Slashdot Mirror
NetBSD Ported to AMD x86-64 (Sledgehammer)
fvdl writes: "Last week, a port of NetBSD to the x86-64 (tm) architecture was committed to the NetBSD CVS repository. The x86-64 is AMD's upcoming 64bit line of CPUs. For now, it is only known to work on the Virtutech simulator, since no x86-64 hardware is available yet. In this environment, it runs multi-user. NetBSD/x86_64 is the 44th architecture that NetBSD runs on (12 different families of CPUs). The porting was done by Frank van der Linden of Wasabi Systems, with
kind support from AMD, who provided the simulator and fast machines
on which ro run it. The Wasabi press release is
here. For more information on the x86-64, see of course
AMD's website and x86-64.org"
Also, different versions of CPUs differ too -- so there is often a little work to be done when a new CPU is actually produced.
But these changes are minor compared to getting the whole operating system to run, and hence don't take very long.
And as a short note: NetBSD is already 64 bit clean and has been for many years. (check out the alpha port, which has been available since 1995.) At that time, the code-base was made 64 bit clean.
There are already 256 interrupt lines on the x86. Has been ever since the 8088.
What are you smoking? Can I have some?
There is one interrupt line on the 8086. Not surprisingly it is called INTR. When it is asserted the processor finishes its current instruction, releases the bus and asserts INTA.
When INTA is received the device requesting the interrupt places a 3-bit vector address on the data bus. The processor multiples this by 4 and grabs the vector's address from this calculated offset + the vector base address (usually 0x00000, but can be anywhere in protected mode).
I'm a little rusty here (haven't done hardware 80x86 development in some years now) but your statement is false. There's only one maskable interrupt line on the x86. There's NMI but it's hardwired to vector 2. So I guess you could argue for a total of TWO interrupt lines. RESET doesn't count. :-)
Perhaps what you're thinking of is the 256 "software" interrupt vectors? i.e. the 0xcd opcode?
I was always curious as to why the IBM PC/AT designers decided to cascade they way they did. I would have had the first PIC (Programmable Interrupt Controller) used as a first-stage cascader. I.e. there are no 1st-level interrupts. Then you put the 2nd, 3rd, 4th, etc. PICs on the eight IRQ lines of the 1st PIC. That way you could have gotten to 256 HARDWARE interrupts and not had this "if it comes in on INT9, check the 2nd PIC" problem -- they'd ALL have to be checked.
Mind you, the task of routing 256 interrupt lines to a 68-pin header would have been tricky. :-)
I believe they (I assume with AMDs help) did this last year. For exactly the reasons you state. I expect they skipped actually supporting more then 4G of physical RAM (or whatever the normal TM chips support) because they would have required real hardware work. I also expect any of the 64 bit instructions are actually kind of slow, but hay probably a lot faster then the pure software emulator (as it is a mostly software emulator with modest hardware assist).
You might try a google search on comp.arch for more info.
Sure, unless the deciding factor turns out to be how slow the 64 bit code is...
(not it isn't painfully slow, but it isn't as fast as some of the already shipping RISC machines! It's not called the iTanic for nothing...)
I would assume they did it to save $27.38 or whatever another PIC and more board traces would have cost. Much like the reason they did the A20 address rollover hack with the keyboard controller.
FYI this problem is "fixed" on systems that have and use the IO APIC, but they still have to emulate the old two level PIC design so they can run "old" OSes and drivers. (I think the APIC may predate the P-I, but it wasn't integrated on all P-II or even PPro CPUs, and I don't know if Intel clones have them or not...)
They're already working on it.
;-)
No idea when/if they'll finish the project, tho. Spill it, Linus!
64-bit *nix varients have used a 64-bit time_t since time immemorial. The 32->64 bit migration was anticipated.
FWIW, 64 bits puts the new rollover date well past the probable heat death of the universe.
DNA just wants to be free...
It will run them, and run them damn well. They run on the same core (no emulation, seperate core like merced, etc) as the 64-bit, just not using the 64-bit stuff.
