On Mon, Jun 23, 2025 at 03:23:27PM +0100, Mark de Vries wrote:
> Greetings,
>
> I am also seeing increased latency and jitter using an Octeon system -
> Edgerouter 6p. It's the same issue for both OpenBSD 7.6 and 7.7. I don't see
> it on a amd64 machine with the same configuration.
>
> Attached is a graph showing on the left ping data collected with traffic
> going through a PC-Engines APU1 board with OpenBSD 7.7. To the right is
> traffic going through an Edgerouter 6p with OpenBSD 7.6 . It's the same for
> OpenBSD 7.7.
>
> In my case the config is with OpenBGPD on a vlan. This data is through NAT
> but it's the same without NAT and it's also the same using PPPoE rather than
> BGPD.
>
> From what I've read on the list it has been observed for Octeon, and also
> occasionally on other architectures. As my data shows there are cases where
> the amd64 architecture performs as expected.
>
> Is it time for a bug report?
>
> Here is some further info on my system, though note it's 7.6 at the moment,
> which I put on temporarily to check if it is unique to version 7.7, which it
> isn't.
>
>
> > pfctl -si
>
> Status: Enabled for 0 days 01:20:57 Debug: err
>
> State Table Total Rate
> current entries 10
> half-open tcp 0
> searches 67352 13.9/s
> inserts 1262 0.3/s
> removals 1252 0.3/s
> Counters
> match 15580 3.2/s
> bad-offset 0 0.0/s
> fragment 0 0.0/s
> short 0 0.0/s
> normalize 4 0.0/s
> memory 0 0.0/s
> bad-timestamp 0 0.0/s
> congestion 0 0.0/s
> ip-option 0 0.0/s
> proto-cksum 0 0.0/s
> state-mismatch 22 0.0/s
> state-insert 0 0.0/s
> state-limit 0 0.0/s
> src-limit 1 0.0/s
> synproxy 0 0.0/s
> translate 0 0.0/s
> no-route 0 0.0/s
>
> > netstat -m
> 1551 mbufs in use:
> 1538 mbufs allocated to data
> 5 mbufs allocated to packet headers
> 8 mbufs allocated to socket names and addresses
> 1536/1712 mbuf 2048 byte clusters in use (current/peak)
> 0/0 mbuf 2112 byte clusters in use (current/peak)
> 0/40 mbuf 4096 byte clusters in use (current/peak)
> 0/24 mbuf 8192 byte clusters in use (current/peak)
> 0/0 mbuf 9216 byte clusters in use (current/peak)
> 0/0 mbuf 12288 byte clusters in use (current/peak)
> 0/0 mbuf 16384 byte clusters in use (current/peak)
> 0/8 mbuf 65536 byte clusters in use (current/peak)
> 4672/4720/131072 Kbytes allocated to network (current/peak/max)
> 0 requests for memory denied
> 0 requests for memory delayed
> 0 calls to protocol drain routines
> 0 defrag mbuf allocation
> 1001 prepend mbuf allocation
> 0 pullup mbuf allocation
> 0 pullup memory copy
> 0 pulldown mbuf allocation
> 0 pulldown memory copy
>
> > systat vm
>
>
> 2 users Load 0.00 0.00 0.00 15:14:23
>
> memory totals (in KB) PAGING SWAPPING Interrupts
> real virtual free in out in out 576
> total
> Active 95280 95280 630160 ops soft
> All 380976 380976 841072 pages 528
> clock
> 3
> cnmac1
> Proc:r d s w Csw Trp Sys Int Sof Flt forks 6
> cnmac2
> 58 57 7 63 576 100 13 fkppw com0
> fksvm octmmc0
> 0.0%Int 0.0%Spn 0.1%Sys 0.0%Usr 99.9%Idle pwait 39 ipi
> | | | | | | | | | | | relck
> rlkok
> noram
> Namei Sys-cache Proc-cache No-cache 1 ndcpy
> Calls hits % hits % miss % fltcp
> 10 10 100 2 zfod
> cow
> Disks sd0 2106 fmin
> seeks 2808 ftarg
> xfers itarg
> speed 2 wired
> sec pdfre
> pdscn
> pzidl 14
> IPKTS
> 8 kmape 12
> OPKTS
>
> > systat mbufs
>
>
> cnmac4 2 users Load 0.00 0.00 0.00 15:15:15
>
> IFACE RING LIVELOCKS SIZE ALIVE LWM HWM CWM
> System mbufs 0 256 1551 27
> mcl2k 2048 1536 214
> mcl4k 4096 0 5
> mcl8k 8192 0 3
> mcl64k 65536 0 1
> lo0
> cnmac0
> cnmac1
> cnmac2
> cnmac3
> cnmac4
> cnmac5
> enc0
> vether0
> vlan100
>
> dmesg:
> root@hrr67:~$ dmesg
> [ using 773576 bytes of bsd ELF symbol table ]
> Copyright (c) 1982, 1986, 1989, 1991, 1993
> The Regents of the University of California. All rights reserved.
