use 2k clusters on the rx ring instead of 9k clusters.

rge can chain mbufs from multiple rx descriptors together to build
a jumbo packet. 1514 byte packets are still the most common, so
avoiding the waste of most of a 9k cluster if we can still makes
sense.

diff --git a/sys/dev/pci/if_rge.c b/sys/dev/pci/if_rge.c
index 44ce8f0..b4a09c1 100644 (file)
--- a/sys/dev/pci/if_rge.c
+++ b/sys/dev/pci/if_rge.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: if_rge.c,v 1.32 2024/08/21 01:12:52 dlg Exp $ */
+/*     $OpenBSD: if_rge.c,v 1.33 2024/08/21 01:17:50 dlg Exp $ */
 
 /*
  * Copyright (c) 2019, 2020, 2023, 2024
@@ -566,7 +566,7 @@ rge_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
                break;
        case SIOCGIFRXR:
                error = if_rxr_ioctl((struct if_rxrinfo *)ifr->ifr_data,
-                   NULL, RGE_JUMBO_FRAMELEN, &sc->sc_queues->q_rx.rge_rx_ring);
+                   NULL, MCLBYTES, &sc->sc_queues->q_rx.rge_rx_ring);
                break;
        default:
                error = ether_ioctl(ifp, &sc->sc_arpcom, cmd, data);
@@ -1144,11 +1144,11 @@ rge_newbuf(struct rge_queues *q)
        uint32_t cmdsts;
        int idx;
 
-       m = MCLGETL(NULL, M_DONTWAIT, RGE_JUMBO_FRAMELEN);
+       m = MCLGETL(NULL, M_DONTWAIT, MCLBYTES);
        if (m == NULL)
                return (ENOBUFS);
 
-       m->m_len = m->m_pkthdr.len = RGE_JUMBO_FRAMELEN;
+       m->m_len = m->m_pkthdr.len = MCLBYTES;
 
        idx = q->q_rx.rge_rxq_prodidx;
        rxq = &q->q_rx.rge_rxq[idx];