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

Limit the size of the posted receive buffers in Rx Rings to MJUMPAGESIZE. Previously for jumbo MTUs, the rx ring buffers were

Browse Source 
MTU + any required pad. Now when this size greater than MJUMPAGESIZE, the packet is spanned across multiple buffers and the
mbufs are stiched together.

Submitted by:gary.zambrano@qlogic.com
Approved by:davidcs@freebsd.org
This commit is contained in:
David C Somayajulu 2015-05-21 20:47:19 +00:00
parent 3e99786994
commit 3d60b1448f
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=283269
1 changed files with 79 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

106
sys/dev/bxe/bxe.c
View File

@ -3100,7 +3100,7 @@ static inline void
bxe_update_sge_prod(struct bxe_softc *sc,
struct bxe_fastpath *fp,
uint16_t sge_len,
struct eth_end_agg_rx_cqe *cqe)
union eth_sgl_or_raw_data *cqe)
{
uint16_t last_max, last_elem, first_elem;
uint16_t delta = 0;
@ -3113,17 +3113,17 @@ bxe_update_sge_prod(struct bxe_softc *sc,
/* first mark all used pages */
for (i = 0; i < sge_len; i++) {
BIT_VEC64_CLEAR_BIT(fp->sge_mask,
RX_SGE(le16toh(cqe->sgl_or_raw_data.sgl[i])));
RX_SGE(le16toh(cqe->sgl[i])));
}
BLOGD(sc, DBG_LRO,
"fp[%02d] fp_cqe->sgl[%d] = %d\n",
fp->index, sge_len - 1,
le16toh(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
le16toh(cqe->sgl[sge_len - 1]));
/* assume that the last SGE index is the biggest */
bxe_update_last_max_sge(fp,
le16toh(cqe->sgl_or_raw_data.sgl[sge_len - 1]));
le16toh(cqe->sgl[sge_len - 1]));
last_max = RX_SGE(fp->last_max_sge);
last_elem = last_max >> BIT_VEC64_ELEM_SHIFT;
@ -3238,6 +3238,53 @@ bxe_tpa_stop(struct bxe_softc *sc,
fp->rx_tpa_queue_used &= ~(1 << queue);
}
static uint8_t
bxe_service_rxsgl(
struct bxe_fastpath *fp,
uint16_t len,
uint16_t lenonbd,
struct mbuf *m,
struct eth_fast_path_rx_cqe *cqe_fp)
{
struct mbuf *m_frag;
uint16_t frags, frag_len;
uint16_t sge_idx = 0;
uint16_t j;
uint8_t i, rc = 0;
uint32_t frag_size;
/* adjust the mbuf */
m->m_len = lenonbd;
frag_size = len - lenonbd;
frags = SGE_PAGE_ALIGN(frag_size) >> SGE_PAGE_SHIFT;
for (i = 0, j = 0; i < frags; i += PAGES_PER_SGE, j++) {
sge_idx = RX_SGE(le16toh(cqe_fp->sgl_or_raw_data.sgl[j]));
m_frag = fp->rx_sge_mbuf_chain[sge_idx].m;
frag_len = min(frag_size, (uint32_t)(SGE_PAGE_SIZE));
m_frag->m_len = frag_len;
/* allocate a new mbuf for the SGE */
rc = bxe_alloc_rx_sge_mbuf(fp, sge_idx);
if (rc) {
/* Leave all remaining SGEs in the ring! */
return (rc);
}
fp->eth_q_stats.mbuf_alloc_sge--;
/* concatenate the fragment to the head mbuf */
m_cat(m, m_frag);
frag_size -= frag_len;
}
bxe_update_sge_prod(fp->sc, fp, frags, &cqe_fp->sgl_or_raw_data);
return rc;
}
static uint8_t
bxe_rxeof(struct bxe_softc *sc,
struct bxe_fastpath *fp)
@ -3278,7 +3325,7 @@ bxe_rxeof(struct bxe_softc *sc,
struct eth_fast_path_rx_cqe *cqe_fp;
