov wrote:
> On 8/13/24 15:39, Jakub Kicinski wrote:
> > On Tue, 13 Aug 2024 03:31:13 +0100 Pavel Begunkov wrote:  
> >> I'm getting lost, so repeating myself a bit. What I think
> >> would be a good approach is if we get an error back from
> >> the driver if it doesn't support netiov / providers.
> >>
> >> netdev_rx_queue_restart() {
> >> 	...
> >> 	err = dev->queue_mgmt_ops->ndo_queue_mem_alloc();
> >> 	if (err == -EOPNOTSUPP) // the driver doesn't support netiov
> >> 		return -EOPNOTSUPP;
> >> 	...
> >> }
> >>
> >> That can be done if drivers opt in to support providers,
> >> e.g. via a page pool flag.
> >>
> >> What I think wouldn't be a great option is getting back a
> >> "success" from the driver even though it ignored  
> > 
> > page pool params are not the right place for a supported flag.
> > Sooner or later we'll want to expose this flag to user space.  
> 
> Fair enough, it appeared to me that's what you was suggesting
> 
> "What comes to mind is adding an "I can gobble up net_iovs from this
> pool" flag in page pool params (the struct that comes from the driver),
> and then on the installation path we can check ..."

Yes, we still need one flag in page pool params -- functioning like 
the inverse of PP_IGNORE_PROVIDERS, I'd call it something like
PP_CAP_NET(MEM|IOV). To distinguish the header and data pools.

> We can also move it from pp flags to queue API callbacks, however if we
> want to expose it to the userspace, I'd imagine we need a queue flag set
> by the driver, which then can be queried by netlink or whichever
> interface is appropriate. And it can be used can be used to fail
> netdev_rx_queue_restart() for queues/drivers that don't support mp.
> 
> netdev_rx_queue_restart() {
> 	if (rxq->mp_params && !rxq->netiov_supported)
> 		fail;
> }

Yup!


More information about the dri-devel mailing list