d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Increase EM_MAX_SCATTER to 64 such that the size of em_xmit()::segs[EM_MAX_SCATTER]

Browse Source 
doesn't get overrun by things like NFS that can and do shove more than 32 segs when
being used with em(4) and TSO4.

Update tso handling code in em_xmit() with update from jhb@ in email thread:
https://lists.freebsd.org/pipermail/freebsd-net/2014-July/039306.html

set ifp->if_hw_tsomax, ifp->if_hw_tsomaxsegcount & ifp->if_hw_tsomaxsegsize
to appropriate values.

Define a TSO workaround "magic" number  of 4 that is used to avoid an
alignment issue in hardware.

Change a couple of integer values that were used as booleans to actual
bool types.

Ensure that em_enable_intr() enables the appropriate mask of interrupts
and not just a hardcoded define of values.

PR:		200221 199174 195078
Differential Revision:	https://reviews.freebsd.org/D3192
Reviewed by:	erj jhb hiren
MFC after:	2 weeks
Sponsored by:	Limelight Networks
This commit is contained in:
Sean Bruno 2015-08-16 19:43:44 +00:00
parent 842ad8ac0b
commit df40405fab
2 changed files with 82 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

148
sys/dev/e1000/if_em.c
View File

@ -364,6 +364,8 @@ MODULE_DEPEND(em, netmap, 1, 1, 1);
#define CSUM_TSO 0
#endif
#define TSO_WORKAROUND 4
static SYSCTL_NODE(_hw, OID_AUTO, em, CTLFLAG_RD, 0, "EM driver parameters");
static int em_disable_crc_stripping = 0;
@ -1876,13 +1878,15 @@ em_xmit(struct tx_ring *txr, struct mbuf **m_headp)
struct ether_header *eh;
struct ip *ip = NULL;
struct tcphdr *tp = NULL;
u32 txd_upper = 0, txd_lower = 0, txd_used = 0;
u32 txd_upper = 0, txd_lower = 0;
int ip_off, poff;
int nsegs, i, j, first, last = 0;
int error, do_tso, tso_desc = 0, remap = 1;
int error;
bool do_tso, tso_desc, remap = TRUE;
m_head = *m_headp;
do_tso = ((m_head->m_pkthdr.csum_flags & CSUM_TSO) != 0);
do_tso = (m_head->m_pkthdr.csum_flags & CSUM_TSO);
tso_desc = FALSE;
ip_off = poff = 0;
/*
@ -1918,74 +1922,82 @@ em_xmit(struct tx_ring *txr, struct mbuf **m_headp)
* for IPv6 yet.
*/
ip_off = sizeof(struct ether_header);
m_head = m_pullup(m_head, ip_off);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
eh = mtod(m_head, struct ether_header *);
if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
ip_off = sizeof(struct ether_vlan_header);
if (m_head->m_len < ip_off) {
m_head = m_pullup(m_head, ip_off);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
m_head = m_pullup(m_head, ip_off + sizeof(struct ip));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
eh = mtod(m_head, struct ether_header *);
if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
ip_off = sizeof(struct ether_vlan_header);
if (m_head->m_len < ip_off) {
m_head = m_pullup(m_head, ip_off);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
poff = ip_off + (ip->ip_hl << 2);
if (do_tso) {
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head->m_len < ip_off + sizeof(struct ip)) {
m_head = m_pullup(m_head, ip_off + sizeof(struct ip));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
poff = ip_off + (ip->ip_hl << 2);
if (do_tso || (m_head->m_pkthdr.csum_flags & CSUM_TCP)) {
if (m_head->m_len < poff + sizeof(struct tcphdr)) {
m_head = m_pullup(m_head, poff +
sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
/*
* TSO workaround:
* pull 4 more bytes of data into it.
*/
m_head = m_pullup(m_head, poff + (tp->th_off << 2) + 4);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
ip->ip_len = 0;
ip->ip_sum = 0;
/*
* The pseudo TCP checksum does not include TCP payload
* length so driver should recompute the checksum here
* what hardware expect to see. This is adherence of
* Microsoft's Large Send specification.
*/
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
tp->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
} else if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
m_head = m_pullup(m_head, poff + (tp->th_off << 2));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
