freebsd-dev/sys/dev/lnc/if_lnc.c
Ruslan Ermilov 4a0d6638b3 - Store pointer to the link-level address right in "struct ifnet"
rather than in ifindex_table[]; all (except one) accesses are
  through ifp anyway.  IF_LLADDR() works faster, and all (except
  one) ifaddr_byindex() users were converted to use ifp->if_addr.

- Stop storing a (pointer to) Ethernet address in "struct arpcom",
  and drop the IFP2ENADDR() macro; all users have been converted
  to use IF_LLADDR() instead.
2005-11-11 16:04:59 +00:00

1561 lines
40 KiB
C

/*-
* Copyright (c) 1994-2000
* Paul Richards. 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,
* verbatim and that no modifications are made prior to this
* point in the file.
* 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.
* 3. The name Paul Richards may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY PAUL RICHARDS ``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 PAUL RICHARDS 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.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
#define DIAGNOSTIC
#define DEBUG
*
* TODO ----
*
* Check all the XXX comments -- some of them are just things I've left
* unfinished rather than "difficult" problems that were hacked around.
*
* Check log settings.
*
* Check how all the arpcom flags get set and used.
*
* Re-inline and re-static all routines after debugging.
*
* Remember to assign iobase in SHMEM probe routines.
*
* Replace all occurences of LANCE-controller-card etc in prints by the name
* strings of the appropriate type -- nifty window dressing
*
* Add DEPCA support -- mostly done.
*
*/
#include "opt_inet.h"
/* Some defines that should really be in generic locations */
#define FCS_LEN 4
#define MULTICAST_FILTER_LEN 8
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/syslog.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <sys/rman.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <netinet/in.h>
#include <netinet/if_ether.h>
#include <net/bpf.h>
#include <machine/md_var.h>
#include <dev/lnc/if_lncvar.h>
#include <dev/lnc/if_lncreg.h>
devclass_t lnc_devclass;
static char const * const nic_ident[] = {
"Unknown",
"BICC",
"NE2100",
"DEPCA",
"CNET98S", /* PC-98 */
};
static char const * const ic_ident[] = {
"Unknown",
"LANCE",
"C-LANCE",
"PCnet-ISA",
"PCnet-ISA+",
"PCnet-ISA II",
"PCnet-32 VL-Bus",
"PCnet-PCI",
"PCnet-PCI II",
"PCnet-FAST",
"PCnet-FAST+",
"PCnet-Home",
};
static void lnc_setladrf(struct lnc_softc *sc);
static void lnc_reset(struct lnc_softc *sc);
static void lnc_free_mbufs(struct lnc_softc *sc);
static __inline int alloc_mbuf_cluster(struct lnc_softc *sc,
struct host_ring_entry *desc);
static __inline struct mbuf *chain_mbufs(struct lnc_softc *sc,
int start_of_packet,
int pkt_len);
static __inline struct mbuf *mbuf_packet(struct lnc_softc *sc,
int start_of_packet,
int pkt_len);
static void lnc_rint(struct lnc_softc *sc);
static void lnc_tint(struct lnc_softc *sc);
static void lnc_init(void *);
static __inline int mbuf_to_buffer(struct mbuf *m, char *buffer);
static __inline struct mbuf *chain_to_cluster(struct mbuf *m);
static void lnc_start(struct ifnet *ifp);
static int lnc_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
static void lnc_watchdog(struct ifnet *ifp);
#ifdef DEBUG
void lnc_dump_state(struct lnc_softc *sc);
void mbuf_dump_chain(struct mbuf *m);
#endif
u_short
read_csr(struct lnc_softc *sc, u_short port)
{
lnc_outw(sc->rap, port);
return (lnc_inw(sc->rdp));
}
void
write_csr(struct lnc_softc *sc, u_short port, u_short val)
{
lnc_outw(sc->rap, port);
lnc_outw(sc->rdp, val);
}
static __inline void
write_bcr(struct lnc_softc *sc, u_short port, u_short val)
{
lnc_outw(sc->rap, port);
lnc_outw(sc->bdp, val);
}
static __inline u_short
read_bcr(struct lnc_softc *sc, u_short port)
{
lnc_outw(sc->rap, port);
return (lnc_inw(sc->bdp));
}
int
lance_probe(struct lnc_softc *sc)
{
write_csr(sc, CSR0, STOP);
if ((lnc_inw(sc->rdp) & STOP) && ! (read_csr(sc, CSR3))) {
/*
* Check to see if it's a C-LANCE. For the LANCE the INEA bit
* cannot be set while the STOP bit is. This restriction is
* removed for the C-LANCE.
*/
write_csr(sc, CSR0, INEA);
if (read_csr(sc, CSR0) & INEA)
return (C_LANCE);
else
return (LANCE);
} else
return (UNKNOWN);
}
void
lnc_release_resources(device_t dev)
{
lnc_softc_t *sc = device_get_softc(dev);
if (sc->irqres) {
bus_teardown_intr(dev, sc->irqres, sc->intrhand);
bus_release_resource(dev, SYS_RES_IRQ, sc->irqrid, sc->irqres);
}
if (sc->portres)
bus_release_resource(dev, SYS_RES_IOPORT,
sc->portrid, sc->portres);
if (sc->drqres)
bus_release_resource(dev, SYS_RES_DRQ, sc->drqrid, sc->drqres);
if (sc->dmat) {
if (sc->dmamap) {
bus_dmamap_unload(sc->dmat, sc->dmamap);
bus_dmamem_free(sc->dmat, sc->recv_ring, sc->dmamap);
}
bus_dma_tag_destroy(sc->dmat);
}
if (sc->ifp)
if_free(sc->ifp);
}
/*
* Set up the logical address filter for multicast packets
*/
static __inline void
lnc_setladrf(struct lnc_softc *sc)
{
struct ifnet *ifp = sc->ifp;
struct ifmultiaddr *ifma;
u_long index;
int i;
if (sc->flags & IFF_ALLMULTI) {
for (i=0; i < MULTICAST_FILTER_LEN; i++)
sc->init_block->ladrf[i] = 0xFF;
return;
}
/*
* For each multicast address, calculate a crc for that address and
* then use the high order 6 bits of the crc as a hash code where
* bits 3-5 select the byte of the address filter and bits 0-2 select
* the bit within that byte.
