800km (500 mi) from San Francisco to Los Angeles

Took my bicycle overseas to ride from San Francisco to Los Angeles. We did it in 7 days. Was a great trip! Not writing the whole story here, just a couple of pictures.

The bike arrived

Using QEMU to navigate a 3000 tons ship

Some time ago I was approached by a passionate engineer Jean Michel Schramm, who asked whether QEMU would help saving public finances and a navigation system of a 3000 tons ship.

The project looked interesting, so we gave it a try and succeeded!

More details for those who plan to use QEMU for replacement of
hardware becoming extinct.

Moscow

Had a really great time at Moscow Center of SPARC Technologies. They invited me to talk about QEMU. The MCST people are very bright and motivated. And their SPARC CPUs are much faster than QEMU can currently emulate. :-) The machines are not Sun clones. For instance it was interesting to see a SPARC V8 machine with a PCI bus.

Despite the company name, they produce not only SPARC systems, but their own VLIW chips, compilers and binary translators as well. Russian readers can find out more on MCST web-site.

Байкал

Debian/sparc64 Wheezy under QEMU How-To

The steps to install Debian Wheezy RC1 / SPARC64 under QEMU.

Since the installation process is not obvious for the current QEMU and Debian versions, I gathered this How-To. Feel free to send any feedback.

Virtio performance and filesystems

Some time ago there was a comment to my post about booting Linux/sparc64 under QEMU, saying that virtio performance sucks.

At first I couldn't reproduce the problem. Tried a few caching strategies, but the I/O performance was not drastically affected by them. And then it came to my mind that the reason might be the guest file system.

Let's take a look:

$ dd if=/dev/zero of=disk-tmp bs=1024k count=300
300+0 records in
300+0 records out
314572800 bytes (315 MB) copied, 0.430909 s, 730 MB/s

  # so, the theoretical maximal raw performance of this partition on my disk is ~ 730 MB/s.

$ sparc64-softmmu/qemu-system-sparc64 -drive file=../boot-disk.qcow2,if=virtio,index=0 -drive file=disk-tmp,if=virtio,index=1
[...]
root@debian-wheezy:~#  mkfs.ext4dev /dev/vdb1
[...]
root@debian-wheezy:~#  mount /dev/vdb1  /mnt/
root@debian-wheezy:~#  dd if=/dev/zero of=/mnt/100M-file bs=1024k count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 3.2384 s, 32.4 MB/s

# while it seems to be 20 times slower than the host speed it is still a lot (the number 730MB/s is the theoretical limit, without performing regular syncs. QEMU does execute syncs to ensure the data integrity for the unlikely case of the crash.

Let's take a look on the read performance:

root@debian-wheezy:~#  umount /mnt/
root@debian-wheezy:~#  mount /dev/vdb1  /mnt/
[  582.673975] EXT4-fs (vdb1): mounted filesystem with ordered data mode. Opts: (null)
root@debian-wheezy:~#  dd if=/mnt/100M-file of=/dev/null bs=1024k
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 1.08423 s, 96.7 MB/s
root@debian-wheezy:~#

This looks even better. But what happens with the usual ext4?

root@debian-wheezy:~#  mkfs.ext4 /dev/vdb1
[...]
root@debian-wheezy:~#  mount /dev/vdb1  /mnt/
root@debian-wheezy:~#  dd if=/dev/zero of=/mnt/100M-file bs=1024k count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 3.54709 s, 29.6 MB/s

Ok, that's a bit less than the newer version. And what about ext3?

root@debian-wheezy:~#  mkfs.ext3 /dev/vdb1
root@debian-wheezy:~#  dd if=/dev/zero of=/mnt/100M-file bs=1024k count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 13.8049 s, 7.6 MB/s

Wow. That's slow. The ext3 partition access with the default options is more than 4 times slower than the ext4 partition access with the default options.

The short conclusion: if you need performance, don't use the ext3 file system with the default settings for your virtual machines.

/ happy hacking

CruiseControl 2.8.4 and JDK7

Upgraded system JDK to Java 7 and noticed that the dashboard in your CruiseControl stays gray?
Well there is even an official bug CC-1049 in the JIRA (currently it seems to be down, I used Google cache to access it).

With -debug switch on there is a hint error message:

Cannot construct net.sourceforge.cruisecontrol.BuildLoopInformation as it does not have a no-args constructor.

