These are my notes for NetBSD. Originally written for NetBSD 7, they have been slightly extended and do also apply to NetBSD 8 and 9.
Please note the prompts:
$ sysctl -a | grep freq machdep.est.frequency.current = 1333 machdep.est.frequency.available = 2167 1667 1333 1000
# sysctl -w machdep.est.frequency.current=1333
/etc/rc.conf:
estd=YES estd_flags="-d -b -M 1667" # upper limit 1667 MHzTODO: Investigate C-state and P-state support in NetBSD.
# sysctl -w hw.acpi.sleep.state=3
# zzz
Suspend to disk is not supported by {Net|Open|Free}BSD
# atactl wd0 apm set 254
PS: atactl wd0 apm disable does not do the job.
# atactl wd0 apm set <value>
# smartctl -g apm /dev/rwd0d APM level is: 254 (maximum performance)
On my system power management is disabled in /etc/rc.local via atactl wd0 apm set 254. This setting does not survive suspend to mem, so it must be re-executed, e.g. via:
# zzz; date; sleep 30; date; . /etc/rc.local; smartctl -g apm /dev/rwd0d
Note the sleep: after executing zzz, the machine is still running for a few sec and without the sleep, the execution of /etc/rc.local would get executed before the actual suspension and not after the suspension. 30 sec work for my machine.
Package sysutils/intel-microcode-netbsd dynamically loads the latest microcode from Intel.
Manual load
# /usr/pkg/share/examples/rc.d/intel-microcode onestart
Autoload on boot
# cp /usr/pkg/share/examples/rc.d/intel-microcode /etc/rc.d # echo microcode=YES >> /etc/rc.conf
My original microcode version is 0xc2
$ cpuctl identify 0 | grep microcode cpu0: microcode version 0xc2, platform ID 1 $ cpuctl identify 1 | grep microcode cpu1: microcode version 0xc2, platform ID 1
Updated microcode to version 0xe2
$ cpuctl identify 0 | grep microcode cpu0: microcode version 0xe2, platform ID 1 $ cpuctl identify 1 | grep microcode cpu1: microcode version 0xe2, platform ID 1
$ envstat
Current CritMax WarnMax WarnMin CritMin Unit
:
cpu0/cpu1 temperature: 69.000 99.000 degC
:
CPU temperature table for cpu0/cpu1 (on a hot summer evening) (set via sysctl -w machdep.est.frequency.target=1000)
| Freq | idle | make-j2 |
|---|---|---|
| 1000 MHz | 47 ℃ | 57 ℃ |
| 1333 MHz | 51 ℃ | 68 ℃ |
| 1667 MHz | 56 ℃ | 80 ℃ |
| 2167 MHz | 65 ℃ | 94 ℃ |
sysutils/smartmontools provides smartctl to display hard drive health:
# smartctl -a /dev/rwd0d :
e.g.
# smartctl -a /dev/rwd0d | grep Load_Cycle_Count ID# ATTRIBUTE_NAME FLAG VALUE WORST THRESH TYPE UPDATED WHEN_FAILED RAW_VALUE 193 Load_Cycle_Count 0x0032 086 086 000 Old_age Always 144467
Display current battery status (via package graphics/gnuplot)
$ while true; do envstat | grep charge: | cut -b 25-35 >> charge; gnuplot -e "set term xterm; plot [0:][0:] 'charge' using 0:1"; sleep 10; done
There is no /sysfs filesystem on BSD. The kernel interface is exposed to userland via sysctl, e.g. sysctl -a, sysctl -w xxx.yyy.zzz=3.
Kernel compilation time NetBSD 7.0-RC2 GENERIC, core duo, system compiler gcc-4.8.4, X11 running:
| parallelization | 1000 Mhz | 1333 Mhz | 1667 MHz | 2167 MHz |
|---|---|---|---|---|
| make -j1 depend | 4m13.150s | 3m11.536s | 3m14.164s | 2m11.462s |
| make -j1 | 25m25.067s | 19m15.233s | 16m16.665s | 12m36.376s |
| make -j2 depend | 2m40.423s | 2m10.196s | 1m48.069s | 1m29.399s |
| make -j2 | 14m27.399s | 11m20.775s | 9m20.556s | 8m00.394s |
XXX make -j1 depend discrepancy between 1333 and 1667 MHz XXX
See Userland section. Config file is /usr/src/sys/arch/*/conf/GENERIC.
