[PATCH net-next v22 03/13] netdev: support binding dma-buf to netdevice

[Jakub Kicinski]

On Sun, 25 Aug 2024 04:15:01 +0000 Mina Almasry wrote:
> +u32 dev_get_min_mp_channel_count(const struct net_device *dev)
> +{
> +	u32 i, max = 0;
> +
> +	ASSERT_RTNL();
> +
> +	for (i = 0; i < dev->real_num_rx_queues; i++)
> +		if (dev->_rx[i].mp_params.mp_priv)
> +			/* The channel count is the idx plus 1. */
> +			max = i + 1;

invert the loop so you're walking from highest indexes and you can

			return i + 1;
	return 0;

> +	return max;
> +}
> +
>  /**
>   * dev_index_reserve() - allocate an ifindex in a namespace
>   * @net: the applicable net namespace

> diff --git a/net/core/devmem.c b/net/core/devmem.c

> +#include <linux/types.h>
> +#include <linux/mm.h>
> +#include <linux/netdevice.h>
> +#include <trace/events/page_pool.h>
> +#include <net/netdev_rx_queue.h>
> +#include <net/page_pool/types.h>
> +#include <net/page_pool/helpers.h>
> +#include <linux/genalloc.h>
> +#include <linux/dma-buf.h>
> +#include <net/devmem.h>
> +#include <net/netdev_queues.h>

Please sort include files alphabetically.

> +#if defined(CONFIG_DMA_SHARED_BUFFER) && defined(CONFIG_GENERIC_ALLOCATOR)

Could you create a hidden Kconfig for this feature and use it to make
building this entire file conditional? Hidden Kconfig has no
description and no help, like config NET_DEVLINK, but it can have
dependencies.

> +void net_devmem_unbind_dmabuf(struct net_devmem_dmabuf_binding *binding)
> +{
> +	struct netdev_rx_queue *rxq;
> +	unsigned long xa_idx;
> +	unsigned int rxq_idx;
> +
> +	if (binding->list.next)
> +		list_del(&binding->list);
> +
> +	xa_for_each(&binding->bound_rxqs, xa_idx, rxq) {
> +		if (rxq->mp_params.mp_priv == binding) {

WARN_ON(rxq->mp_params.mp_priv != binding) ?
We know we're bound to this queue, nobody should be able to replace 
the mp, right?

> +			rxq->mp_params.mp_priv = NULL;
> +
> +			rxq_idx = get_netdev_rx_queue_index(rxq);
> +
> +			WARN_ON(netdev_rx_queue_restart(binding->dev, rxq_idx));
> +		}
> +	}
> +
> +	xa_erase(&net_devmem_dmabuf_bindings, binding->id);
> +
> +	net_devmem_dmabuf_binding_put(binding);
> +}
> +
> +int net_devmem_bind_dmabuf_to_queue(struct net_device *dev, u32 rxq_idx,
> +				    struct net_devmem_dmabuf_binding *binding)
> +{
> +	struct netdev_rx_queue *rxq;
> +	u32 xa_idx;
> +	int err;
> +
> +	if (rxq_idx >= dev->real_num_rx_queues)
> +		return -ERANGE;
> +
> +	rxq = __netif_get_rx_queue(dev, rxq_idx);
> +	if (rxq->mp_params.mp_priv)
> +		return -EEXIST;
> +
> +#ifdef CONFIG_XDP_SOCKETS
> +	if (rxq->pool)
> +		return -EEXIST;

EBUSY plus extack "designated queue already in use by AF_XDP"

> +#endif
> +
> +	if (dev_xdp_prog_count(dev))
> +		return -EEXIST;

Also needs an extack, but since it's not queue-specific should 
it not live inside net_devmem_bind_dmabuf() ? Or do you anticipate
reuse of this function by non-dmabuf code?

> +void dev_dmabuf_uninstall(struct net_device *dev)
> +{
> +	struct net_devmem_dmabuf_binding *binding;
> +	struct netdev_rx_queue *rxq;
> +	unsigned long xa_idx;
> +	unsigned int i;
> +
> +	for (i = 0; i < dev->real_num_rx_queues; i++) {
> +		binding = dev->_rx[i].mp_params.mp_priv;
> +		if (!binding)
> +			continue;
> +
> +		xa_for_each(&binding->bound_rxqs, xa_idx, rxq)
> +			if (rxq == &dev->_rx[i])
> +				xa_erase(&binding->bound_rxqs, xa_idx);

break;

I don't think we can store the same queue twice

> +	}
> +}
> +#endif

> diff --git a/net/core/netdev-genl.c b/net/core/netdev-genl.c
> index 2d726e65211d..269faa37f84e 100644
> --- a/net/core/netdev-genl.c
> +++ b/net/core/netdev-genl.c
> @@ -10,6 +10,7 @@
>  #include <net/netdev_rx_queue.h>
>  #include <net/netdev_queues.h>
>  #include <net/busy_poll.h>
> +#include <net/devmem.h>

include order

> +	return genlmsg_reply(rsp, info);

Should we goto err_unbind if genlmsg_reply() fails?
Shouldn't really happen unless socket is full but simple enough to fix.