The root of the problem: Oracle changed the vendor name from "Sun Microsystems Inc." to “Oracle Corporation”. The solution is really simple, either add

-Djava.vm.vendor="Sun Microsystems Inc." -XX:+OverrideVMProperties

to your java call in cruisecontrol.sh , or replace xstream-1.2.2.jar and xpp3_min-1.1.3.4.O.jar files both in lib/ and webapps/dashboard/WEB-INF/lib/ with xstream-1.4.1.jar and kxml2-2.3.0.jar respectively.

Networking in Linux/sparc64 under qemu...

... is also possible. The bulit-in ne2k-pci network card doesn't work, but hey there is an even faster virtio-net alternative. At the OpenBIOS "ok" prompt, before the "boot" command in the previous post, type

cd /pci@1fe,0/pci1af4,1000
0 encode-int " interrupts" property
device-end

Will send this patch upstream as soon as we clarify whether "0" is allowed for the "interrupts" property.

Booting Linux/sparc64 on todays OpenBIOS

Historical day for everyone interested in the vanilla qemu-system-sparc64 emulator. For the first time it can boot Linux!

Although the patches I sent upstream are  fixing CPU and IOMMU, there is still one missing piece: cmd646 IDE. Luckily it's not a showstopper at all: qemu-sparc64 is a PCI machine, which means one could use the fast virtio instead! Now, this is the command line:

$  sparc64-softmmu/qemu-system-sparc64 -m 256 -nographic -prom-env 'auto-boot?=false' -kernel /path/to/kernel/image -drive file=/path/to/debian-disk,if=virtio,index=0 -append 'root=/dev/vda1 init=/bin/sh'

Some time ago, I used Forth to workaround missing qemu features  to get OBP working.
Now it's pretty similar: OpenBIOS doesn't have all the properties necessary for Linux to get the interrupt mapping.

So at the ok command prompt where you'd get after the command above, type:
 
cd /
1 encode-int " #interrupt-cells" property
cd /pci@1fe,0/ebus
000001fe encode-int 020003f8 encode-int encode+ 1 encode-int encode+
ffe29140 encode-int encode+ 2b encode-int encode+ " interrupt-map" property
000001ff encode-int ffffffff encode-int encode+ 00000003 encode-int encode+
" interrupt-map-mask" property

cd /pci@1fe,0/ebus@3/su
1 encode-int " interrupts" property
cd /pci@1fe,0/pci-ata@5
0 encode-int " interrupts" property

cd /pci@1fe,0/pci1af4,1001
0 encode-int " interrupts" property
device-end
boot


The only magical constant above is actually "ffe29140" - the address of the root pci node in the OpenBIOS device hierarchy. Although it the same in all the recent builds, in theory it could be moved somewhere else one day. But I guess till that day OpenBIOS will have the missing properties... ;-)

Update:

Oh, and a few words to /path/to/kernel/image and /path/to/debian-disk:

I found no Linux/sparc64 distribution which has a built in virtio driver. This makes installing from a  CD/DVD image not possible. The solution is build your own kernel with the virtio driver compiled in and put it at /path/to/kernel/image.

As for the disk, the user space utilities from the sparc32 world can be used in the sparc64 world as is. So, you can install Debian (or your favorite sparc32 distribution) in the /path/to/debian-disk, using qemu-system-sparc (no 64 at the end), and then use it with qemu-system-sparc64 as described above.

If you know a Linux/sparc64 distribution with the virtio support, please let me know.

/Happy hacking

Qemu is going to boot Linux/sparc64

After considering it a bit, I thought, it's a good marketing strategy: the free QEMU version shall run the free OS - Linux. And if anyone is interested in running something else, feel free to ask me for a [paid] support. :-)

So, Linux is going to be the second OS which vanilla qemu-system-sparc64 would boot - HelenOS was the first one. But, unlike HelenOS, Linux will be fully functional, having not just a stdout, but a stdin as well. And a serial port support!

At the moment OpenBIOS has some missing features - it doesn't describe the interrupt mappings - and a regression - currently it can't even boot HelenOS from a command line. But both are not show-stoppers.

Once my patches are accepted I'll publish the OpenBIOS command to boot Linux (not because it's top secret, but due to the dependency to a certain version).

Sent y2k10 fix upstream

It occurred to me that I've described the fix for y2k10 bug in qemu-system-sparc which makes Solaris 2.5.1 (and prior) hang, but never managed to submit it upstream.

So, let's fix it. For those one and half users who care ;-) .

The 100th post (cycling in a City is dangerous)

Allright, this is my 100th post. I was going to look back, summarize and talk about the future.

But I won't.

