freebsd-skq/sys/dev/netmap/if_re_netmap.h
Luigi Rizzo 64ae02c365 A bunch of netmap fixes:
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.
2012-02-27 19:05:01 +00:00

422 lines
12 KiB
C

/*
* Copyright (C) 2011 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: if_re_netmap.h 10609 2012-02-22 19:44:58Z luigi $
*
* netmap support for "re"
* For details on netmap support please see ixgbe_netmap.h
*/
#include <net/netmap.h>
#include <sys/selinfo.h>
#include <vm/vm.h>
#include <vm/pmap.h> /* vtophys ? */
#include <dev/netmap/netmap_kern.h>
/*
* wrapper to export locks to the generic code
* We should not use the tx/rx locks
*/
static void
re_netmap_lock_wrapper(struct ifnet *ifp, int what, u_int queueid)
{
struct rl_softc *adapter = ifp->if_softc;
switch (what) {
case NETMAP_CORE_LOCK:
RL_LOCK(adapter);
break;
case NETMAP_CORE_UNLOCK:
RL_UNLOCK(adapter);
break;
case NETMAP_TX_LOCK:
case NETMAP_RX_LOCK:
case NETMAP_TX_UNLOCK:
case NETMAP_RX_UNLOCK:
D("invalid lock call %d, no tx/rx locks here", what);
break;
}
}
/*
* support for netmap register/unregisted. We are already under core lock.
* only called on the first register or the last unregister.
*/
static int
re_netmap_reg(struct ifnet *ifp, int onoff)
{
struct rl_softc *adapter = ifp->if_softc;
struct netmap_adapter *na = NA(ifp);
int error = 0;
if (na == NULL)
return EINVAL;
/* Tell the stack that the interface is no longer active */
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
re_stop(adapter);
if (onoff) {
ifp->if_capenable |= IFCAP_NETMAP;
/* save if_transmit to restore it later */
na->if_transmit = ifp->if_transmit;
ifp->if_transmit = netmap_start;
re_init_locked(adapter);
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) == 0) {
error = ENOMEM;
goto fail;
}
} else {
fail:
/* restore if_transmit */
ifp->if_transmit = na->if_transmit;
ifp->if_capenable &= ~IFCAP_NETMAP;
re_init_locked(adapter); /* also enables intr */
}
return (error);
}
/*
* Reconcile kernel and user view of the transmit ring.
*/
static int
re_netmap_txsync(struct ifnet *ifp, u_int ring_nr, int do_lock)
{
struct rl_softc *sc = ifp->if_softc;
struct rl_txdesc *txd = sc->rl_ldata.rl_tx_desc;
struct netmap_adapter *na = NA(sc->rl_ifp);
struct netmap_kring *kring = &na->tx_rings[ring_nr];
struct netmap_ring *ring = kring->ring;
int j, k, l, n, lim = kring->nkr_num_slots - 1;
k = ring->cur;
if (k > lim)
return netmap_ring_reinit(kring);
if (do_lock)
RL_LOCK(sc);
/* Sync the TX descriptor list */
bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
/* XXX move after the transmissions */
/* record completed transmissions */
for (n = 0, l = sc->rl_ldata.rl_tx_considx;
l != sc->rl_ldata.rl_tx_prodidx;
n++, l = RL_TX_DESC_NXT(sc, l)) {
uint32_t cmdstat =
le32toh(sc->rl_ldata.rl_tx_list[l].rl_cmdstat);
if (cmdstat & RL_TDESC_STAT_OWN)
break;
}
if (n > 0) {
sc->rl_ldata.rl_tx_considx = l;
sc->rl_ldata.rl_tx_free += n;
kring->nr_hwavail += n;
}
/* update avail to what the kernel knows */
ring->avail = kring->nr_hwavail;
j = kring->nr_hwcur;
if (j != k) { /* we have new packets to send */
l = sc->rl_ldata.rl_tx_prodidx;
for (n = 0; j != k; n++) {
struct netmap_slot *slot = &ring->slot[j];
struct rl_desc *desc = &sc->rl_ldata.rl_tx_list[l];
int cmd = slot->len | RL_TDESC_CMD_EOF |
RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF ;
uint64_t paddr;
void *addr = PNMB(slot, &paddr);
int len = slot->len;
if (addr == netmap_buffer_base || len > NETMAP_BUF_SIZE) {
if (do_lock)
RL_UNLOCK(sc);
// XXX what about prodidx ?