*/
bzero(sc->init_block->ladrf, MULTICAST_FILTER_LEN);
IF_ADDR_LOCK(ifp);
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
index = ether_crc32_le(LLADDR((struct sockaddr_dl *)
ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
sc->init_block->ladrf[index >> 3] |= 1 << (index & 7);
}
IF_ADDR_UNLOCK(ifp);
}
void
lnc_stop(struct lnc_softc *sc)
{
write_csr(sc, CSR0, STOP);
}
static void
lnc_reset(struct lnc_softc *sc)
{
lnc_init(sc);
}
static void
lnc_free_mbufs(struct lnc_softc *sc)
{
int i;
/*
* We rely on other routines to keep the buff.mbuf field valid. If
* it's not NULL then we assume it points to an allocated mbuf.
*/
for (i = 0; i < NDESC(sc->nrdre); i++)
if ((sc->recv_ring + i)->buff.mbuf)
m_free((sc->recv_ring + i)->buff.mbuf);
for (i = 0; i < NDESC(sc->ntdre); i++)
if ((sc->trans_ring + i)->buff.mbuf)
m_free((sc->trans_ring + i)->buff.mbuf);
if (sc->mbuf_count)
m_freem(sc->mbufs);
}
static __inline int
alloc_mbuf_cluster(struct lnc_softc *sc, struct host_ring_entry *desc)
{
register struct mds *md = desc->md;
struct mbuf *m=0;
int addr;
/* Try and get cluster off local cache */
if (sc->mbuf_count) {
sc->mbuf_count--;
m = sc->mbufs;
sc->mbufs = m->m_next;
/* XXX m->m_data = m->m_ext.ext_buf;*/
} else {
MGET(m, M_DONTWAIT, MT_DATA);
if (!m)
return(1);
MCLGET(m, M_DONTWAIT);
if (!m->m_ext.ext_buf) {
m_free(m);
return(1);
}
}
desc->buff.mbuf = m;
addr = kvtop(m->m_data);
md->md0 = addr;
md->md1= ((addr >> 16) & 0xff) | OWN;
md->md2 = -(short)(MCLBYTES - sizeof(struct pkthdr));
md->md3 = 0;
return(0);
}
static __inline struct mbuf *
chain_mbufs(struct lnc_softc *sc, int start_of_packet, int pkt_len)
{
struct mbuf *head, *m;
struct host_ring_entry *desc;
/*
* Turn head into a pkthdr mbuf --
* assumes a pkthdr type mbuf was
* allocated to the descriptor
* originally.
*/
desc = sc->recv_ring + start_of_packet;
head = desc->buff.mbuf;
head->m_flags |= M_PKTHDR;
bzero(&head->m_pkthdr, sizeof(head->m_pkthdr));
m = head;
do {
m = desc->buff.mbuf;
m->m_len = min((MCLBYTES - sizeof(struct pkthdr)), pkt_len);
pkt_len -= m->m_len;
if (alloc_mbuf_cluster(sc, desc))
return((struct mbuf *)NULL);
INC_MD_PTR(start_of_packet, sc->nrdre)
desc = sc->recv_ring + start_of_packet;
m->m_next = desc->buff.mbuf;
} while (start_of_packet != sc->recv_next);
m->m_next = 0;
return(head);
}
static __inline struct mbuf *
mbuf_packet(struct lnc_softc *sc, int start_of_packet, int pkt_len)
{
struct host_ring_entry *start;
struct mbuf *head,*m,*m_prev;
char *data,*mbuf_data;
short blen;
int amount;
/* Get a pkthdr mbuf for the start of packet */
MGETHDR(head, M_DONTWAIT, MT_DATA);
if (!head) {
LNCSTATS(drop_packet)
return(0);
}
m = head;
m->m_len = 0;
start = sc->recv_ring + start_of_packet;
/*blen = -(start->md->md2);*/
blen = RECVBUFSIZE; /* XXX More PCnet-32 crap */
data = start->buff.data;
mbuf_data = m->m_data;
while (start_of_packet != sc->recv_next) {
/*
* If the data left fits in a single buffer then set
* blen to the size of the data left.
*/
if (pkt_len < blen)
blen = pkt_len;
/*
* amount is least of data in current ring buffer and
* amount of space left in current mbuf.
*/
amount = min(blen, M_TRAILINGSPACE(m));
if (amount == 0) {
/* mbuf must be empty */
m_prev = m;
MGET(m, M_DONTWAIT, MT_DATA);
if (!m) {
m_freem(head);
return(0);
}
if (pkt_len >= MINCLSIZE)
MCLGET(m, M_DONTWAIT);
m->m_len = 0;
m_prev->m_next = m;
amount = min(blen, M_TRAILINGSPACE(m));
mbuf_data = m->m_data;
}
bcopy(data, mbuf_data, amount);
blen -= amount;
pkt_len -= amount;
m->m_len += amount;
data += amount;
mbuf_data += amount;
if (blen == 0) {
start->md->md1 &= HADR;
start->md->md1 |= OWN;
start->md->md2 = -RECVBUFSIZE; /* XXX - shouldn't be necessary */
INC_MD_PTR(start_of_packet, sc->nrdre)
start = sc->recv_ring + start_of_packet;
data = start->buff.data;
/*blen = -(start->md->md2);*/
blen = RECVBUFSIZE; /* XXX More PCnet-32 crap */
}
}
return(head);
}
static void
lnc_rint(struct lnc_softc *sc)
{
struct ifnet *ifp = sc->ifp;
struct host_ring_entry *next, *start;
int start_of_packet;
struct mbuf *head;
struct ether_header *eh;
int lookahead;
int flags;
int pkt_len;
/*
* The LANCE will issue a RINT interrupt when the ownership of the
* last buffer of a receive packet has been relinquished by the LANCE.
