2 * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * netmap support for: lem
32 * For details on netmap support please see ixgbe_netmap.h
36 #include <net/netmap.h>
37 #include <sys/selinfo.h>
39 #include <vm/pmap.h> /* vtophys ? */
40 #include <dev/netmap/netmap_kern.h>
42 extern int netmap_adaptive_io;
45 * Register/unregister. We are already under netmap lock.
48 lem_netmap_reg(struct netmap_adapter *na, int onoff)
50 struct ifnet *ifp = na->ifp;
51 struct adapter *adapter = ifp->if_softc;
53 EM_CORE_LOCK(adapter);
55 lem_disable_intr(adapter);
57 /* Tell the stack that the interface is no longer active */
58 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
60 #ifndef EM_LEGACY_IRQ // XXX do we need this ?
61 taskqueue_block(adapter->tq);
62 taskqueue_drain(adapter->tq, &adapter->rxtx_task);
63 taskqueue_drain(adapter->tq, &adapter->link_task);
64 #endif /* !EM_LEGCY_IRQ */
66 /* enable or disable flags and callbacks in na and ifp */
68 nm_set_native_flags(na);
70 nm_clear_native_flags(na);
72 lem_init_locked(adapter); /* also enable intr */
75 taskqueue_unblock(adapter->tq); // XXX do we need this ?
76 #endif /* !EM_LEGCY_IRQ */
78 EM_CORE_UNLOCK(adapter);
80 return (ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1);
85 * Reconcile kernel and user view of the transmit ring.
88 lem_netmap_txsync(struct netmap_kring *kring, int flags)
90 struct netmap_adapter *na = kring->na;
91 struct ifnet *ifp = na->ifp;
92 struct netmap_ring *ring = kring->ring;
93 u_int nm_i; /* index into the netmap ring */
94 u_int nic_i; /* index into the NIC ring */
95 u_int const lim = kring->nkr_num_slots - 1;
96 u_int const head = kring->rhead;
97 /* generate an interrupt approximately every half ring */
98 u_int report_frequency = kring->nkr_num_slots >> 1;
100 /* device-specific */
101 struct adapter *adapter = ifp->if_softc;
103 struct paravirt_csb *csb = adapter->csb;
104 uint64_t *csbd = (uint64_t *)(csb + 1);
105 #endif /* NIC_PARAVIRT */
107 bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
108 BUS_DMASYNC_POSTREAD);
111 * First part: process new packets to send.
114 nm_i = kring->nr_hwcur;
115 if (nm_i != head) { /* we have new packets to send */
118 uint64_t t = 0; // timestamp
123 t = rdtsc(); /* last timestamp */
124 csbd[16] += t - csbd[0]; /* total Wg */
125 csbd[17] += n; /* Wg count */
128 #endif /* NIC_PARAVIRT */
129 nic_i = netmap_idx_k2n(kring, nm_i);
130 while (nm_i != head) {
131 struct netmap_slot *slot = &ring->slot[nm_i];
132 u_int len = slot->len;
134 void *addr = PNMB(na, slot, &paddr);
136 /* device-specific */
137 struct e1000_tx_desc *curr = &adapter->tx_desc_base[nic_i];
138 struct em_buffer *txbuf = &adapter->tx_buffer_area[nic_i];
139 int flags = (slot->flags & NS_REPORT ||
140 nic_i == 0 || nic_i == report_frequency) ?
141 E1000_TXD_CMD_RS : 0;
143 NM_CHECK_ADDR_LEN(na, addr, len);
145 if (slot->flags & NS_BUF_CHANGED) {
146 /* buffer has changed, reload map */
147 curr->buffer_addr = htole64(paddr);
148 netmap_reload_map(na, adapter->txtag, txbuf->map, addr);
150 slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED);
152 /* Fill the slot in the NIC ring. */
153 curr->upper.data = 0;
154 curr->lower.data = htole32(adapter->txd_cmd | len |
155 (E1000_TXD_CMD_EOP | flags) );
156 bus_dmamap_sync(adapter->txtag, txbuf->map,
157 BUS_DMASYNC_PREWRITE);
159 nm_i = nm_next(nm_i, lim);
160 nic_i = nm_next(nic_i, lim);
161 // XXX might try an early kick
163 kring->nr_hwcur = head;
165 /* synchronize the NIC ring */
166 bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
167 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
170 /* set unconditionally, then also kick if needed */
173 if (csb->host_need_txkick == 2) {
174 /* can compute an update of delta */
175 int64_t delta = t - csbd[3];
178 if (csbd[8] == 0 || delta < csbd[8]) {
184 csb->guest_tdt = nic_i;
185 csbd[18] += t - csbd[0]; // total wp
188 if (!csb || !csb->guest_csb_on || (csb->host_need_txkick & 1))
191 #endif /* NIC_PARAVIRT */
192 /* (re)start the tx unit up to slot nic_i (excluded) */
193 E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), nic_i);
196 uint64_t t1 = rdtsc();
197 csbd[20] += t1 - t; // total Np
200 #endif /* NIC_PARAVIRT */
204 * Second part: reclaim buffers for completed transmissions.