if (m_head->m_len < poff + (tp->th_off << 2)) {
m_head = m_pullup(m_head, poff +
(tp->th_off << 2) +
TSO_WORKAROUND);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
if (do_tso) {
ip->ip_len = htons(m_head->m_pkthdr.tso_segsz +
(ip->ip_hl << 2) +
(tp->th_off << 2));
ip->ip_sum = 0;
/*
* The pseudo TCP checksum does not include TCP
* payload length so driver should recompute
* the checksum here what hardware expect to
* see. This is adherence of Microsoft's Large
* Send specification.
*/
tp->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
}
} else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
m_head = m_pullup(m_head, poff + sizeof(struct udphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
if (m_head->m_len < poff + sizeof(struct udphdr)) {
m_head = m_pullup(m_head, poff +
sizeof(struct udphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
}
@ -2031,7 +2043,7 @@ retry:
*m_headp = m;
/* Try it again, but only once */
remap = 0;
remap = FALSE;
goto retry;
} else if (error != 0) {
adapter->no_tx_dma_setup++;
@ -2046,13 +2058,13 @@ retry:
* it follows a TSO burst, then we need to add a
* sentinel descriptor to prevent premature writeback.
*/
if ((do_tso == 0) && (txr->tx_tso == TRUE)) {
if ((!do_tso) && (txr->tx_tso == TRUE)) {
if (nsegs == 1)
tso_desc = TRUE;
txr->tx_tso = FALSE;
}
if (nsegs > (txr->tx_avail - 2)) {
if (nsegs > (txr->tx_avail - EM_MAX_SCATTER)) {
txr->no_desc_avail++;
bus_dmamap_unload(txr->txtag, map);
return (ENOBUFS);
@ -2092,23 +2104,23 @@ retry:
** If this is the last descriptor, we want to
** split it so we have a small final sentinel
*/
if (tso_desc && (j == (nsegs -1)) && (seg_len > 8)) {
seg_len -= 4;
if (tso_desc && (j == (nsegs - 1)) && (seg_len > 8)) {
seg_len -= TSO_WORKAROUND;
ctxd->buffer_addr = htole64(seg_addr);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data =
htole32(txd_upper);
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data = htole32(txd_upper);
if (++i == adapter->num_tx_desc)
i = 0;
/* Now make the sentinel */
++txd_used; /* using an extra txd */
txr->tx_avail--;
ctxd = &txr->tx_base[i];
tx_buffer = &txr->tx_buffers[i];
ctxd->buffer_addr =
htole64(seg_addr + seg_len);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | 4);
adapter->txd_cmd | txd_lower | TSO_WORKAROUND);
ctxd->upper.data =
htole32(txd_upper);
last = i;
@ -2118,8 +2130,7 @@ retry:
ctxd->buffer_addr = htole64(seg_addr);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data =
htole32(txd_upper);
ctxd->upper.data = htole32(txd_upper);
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
@ -2130,8 +2141,6 @@ retry:
txr->next_avail_desc = i;
txr->tx_avail -= nsegs;
if (tso_desc) /* TSO used an extra for sentinel */
txr->tx_avail -= txd_used;
tx_buffer->m_head = m_head;
/*
@ -3034,6 +3043,11 @@ em_setup_interface(device_t dev, struct adapter *adapter)
if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
if_setioctlfn(ifp, em_ioctl);
if_setgetcounterfn(ifp, em_get_counter);
/* TSO parameters */
ifp->if_hw_tsomax = EM_TSO_SIZE;
ifp->if_hw_tsomaxsegcount = EM_MAX_SCATTER;
ifp->if_hw_tsomaxsegsize = EM_TSO_SEG_SIZE;
#ifdef EM_MULTIQUEUE
/* Multiqueue stack interface */
if_settransmitfn(ifp, em_mq_start);
@ -4893,8 +4907,8 @@ em_enable_intr(struct adapter *adapter)
u32 ims_mask = IMS_ENABLE_MASK;
if (hw->mac.type == e1000_82574) {
E1000_WRITE_REG(hw, EM_EIAC, EM_MSIX_MASK);
ims_mask |= EM_MSIX_MASK;
E1000_WRITE_REG(hw, EM_EIAC, adapter->ims);
ims_mask |= adapter->ims;
}
E1000_WRITE_REG(hw, E1000_IMS, ims_mask);
}

2
sys/dev/e1000/if_em.h
View File

@ -266,7 +266,7 @@
#define HW_DEBUGOUT1(S, A) if (DEBUG_HW) printf(S "\n", A)
#define HW_DEBUGOUT2(S, A, B) if (DEBUG_HW) printf(S "\n", A, B)
#define EM_MAX_SCATTER 32
#define EM_MAX_SCATTER 64
#define EM_VFTA_SIZE 128
#define EM_TSO_SIZE (65535 + sizeof(struct ether_vlan_header))
#define EM_TSO_SEG_SIZE 4096 /* Max dma segment size */