64ae02c365
USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
327 lines
10 KiB
C
327 lines
10 KiB
C
/*
|
|
* Copyright (C) 2011-2012 Matteo Landi, Luigi Rizzo. All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
* SUCH DAMAGE.
|
|
*/
|
|
|
|
/*
|
|
* $FreeBSD$
|
|
* $Id: netmap_kern.h 10602 2012-02-21 16:47:55Z luigi $
|
|
*
|
|
* The header contains the definitions of constants and function
|
|
* prototypes used only in kernelspace.
|
|
*/
|
|
|
|
#ifndef _NET_NETMAP_KERN_H_
|
|
#define _NET_NETMAP_KERN_H_
|
|
|
|
#if defined(__FreeBSD__)
|
|
#define NM_LOCK_T struct mtx
|
|
#define NM_SELINFO_T struct selinfo
|
|
#define MBUF_LEN(m) ((m)->m_pkthdr.len)
|
|
#define NM_SEND_UP(ifp, m) ((ifp)->if_input)(ifp, m)
|
|
#elif defined (linux)
|
|
#define NM_LOCK_T spinlock_t
|
|
#define NM_SELINFO_T wait_queue_head_t
|
|
#define MBUF_LEN(m) ((m)->len)
|
|
#define NM_SEND_UP(ifp, m) netif_rx(m)
|
|
#else
|
|
#error unsupported platform
|
|
#endif
|
|
|
|
#ifdef MALLOC_DECLARE
|
|
MALLOC_DECLARE(M_NETMAP);
|
|
#endif
|
|
|
|
#define ND(format, ...)
|
|
#define D(format, ...) \
|
|
do { \
|
|
struct timeval __xxts; \
|
|
microtime(&__xxts); \
|
|
printf("%03d.%06d %s [%d] " format "\n", \
|
|
(int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \
|
|
__FUNCTION__, __LINE__, ##__VA_ARGS__); \
|
|
} while (0)
|
|
|
|
struct netmap_adapter;
|
|
|
|
/*
|
|
* private, kernel view of a ring. Keeps track of the status of
|
|
* a ring across system calls.
|
|
*
|
|
* nr_hwcur index of the next buffer to refill.
|
|
* It corresponds to ring->cur - ring->reserved
|
|
*
|
|
* nr_hwavail the number of slots "owned" by userspace.
|
|
* nr_hwavail =:= ring->avail + ring->reserved
|
|
*
|
|
* The indexes in the NIC and netmap rings are offset by nkr_hwofs slots.
|
|
* This is so that, on a reset, buffers owned by userspace are not
|
|
* modified by the kernel. In particular:
|
|
* RX rings: the next empty buffer (hwcur + hwavail + hwofs) coincides with
|
|
* the next empty buffer as known by the hardware (next_to_check or so).
|
|
* TX rings: hwcur + hwofs coincides with next_to_send
|
|
*/
|
|
struct netmap_kring {
|
|
struct netmap_ring *ring;
|
|
u_int nr_hwcur;
|
|
int nr_hwavail;
|
|
u_int nr_kflags; /* private driver flags */
|
|
#define NKR_PENDINTR 0x1 // Pending interrupt.
|
|
u_int nkr_num_slots;
|
|
|
|
int nkr_hwofs; /* offset between NIC and netmap ring */
|
|
struct netmap_adapter *na;
|
|
NM_SELINFO_T si; /* poll/select wait queue */
|
|
NM_LOCK_T q_lock; /* used if no device lock available */
|
|
} __attribute__((__aligned__(64)));
|
|
|
|
/*
|
|
* This struct extends the 'struct adapter' (or
|
|
* equivalent) device descriptor. It contains all fields needed to
|
|
* support netmap operation.
|
|
*/
|
|
struct netmap_adapter {
|
|
int refcount; /* number of user-space descriptors using this
|
|
interface, which is equal to the number of
|
|
struct netmap_if objs in the mapped region. */
|
|
|
|
int separate_locks; /* set if the interface suports different
|
|
locks for rx, tx and core. */
|
|
|
|
u_int num_rx_queues; /* number of tx/rx queue pairs: this is
|
|
a duplicate field needed to simplify the
|
|
signature of ``netmap_detach``. */
|
|
u_int num_tx_queues; // if nonzero, overrides num_queues XXX
|
|
|
|
u_int num_tx_desc; /* number of descriptor in each queue */
|
|
u_int num_rx_desc;
|
|
u_int buff_size; // XXX deprecate, use NETMAP_BUF_SIZE
|
|
|
|
//u_int flags; // XXX unused
|
|
/* tx_rings and rx_rings are private but allocated
|
|
* as a contiguous chunk of memory. Each array has
|
|
* N+1 entries, for the adapter queues and for the host queue.