* Therefore, it can be assumed that a complete packet can be found
* before hitting buffers that are still owned by the LANCE, if not
* then there is a bug in the driver that is causing the descriptors
* to get out of sync.
*/
#ifdef DIAGNOSTIC
if ((sc->recv_ring + sc->recv_next)->md->md1 & OWN) {
log(LOG_ERR, "%s: Receive interrupt with buffer still owned by controller -- Resetting\n", ifp->if_xname);
lnc_reset(sc);
return;
}
if (!((sc->recv_ring + sc->recv_next)->md->md1 & STP)) {
log(LOG_ERR, "%s: Receive interrupt but not start of packet -- Resetting\n", ifp->if_xname);
lnc_reset(sc);
return;
}
#endif
lookahead = 0;
next = sc->recv_ring + sc->recv_next;
while ((flags = next->md->md1) & STP) {
/* Make a note of the start of the packet */
start_of_packet = sc->recv_next;
/*
* Find the end of the packet. Even if not data chaining,
* jabber packets can overrun into a second descriptor.
* If there is no error, then the ENP flag is set in the last
* descriptor of the packet. If there is an error then the ERR
* flag will be set in the descriptor where the error occured.
* Therefore, to find the last buffer of a packet we search for
* either ERR or ENP.
*/
if (!(flags & (ENP | MDERR))) {
do {
INC_MD_PTR(sc->recv_next, sc->nrdre)
next = sc->recv_ring + sc->recv_next;
flags = next->md->md1;
} while (!(flags & (STP | OWN | ENP | MDERR)));
if (flags & STP) {
log(LOG_ERR, "%s: Start of packet found before end of previous in receive ring -- Resetting\n", ifp->if_xname);
lnc_reset(sc);
return;
}
if (flags & OWN) {
if (lookahead) {
/*
* Looked ahead into a packet still
* being received
*/
sc->recv_next = start_of_packet;
break;
} else {
log(LOG_ERR, "%s: End of received packet not found-- Resetting\n", ifp->if_xname);
lnc_reset(sc);
return;
}
}
}
pkt_len = (next->md->md3 & MCNT) - FCS_LEN;
/* Move pointer onto start of next packet */
INC_MD_PTR(sc->recv_next, sc->nrdre)
next = sc->recv_ring + sc->recv_next;
if (flags & MDERR) {
const char *if_xname = ifp->if_xname;
if (flags & RBUFF) {
LNCSTATS(rbuff)
log(LOG_ERR, "%s: Receive buffer error\n", if_xname);
}
if (flags & OFLO) {
/* OFLO only valid if ENP is not set */
if (!(flags & ENP)) {
LNCSTATS(oflo)
log(LOG_ERR, "%s: Receive overflow error \n", if_xname);
}
} else if (flags & ENP) {
if ((ifp->if_flags & IFF_PROMISC)==0) {
/*
* FRAM and CRC are valid only if ENP
* is set and OFLO is not.
*/
if (flags & FRAM) {
LNCSTATS(fram)
log(LOG_ERR, "%s: Framing error\n", if_xname);
/*
* FRAM is only set if there's a CRC
* error so avoid multiple messages
*/
} else if (flags & CRC) {
LNCSTATS(crc)
log(LOG_ERR, "%s: Receive CRC error\n", if_xname);
}
}
}
/* Drop packet */
LNCSTATS(rerr)
ifp->if_ierrors++;
while (start_of_packet != sc->recv_next) {
start = sc->recv_ring + start_of_packet;
start->md->md2 = -RECVBUFSIZE; /* XXX - shouldn't be necessary */
start->md->md1 &= HADR;
start->md->md1 |= OWN;
INC_MD_PTR(start_of_packet, sc->nrdre)
}
} else { /* Valid packet */
ifp->if_ipackets++;
if (sc->nic.mem_mode == DMA_MBUF)
head = chain_mbufs(sc, start_of_packet, pkt_len);
else
head = mbuf_packet(sc, start_of_packet, pkt_len);
if (head) {
/*
* First mbuf in packet holds the
* ethernet and packet headers
*/
head->m_pkthdr.rcvif = ifp;
head->m_pkthdr.len = pkt_len ;
eh = (struct ether_header *) head->m_data;
/*
* vmware ethernet hardware emulation loops
* packets back to itself, violates IFF_SIMPLEX.
* drop it if it is from myself.
*/
if (bcmp(eh->ether_shost,
IF_LLADDR(sc->ifp), ETHER_ADDR_LEN) == 0) {
m_freem(head);
} else {
(*ifp->if_input)(ifp, head);
}
} else {
log(LOG_ERR,"%s: Packet dropped, no mbufs\n",ifp->if_xname);
LNCSTATS(drop_packet)
}
}
lookahead++;
}
/*
* At this point all completely received packets have been processed
* so clear RINT since any packets that have arrived while we were in
* here have been dealt with.
*/
lnc_outw(sc->rdp, RINT | INEA);
}
static void
lnc_tint(struct lnc_softc *sc)
{
struct host_ring_entry *next, *start;
int start_of_packet;
int lookahead;
/*
* If the driver is reset in this routine then we return immediately to
* the interrupt driver routine. Any interrupts that have occured
* since the reset will be dealt with there. sc->trans_next
* should point to the start of the first packet that was awaiting
* transmission after the last transmit interrupt was dealt with. The
* LANCE should have relinquished ownership of that descriptor before
* the interrupt. Therefore, sc->trans_next should point to a
* descriptor with STP set and OWN cleared. If not then the driver's
* pointers are out of sync with the LANCE, which signifies a bug in
* the driver. Therefore, the following two checks are really
* diagnostic, since if the driver is working correctly they should
* never happen.
*/
#ifdef DIAGNOSTIC
if ((sc->trans_ring + sc->trans_next)->md->md1 & OWN) {
log(LOG_ERR, "%s: Transmit interrupt with buffer still owned by controller -- Resetting\n", sc->ifp->if_xname);
lnc_reset(sc);
return;
}
#endif
/*
* The LANCE will write the status information for the packet it just
* tried to transmit in one of two places. If the packet was
* transmitted successfully then the status will be written into the
* last descriptor of the packet. If the transmit failed then the
* status will be written into the descriptor that was being accessed
* when the error occured and all subsequent descriptors in that
* packet will have been relinquished by the LANCE.
