This is a summary of the "network interface becomes inoperable - No buffer space" thread from misc because something seems wrong to me (also I have some fibre ix(4) myself in places that would be a pain for them to break and I don't really want to run into it if possible ;)
The reporter has 82599 ix(4) with 10G LR SFP+, and getting enobufs when trying to transmit on the interface at times, so far they have been clearing the fault with a reboot (unclear yet whether down+up the interface would help but in any event that shouldn't be needed either) I'm guessing possibly the tx ring may possibly be stuck somehow? Anyone have ideas? rx looks ok I think. Thread https://marc.info/?t=165588707200005&r=1&w=2 - summary of what's in the mails https://marc.info/?l=openbsd-misc&m=165588703222265&w=2 netstat -i during failure (bunch of Ofail) netstat -m (looks normal) sysctl | grep buf dmesg > r...@syspatch-71-amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/GENERIC.MP > real mem = 17058750464 (16268MB) > bios0: Supermicro Super Server > cpu0: Intel(R) Xeon(R) E-2126G CPU @ 3.30GHz, 3292.33 MHz, 06-9e-0a > ix0 at pci1 dev 0 function 0 "Intel 82599" rev 0x01, msix, 6 queues, https://marc.info/?l=openbsd-misc&m=165588793522744&w=2 ifconfig (looks normal) > ix0: flagsĖ43<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500 https://marc.info/?l=openbsd-misc&m=165607469228799&w=2 netstat -i (normal and failed states), no Ofail in normal, lots in failed pfctl -si (looks normal) systat mbuf (normal and failed states), fairly high-ish anyway (user reports box normally does about 3G/s), 2.5x more mcl2k2 in failed state, though the stats for individual interface (alive/hwm/cwm) all look pretty similar between normal+failed vmstat -m (i only had a quick look through but no Fail in pools at lesat) Any ideas? I had a quick look at what's exported in kstat for ix on -current but I didn't spot anything in there that looked like it would be worth asking them to switch from -stable to obtain.