Yesterday on my way to the office I broke my leg. Multiple fractures. After all cycling in a city on an asphalt is more dangerous than cycling on an ice.

The good news: I'm going to have more time to read books and work on Open Source projects...

Making a Vacation

Far away from the Internet and cellular networks. Crossing the Great Baikal Lake on a bicycle.

Of course the spikes are required

A couple more pictures

Power management in SPARC & Solaris

Few days ago someone asked me in the comments to the How-To post, whether there is a way to reduce CPU consumption of the qemu-system-sparc. I assumed qemu lacks some device necessary for the power management. So, if we implement it, qemu-system-sparc won't take 100% of CPU all the time.

And then after looking at the qemu source I was really surprised: qemu does emulate the 'power-management' device. The reason why Solaris consumes so much CPU is Solaris itself! It's just doesn't have a driver for the 'power-management' device on SPARCStation-5. In all 'prtconf' outputs I could google out, there is a line:

power-management (driver not attached)

And then comes a bigger surprise: Linux perfectly supports CPU power management on SPARCStation-5. Even under qemu: after Debian/SPARC is booted, 'top' shows that qemu-system-sparc takes only 3% of a host  CPU.
 
How came that Sun didn't support CPU power management on their own sun4m machines?

P.S. By the way, NetBSD supports CPU power management too.

Does anybody use recent qemu versions?

After a long break I tried a current qemu git/master today. What can I say? The SPARC-related stuff is broken in a few places since a few months. SCSI signals an error for the "Inquiry" command, Leon3 test hangs, sparc64 is broken too. Bisecting is tricky though, because at least sparc64 seems to broken in different ways in subsequent commits.

Don't have time to fix it myself, but will send a couple of bug reports to the mailing list. Meanwhile, what is your favorite qemu version?

Update: meanwhile (7/Feb/2012) SCSI and Leon3 issues are fixed, thanks to Thomas Higdon and Fabien Chouteau.

Of Course, It Runs NetBSD!™

NetBSD boot was almost a piece of cake. It tries to detect more things than Solaris and Linux, so I had to implement a couple of device registers more. At the first glance using more registers contradicts with the declared portability. On another hand, it would work on some modified/weird chipsets having non-standard interrupt controllers. Don't know if such chipsets were ever produced though.

NetBSD 4.0.1 (INSTALL) #0: Wed Oct  8 01:13:04 PDT 2008

builds@wb32:/home/builds/ab/netbsd-4-0-1-RELEASE/sparc64/200810080053Z-obj/home/builds/ab/netbsd-4-0-1-RELEASE/src/sys/arch/sparc64/compile/INSTALL

total memory = 256 MB

avail memory = 234 MB

timecounter: Timecounters tick every 10.000 msec

mainbus0 (root): QEMU,Ultra-3/2: hostid 80000000

cpu0 at mainbus0: SUNW,UltraSPARC @ 100.681 MHz, UPA id 0

cpu0: 32K instruction (32 b/l), 16K data (32 b/l), 512K external (64 b/l)

...

# ping 10.0.2.2

PING 10.0.2.2 (10.0.2.2): 56 data bytes

64 bytes from 10.0.2.2: icmp_seq=0 ttl=255 time=1.575 ms

64 bytes from 10.0.2.2: icmp_seq=1 ttl=255 time=1.150 ms

^C

----10.0.2.2 PING Statistics----

2 packets transmitted, 2 packets received, 0.0% packet loss

round-trip min/avg/max/stddev = 1.150/1.363/1.575/0.301 ms

#

Haven't found any CPU bugs so far, only interrupt processing in the serial port (not relevant to [Open]Solaris). Surprisingly sparc32 and sparc64 serial drivers diverge quite a lot.

Next stop - FreeBSD/sparc64.

Hiding the boot device from an OS

Suspecting a bug in SCSI subsystem, I was wondering how to prevent an OS from detecting the device it's booting from. At first it sounds weird, but it's possible for most of OS/distributions in the SPARC (and probably PPC) world, because they

• boot  init scripts from a RAM disk
• use IEEE 1275 device tree to find out available devices *

*) with exception of  Linux. At least the Linux uniform IDE driver tries to detect a hardware by probing.

For FreeBSD, NetBSD, OpenBSD and Solaris it's possible to hide the boot device from the OS install CD with the following commands:

ok cd scsi " unknown" name device-end
ok boot /unknown/sd@6,0:f

Use the Forth, Luke!