*
* At this point we know that sc->trans_next points to the start
* of a packet that the LANCE has just finished trying to transmit.
* We now search for a buffer with either ENP or ERR set.
*/
lookahead = 0;
do {
start_of_packet = sc->trans_next;
next = sc->trans_ring + sc->trans_next;
#ifdef DIAGNOSTIC
if (!(next->md->md1 & STP)) {
log(LOG_ERR, "%s: Transmit interrupt but not start of packet -- Resetting\n", sc->ifp->if_xname);
lnc_reset(sc);
return;
}
#endif
/*
* Find end of packet.
*/
if (!(next->md->md1 & (ENP | MDERR))) {
do {
INC_MD_PTR(sc->trans_next, sc->ntdre)
next = sc->trans_ring + sc->trans_next;
} while (!(next->md->md1 & (STP | OWN | ENP | MDERR)));
if (next->md->md1 & STP) {
log(LOG_ERR, "%s: Start of packet found before end of previous in transmit ring -- Resetting\n", sc->ifp->if_xname);
lnc_reset(sc);
return;
}
if (next->md->md1 & OWN) {
if (lookahead) {
/*
* Looked ahead into a packet still
* being transmitted
*/
sc->trans_next = start_of_packet;
break;
} else {
log(LOG_ERR, "%s: End of transmitted packet not found -- Resetting\n", sc->ifp->if_xname);
lnc_reset(sc);
return;
}
}
}
/*
* Check for ERR first since other flags are irrelevant if an
* error occurred.
*/
if (next->md->md1 & MDERR) {
LNCSTATS(terr)
sc->ifp->if_oerrors++;
if (next->md->md3 & LCOL) {
LNCSTATS(lcol)
log(LOG_ERR, "%s: Transmit late collision -- Net error?\n", sc->ifp->if_xname);
sc->ifp->if_collisions++;
/*
* Clear TBUFF since it's not valid when LCOL
* set
*/
next->md->md3 &= ~TBUFF;
}
if (next->md->md3 & LCAR) {
LNCSTATS(lcar)
log(LOG_ERR, "%s: Loss of carrier during transmit -- Net error?\n", sc->ifp->if_xname);
}
if (next->md->md3 & RTRY) {
LNCSTATS(rtry)
log(LOG_ERR, "%s: Transmit of packet failed after 16 attempts -- TDR = %d\n", sc->ifp->if_xname, ((sc->trans_ring + sc->trans_next)->md->md3 & TDR));
sc->ifp->if_collisions += 16;
/*
* Clear TBUFF since it's not valid when RTRY
* set
*/
next->md->md3 &= ~TBUFF;
}
/*
* TBUFF is only valid if neither LCOL nor RTRY are set.
* We need to check UFLO after LCOL and RTRY so that we
* know whether or not TBUFF is valid. If either are
* set then TBUFF will have been cleared above. A
* UFLO error will turn off the transmitter so we
* have to reset.
*
*/
if (next->md->md3 & UFLO) {
LNCSTATS(uflo)
/*
* If an UFLO has occured it's possibly due
* to a TBUFF error
*/
if (next->md->md3 & TBUFF) {
LNCSTATS(tbuff)
log(LOG_ERR, "%s: Transmit buffer error -- Resetting\n", sc->ifp->if_xname);
} else
log(LOG_ERR, "%s: Transmit underflow error -- Resetting\n", sc->ifp->if_xname);
lnc_reset(sc);
return;
}
do {
INC_MD_PTR(sc->trans_next, sc->ntdre)
next = sc->trans_ring + sc->trans_next;
} while (!(next->md->md1 & STP) && (sc->trans_next != sc->next_to_send));
} else {
/*
* Since we check for ERR first then if we get here
* the packet was transmitted correctly. There may
* still have been non-fatal errors though.
* Don't bother checking for DEF, waste of time.
*/
sc->ifp->if_opackets++;
if (next->md->md1 & MORE) {
LNCSTATS(more)
sc->ifp->if_collisions += 2;
}
/*
* ONE is invalid if LCOL is set. If LCOL was set then
* ERR would have also been set and we would have
* returned from lnc_tint above. Therefore we can
* assume if we arrive here that ONE is valid.
*
*/
if (next->md->md1 & ONE) {
LNCSTATS(one)
sc->ifp->if_collisions++;
}
INC_MD_PTR(sc->trans_next, sc->ntdre)
next = sc->trans_ring + sc->trans_next;
}
/*
* Clear descriptors and free any mbufs.
*/
do {
start = sc->trans_ring + start_of_packet;
start->md->md1 &= HADR;
if (sc->nic.mem_mode == DMA_MBUF) {
/* Cache clusters on a local queue */
if ((start->buff.mbuf->m_flags & M_EXT) && (sc->mbuf_count < MBUF_CACHE_LIMIT)) {
if (sc->mbuf_count) {
start->buff.mbuf->m_next = sc->mbufs;
sc->mbufs = start->buff.mbuf;
} else
sc->mbufs = start->buff.mbuf;
sc->mbuf_count++;
start->buff.mbuf = 0;
} else {
/*
* XXX should this be m_freem()?
*/
m_free(start->buff.mbuf);
start->buff.mbuf = NULL;
}
}
sc->pending_transmits--;
INC_MD_PTR(start_of_packet, sc->ntdre)
}while (start_of_packet != sc->trans_next);
/*
* There's now at least one free descriptor
* in the ring so indicate that we can accept
* more packets again.
*/
sc->ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
lookahead++;
} while (sc->pending_transmits && !(next->md->md1 & OWN));
/*
* Clear TINT since we've dealt with all
* the completed transmissions.
