linux_dsm_epyc7002/include/net/xdp.h
Jesper Dangaard Brouer 039930945a xdp: transition into using xdp_frame for return API
Changing API xdp_return_frame() to take struct xdp_frame as argument,
seems like a natural choice. But there are some subtle performance
details here that needs extra care, which is a deliberate choice.

When de-referencing xdp_frame on a remote CPU during DMA-TX
completion, result in the cache-line is change to "Shared"
state. Later when the page is reused for RX, then this xdp_frame
cache-line is written, which change the state to "Modified".

This situation already happens (naturally) for, virtio_net, tun and
cpumap as the xdp_frame pointer is the queued object.  In tun and
cpumap, the ptr_ring is used for efficiently transferring cache-lines
(with pointers) between CPUs. Thus, the only option is to
de-referencing xdp_frame.

It is only the ixgbe driver that had an optimization, in which it can
avoid doing the de-reference of xdp_frame.  The driver already have
TX-ring queue, which (in case of remote DMA-TX completion) have to be
transferred between CPUs anyhow.  In this data area, we stored a
struct xdp_mem_info and a data pointer, which allowed us to avoid
de-referencing xdp_frame.

To compensate for this, a prefetchw is used for telling the cache
coherency protocol about our access pattern.  My benchmarks show that
this prefetchw is enough to compensate the ixgbe driver.

V7: Adjust for commit d9314c474d ("i40e: add support for XDP_REDIRECT")
V8: Adjust for commit bd658dda42 ("net/mlx5e: Separate dma base address
and offset in dma_sync call")

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-04-17 10:50:29 -04:00

132 lines
3.7 KiB
C

/* include/net/xdp.h
*
* Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
* Released under terms in GPL version 2. See COPYING.
*/
#ifndef __LINUX_NET_XDP_H__
#define __LINUX_NET_XDP_H__
/**
* DOC: XDP RX-queue information
*
* The XDP RX-queue info (xdp_rxq_info) is associated with the driver
* level RX-ring queues. It is information that is specific to how
* the driver have configured a given RX-ring queue.
*
* Each xdp_buff frame received in the driver carry a (pointer)
* reference to this xdp_rxq_info structure. This provides the XDP
* data-path read-access to RX-info for both kernel and bpf-side
* (limited subset).
*
* For now, direct access is only safe while running in NAPI/softirq
* context. Contents is read-mostly and must not be updated during
* driver NAPI/softirq poll.
*
* The driver usage API is a register and unregister API.
*
* The struct is not directly tied to the XDP prog. A new XDP prog
* can be attached as long as it doesn't change the underlying
* RX-ring. If the RX-ring does change significantly, the NIC driver
* naturally need to stop the RX-ring before purging and reallocating
* memory. In that process the driver MUST call unregistor (which
* also apply for driver shutdown and unload). The register API is
* also mandatory during RX-ring setup.
*/
enum xdp_mem_type {
MEM_TYPE_PAGE_SHARED = 0, /* Split-page refcnt based model */
MEM_TYPE_PAGE_ORDER0, /* Orig XDP full page model */
MEM_TYPE_PAGE_POOL,
MEM_TYPE_MAX,
};
struct xdp_mem_info {
u32 type; /* enum xdp_mem_type, but known size type */
u32 id;
};
struct page_pool;
struct xdp_rxq_info {
struct net_device *dev;
u32 queue_index;
u32 reg_state;
struct xdp_mem_info mem;
} ____cacheline_aligned; /* perf critical, avoid false-sharing */
struct xdp_buff {
void *data;
void *data_end;
void *data_meta;
void *data_hard_start;
struct xdp_rxq_info *rxq;
};
struct xdp_frame {
void *data;
u16 len;
u16 headroom;
u16 metasize;
/* Lifetime of xdp_rxq_info is limited to NAPI/enqueue time,
* while mem info is valid on remote CPU.
*/
struct xdp_mem_info mem;
struct net_device *dev_rx; /* used by cpumap */
};
/* Convert xdp_buff to xdp_frame */
static inline
struct xdp_frame *convert_to_xdp_frame(struct xdp_buff *xdp)
{
struct xdp_frame *xdp_frame;
int metasize;
int headroom;
/* Assure headroom is available for storing info */
headroom = xdp->data - xdp->data_hard_start;
metasize = xdp->data - xdp->data_meta;
metasize = metasize > 0 ? metasize : 0;
if (unlikely((headroom - metasize) < sizeof(*xdp_frame)))
return NULL;
/* Store info in top of packet */
xdp_frame = xdp->data_hard_start;
xdp_frame->data = xdp->data;
xdp_frame->len = xdp->data_end - xdp->data;
xdp_frame->headroom = headroom - sizeof(*xdp_frame);
xdp_frame->metasize = metasize;
/* rxq only valid until napi_schedule ends, convert to xdp_mem_info */
xdp_frame->mem = xdp->rxq->mem;
return xdp_frame;
}
void xdp_return_frame(struct xdp_frame *xdpf);
int xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
struct net_device *dev, u32 queue_index);
void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq);
void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq);
bool xdp_rxq_info_is_reg(struct xdp_rxq_info *xdp_rxq);
int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info *xdp_rxq,
enum xdp_mem_type type, void *allocator);
/* Drivers not supporting XDP metadata can use this helper, which
* rejects any room expansion for metadata as a result.
*/
static __always_inline void
xdp_set_data_meta_invalid(struct xdp_buff *xdp)
{
xdp->data_meta = xdp->data + 1;
}
static __always_inline bool
xdp_data_meta_unsupported(const struct xdp_buff *xdp)
{
return unlikely(xdp->data_meta > xdp->data);
}
#endif /* __LINUX_NET_XDP_H__ */