User mode emulation for Linux/SPARC64

As you know, qemu has a user-mode emulation. This means binaries for one CPU (for example SPARC) can be executed on another CPU (for example i686) under the same OS. The system calls are executed directly on the host (which means they are executed as fast as for native binaries), and the executable code itself is translated with TCG.
After I fixed ELF loading for SPARC64 binaries, qemu can load not only static Linux/sparc64 binaries, but dynamically linked ones too. To achieve that qemu has to be statically linked (it may sound confusing, for launching statically linked binaries qemu doesn't have to be built statically but for the dynamically one it has to be) and chrooted to the guest OS file system image:
 qemu$ ./configure --target-list=sparc-linux-user,sparc64-linux-user,sparc32plus-linux-user --static && make
 ...
 qemu$  mv -i sparc32plus-linux-user/qemu-sparc32plus ../debian-6-sparc64-initrd/
 qemu$  su
 Password:
 #  /usr/sbin/chroot ../debian-6-sparc64-initrd/ /qemu-sparc32plus -L . /bin/busybox
BusyBox v1.17.1 (Debian 1:1.17.1-8) multi-call binary.
Copyright (C) 1998-2009 Erik Andersen, Rob Landley, Denys Vlasenko
and others. Licensed under GPLv2.
See source distribution for full notice.

Usage: busybox [function] [arguments]...
   or: function [arguments]...

        BusyBox is a multi-call binary that combines many common Unix
        utilities into a single executable.  Most people will create a
        link to busybox for each function they wish to use and BusyBox
        will act like whatever it was invoked as.

Currently defined functions:
        [, [[, ar, ash, basename, blockdev, cat, chmod, chown, chroot, cp, cut, dd, df, dirname, dmesg, dnsdomainname, echo, egrep, env, expr, false,
        find, free, freeramdisk, grep, gunzip, halt, head, hostname, id, init, ip, kill, klogd, ln, logger, ls, md5sum, mkdir, mknod, mkswap, modinfo,
        more, mount, mv, nc, pidof, pivot_root, poweroff, printf, ps, pwd, readlink, realpath, reboot, rm, rmdir, route, sed, sh, sleep, sort,
        swapoff, swapon, sync, syslogd, tail, tar, test, tftp, touch, tr, true, tty, udhcpc, umount, uname, uniq, wc, wget, zcat


This is a great tool to find CPU bugs! One can use existing binaries for example to check that emulated CPU produces the same md5 or sha512 sum for a certain binary as the host does, or pack/unpack using gzip and bzip, or just observe weird unames:

 /usr/sbin/chroot ../debian-6-sparc64-initrd/ /qemu-sparc32plus -L . /bin/uname -a
Linux localhost 2.6.34.9-69.fc13.x86_64 #1 SMP Tue May 3 09:23:03 UTC 2011 sun4 GNU/Linux

NetBSD vfs and MMU emulation

Since Solaris works perfectly, I turned again to NetBSD for the further test cases.

root on md0a dumps on md0b

root file system type: ffs

WARNING: clock gained 1005 days

WARNING: CHECK AND RESET THE DATE!

exec /sbin/init: error 8

init: trying /sbin/oinit

exec /sbin/oinit: error 2

init: trying /sbin/init.bak

exec /sbin/init.bak: error 2

init: not found panic: no init

cpu0: kdb breakpoint at 1362e60

Stopped in pid 1.1 (init) at 0x1362e64: nop

db>

Hmm what would be a reason for such a behavior? Sounds familiar, right? I hear you saying CPU math bug? Wrong. :)
The previous time I saw the message it was really a math bug. This time it's a MMU problem. Unlike SPARC v8, the v9 has a NFO mode for page mappings, which means No Fault Only. The 'only' part is crucial: all other page loads should fault.

Solaris knows what pages has been marked NFO (after all they can be marked by the OS only) and uses only the appropriate loads.

On the other side NetBSD vfs implementation explicitly provokes such faults to load RAM pages from a file. Had to read the NetBSD documentation to understand how its vfs works. The sources are not that good documented.

After fixing the bug (a sort of a counterpart to the one Tsuneo Saito mentioned on the mailing list), NetBSD gets a bit further:

Kernelized RAIDframe activated md0:

internal 5120 KB image area

root on md0a dumps on md0b

root file system type: ffs

WARNING: clock gained 1005 days

WARNING: CHECK AND RESET THE DATE!

erase ^?, werase ^W, kill ^U, intr ^C

The only part is missing here is the '#' prompt after the message. :) Working on it.

More math bugs

 Why would one file system work, and another one - not?

Because the fs driver is buggy? Of course not! The ones who read this blog on a regular basis, know that inability to read the file has to do with a bug in a math emulation.