*/
lnc_outw(sc->rdp, TINT | INEA);
}
int
lnc_attach_common(device_t dev)
{
lnc_softc_t *sc = device_get_softc(dev);
int i;
int skip;
u_char eaddr[6];
switch (sc->nic.ident) {
case BICC:
case CNET98S:
skip = 2;
break;
default:
skip = 1;
break;
}
/* Set default mode */
sc->nic.mode = NORMAL;
/* Fill in arpcom structure entries */
sc->ifp = if_alloc(IFT_ETHER);
if (sc->ifp == NULL) {
device_printf(dev, "can not if_alloc()\n");
return (0);
}
sc->ifp->if_softc = sc;
if_initname(sc->ifp, device_get_name(dev),
device_get_unit(dev));
sc->ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX |
IFF_MULTICAST | IFF_NEEDSGIANT;
sc->ifp->if_timer = 0;
sc->ifp->if_start = lnc_start;
sc->ifp->if_ioctl = lnc_ioctl;
sc->ifp->if_watchdog = lnc_watchdog;
sc->ifp->if_init = lnc_init;
IFQ_SET_MAXLEN(&sc->ifp->if_snd, IFQ_MAXLEN);
sc->ifp->if_snd.ifq_drv_maxlen = IFQ_MAXLEN;
IFQ_SET_READY(&sc->ifp->if_snd);
/* Extract MAC address from PROM */
for (i = 0; i < ETHER_ADDR_LEN; i++)
eaddr[i] = lnc_inb(i * skip);
ether_ifattach(sc->ifp, eaddr);
if (sc->nic.ic == LANCE || sc->nic.ic == C_LANCE)
if_printf(sc->ifp, "%s (%s)\n",
nic_ident[sc->nic.ident], ic_ident[sc->nic.ic]);
else
if_printf(sc->ifp, "%s\n", ic_ident[sc->nic.ic]);
return (1);
}
int
lnc_detach_common(device_t dev) {
lnc_softc_t *sc = device_get_softc(dev);
int s = splimp();
ether_ifdetach(sc->ifp);
lnc_stop(sc);
lnc_release_resources(dev);
splx(s);
return (0);
}
static void
lnc_init(xsc)
void *xsc;
{
struct lnc_softc *sc = xsc;
int s, i;
char *lnc_mem;
/* Shut down interface */
s = splimp();
lnc_stop(sc);
sc->ifp->if_flags |= IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; /* XXX??? */
/*
* This sets up the memory area for the controller. Memory is set up for
* the initialisation block (12 words of contiguous memory starting
* on a word boundary),the transmit and receive ring structures (each
* entry is 4 words long and must start on a quadword boundary) and
* the data buffers.
*
* The alignment tests are particularly paranoid.
*/
sc->recv_next = 0;
sc->trans_ring = sc->recv_ring + NDESC(sc->nrdre);
sc->trans_next = 0;
if (sc->nic.mem_mode == SHMEM)
lnc_mem = (char *)(uintptr_t)sc->nic.iobase;
else
lnc_mem = (char *) (sc->trans_ring + NDESC(sc->ntdre));
lnc_mem = (char *)(((long)lnc_mem + 1) & ~1);
sc->init_block = (struct init_block *) ((long) lnc_mem & ~1);
lnc_mem = (char *) (sc->init_block + 1);
lnc_mem = (char *)(((long)lnc_mem + 7) & ~7);
/* Initialise pointers to descriptor entries */
for (i = 0; i < NDESC(sc->nrdre); i++) {
(sc->recv_ring + i)->md = (struct mds *) lnc_mem;
lnc_mem += sizeof(struct mds);
}
for (i = 0; i < NDESC(sc->ntdre); i++) {
(sc->trans_ring + i)->md = (struct mds *) lnc_mem;
lnc_mem += sizeof(struct mds);
}
/* Initialise the remaining ring entries */
if (sc->nic.mem_mode == DMA_MBUF) {
sc->mbufs = 0;
sc->mbuf_count = 0;
/* Free previously allocated mbufs */
if (sc->flags & LNC_INITIALISED)
lnc_free_mbufs(sc);
for (i = 0; i < NDESC(sc->nrdre); i++) {
if (alloc_mbuf_cluster(sc, sc->recv_ring+i)) {
log(LOG_ERR, "Initialisation failed -- no mbufs\n");
splx(s);
return;
}
}
for (i = 0; i < NDESC(sc->ntdre); i++) {
(sc->trans_ring + i)->buff.mbuf = 0;
(sc->trans_ring + i)->md->md0 = 0;
(sc->trans_ring + i)->md->md1 = 0;
(sc->trans_ring + i)->md->md2 = 0;
(sc->trans_ring + i)->md->md3 = 0;
}
} else {
for (i = 0; i < NDESC(sc->nrdre); i++) {
(sc->recv_ring + i)->md->md0 = kvtop(lnc_mem);
(sc->recv_ring + i)->md->md1 = ((kvtop(lnc_mem) >> 16) & 0xff) | OWN;
(sc->recv_ring + i)->md->md2 = -RECVBUFSIZE;
(sc->recv_ring + i)->md->md3 = 0;
(sc->recv_ring + i)->buff.data = lnc_mem;
lnc_mem += RECVBUFSIZE;
}
for (i = 0; i < NDESC(sc->ntdre); i++) {
(sc->trans_ring + i)->md->md0 = kvtop(lnc_mem);
(sc->trans_ring + i)->md->md1 = ((kvtop(lnc_mem) >> 16) & 0xff);
(sc->trans_ring + i)->md->md2 = 0;
(sc->trans_ring + i)->md->md3 = 0;
(sc->trans_ring + i)->buff.data = lnc_mem;
lnc_mem += TRANSBUFSIZE;
}
}
sc->next_to_send = 0;
/* Set up initialisation block */
sc->init_block->mode = sc->nic.mode;
for (i = 0; i < ETHER_ADDR_LEN; i++)
sc->init_block->padr[i] = IF_LLADDR(sc->ifp)[i];
lnc_setladrf(sc);
sc->init_block->rdra = kvtop(sc->recv_ring->md);
sc->init_block->rlen = ((kvtop(sc->recv_ring->md) >> 16) & 0xff) | (sc->nrdre << 13);
sc->init_block->tdra = kvtop(sc->trans_ring->md);
sc->init_block->tlen = ((kvtop(sc->trans_ring->md) >> 16) & 0xff) | (sc->ntdre << 13);
/* Set flags to show that the memory area is valid */
sc->flags |= LNC_INITIALISED;
sc->pending_transmits = 0;
/* Give the LANCE the physical address of the initialisation block */
if (sc->nic.ic == PCnet_Home) {
u_short media;
/* Set PHY_SEL to HomeRun */
media = read_bcr(sc, BCR49);
media &= ~3;
media |= 1;
write_bcr(sc, BCR49, media);
}
write_csr(sc, CSR1, kvtop(sc->init_block));
write_csr(sc, CSR2, (kvtop(sc->init_block) >> 16) & 0xff);
/*
* Depending on which controller this is, CSR3 has different meanings.