First time checkout (-7 is always latest stable, e.g. 7.0, 7.1, etc):
# cd /usr # export CVS_RSH=ssh # cvs -d anoncvs@anoncvs.NetBSD.org:/cvsroot co -r netbsd-7 -P src # cvs -d anoncvs@anoncvs.NetBSD.org:/cvsroot co -r netbsd-7 -P xsrc
Further source updates of tools, kernel and userland:
# cd /usr/src # export CVS_RSH=ssh # cvs update -dP
Further source updates of x-window:
# cd /usr/xsrc # export CVS_RSH=ssh # cvs update -dP
Building tools, kernel, userland and x-window [timings with 1.667 GHz][-u: don't clean][-x: build and install x-window as well (this is not done from /usr/xsrc)][MKDEBUG: build debug symbols and install into /usr/XXX]:
# mkdir /usr/obj /usr/tools # cd /usr/src # PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools tools # 24 min 08 sec # PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools kernel=GENERIC # 21 min 57 sec # PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools modules # 29 min 54 sec # PATH=/sbin:/usr/sbin:/bin:/usr/bin MKDEBUG=1 MKDEBUGLIB=1 ./build.sh -x -u -O ../obj -T ../tools distribution # 5 hour 15 min 59 sec
Installing (tools?,) kernel, userland and x-window:
# cd /usr/src # mv /netbsd /netbsd.old # mv /usr/obj/sys/arch/amd64/compile/GENERIC/netbsd / # PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -O ../obj -T ../tools installmodules=/ # PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -x -O ../obj -T ../tools install=/ # rm -rf /usr/tools /usr/obj # shutdown -r now
Building and installing everything as a one-liner:
PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools tools && PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools kernel=GENERIC && PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -u -O ../obj -T ../tools modules && PATH=/sbin:/usr/sbin:/bin:/usr/bin MKDEBUG=1 MKDEBUGLIB=1 ./build.sh -x -u -O ../obj -T ../tools distribution && mv /netbsd /netbsd.old && mv /usr/obj/sys/arch/amd64/compile/GENERIC/netbsd / && PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -O ../obj -T ../tools installmodules=/ && PATH=/sbin:/usr/sbin:/bin:/usr/bin ./build.sh -x -O ../obj -T ../tools install=/
My steps to fix bug lib/50367:
cscope /usr/src (indexes also #includes from /usr/include):
# cd /usr/src # cscope-indexer -r -v
Build optimized bins/libs and debug symbol files:
# cd /usr/src [-u doesn't clean - omit the first time] # PATH=/sbin:/usr/sbin:/bin:/usr/bin MKDEBUG=1 MKDEBUGLIB=1 ./build.sh -u -O ../obj -T ../tools distribution
XXX debug versions don't seem to be installed - perhaps set MKDEBUG environment, variables for installation, as well
# ls /usr/obj/usr.bin/getaddrinfo/ getaddrinfo.c getaddrinfo.o getaddrinfo getaddrinfo.debug
Debug new Binary:
# LD_LIBRARY_PATH=/usr/obj/lib/libc gdb getaddrinfo (gdb) set solib-search-path /usr/obj/lib/libc Reading symbols from /usr/obj/lib/libc/libc.so.12.193.1... Reading symbols from /usr/obj/lib/libc/libc.so.12.193.1.debug...
