Re: CVS commit: src/sys/net

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

Thanks for kind words!

Apparently, we need some test to detect kernel and userland mismatch
for bpf(4) header...

I'd also like to fix problems by which ATF does not complete on ERL3.
Some CPU/memory consuming tests, e.g., lib/libc/regex/t_exhaust,
trigger watchdog.

I guess something wrong in interrupt handler for mips (IIRC, there
are similar kind of behaviors for hpcmips machines). Disabling
__HAVE_FAST_SOFTINT by hand does not work around the problem...

For n32 userland, most rump-based tests in /usr/tests/net crash.
But for n64 userland, on the other hand, they seem to work just fine.

This may be due to bugs for n32 kernel. IIUC, most kernel codes are
compiled for o32 or n64, and only the exception is n32 rump kernel for
mips64e[bl] userland.

Thanks,
rin

On 2023/08/26 9:10, David H. Gutteridge wrote:
> > Module Name:    src
> > Committed By:   rin
> > Date:           Wed Aug 23 13:21:17 UTC 2023
> >
> > Modified Files:
> >        src/sys/net: bpf.h
> >
> > Log Message:
> > bpf: Fix SIZEOF_BPF_HDR (for LP64 userland) on mips64
> >
> > It cannot fit within 18 bytes, of course ;)
> >
> > As we had never provided working bpf(4) implementation for LP64
> > userland on mips, just use natural structure size here.
>
> Thanks for fixing this. This means dhcpcd now works on Octeon hardware
> using the n64 userland. Previously, it would fail with "ps_bpf_recvbpf:
> Resource temporarily unavailable". (Perhaps you'd noticed this?) This
> was reported a while ago[1] by an end user on port-mips@, so we should
> presumably pull this up.
>
> Where you say "we had never provided working bpf(4) implementation", I
> had been looking into the dhcpcd issue myself, but hadn't thought to
> look here. That is, I had run the tests under net/bpf and net/t_ip_reass
> (the latter modified to open /dev/bpf directly, not via rump) on my
> ERL3, and these all passed, giving me the impression we did indeed have
> a working bpf(4) implementation on n64. Since we didn't, it seems we
> need more test coverage here.
>
> Thanks,
>
> Dave
>
> 1. http://mail-index.netbsd.org/port-mips/2021/07/04/msg001112.html