Now, there may be some issues with the 32-bit addressing and extra instructions not being as heavily optimized as they might be on a purely 32-bit processor... But given that this is meant to replace the K7 line across the board, I'd say they will be putting a lot of effort into making the 32-bit smoke. Shouldn't be too hard, since the 64 is just a bit-extension and simplification of the 32 stuff. Time will tell, of course.
The enemies of Democracy are
Put the crack away. The ia32 architecture is currently very problematic for those of us who need to program for it. The basic problem is the 32-bit memory space. Sure, the kernel can address more than that, but no individual process can address that because the pointer is only 32 bits wide. Want to mmap a 2.1GB file? Better get a 64-bit machine. The move to 64 bits gives programmers the ability to write the obvious and natural algorithm, instead of spending their time fretting over their implementation of a sliding mmap window or some other hack.
Well, I manage to earn enough to meet a pretty large payroll off of what real companies pay us to make the "toy" work on their embedded platforms, and our customers (of which there are a lot) seem pretty happy.
However, what would I know? I just run a company that exclusively focuses on NetBSD.
Perry Metzger
CEO
Wasabi Systems
Replying to trolls is always a deadly business, but what the hell. I'm bored.
I don't know how many NetBSD users there are, but somehow, I suspect you don't either. NetBSD is in use every day in far more places than you think. Your DSL router or a cache box at your ISP may very well run NetBSD without your even knowing it. Ditto for large numbers of users of NCs -- NC/OS is NetBSD 1.3. There are users of hundreds of thousands of things like airline reservation terminals and such that use NetBSD every day and don't even know it. NetBSD is a very portable, very clean OS with a BSD license attached, so it gets put into LOTS of embedded hardware.
There are also a pretty large number of people who use NetBSD day to day and are very well aware of the fact. Just look at the NetBSD mailing lists if you don't believe me.
Why do people like NetBSD? Well, that would be a very long discussion. Suffice it to say, though, we've done a lot of careful architectural work on NetBSD over the years and it has paid off handsomely. There is a reason we port to new hardware quickly for instance, and it isn't that we have more fanatics than the Linux crowd. We've also got a BSD license on the code, and without trying to start a fight some people prefer the BSD license to the GPL. There are also other people who like having their whole OS build out of a single coherent source tree -- Linux code integration is a big pain which is why most people don't build full Linux systems from sources on their own.
Why is it news that there's now an OS that's gone multiuser on the x86-64? Well, that's probably more to do with the x86-64 being a neat new design than to do with us. We do a lot of ports and most of them don't get slashdotted. However, I'd say that given the fight brewing between Intel and AMD, this wasn't the silliest story for slashdot to cover.
Why do lots of people seem to think there are no BSD users in the world? I don't think anyone but a troll would claim that with a straight face...
Perry
So, I'm fishing around in my patented /. Flippant Response Generator, and I'm seeing a 'windows 2005' comment and some obscure '640k' reference.
There is no connection to Satanism. The daemon is not anti-christian or neo-pagan. He's just a mascot. I don't see you complaining about the New Jersey Blue Devils hockey team, or any of the other countless examples in todays society. Some so-called Christians will have you believe that all things that aren't god are evil, or that anything that matches their silly little filters for a devil is wrong. They are not to be trusted as they distort the truth and burn people at the stakes.
See, old stereo types are easy to pin on people, but utterly without basis in fact. There hasn't been a witch burning in over a hundred years.
No, no, no, no............cancer is NOT a virus. It's more like a celluar Pac-Man, running around and eating all the other cells.
--
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
It would be nice if 32 bit *nix/*BSD distribs be out 64 bit MS product anyhow
Can anyone translate this to plain old everybody-talk? Babel Fish garbled it up pretty bad.
--
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
The fact that this impossible OS exits is just causing the earth to collapse on itself.
Well, we'll just have to lock the door, so it can't get out.
--
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
MS Rep: Meecrob!
--
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
The AMD 64 chip is backwards compatible, it depends on the level of OS support.