> Copyright (c) 1995-2024 OpenBSD. All rights reserved.
> https://www.OpenBSD.org
>
> OpenBSD 7.6 (GENERIC.MP) #233: Mon Sep 30 09:41:48 MDT 2024
> dera...@octeon.openbsd.org:/usr/src/sys/arch/octeon/compile/GENERIC.MP
> real mem = 1073741824 (1024MB)
> avail mem = 1035321344 (987MB)
> random: good seed from bootblocks
> mainbus0 at root: board 20300 rev 1.23, model cavium,ubnt_e300
> cpu0 at mainbus0: CN70xx/CN71xx CPU rev 0.2 1000 MHz, CN70xx/CN71xx FPU rev
> 0.0
> cpu0: cache L1-I 78KB 39 way D 32KB 32 way, L2 1024KB 8 way
> cpu1 at mainbus0: CN70xx/CN71xx CPU rev 0.2 1000 MHz, CN70xx/CN71xx FPU rev
> 0.0
> cpu1: cache L1-I 78KB 39 way D 32KB 32 way, L2 1024KB 8 way
> cpu2 at mainbus0: CN70xx/CN71xx CPU rev 0.2 1000 MHz, CN70xx/CN71xx FPU rev
> 0.0
> cpu2: cache L1-I 78KB 39 way D 32KB 32 way, L2 1024KB 8 way
> cpu3 at mainbus0: CN70xx/CN71xx CPU rev 0.2 1000 MHz, CN70xx/CN71xx FPU rev
> 0.0
> cpu3: cache L1-I 78KB 39 way D 32KB 32 way, L2 1024KB 8 way
> clock0 at mainbus0: int 5
> octcrypto0 at mainbus0
> iobus0 at mainbus0
> simplebus0 at iobus0: "soc"
> "bootbus" at simplebus0 not configured
> octciu0 at simplebus0
> octcib0 at simplebus0: max-bits 23
> octcib1 at simplebus0: max-bits 12
> octcib2 at simplebus0: max-bits 6
> octcib3 at simplebus0: max-bits 15
> octcib4 at simplebus0: max-bits 4
> octcib5 at simplebus0: max-bits 11
> octcib6 at simplebus0: max-bits 11
> octgpio0 at simplebus0: 20 pins, xbit 16
> octsmi0 at simplebus0
> octsmi1 at simplebus0
> octpip0 at simplebus0
> octgmx0 at octpip0 interface 0
> cnmac0 at octgmx0: port 0 SGMII, address 18:e8:29:ba:19:1b
> ukphy0 at cnmac0 phy 4: Generic IEEE 802.3u media interface, rev. 2: OUI
> 0x0001c1, model 0x000c
> cnmac1 at octgmx0: port 1 SGMII, address 18:e8:29:ba:19:1c
> ukphy1 at cnmac1 phy 5: Generic IEEE 802.3u media interface, rev. 2: OUI
> 0x0001c1, model 0x000c
> cnmac2 at octgmx0: port 2 SGMII, address 18:e8:29:ba:19:1d
> ukphy2 at cnmac2 phy 6: Generic IEEE 802.3u media interface, rev. 2: OUI
> 0x0001c1, model 0x000c
> cnmac3 at octgmx0: port 3 SGMII, address 18:e8:29:ba:19:1e
> ukphy3 at cnmac3 phy 7: Generic IEEE 802.3u media interface, rev. 2: OUI
> 0x0001c1, model 0x000c
> octgmx1 at octpip0 interface 1
> cnmac4 at octgmx1: port 16 SGMII, address 18:e8:29:ba:19:1f
> ukphy4 at cnmac4 phy 8: Generic IEEE 802.3u media interface, rev. 0: OUI
> 0x0001c1, model 0x0027
> cnmac5 at octgmx1: port 17 SGMII, address 18:e8:29:ba:19:20
> ukphy5 at cnmac5 phy 9: Generic IEEE 802.3u media interface, rev. 0: OUI
> 0x0001c1, model 0x0027
> octsctl0 at simplebus0: disabled
> octxctl0 at simplebus0: DWC3 rev 0x250a
> xhci0 at octxctl0, xHCI 1.0
> usb0 at xhci0: USB revision 3.0
> uhub0 at usb0 configuration 1 interface 0 "Generic xHCI root hub" rev
> 3.00/1.00 addr 1
> octxctl1 at simplebus0: DWC3 rev 0x250a
> xhci1 at octxctl1, xHCI 1.0
> usb1 at xhci1: USB revision 3.0
> uhub1 at usb1 configuration 1 interface 0 "Generic xHCI root hub" rev
> 3.00/1.00 addr 1
> "i2c" at simplebus0 not configured
> "i2c" at simplebus0 not configured
> com0 at simplebus0: ns16550a, 64 byte fifo
> com0: console
> com1 at simplebus0: ns16550a, 64 byte fifo
> com1: probed fifo depth: 0 bytes
> octmmc0 at simplebus0
> sdmmc0 at octmmc0: 8-bit, mmc high-speed
> "spi" at simplebus0 not configured
> "ocla0" at simplebus0 not configured
> "dma-engine" at simplebus0 not configured
> "dma-engine" at simplebus0 not configured
> octrng0 at iobus0 base 0x1400000000000 irq 0
> octpcie0 at iobus0: 3 ports
> octpcie0 port 0: link timeout
> octpcie0 port 1: reset timeout
> octpcie0 port 2: reset timeout
> scsibus0 at sdmmc0: 2 targets, initiator 0
> sd0 at scsibus0 targ 1 lun 0: <Kingston, MMC4GB, 0000> removable
> sd0: 3728MB, 512 bytes/sector, 7634944 sectors
> vscsi0 at root
> scsibus1 at vscsi0: 256 targets
> softraid0 at root
> scsibus2 at softraid0: 256 targets
> root on sd0a (285e489508f0f02b.a) swap on sd0b dump on sd0b
can you try this?