return netmap_ring_reinit(kring);
}
if (l == lim) /* mark end of ring */
cmd |= RL_TDESC_CMD_EOR;
if (slot->flags & NS_BUF_CHANGED) {
desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(paddr));
desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(paddr));
/* buffer has changed, unload and reload map */
netmap_reload_map(sc->rl_ldata.rl_tx_mtag,
txd[l].tx_dmamap, addr);
slot->flags &= ~NS_BUF_CHANGED;
}
slot->flags &= ~NS_REPORT;
desc->rl_cmdstat = htole32(cmd);
bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
txd[l].tx_dmamap, BUS_DMASYNC_PREWRITE);
j = (j == lim) ? 0 : j + 1;
l = (l == lim) ? 0 : l + 1;
}
sc->rl_ldata.rl_tx_prodidx = l;
kring->nr_hwcur = k; /* the saved ring->cur */
ring->avail -= n; // XXX see others
kring->nr_hwavail = ring->avail;
bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
/* start ? */
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
}
if (do_lock)
RL_UNLOCK(sc);
return 0;
}
/*
* Reconcile kernel and user view of the receive ring.
*/
static int
re_netmap_rxsync(struct ifnet *ifp, u_int ring_nr, int do_lock)
{
struct rl_softc *sc = ifp->if_softc;
struct rl_rxdesc *rxd = sc->rl_ldata.rl_rx_desc;
struct netmap_adapter *na = NA(sc->rl_ifp);
struct netmap_kring *kring = &na->rx_rings[ring_nr];
struct netmap_ring *ring = kring->ring;
int j, l, n, lim = kring->nkr_num_slots - 1;
int force_update = do_lock || kring->nr_kflags & NKR_PENDINTR;
u_int k = ring->cur, resvd = ring->reserved;
k = ring->cur;
if (k > lim)
return netmap_ring_reinit(kring);
if (do_lock)
RL_LOCK(sc);
/* XXX check sync modes */
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
/*
* Import newly received packets into the netmap ring.
* j is an index in the netmap ring, l in the NIC ring.
*
* The device uses all the buffers in the ring, so we need
* another termination condition in addition to RL_RDESC_STAT_OWN
* cleared (all buffers could have it cleared. The easiest one
* is to limit the amount of data reported up to 'lim'
*/
l = sc->rl_ldata.rl_rx_prodidx; /* next pkt to check */
j = netmap_idx_n2k(kring, l); /* the kring index */
if (netmap_no_pendintr || force_update) {
for (n = kring->nr_hwavail; n < lim ; n++) {
struct rl_desc *cur_rx = &sc->rl_ldata.rl_rx_list[l];
uint32_t rxstat = le32toh(cur_rx->rl_cmdstat);
uint32_t total_len;
if ((rxstat & RL_RDESC_STAT_OWN) != 0)
break;
total_len = rxstat & sc->rl_rxlenmask;
/* XXX subtract crc */
total_len = (total_len < 4) ? 0 : total_len - 4;
kring->ring->slot[j].len = total_len;
/* sync was in re_newbuf() */
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
rxd[l].rx_dmamap, BUS_DMASYNC_POSTREAD);
j = (j == lim) ? 0 : j + 1;
l = (l == lim) ? 0 : l + 1;
}
if (n != kring->nr_hwavail) {
sc->rl_ldata.rl_rx_prodidx = l;
sc->rl_ifp->if_ipackets += n - kring->nr_hwavail;
kring->nr_hwavail = n;
}
kring->nr_kflags &= ~NKR_PENDINTR;
}
/* skip past packets that userspace has released */
j = kring->nr_hwcur;
if (resvd > 0) {
if (resvd + ring->avail >= lim + 1) {
D("XXX invalid reserve/avail %d %d", resvd, ring->avail);
ring->reserved = resvd = 0; // XXX panic...