My understanding is that AMD's 64bit CPUs will run old code beautifully. This may be the deciding factor against Intel's Itanium, which by all the accounts I've seen, runs 32 bit code terribly slow.
-B
Some days, I feel like upgrading from Windows 2000 to Windows 3.11. . .
of COURSE it runs NetBSD!!
(sorry, obligitory
sean
...if you count Sun Solaris and Digital/Compaq DigitalUnix/Tru64, there are popular *nixes that have done native 64-bit hardware and OSes for many years. I remember seeing many early programs which were hard to port to DEC Alphas running Digital Unix because they assumed that 'long int' and pointers were 32 bits instead of 64.
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
- Port NetBSD to our chips.CHECK
- Actually CREATE one of the chips.
- Convince someone to use it in a server.
They're still working on the last two...B1ood
Note to self: pasty-skinned programmers ought not stand in the Mojave desert for multiple hours. -- John Carmack
Programer: Waaasaaabi!
User: Wasabi?
Programer: Waaaaasaaaaabi!
User: Wasabi?
BSD Guys: Waaaaasaaaaabi!
User: Waasaabi!
Everybody: Waaasaaabi!
Wasabi!
http://www.freebsd.org
This is actually becoming a standard practice ... a few mos ago I saw an add for the Sony Playstation 9 :)
Free Techno/Jazz/DNB/MI Music by guys obsessed with monkeys!
Yes. It'll run 32-bit applications, so anything compiled for NetBSD/x86 will run just fine as long as you have all of the shared libraries and COMPAT_NETBSD32 in your kernel.
Since the Transmeta "code morpher" is closed source, and the actual machine interface is propretary, only Transmeta can do this port.
NetBSD already runs on several other 64bit :)
architectures (alpha, sparc64, dunno about MIPS
and PPC), so this is not the first stab at
getting NetBSD 64bit clean.
- Hubert
Here's a screenshot
and a dmesg output, right off the NetBSD site.
Also, check out the NetBSD/x86_64 port page!
- Hubert
For NetBSD to be portable to all the various platforms, it has to abstract the properties of these platforms, and provide interfaces between machine dependent and machine independent code, so that not every port to a new platform results in copying the whole code, and modifying it until it works on the platform, as that would give you a lot of code redundancy.
Instead, NetBSD does a (IMHO) pretty good job to avoid code redundancy, and with abstract interfaces for bus-access, DMA etc., it's amazing to see lots of code written once, and running on platforms of either endianness, CPU, bus structure, etc.
NetBSD currently runs on 44 different hardware platforms, and 12 different CPUs. If you think adding a new one is "just" a port, you miss something.
If you feel bored, you can read a bit more about what makes an operating system here.
- Hubert
Well, yeah, I guess there's plenty of time until 2038 when 32-bit time() will suddenly start telling us it's 1901. ;-P
Sheer curiosity of meeself: how wide are "short", "int", "long" and "long long" in this port? I would guess that decision was made by GCC actually. Right?
time_t when;
when = 0x7FFFFFFF;
printf("%s\n", ctime(&when));
when = 0x80000000;
printf("%s\n", ctime(&when));
Also, if time_t was unsigned, the wraparound would occur at 4 billion secons instead of 2 -- well into the XXII century.
On the other hand, It would be nice if 32 bit *nix/*BSD distribs be out 64 bit MS product anyhow.
Not that this would be *that* hard to do, for those reasonably expert.
Check out the Vinny the Vampire comic strip
"It is a greater offense to steal men's labor, than their clothes"
hmmm... ok troll I'll bite. First, I have a very hard time believing that you really and truly think this way. The first thought in my head when I read this was "This guy has to be kidding right?" So now, let's go through some of your complaints.
"I am very uncomfortable with *BSD's connection to Satanism and demonic figures."
Exactly how is BSD connected with Satanism? I don't know any Satanists who use *BSD nor are any of the original developers satanists. The demonic figure of which you speak is a daemon, not a demon. The symbol is meant to be satirical. Daemons are processes that run in the background and keep the system running clean or offer services. Examples would be syslogd, inetd, httpd, etc.. The pitchfork he's holding represents the fork() function which is how processes are executed in the first place.