* For the Am7990 it controls DMA operations, for the Am79C960 it
* controls interrupt masks and transmitter algorithms. In either
* case, none of the flags are set.
*
*/
write_csr(sc, CSR3, 0);
/* Let's see if it starts */
/*
printf("Enabling lnc interrupts\n");
sc->ifp->if_timer = 10;
write_csr(sc, CSR0, INIT|INEA);
*/
/*
* Now that the initialisation is complete there's no reason to
* access anything except CSR0, so we leave RAP pointing there
* so we can just access RDP from now on, saving an outw each
* time.
*/
write_csr(sc, CSR0, INIT);
for(i=0; i < 1000; i++)
if (read_csr(sc, CSR0) & IDON)
break;
if (read_csr(sc, CSR0) & IDON) {
/*
* Enable interrupts, start the LANCE, mark the interface as
* running and transmit any pending packets.
*/
write_csr(sc, CSR0, STRT | INEA);
sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
sc->ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
lnc_start(sc->ifp);
} else
log(LOG_ERR, "%s: Initialisation failed\n",
sc->ifp->if_xname);
splx(s);
}
/*
* The interrupt flag (INTR) will be set and provided that the interrupt enable
* flag (INEA) is also set, the interrupt pin will be driven low when any of
* the following occur:
*
* 1) Completion of the initialisation routine (IDON). 2) The reception of a
* packet (RINT). 3) The transmission of a packet (TINT). 4) A transmitter
* timeout error (BABL). 5) A missed packet (MISS). 6) A memory error (MERR).
*
* The interrupt flag is cleared when all of the above conditions are cleared.
*
* If the driver is reset from this routine then it first checks to see if any
* interrupts have ocurred since the reset and handles them before returning.
* This is because the NIC may signify a pending interrupt in CSR0 using the
* INTR flag even if a hardware interrupt is currently inhibited (at least I
* think it does from reading the data sheets). We may as well deal with
* these pending interrupts now rather than get the overhead of another
* hardware interrupt immediately upon returning from the interrupt handler.
*
*/
void
lncintr(void *arg)
{
lnc_softc_t *sc = arg;
u_short csr0;
/*
* INEA is the only bit that can be cleared by writing a 0 to it so
* we have to include it in any writes that clear other flags.
*/
while ((csr0 = lnc_inw(sc->rdp)) & INTR) {
/*
* Clear interrupt flags early to avoid race conditions. The
* controller can still set these flags even while we're in
* this interrupt routine. If the flag is still set from the
* event that caused this interrupt any new events will
* be missed.
*/
lnc_outw(sc->rdp, csr0);
/*lnc_outw(sc->rdp, IDON | CERR | BABL | MISS | MERR | RINT | TINT | INEA);*/
#ifdef notyet
if (csr0 & IDON) {
printf("IDON\n");
sc->ifp->if_timer = 0;
write_csr(sc, CSR0, STRT | INEA);
sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
sc->ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
lnc_start(sc->ifp);
continue;
}
#endif
if (csr0 & ERR) {
if (csr0 & CERR) {
log(LOG_ERR, "%s: Heartbeat error -- SQE test failed\n", sc->ifp->if_xname);
LNCSTATS(cerr)
}
if (csr0 & BABL) {
log(LOG_ERR, "%s: Babble error - more than 1519 bytes transmitted\n", sc->ifp->if_xname);
LNCSTATS(babl)
sc->ifp->if_oerrors++;
}
if (csr0 & MISS) {
log(LOG_ERR, "%s: Missed packet -- no receive buffer\n", sc->ifp->if_xname);
LNCSTATS(miss)
sc->ifp->if_ierrors++;
}
if (csr0 & MERR) {
log(LOG_ERR, "%s: Memory error -- Resetting\n", sc->ifp->if_xname);
LNCSTATS(merr)
lnc_reset(sc);
continue;
}
}
if (csr0 & RINT) {
LNCSTATS(rint)
lnc_rint(sc);
}
if (csr0 & TINT) {
LNCSTATS(tint)
sc->ifp->if_timer = 0;
lnc_tint(sc);
}
/*
* If there's room in the transmit descriptor ring then queue
* some more transmit packets.
*/
if (!(sc->ifp->if_drv_flags & IFF_DRV_OACTIVE))
lnc_start(sc->ifp);
}
}
static __inline int
mbuf_to_buffer(struct mbuf *m, char *buffer)
{
int len=0;
for( ; m; m = m->m_next) {
bcopy(mtod(m, caddr_t), buffer, m->m_len);
buffer += m->m_len;
len += m->m_len;
}
return(len);
}
static __inline struct mbuf *
chain_to_cluster(struct mbuf *m)
{
struct mbuf *new;
MGET(new, M_DONTWAIT, MT_DATA);
if (new) {
MCLGET(new, M_DONTWAIT);
if (new->m_ext.ext_buf) {
new->m_len = mbuf_to_buffer(m, new->m_data);
m_freem(m);
return(new);
} else
m_free(new);
}
return(0);
}
/*
* IFF_DRV_OACTIVE and IFF_DRV_RUNNING are checked in ether_output so it's
* redundant to check them again since we wouldn't have got here if they were
* not appropriately set. This is also called from lnc_init and lncintr but the
* flags should be ok at those points too.