}
k = (k >= resvd) ? k - resvd : k + lim + 1 - resvd;
}
if (j != k) { /* userspace has released some packets. */
l = netmap_idx_k2n(kring, j); /* the NIC index */
for (n = 0; j != k; n++) {
struct netmap_slot *slot = ring->slot + j;
struct rl_desc *desc = &sc->rl_ldata.rl_rx_list[l];
int cmd = na->buff_size | RL_RDESC_CMD_OWN;
uint64_t paddr;
void *addr = PNMB(slot, &paddr);
if (addr == netmap_buffer_base) { /* bad buf */
if (do_lock)
RL_UNLOCK(sc);
return netmap_ring_reinit(kring);
}
if (l == lim) /* mark end of ring */
cmd |= RL_RDESC_CMD_EOR;
slot->flags &= ~NS_REPORT;
if (slot->flags & NS_BUF_CHANGED) {
netmap_reload_map(sc->rl_ldata.rl_rx_mtag,
rxd[l].rx_dmamap, addr);
desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(paddr));
desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(paddr));
slot->flags &= ~NS_BUF_CHANGED;
}
desc->rl_cmdstat = htole32(cmd);
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
rxd[l].rx_dmamap, BUS_DMASYNC_PREREAD);
j = (j == lim) ? 0 : j + 1;
l = (l == lim) ? 0 : l + 1;
}
kring->nr_hwavail -= n;
kring->nr_hwcur = k;
/* Flush the RX DMA ring */
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
}
/* tell userspace that there are new packets */
ring->avail = kring->nr_hwavail - resvd;
if (do_lock)
RL_UNLOCK(sc);
return 0;
}
/*
* Additional routines to init the tx and rx rings.
* In other drivers we do that inline in the main code.
*/
static void
re_netmap_tx_init(struct rl_softc *sc)
{
struct rl_txdesc *txd;
struct rl_desc *desc;
int i, n;
struct netmap_adapter *na = NA(sc->rl_ifp);
struct netmap_slot *slot = netmap_reset(na, NR_TX, 0, 0);
/* slot is NULL if we are not in netmap mode */
if (!slot)
return;
/* in netmap mode, overwrite addresses and maps */
txd = sc->rl_ldata.rl_tx_desc;
desc = sc->rl_ldata.rl_tx_list;
n = sc->rl_ldata.rl_tx_desc_cnt;
/* l points in the netmap ring, i points in the NIC ring */
for (i = 0; i < n; i++) {
uint64_t paddr;
int l = netmap_idx_n2k(&na->tx_rings[0], i);
void *addr = PNMB(slot + l, &paddr);
desc[i].rl_bufaddr_lo = htole32(RL_ADDR_LO(paddr));
desc[i].rl_bufaddr_hi = htole32(RL_ADDR_HI(paddr));
netmap_load_map(sc->rl_ldata.rl_tx_mtag,
txd[i].tx_dmamap, addr);
}
}
static void
re_netmap_rx_init(struct rl_softc *sc)
{
struct netmap_adapter *na = NA(sc->rl_ifp);
struct netmap_slot *slot = netmap_reset(na, NR_RX, 0, 0);
struct rl_desc *desc = sc->rl_ldata.rl_rx_list;
uint32_t cmdstat;
int i, n, max_avail;
if (!slot)
return;
n = sc->rl_ldata.rl_rx_desc_cnt;
/*
* Userspace owned hwavail packets before the reset,
* so the NIC that last hwavail descriptors of the ring
* are still owned by the driver (and keep one empty).
*/
max_avail = n - 1 - na->rx_rings[0].nr_hwavail;
for (i = 0; i < n; i++) {
void *addr;
uint64_t paddr;
int l = netmap_idx_n2k(&na->rx_rings[0], i);
addr = PNMB(slot + l, &paddr);
netmap_reload_map(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_desc[i].rx_dmamap, addr);
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_desc[i].rx_dmamap, BUS_DMASYNC_PREREAD);
desc[i].rl_bufaddr_lo = htole32(RL_ADDR_LO(paddr));
desc[i].rl_bufaddr_hi = htole32(RL_ADDR_HI(paddr));
cmdstat = na->buff_size;
if (i == n - 1) /* mark the end of ring */
cmdstat |= RL_RDESC_CMD_EOR;
if (i < max_avail)
cmdstat |= RL_RDESC_CMD_OWN;
desc[i].rl_cmdstat = htole32(cmdstat);
}
}
static void
re_netmap_attach(struct rl_softc *sc)
{
struct netmap_adapter na;
bzero(&na, sizeof(na));
na.ifp = sc->rl_ifp;
na.separate_locks = 0;
na.num_tx_desc = sc->rl_ldata.rl_tx_desc_cnt;
na.num_rx_desc = sc->rl_ldata.rl_rx_desc_cnt;
na.nm_txsync = re_netmap_txsync;
na.nm_rxsync = re_netmap_rxsync;
na.nm_lock = re_netmap_lock_wrapper;
na.nm_register = re_netmap_reg;
netmap_attach(&na, 1);
}
/* end of file */