Index: uipc_socket.c
===================================================================
RCS file: /cvs/src/sys/kern/uipc_socket.c,v
diff -u -p -r1.344 uipc_socket.c
--- uipc_socket.c 31 Oct 2024 12:51:55 -0000 1.344
+++ uipc_socket.c 1 Nov 2024 23:53:25 -0000
@@ -787,48 +787,36 @@ m_getuio(struct mbuf **mp, int atomic, l
{
struct mbuf *m, *top = NULL;
struct mbuf **nextp = ⊤
- u_long len, mlen;
- size_t resid = uio->uio_resid;
+ u_long len, mlen, alen;
+ int align = atomic ? roundup(max_hdr, sizeof(long)) : 0;
int error;
- do {
- if (top == NULL) {
- MGETHDR(m, M_WAIT, MT_DATA);
- mlen = MHLEN;
- } else {
- MGET(m, M_WAIT, MT_DATA);
- mlen = MLEN;
- }
+ m = m_gethdr(M_WAIT, MT_DATA);
+ mlen = MHLEN;
+
+ for (;;) {
/* chain mbuf together */
*nextp = m;
nextp = &m->m_next;
- resid = ulmin(resid, space);
- if (resid >= MINCLSIZE) {
- MCLGETL(m, M_NOWAIT, ulmin(resid, MAXMCLBYTES));
- if ((m->m_flags & M_EXT) == 0)
+ /* How much data we want to put in this mbuf? */
+ len = ulmin(uio->uio_resid, space);
+ /* How much space are we allocating for that data? */
+ alen = align + len;
+ if (alen > mlen) {
+ MCLGETL(m, M_NOWAIT, ulmin(alen, MAXMCLBYTES));
+ if (!ISSET(m->m_flags, M_EXT) && alen > MCLBYTES)
MCLGETL(m, M_NOWAIT, MCLBYTES);
- if ((m->m_flags & M_EXT) == 0)
- goto nopages;
- mlen = m->m_ext.ext_size;
- len = ulmin(mlen, resid);
- /*
- * For datagram protocols, leave room
- * for protocol headers in first mbuf.
- */
- if (atomic && m == top && len < mlen - max_hdr)
- m->m_data += max_hdr;
- } else {
-nopages:
- len = ulmin(mlen, resid);
- /*
- * For datagram protocols, leave room
- * for protocol headers in first mbuf.
- */
- if (atomic && m == top && len < mlen - max_hdr)
- m_align(m, len);
+ if (ISSET(m->m_flags, M_EXT))
+ mlen = m->m_ext.ext_size;
}
+ /* Avoid pain from a stupid max_hdr value */
+ if (align < mlen)
+ m->m_data += align;
+
+ /* How much data can we put in this mbuf? */
+ len = ulmin(mlen, len);
error = uiomove(mtod(m, caddr_t), len, uio);
if (error) {
m_freem(top);
@@ -836,13 +824,19 @@ nopages:
}
/* adjust counters */
- resid = uio->uio_resid;
space -= len;
m->m_len = len;
top->m_pkthdr.len += len;
/* Is there more space and more data? */
- } while (space > 0 && resid > 0);
+ if (space == 0 || uio->uio_resid == 0)
+ break;
+
+ align = 0;
+
+ m = m_get(M_WAIT, MT_DATA);
+ mlen = MLEN;
+ }
*mp = top;
return 0;