|
|
*/
|
|
struct netmap_kring *tx_rings; /* array of TX rings. */
|
|
struct netmap_kring *rx_rings; /* array of RX rings. */
|
|
|
|
NM_SELINFO_T tx_si, rx_si; /* global wait queues */
|
|
|
|
/* copy of if_qflush and if_transmit pointers, to intercept
|
|
* packets from the network stack when netmap is active.
|
|
* XXX probably if_qflush is not necessary.
|
|
*/
|
|
//void (*if_qflush)(struct ifnet *); // XXX unused
|
|
int (*if_transmit)(struct ifnet *, struct mbuf *);
|
|
|
|
/* references to the ifnet and device routines, used by
|
|
* the generic netmap functions.
|
|
*/
|
|
struct ifnet *ifp; /* adapter is ifp->if_softc */
|
|
|
|
NM_LOCK_T core_lock; /* used if no device lock available */
|
|
|
|
int (*nm_register)(struct ifnet *, int onoff);
|
|
void (*nm_lock)(struct ifnet *, int what, u_int ringid);
|
|
int (*nm_txsync)(struct ifnet *, u_int ring, int lock);
|
|
int (*nm_rxsync)(struct ifnet *, u_int ring, int lock);
|
|
#ifdef linux
|
|
struct net_device_ops nm_ndo;
|
|
#endif /* linux */
|
|
};
|
|
|
|
/*
|
|
* The combination of "enable" (ifp->if_capabilities &IFCAP_NETMAP)
|
|
* and refcount gives the status of the interface, namely:
|
|
*
|
|
* enable refcount Status
|
|
*
|
|
* FALSE 0 normal operation
|
|
* FALSE != 0 -- (impossible)
|
|
* TRUE 1 netmap mode
|
|
* TRUE 0 being deleted.
|
|
*/
|
|
|
|
#define NETMAP_DELETING(_na) ( ((_na)->refcount == 0) && \
|
|
( (_na)->ifp->if_capenable & IFCAP_NETMAP) )
|
|
|
|
/*
|
|
* parameters for (*nm_lock)(adapter, what, index)
|
|
*/
|
|
enum {
|
|
NETMAP_NO_LOCK = 0,
|
|
NETMAP_CORE_LOCK, NETMAP_CORE_UNLOCK,
|
|
NETMAP_TX_LOCK, NETMAP_TX_UNLOCK,
|
|
NETMAP_RX_LOCK, NETMAP_RX_UNLOCK,
|
|
#ifdef __FreeBSD__
|
|
#define NETMAP_REG_LOCK NETMAP_CORE_LOCK
|
|
#define NETMAP_REG_UNLOCK NETMAP_CORE_UNLOCK
|
|
#else
|
|
NETMAP_REG_LOCK, NETMAP_REG_UNLOCK
|
|
#endif
|
|
};
|
|
|
|
/*
|
|
* The following are support routines used by individual drivers to
|
|
* support netmap operation.
|
|
*
|
|
* netmap_attach() initializes a struct netmap_adapter, allocating the
|
|
* struct netmap_ring's and the struct selinfo.
|
|
*
|
|
* netmap_detach() frees the memory allocated by netmap_attach().
|
|
*
|
|
* netmap_start() replaces the if_transmit routine of the interface,
|
|
* and is used to intercept packets coming from the stack.
|
|
*
|
|
* netmap_load_map/netmap_reload_map are helper routines to set/reset
|
|
* the dmamap for a packet buffer
|
|
*
|
|
* netmap_reset() is a helper routine to be called in the driver
|
|
* when reinitializing a ring.