206 if (ticks != kring->last_reclaim || flags & NAF_FORCE_RECLAIM || nm_kr_txempty(kring)) {
207 kring->last_reclaim = ticks;
208 /* record completed transmissions using TDH */
210 /* host updates tdh unconditionally, and we have
211 * no side effects on reads, so we can read from there
212 * instead of exiting.
215 static int drain = 0, nodrain=0, good = 0, bad = 0, fail = 0;
216 u_int x = adapter->next_tx_to_clean;
217 csbd[19]++; // XXX count reclaims
218 nic_i = csb->host_tdh;
219 if (csb->guest_csb_on) {
222 csbd[24]++; // failed reclaims
223 /* no progress, request kick and retry */
224 csb->guest_need_txkick = 1;
226 nic_i = csb->host_tdh;
231 csb->guest_need_txkick = 2;
232 if (nic_i == csb->guest_tdt)
237 if (netmap_adaptive_io) {
238 /* new mechanism: last half ring (or so)
239 * released one slot at a time.
240 * This effectively makes the system spin.
242 * Take next_to_clean + 1 as a reference.
243 * tdh must be ahead or equal
244 * On entry, the logical order is
246 * We first push tdh up to avoid wraps.
247 * The limit is tdh-ll (half ring).
248 * if tdh-256 < x we report x;
249 * else we report tdh-256
254 if (netmap_adaptive_io == 2 || ll > delta)
255 csbd[15] = ll = delta;
256 else if (netmap_adaptive_io == 1 && ll > 1) {
260 if (nic_i >= kring->nkr_num_slots) {
261 RD(5, "bad nic_i %d on input", nic_i);
268 csbd[25]++; //report n + 1;
274 csbd[26]++; // report tdh - ll
279 /* we stop, count whether we are idle or not */
280 int bh_active = csb->host_need_txkick & 2 ? 4 : 0;
281 csbd[27+ csb->host_need_txkick]++;
282 if (netmap_adaptive_io == 1) {
283 if (bh_active && csbd[15] > 1)
285 else if (!bh_active && csbd[15] < lim/2)
292 RD(1, "drain %d nodrain %d good %d retry %d fail %d",
293 drain, nodrain, good, bad, fail);
295 #endif /* !NIC_PARAVIRT */
296 nic_i = E1000_READ_REG(&adapter->hw, E1000_TDH(0));
297 if (nic_i >= kring->nkr_num_slots) { /* XXX can it happen ? */
298 D("TDH wrap %d", nic_i);
299 nic_i -= kring->nkr_num_slots;
301 adapter->next_tx_to_clean = nic_i;
302 kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
310 * Reconcile kernel and user view of the receive ring.
313 lem_netmap_rxsync(struct netmap_kring *kring, int flags)
315 struct netmap_adapter *na = kring->na;
316 struct ifnet *ifp = na->ifp;
317 struct netmap_ring *ring = kring->ring;
318 u_int nm_i; /* index into the netmap ring */
319 u_int nic_i; /* index into the NIC ring */
321 u_int const lim = kring->nkr_num_slots - 1;
322 u_int const head = kring->rhead;
323 int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR;
325 /* device-specific */
326 struct adapter *adapter = ifp->if_softc;
328 struct paravirt_csb *csb = adapter->csb;
329 uint32_t csb_mode = csb && csb->guest_csb_on;
330 uint32_t do_host_rxkick = 0;
331 #endif /* NIC_PARAVIRT */
334 return netmap_ring_reinit(kring);
339 csb->guest_need_rxkick = 0;
341 #endif /* NIC_PARAVIRT */
342 /* XXX check sync modes */
343 bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
344 BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
347 * First part: import newly received packets.