*/
static void
lnc_start(struct ifnet *ifp)
{
struct lnc_softc *sc = ifp->if_softc;
struct host_ring_entry *desc;
int tmp;
int end_of_packet;
struct mbuf *head, *m;
int len, chunk;
int addr;
int no_entries_needed;
do {
IFQ_DRV_DEQUEUE(&sc->ifp->if_snd, head);
if (!head)
return;
if (sc->nic.mem_mode == DMA_MBUF) {
no_entries_needed = 0;
for (m=head; m; m = m->m_next)
no_entries_needed++;
/*
* We try and avoid bcopy as much as possible
* but there are two cases when we use it.
*
* 1) If there are not enough free entries in the ring
* to hold each mbuf in the chain then compact the
* chain into a single cluster.
*
* 2) The Am7990 and Am79C90 must not have less than
* 100 bytes in the first descriptor of a chained
* packet so it's necessary to shuffle the mbuf
* contents to ensure this.
*/
if (no_entries_needed > (NDESC(sc->ntdre) - sc->pending_transmits)) {
if (!(head = chain_to_cluster(head))) {
log(LOG_ERR, "%s: Couldn't get mbuf for transmit packet -- Resetting \n ",ifp->if_xname);
lnc_reset(sc);
return;
}
} else if ((sc->nic.ic == LANCE) || (sc->nic.ic == C_LANCE)) {
if ((head->m_len < 100) && (head->m_next)) {
len = 100 - head->m_len;
if (M_TRAILINGSPACE(head) < len) {
/*
* Move data to start of data
* area. We assume the first
* mbuf has a packet header
* and is not a cluster.
*/
bcopy((caddr_t)head->m_data, (caddr_t)head->m_pktdat, head->m_len);
head->m_data = head->m_pktdat;
}
m = head->m_next;
while (m && (len > 0)) {
chunk = min(len, m->m_len);
bcopy(mtod(m, caddr_t), mtod(head, caddr_t) + head->m_len, chunk);
len -= chunk;
head->m_len += chunk;
m->m_len -= chunk;
m->m_data += chunk;
if (m->m_len <= 0) {
m = m_free(m);
head->m_next = m;
}
}
}
}
tmp = sc->next_to_send;
/*
* On entering this loop we know that tmp points to a
* descriptor with a clear OWN bit.
*/
desc = sc->trans_ring + tmp;
len = ETHER_MIN_LEN;
for (m = head; m; m = m->m_next) {
desc->buff.mbuf = m;
addr = kvtop(m->m_data);
desc->md->md0 = addr;
desc->md->md1 = ((addr >> 16) & 0xff);
desc->md->md3 = 0;
desc->md->md2 = -m->m_len;
sc->pending_transmits++;
len -= m->m_len;
INC_MD_PTR(tmp, sc->ntdre)
desc = sc->trans_ring + tmp;
}
end_of_packet = tmp;
DEC_MD_PTR(tmp, sc->ntdre)
desc = sc->trans_ring + tmp;
desc->md->md1 |= ENP;
if (len > 0)
desc->md->md2 -= len;
/*
* Set OWN bits in reverse order, otherwise the Lance
* could start sending the packet before all the
* buffers have been relinquished by the host.
*/
while (tmp != sc->next_to_send) {
desc->md->md1 |= OWN;
DEC_MD_PTR(tmp, sc->ntdre)
desc = sc->trans_ring + tmp;
}
sc->next_to_send = end_of_packet;
desc->md->md1 |= STP | OWN;
} else {
sc->pending_transmits++;
desc = sc->trans_ring + sc->next_to_send;
len = mbuf_to_buffer(head, desc->buff.data);
desc->md->md3 = 0;
desc->md->md2 = -max(len, ETHER_MIN_LEN - ETHER_CRC_LEN);
desc->md->md1 |= OWN | STP | ENP;
INC_MD_PTR(sc->next_to_send, sc->ntdre)
}
/* Force an immediate poll of the transmit ring */
lnc_outw(sc->rdp, TDMD | INEA);
/*
* Set a timer so if the buggy Am7990.h shuts
* down we can wake it up.
*/
ifp->if_timer = 2;
BPF_MTAP(sc->ifp, head);
if (sc->nic.mem_mode != DMA_MBUF)
m_freem(head);
} while (sc->pending_transmits < NDESC(sc->ntdre));
/*
* Transmit ring is full so set IFF_DRV_OACTIVE
* since we can't buffer any more packets.
*/
sc->ifp->if_drv_flags |= IFF_DRV_OACTIVE;
LNCSTATS(trans_ring_full)
}
static int
lnc_ioctl(struct ifnet * ifp, u_long command, caddr_t data)
{
struct lnc_softc *sc = ifp->if_softc;
int s, error = 0;
s = splimp();
switch (command) {
case SIOCSIFFLAGS:
#ifdef DEBUG
if (ifp->if_flags & IFF_DEBUG)
sc->lnc_debug = 1;
else
sc->lnc_debug = 0;
#endif
if (ifp->if_flags & IFF_PROMISC) {
if (!(sc->nic.mode & PROM)) {
sc->nic.mode |= PROM;
lnc_init(sc);
}
} else if (sc->nic.mode & PROM) {
sc->nic.mode &= ~PROM;
lnc_init(sc);
}
if ((ifp->if_flags & IFF_ALLMULTI) &&
!(sc->flags & LNC_ALLMULTI)) {
sc->flags |= LNC_ALLMULTI;
lnc_init(sc);
} else if (!(ifp->if_flags & IFF_ALLMULTI) &&
(sc->flags & LNC_ALLMULTI)) {
sc->flags &= ~LNC_ALLMULTI;
lnc_init(sc);
}
if ((ifp->if_flags & IFF_UP) == 0 &&
(ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
/*
* If interface is marked down and it is running,
* then stop it.
*/
lnc_stop(sc);
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
} else if ((ifp->if_flags & IFF_UP) != 0 &&
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
/*
* If interface is marked up and it is stopped, then
* start it.