Run Program against newly build Shared Library:
# LD_LIBRARY_PATH=/usr/obj/lib/libc getaddrinfo heise.de
Save Changes:
# cd /usr/src # cvs diff > changes.diff # diff against cvs server version # patch -R < changes.diff # undo changes
/etc/mk.conf, /root/.profile (partial) and /root/.cvsrc. Please run all commands in /usr/pkgsrc:
# pkg_admin check
# cvs update
check packages that need updates (cmp pkgsrc vs installed packages):
# pkg_chk -u -q
update these packages (dependent packages are recompiled as well):
# pkg_chk -u -s
# cvs update -dP -rpkgsrc-2013Q2
# cvs update -dPA
# rm -rf /var/tmp/pkgsrc/ # this is my /etc/mk.conf WRKOBJDIR PATH # rm /usr/pkgsrc/distfiles/* # rm /usr/pkgsrc/packages/All/*
TODO. For now, try ifconfig wpi0 list scan and wap_gui.
Since NetBSD 9.0 and pkgsrc 2020Q1, NetBSD provides qemu acceleration via nvmm [details]. Speed improvement over plain qemu is around factor 20 for compiling the Linux kernel in a Debian 10 guest.
Make nvmm accessible to a user here sschaeck
# echo "nvmm:*:34:sschaeck" >> /etc/group # chown root:nvmm /usr/pkg/bin/qemu-system-x86_64 # chmod g+s /usr/pkg/bin/qemu-system-x86_64
Boot-time setup (or add module name to /etc/modules.conf)
# modload nvmm
Optional: create this alias
$ alias qemu="qemu-system-x86_64 -accel nvmm"
Qemu 101 (assumes above alias)
Boot an iso
$ qemu -m 512M -cdrom /opt/isos/KNOPPIX*.iso
Create an empty 10GB disk
$ qemu-img create -f qcow2 /opt/qemu/debian.img 10G
Install an operating system from a bootable iso onto the just created empty disk (the booted iso will see the empty disk)
$ qemu -hda /opt/qemu/debian.img -cdrom /opt/isos/firmware-10.3.0-amd64-netinst.iso -boot d -m 512M -netdev user,id=mynet0,host=10.0.2.10 -device e1000,netdev=mynet0
Boot an operating system from a disk (guest ip 10.0.2.15 / ssh from host to guest ssh sschaeck@127.0.0.1 -p 7001)
$ qemu -hda /opt/qemu/debian.img -m 512M -netdev user,id=mynet0,host=10.0.2.10,hostfwd=tcp:127.0.0.1:7001-:22 -device e1000,netdev=mynet0
Mount guest from host (hack hack hack)
# mount_psshfs -O Port=7001 sschaeck@127.0.0.1:/ /mnt/tmp ... # umount /mnt/tmp
Volume control via mixerctl, e.g. mixerctl -a; mixerctl -w monitor.output=200.
Starting X applications as root or any another user (here named otheruser).
$ echo $DISPLAY $ xhost local:otheruser $ su - otheruser $ export DISPLAY=:0 # the DISPLAY is a moving target increasing on each start. Take it from command #1
Provided by packages fonts/cyberbit-ttf and fonts/kochi-ttf (both are required).
TODO
NetBSD provides a seamless Linux emulation. Package meta-pkgs/suse* provides a whole Linux userland in /usr/pkg/emul/linux[32|64]/.
A Linux binary, say downloaded from the Internet with dependencies on libraries from /usr/pkg/emul/linux[32|64]/, can be executed from any directory such as /opt/bin/. One can also chroot into /usr/pkg/emul/linux[32|64]/:
# chroot /usr/pkg/emul/linux32/ /bin/bash bash-3.00# uname -a Linux hostname 3.11.6 #1 SMP PREEMPT Thu Oct 24 16:23:02 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
TODO: Try building the Linux kernel in a chrooted environment.
Opposed to OpenBSD and FreeBSD, NetBSD does not support whole disk encryption. NetBSD's crypto framework cgd encrypts whole slices with one 1 passphrase for all partitions within it. I encrypted a slice holding exactly one partition, the /home partition. Also, swap on yet another slice can be encrypted with an always randomly generated passphrase. Here I succeeded to a certain degree, only.
Update: I learned of cgdroot for encrypting the root filesystem. I have not used it.
You are welcome to send me comments. Contact information is available on the main page.