349 if (netmap_no_pendintr || force_update) {
350 uint16_t slot_flags = kring->nkr_slot_flags;
352 nic_i = adapter->next_rx_desc_to_check;
353 nm_i = netmap_idx_n2k(kring, nic_i);
356 struct e1000_rx_desc *curr = &adapter->rx_desc_base[nic_i];
357 uint32_t staterr = le32toh(curr->status);
362 if ((staterr & E1000_RXD_STAT_DD) == 0) {
363 /* don't bother to retry if more than 1 pkt */
366 csb->guest_need_rxkick = 1;
368 staterr = le32toh(curr->status);
369 if ((staterr & E1000_RXD_STAT_DD) == 0) {
371 } else { /* we are good */
372 csb->guest_need_rxkick = 0;
376 #endif /* NIC_PARAVIRT */
377 if ((staterr & E1000_RXD_STAT_DD) == 0)
379 len = le16toh(curr->length) - 4; // CRC
381 RD(5, "bogus pkt (%d) size %d nic idx %d", n, len, nic_i);
384 ring->slot[nm_i].len = len;
385 ring->slot[nm_i].flags = slot_flags;
386 bus_dmamap_sync(adapter->rxtag,
387 adapter->rx_buffer_area[nic_i].map,
388 BUS_DMASYNC_POSTREAD);
389 nm_i = nm_next(nm_i, lim);
390 nic_i = nm_next(nic_i, lim);
392 if (n) { /* update the state variables */
396 /* leave one spare buffer so we avoid rxkicks */
397 nm_i = nm_prev(nm_i, lim);
398 nic_i = nm_prev(nic_i, lim);
401 csb->guest_need_rxkick = 1;
404 #endif /* NIC_PARAVIRT */
405 ND("%d new packets at nic %d nm %d tail %d",
407 adapter->next_rx_desc_to_check,
408 netmap_idx_n2k(kring, adapter->next_rx_desc_to_check),
410 adapter->next_rx_desc_to_check = nic_i;
411 // if_inc_counter(ifp, IFCOUNTER_IPACKETS, n);
412 kring->nr_hwtail = nm_i;
414 kring->nr_kflags &= ~NKR_PENDINTR;
418 * Second part: skip past packets that userspace has released.
420 nm_i = kring->nr_hwcur;
422 nic_i = netmap_idx_k2n(kring, nm_i);
423 for (n = 0; nm_i != head; n++) {
424 struct netmap_slot *slot = &ring->slot[nm_i];
426 void *addr = PNMB(na, slot, &paddr);
428 struct e1000_rx_desc *curr = &adapter->rx_desc_base[nic_i];
429 struct em_buffer *rxbuf = &adapter->rx_buffer_area[nic_i];
431 if (addr == NETMAP_BUF_BASE(na)) /* bad buf */
434 if (slot->flags & NS_BUF_CHANGED) {
435 /* buffer has changed, reload map */
436 curr->buffer_addr = htole64(paddr);
437 netmap_reload_map(na, adapter->rxtag, rxbuf->map, addr);
438 slot->flags &= ~NS_BUF_CHANGED;
441 bus_dmamap_sync(adapter->rxtag, rxbuf->map,
442 BUS_DMASYNC_PREREAD);
444 if (csb_mode && csb->host_rxkick_at == nic_i)
446 #endif /* NIC_PARAVIRT */
447 nm_i = nm_next(nm_i, lim);
448 nic_i = nm_next(nic_i, lim);
450 kring->nr_hwcur = head;
451 bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
452 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
454 * IMPORTANT: we must leave one free slot in the ring,
455 * so move nic_i back by one unit
457 nic_i = nm_prev(nic_i, lim);
459 /* set unconditionally, then also kick if needed */
461 csb->guest_rdt = nic_i;
462 if (!csb_mode || do_host_rxkick)
463 #endif /* NIC_PARAVIRT */
464 E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), nic_i);
470 return netmap_ring_reinit(kring);
475 lem_netmap_attach(struct adapter *adapter)
477 struct netmap_adapter na;
479 bzero(&na, sizeof(na));
481 na.ifp = adapter->ifp;
482 na.na_flags = NAF_BDG_MAYSLEEP;
483 na.num_tx_desc = adapter->num_tx_desc;
484 na.num_rx_desc = adapter->num_rx_desc;
485 na.nm_txsync = lem_netmap_txsync;
486 na.nm_rxsync = lem_netmap_rxsync;
487 na.nm_register = lem_netmap_reg;
488 na.num_tx_rings = na.num_rx_rings = 1;