*/
lnc_init(sc);
}
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
lnc_init(sc);
error = 0;
break;
default:
error = ether_ioctl(ifp, command, data);
break;
}
(void) splx(s);
return error;
}
static void
lnc_watchdog(struct ifnet *ifp)
{
log(LOG_ERR, "%s: Device timeout -- Resetting\n", ifp->if_xname);
ifp->if_oerrors++;
lnc_reset(ifp->if_softc);
}
#ifdef DEBUG
void
lnc_dump_state(struct lnc_softc *sc)
{
int i;
printf("\nDriver/NIC [%s] state dump\n", sc->ifp->if_xname);
printf("Memory access mode: %b\n", sc->nic.mem_mode, MEM_MODES);
printf("Host memory\n");
printf("-----------\n");
printf("Receive ring: base = %p, next = %p\n",
(void *)sc->recv_ring, (void *)(sc->recv_ring + sc->recv_next));
for (i = 0; i < NDESC(sc->nrdre); i++)
printf("\t%d:%p md = %p buff = %p\n",
i, (void *)(sc->recv_ring + i),
(void *)(sc->recv_ring + i)->md,
(void *)(sc->recv_ring + i)->buff.data);
printf("Transmit ring: base = %p, next = %p\n",
(void *)sc->trans_ring, (void *)(sc->trans_ring + sc->trans_next));
for (i = 0; i < NDESC(sc->ntdre); i++)
printf("\t%d:%p md = %p buff = %p\n",
i, (void *)(sc->trans_ring + i),
(void *)(sc->trans_ring + i)->md,
(void *)(sc->trans_ring + i)->buff.data);
printf("Lance memory (may be on host(DMA) or card(SHMEM))\n");
printf("Init block = %p\n", (void *)sc->init_block);
printf("\tmode = %b rlen:rdra = %x:%x tlen:tdra = %x:%x\n",
sc->init_block->mode, INIT_MODE, sc->init_block->rlen,
sc->init_block->rdra, sc->init_block->tlen, sc->init_block->tdra);
printf("Receive descriptor ring\n");
for (i = 0; i < NDESC(sc->nrdre); i++)
printf("\t%d buffer = 0x%x%x, BCNT = %d,\tMCNT = %u,\tflags = %b\n",
i, ((sc->recv_ring + i)->md->md1 & HADR),
(sc->recv_ring + i)->md->md0,
-(short) (sc->recv_ring + i)->md->md2,
(sc->recv_ring + i)->md->md3,
(((sc->recv_ring + i)->md->md1 & ~HADR) >> 8), RECV_MD1);
printf("Transmit descriptor ring\n");
for (i = 0; i < NDESC(sc->ntdre); i++)
printf("\t%d buffer = 0x%x%x, BCNT = %d,\tflags = %b %b\n",
i, ((sc->trans_ring + i)->md->md1 & HADR),
(sc->trans_ring + i)->md->md0,
-(short) (sc->trans_ring + i)->md->md2,
((sc->trans_ring + i)->md->md1 >> 8), TRANS_MD1,
((sc->trans_ring + i)->md->md3 >> 10), TRANS_MD3);
printf("\nnext_to_send = %x\n", sc->next_to_send);
printf("\n CSR0 = %b CSR1 = %x CSR2 = %x CSR3 = %x\n\n",
read_csr(sc, CSR0), CSR0_FLAGS, read_csr(sc, CSR1),
read_csr(sc, CSR2), read_csr(sc, CSR3));
/* Set RAP back to CSR0 */
lnc_outw(sc->rap, CSR0);
}
void
mbuf_dump_chain(struct mbuf * m)
{
#define MBUF_FLAGS \
"\20\1M_EXT\2M_PKTHDR\3M_EOR\4UNKNOWN\5M_BCAST\6M_MCAST"
if (!m)
log(LOG_DEBUG, "m == NULL\n");
do {
log(LOG_DEBUG, "m = %p\n", (void *)m);
log(LOG_DEBUG, "m_hdr.mh_next = %p\n",
(void *)m->m_hdr.mh_next);
log(LOG_DEBUG, "m_hdr.mh_nextpkt = %p\n",
(void *)m->m_hdr.mh_nextpkt);
log(LOG_DEBUG, "m_hdr.mh_len = %d\n", m->m_hdr.mh_len);
log(LOG_DEBUG, "m_hdr.mh_data = %p\n",
(void *)m->m_hdr.mh_data);
log(LOG_DEBUG, "m_hdr.mh_type = %d\n", m->m_hdr.mh_type);
log(LOG_DEBUG, "m_hdr.mh_flags = %b\n", m->m_hdr.mh_flags,
MBUF_FLAGS);
if (!(m->m_hdr.mh_flags & (M_PKTHDR | M_EXT)))
log(LOG_DEBUG, "M_dat.M_databuf = %p\n",
(void *)m->M_dat.M_databuf);
else {
if (m->m_hdr.mh_flags & M_PKTHDR) {
log(LOG_DEBUG, "M_dat.MH.MH_pkthdr.len = %d\n",
m->M_dat.MH.MH_pkthdr.len);
log(LOG_DEBUG,
"M_dat.MH.MH_pkthdr.rcvif = %p\n",
(void *)m->M_dat.MH.MH_pkthdr.rcvif);
if (!(m->m_hdr.mh_flags & M_EXT))
log(LOG_DEBUG,
"M_dat.MH.MH_dat.MH_databuf = %p\n",
(void *)m->M_dat.MH.MH_dat.MH_databuf);
}
if (m->m_hdr.mh_flags & M_EXT) {
log(LOG_DEBUG,
"M_dat.MH.MH_dat.MH_ext.ext_buff %p\n",
(void *)m->M_dat.MH.MH_dat.MH_ext.ext_buf);
log(LOG_DEBUG,
"M_dat.MH.MH_dat.MH_ext.ext_free %p\n",
(void *)m->M_dat.MH.MH_dat.MH_ext.ext_free);
log(LOG_DEBUG,
"M_dat.MH.MH_dat.MH_ext.ext_size %d\n",
m->M_dat.MH.MH_dat.MH_ext.ext_size);
}
}
} while ((m = m->m_next) != NULL);
}
#endif