Thanks to Jakub's test case, I fixed two bugs with carry flag handling (yes, again). Now HelenOS/sparc64 boot looks like this:

My impressions of HelenOS - it's neat, the sources are good documented and easy readable. Also I needed just a few minutes to set up cross compiling under Linux/x86_64. So, if you need a small, micro-kernel(!) and multi-arch (amd64, arm32, ia32, ia64, mips32, ppc32, sparc64) OS to play with, HelenOS is definitely worth looking at.

Now finally something new

Loading: /platform/sun4u/boot_archiveramdisk-root ufs-file-system
Loading: /platform/sun4u/kernel/sparcv9/unix
module /platform/sun4u/kernel/sparcv9/unix: text at [0x1000000, 0x10bf34d] data at 0x1800000
module /platform/sun4u/kernel/sparcv9/genunix: text at [0x10bf350, 0x12b5c7f] data at 0x1865e00
module /platform/sun4u/kernel/misc/sparcv9/platmod: text at [0x12b5c80, 0x12b5c97] data at 0x18bac30
module /platform/sun4u/kernel/cpu/sparcv9/SUNW,UltraSPARC-II: text at [0x12b5cc0, 0x12c2a37] data at 0x18bb2c0
SunOS Release 5.11 Version MilaX_0.3.2 64-bit
Copyright 1983-2008 Sun Microsystems, Inc.  All rights reserved.
Use is subject to license terms.
os-io Ethernet address = 52:54:0:12:34:56
Using default device instance data
mem = 262144K (0x10000000)
avail mem = 154615808
...
Preparing live image for use
Hostname: milax
Requesting System Maintenance Mode
(See /lib/svc/share/README for more information.)
Console login service(s) cannot run

Root password for system maintenance (control-d to bypass):
single-user privilege assigned to /dev/console.
Entering System Maintenance Mode

May 20 07:26:42 su: 'su root' succeeded for root on /dev/console
-bash: /usr/sbin/quota: No such file or directory
Sun Microsystems Inc.   SunOS 5.11      MilaX_0.3.2     October 2008
-bash: /bin/mail: No such file or directory
-bash: id: command not found
-bash: id: command not found
-bash: [: !=: unary operator expected
(root@milax)# uname -X
System = SunOS
Node = milax
Release = 5.11
KernelID = MilaX_0.3.2
Machine = sun4u
BusType =
Serial =
Users =
OEM# = 0
Origin# = 1
NumCPU = 1

(root@milax)# uname -a
SunOS milax 5.11 MilaX_0.3.2 sun4u sparc sun4u
(root@milax)#

The missing files above are caused by the maintenance mode which in turn is caused by the missing network card. So the next steps are obvious:
- plug in a supported NIC
- find a customer for the sun4u emulation
- ???
- PROFIT!

PS. Btw, are there any live OpenSolaris/Illumos based live CDs newer than MilaX 0.3.2? The later MilaX versions seem to be PC-only.

Seen a really broken pipe

Have you seen a broken pipe? Sounds like a stupid question for everyone who is working with *NIX. Everyone who has English as a mother tongue, or is old enough to use a non-localized OS has seen a "broken pipe" message. A younger generation might have seen the message in their native language.

Well, that's not the sort of brokenness  I'm talking about.  I mean it more literary:

# echo "This pipe works fine" |cat
This pipe works fine
# echo "And this pipe is veeeeeeeeeeeeeeeeeeeery broken. This pipe is really very broken. Broken." | cat
Ths pip eisveTkebn

Ops. Had to spend a lot of of time in Solaris ascending from SCSI driver (at first I didn't realize that the bug appears in pipes, it looked like a DMA bug) to streams, pipe and so on and then descending to the memcpy which was the source of trouble. It turned out, memcpy uses different routines for small and large data chunks. The small chunks are copied in a loop word-wise and the large ones are copied using VIS instructions (that's SPARC equivalent for Intel's MMX). The emulation of VIS instructions in qemu is buggy, so the memory gets corrupted when these instructions are used.

Once again I'm very impressed that Solaris 2.5.1 - Solaris 7 boots without problems on such a broken hardware. Let's see if  the newer Solaris versions would work with this emulation bug fixed.

OpenBIOS strikes back

Mark did it! OpenBIOS svn.r1035 with two patches (escc and lance) can boot a sun4m Solaris. As soon as the patches are reviewed and committed I'll have to update the how-to. Booting Solaris is going to be much simpler. Congratulations the OpenBIOS Team!