uint8_t cqe_fp_flags;
enum eth_rx_cqe_type cqe_fp_type;
uint16_t len, pad;
uint16_t len, lenonbd, pad;
struct mbuf *m = NULL;
comp_ring_cons = RCQ(sw_cq_cons);
@ -3293,7 +3340,7 @@ bxe_rxeof(struct bxe_softc *sc,
BLOGD(sc, DBG_RX,
"fp[%02d] Rx hw_cq_cons=%d hw_sw_cons=%d "
"BD prod=%d cons=%d CQE type=0x%x err=0x%x "
"status=0x%x rss_hash=0x%x vlan=0x%x len=%u\n",
"status=0x%x rss_hash=0x%x vlan=0x%x len=%u lenonbd=%u\n",
fp->index,
hw_cq_cons,
sw_cq_cons,
@ -3304,7 +3351,8 @@ bxe_rxeof(struct bxe_softc *sc,
cqe_fp->status_flags,
le32toh(cqe_fp->rss_hash_result),
le16toh(cqe_fp->vlan_tag),
le16toh(cqe_fp->pkt_len_or_gro_seg_len));
le16toh(cqe_fp->pkt_len_or_gro_seg_len),
le16toh(cqe_fp->len_on_bd));
/* is this a slowpath msg? */
if (__predict_false(CQE_TYPE_SLOW(cqe_fp_type))) {
@ -3351,7 +3399,7 @@ bxe_rxeof(struct bxe_softc *sc,
bxe_tpa_stop(sc, fp, tpa_info, queue, pages,
&cqe->end_agg_cqe, comp_ring_cons);
bxe_update_sge_prod(sc, fp, pages, &cqe->end_agg_cqe);
bxe_update_sge_prod(sc, fp, pages, &cqe->end_agg_cqe.sgl_or_raw_data);
goto next_cqe;
}
@ -3367,6 +3415,7 @@ bxe_rxeof(struct bxe_softc *sc,
}
len = le16toh(cqe_fp->pkt_len_or_gro_seg_len);
lenonbd = le16toh(cqe_fp->len_on_bd);
pad = cqe_fp->placement_offset;
m = rx_buf->m;
@ -3413,6 +3462,12 @@ bxe_rxeof(struct bxe_softc *sc,
m_adj(m, pad);
m->m_pkthdr.len = m->m_len = len;
if (len != lenonbd){
rc = bxe_service_rxsgl(fp, len, lenonbd, m, cqe_fp);
if (rc)
break;
}
/* assign packet to this interface interface */
if_setrcvif(m, ifp);
@ -6210,30 +6265,27 @@ static void
bxe_set_fp_rx_buf_size(struct bxe_softc *sc)
{
int i;
uint32_t rx_buf_size;
BLOGD(sc, DBG_LOAD, "mtu = %d\n", sc->mtu);
rx_buf_size = (IP_HEADER_ALIGNMENT_PADDING + ETH_OVERHEAD + sc->mtu);
for (i = 0; i < sc->num_queues; i++) {
/* get the Rx buffer size for RX frames */
sc->fp[i].rx_buf_size =
(IP_HEADER_ALIGNMENT_PADDING +
ETH_OVERHEAD +
sc->mtu);
BLOGD(sc, DBG_LOAD, "rx_buf_size for fp[%02d] = %d\n",
i, sc->fp[i].rx_buf_size);
/* get the mbuf allocation size for RX frames */
if (sc->fp[i].rx_buf_size <= MCLBYTES) {
if(rx_buf_size <= MCLBYTES){
sc->fp[i].rx_buf_size = rx_buf_size;
sc->fp[i].mbuf_alloc_size = MCLBYTES;
}else if (rx_buf_size <= MJUMPAGESIZE){
sc->fp[i].rx_buf_size = rx_buf_size;
sc->fp[i].mbuf_alloc_size = MJUMPAGESIZE;
}else if (rx_buf_size <= (MJUMPAGESIZE + MCLBYTES)){
sc->fp[i].rx_buf_size = MCLBYTES;
sc->fp[i].mbuf_alloc_size = MCLBYTES;
}else if (rx_buf_size <= (2 * MJUMPAGESIZE)){
sc->fp[i].rx_buf_size = MJUMPAGESIZE;
sc->fp[i].mbuf_alloc_size = MJUMPAGESIZE;
}else {
sc->fp[i].rx_buf_size = MCLBYTES;
sc->fp[i].mbuf_alloc_size = MCLBYTES;
} else if (sc->fp[i].rx_buf_size <= BCM_PAGE_SIZE) {
sc->fp[i].mbuf_alloc_size = PAGE_SIZE;
} else {
sc->fp[i].mbuf_alloc_size = MJUM9BYTES;
}
BLOGD(sc, DBG_LOAD, "mbuf_alloc_size for fp[%02d] = %d\n",
i, sc->fp[i].mbuf_alloc_size);
}
}