OpenBSD cvs log

created 2019-09-18T17:00:54Z
begin 2019-09-12T00:00:00Z
end 2019-09-13T00:00:00Z
path src/sys
commits 11

date 2019-09-12T01:17:38Z
author dlg
files src/sys/net/bpf.c log diff annotate
src/sys/net/bpf.h log diff annotate
message make bpf_mtap_hdr take a const void *, not a caddr_t.

this makes it easier to call at least, and makes it consistent with
bpf_tap_hdr.

ok stsp@ sashan@

date 2019-09-12T01:25:14Z
author dlg
files src/sys/net/bpf.c log diff annotate
message we don't need to cast hdr arguments to caddr_t for bpf_mtap_hdr anymore.

pointed out by naddy@

date 2019-09-12T01:27:02Z
author dlg
files src/sys/net/if_tun.c log diff annotate
message let userland write up to hardmtu bytes, not just mtu bytes.

this brings tun in line with pretty much every other driver we have
where we let interfaces rx whatever they can.

while here make sure userland provides enough bytes for the link
header, which is the 4 byte address family for tun(4), and an
ethernet header for tap(4).

date 2019-09-12T01:28:29Z
author dlg
files src/sys/net/if_tun.c log diff annotate
message knf for the switch statement in tun_dev_kqfilter.

no functional change.

date 2019-09-12T02:02:54Z
author dlg
files src/sys/net/if_tpmr.c log diff annotate
message pf_test can return PF_PASS, but set the mbuf to NULL.

check for this like every other caller to pf_test so we don't go
on to pass NULL to m_prepend. that causes a fault.

chris@ gave me an excellent bug report. -msave-args is pretty great.

date 2019-09-12T03:17:12Z
author jsg
files src/sys/arch/armv7/sunxi/sxie.c log diff annotate
message Prepare for the emac/gmac "phy" property being renamed to "phy-handle"
in a large number of allwinner device trees in arm-soc/for-next which is
expected to be merged into linux 5.4-rc1.

ok kettenis@

date 2019-09-12T03:17:13Z
author jsg
files src/sys/dev/fdt/if_dwge_fdt.c log diff annotate
message Prepare for the emac/gmac "phy" property being renamed to "phy-handle"
in a large number of allwinner device trees in arm-soc/for-next which is
expected to be merged into linux 5.4-rc1.

ok kettenis@

date 2019-09-12T04:23:59Z
author jmatthew
files src/sys/dev/pci/if_mcx.c log diff annotate
message Use unsigned types for queue counters and mask them appropriately before
writing to doorbell registers. Now it should work for more than a couple
of billion packets. While here, actually set up the timeout to refill
the rx ring if it's empty, and remove the unused rx consumer counter.

chris@ reminded me this was still a problem
ok dlg@

date 2019-09-12T12:55:06Z
author stsp
files src/sys/dev/ic/ar5008.c log diff annotate
src/sys/dev/ic/ar9003.c log diff annotate
src/sys/dev/ic/ath.c log diff annotate
message Make wireless drivers call if_input() only once per interrupt.

This reduces drops caused by the ifq pressure drop mechanism and hence
increases throughput. Such drops are visible with e.g. 'netstat -dnI iwm0'.

Not all affected drivers have been tested yet but these changes are largely
mechanical and should be safe. As usual, please report any regressions.

With help from dlg@ and mpi@

Problem found by robert@
Tested by robert, jmc, Tracey Emer, Matthias Schmidt, florian, Björn Ketelaars
ok mpi@

date 2019-09-12T12:55:07Z
author stsp
files src/sys/dev/ic/atw.c log diff annotate
src/sys/dev/ic/bwi.c log diff annotate
src/sys/dev/ic/malo.c log diff annotate
src/sys/dev/ic/pgt.c log diff annotate
src/sys/dev/ic/rt2560.c log diff annotate
src/sys/dev/ic/rt2661.c log diff annotate
src/sys/dev/ic/rt2860.c log diff annotate
src/sys/dev/ic/rtw.c log diff annotate
src/sys/dev/pci/if_ipw.c log diff annotate
src/sys/dev/pci/if_iwi.c log diff annotate
src/sys/dev/pci/if_iwm.c log diff annotate
src/sys/dev/pci/if_iwn.c log diff annotate
src/sys/dev/pci/if_rtwn.c log diff annotate
src/sys/dev/pci/if_wpi.c log diff annotate
src/sys/dev/usb/if_athn_usb.c log diff annotate
src/sys/dev/usb/if_otus.c log diff annotate
src/sys/dev/usb/if_rsu.c log diff annotate
src/sys/dev/usb/if_run.c log diff annotate
src/sys/dev/usb/if_urtwn.c log diff annotate
src/sys/dev/usb/if_zyd.c log diff annotate
src/sys/net80211/ieee80211_input.c log diff annotate
src/sys/net80211/ieee80211_proto.h log diff annotate
message Make wireless drivers call if_input() only once per interrupt.

This reduces drops caused by the ifq pressure drop mechanism and hence
increases throughput. Such drops are visible with e.g. 'netstat -dnI iwm0'.

Not all affected drivers have been tested yet but these changes are largely
mechanical and should be safe. As usual, please report any regressions.

With help from dlg@ and mpi@

Problem found by robert@
Tested by robert, jmc, Tracey Emer, Matthias Schmidt, florian, Björn Ketelaars
ok mpi@

date 2019-09-12T22:22:53Z
author jmatthew
files src/sys/dev/pci/mpii.c log diff annotate
src/sys/dev/pci/mpiireg.h log diff annotate
message SAS3 controllers have a maximum size for sgl chain segments, and the
controller faults if you exceed it, so instead use multiple chain
segments in the same request frame layout.

tested on a SAS3008 (max chain size 8), where MAXPHYS io would occasionally
cause faults
ok dlg@ krw@