|
|
*/
|
|
int netmap_attach(struct netmap_adapter *, int);
|
|
void netmap_detach(struct ifnet *);
|
|
int netmap_start(struct ifnet *, struct mbuf *);
|
|
enum txrx { NR_RX = 0, NR_TX = 1 };
|
|
struct netmap_slot *netmap_reset(struct netmap_adapter *na,
|
|
enum txrx tx, int n, u_int new_cur);
|
|
int netmap_ring_reinit(struct netmap_kring *);
|
|
|
|
extern int netmap_buf_size;
|
|
#define NETMAP_BUF_SIZE netmap_buf_size
|
|
extern int netmap_mitigate;
|
|
extern int netmap_no_pendintr;
|
|
extern u_int netmap_total_buffers;
|
|
extern char *netmap_buffer_base;
|
|
extern int netmap_verbose; // XXX debugging
|
|
enum { /* verbose flags */
|
|
NM_VERB_ON = 1, /* generic verbose */
|
|
NM_VERB_HOST = 0x2, /* verbose host stack */
|
|
NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */
|
|
NM_VERB_TXSYNC = 0x20,
|
|
NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */
|
|
NM_VERB_TXINTR = 0x200,
|
|
NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */
|
|
NM_VERB_NIC_TXSYNC = 0x2000,
|
|
};
|
|
|
|
/*
|
|
* NA returns a pointer to the struct netmap adapter from the ifp,
|
|
* WNA is used to write it.
|
|
*/
|
|
#ifndef WNA
|
|
#define WNA(_ifp) (_ifp)->if_pspare[0]
|
|
#endif
|
|
#define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp))
|
|
|
|
|
|
/* Callback invoked by the dma machinery after a successfull dmamap_load */
|
|
static void netmap_dmamap_cb(__unused void *arg,
|
|
__unused bus_dma_segment_t * segs, __unused int nseg, __unused int error)
|
|
{
|
|
}
|
|
|
|
/* bus_dmamap_load wrapper: call aforementioned function if map != NULL.
|
|
* XXX can we do it without a callback ?
|
|
*/
|
|
static inline void
|
|
netmap_load_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
|
|
{
|
|
if (map)
|
|
bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE,
|
|
netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT);
|
|
}
|
|
|
|
/* update the map when a buffer changes. */
|
|
static inline void
|
|
netmap_reload_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
|
|
{
|
|
if (map) {
|
|
bus_dmamap_unload(tag, map);
|
|
bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE,
|
|
netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT);
|
|
}
|
|
}
|
|
|
|
/*
|
|
* functions to map NIC to KRING indexes (n2k) and vice versa (k2n)
|
|
*/
|
|
static inline int
|
|
netmap_idx_n2k(struct netmap_kring *kr, int idx)
|
|
{
|
|
int n = kr->nkr_num_slots;
|
|
idx += kr->nkr_hwofs;
|
|
if (idx < 0)
|
|
return idx + n;
|
|
else if (idx < n)
|
|
return idx;
|
|
else
|
|
return idx - n;
|
|
}
|
|
|
|
|
|
static inline int
|
|
netmap_idx_k2n(struct netmap_kring *kr, int idx)
|
|
{
|
|
int n = kr->nkr_num_slots;
|
|
idx -= kr->nkr_hwofs;
|
|
if (idx < 0)
|
|
return idx + n;
|
|
else if (idx < n)
|
|
return idx;
|
|
else
|
|
return idx - n;
|
|
}
|
|
|
|
|
|
/*
|
|
* NMB return the virtual address of a buffer (buffer 0 on bad index)
|
|
* PNMB also fills the physical address
|
|
*/
|
|
static inline void *
|
|
NMB(struct netmap_slot *slot)
|
|
{
|
|
uint32_t i = slot->buf_idx;
|
|
return (i >= netmap_total_buffers) ? netmap_buffer_base :
|
|
netmap_buffer_base + (i *NETMAP_BUF_SIZE);
|
|
}
|
|
|
|
static inline void *
|
|
PNMB(struct netmap_slot *slot, uint64_t *pp)
|
|
{
|
|
uint32_t i = slot->buf_idx;
|
|
void *ret = (i >= netmap_total_buffers) ? netmap_buffer_base :
|
|
netmap_buffer_base + (i *NETMAP_BUF_SIZE);
|
|
*pp = vtophys(ret);
|
|
return ret;
|
|
}
|
|
|
|
/* default functions to handle rx/tx interrupts */
|
|
int netmap_rx_irq(struct ifnet *, int, int *);
|
|
#define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL)
|
|
#ifdef __linux__
|
|
#define bus_dmamap_sync(_a, _b, _c) // wmb() or rmb() ?
|
|
netdev_tx_t netmap_start_linux(struct sk_buff *skb, struct net_device *dev);
|
|
#endif
|
|
#endif /* _NET_NETMAP_KERN_H_ */
|