"It is morally wrong to associate a piece of software with such an image, and I sure as hell do not want my children coming into contact with it."
hmmm... this image? Have you ever seen a demon to know exactly what one looks like? Can you even prove that such a creature even exists? Oh I forgot, your religious and cultural background lead you to believe such nonesense. I guess unicorns, dragons, trolls, gnomes, elfs, leprechauns, vampires, zombies, and boogie monsters exist too. After all, by this logic they should exists since they are part of somebody's popular culture. I guess the Taliban was justified in destroying the Buddhist statues in the Bamiyan valley of Afghanistan because graven images are unislamic.
"Witches are to be burned, not honored, and I don't think that most God-fearing Christians out there want their children corrupted by such "gateway" groups."
Yes, typical Christian bigot attitude. I bet you read Arthur Millers "The Crucible" with pride don't you. The Spanish Inquisition which killed more non-Christians and Jews than Hitler probably pleases you. After all, these other people don't deserve to exist do they. The early white Christian settlers killed the native Indian populations because they were heathens and unchristian. Christianity has been responsible for more cruelty and barbaric acts than any other force in history. But, I'm sure this is a source of pride for you. Witches being burned, the Spanish Inquisition, The Crusades, The Holocaust, The destruction of Native American culture, and the enslavement of millions of African Americans are all examples of how Christian bible verses were used to justify the horrible things.
Finally, your statement "I don't think that most God-fearing Christians out there want their children corrupted by such "gateway" groups".
Well, I'm sure most God-fearing Christians don't share your extremist viewpoints. I consider your group, whatever that group is, a "gateway" group. I'll be sure to not allow my children to grow up being close-minded bigots such as yourself and I'll make sure they know all the cruel barbaric acts that your group has committed in the name of God.
--Volrath50
If this goes on we are going to have the Pentium 9 avalible before the Pentium 5, Office 2005 before Office 2003. And when that happens any Pentium 9 computer will collapse on itself, being so fast that the universe will not have sped up to account for Moore's law! Don't you ever wonder why Windows 3.11 on a 386 with 4MB of RAM was as fast as Windows XP on an Athlon 4 with 512MB of RAM???
By running NetBSD on an x86-64 simulator you are creating a gap in time. The more people use the x86-64 simulator the bigger the gap will be. Eventually the gap will become so big that stuff will start coming through. First software such as WindowsZX/2023. Unfoutonetly WindowsZX will require a 2.4 THz Pentium 13(801986) with 512GB of RAM.
So of course Intel or AMD will build a 1986/P13 simulator to run WindowsZX. The motherboard of the computer will collapse causing a HUGE reverse black hole to spit out a a REAL Pentium 13. As well as 512GB or RAM.
If you thought running an operating system from 2023 was bad you can't imagine what a Pentium 13 would do! When they turn the Pentium 13 box on it would completly deplete California's power supply. That is until the entire California from 2099 gets sucked to where California is now.
Now that California has technology from 2099 such as WindowsBLT and the Pentium 86, the Universe hasn't ajusted to Moore's law and part of the world is running 3GGLHz (Googol Hertz) machines. California is now running 300 times slower than the rest of the world, and is causing random stuff from the future to appear. Which sets those parts of the world out of sync.
By now no part of the world is in sync and parts of the future are appearing everywhere. As you probably have figured out, this will eventualy spread to the rest of the universe. Destroying it.
Motto of this story: By running NetBSD on an x86-64 simulator, you have already doomed the universe.
Have a nice day!
--Volrath50
Always funny, when people refer to themselves in the third person:
fvdl writes: "[...] The porting was done by Frank van der Linden of Wasabi Systems [...]"
Well, hey, Frank you deserve the publicity!
-Kraft
-Kraft
Live and let live
Well considering the simulator CAME from AMD, I'd say they're off to a heck of a start - of course there will be some bugs when teh actual chip comes out, but they are WAY ahead of the pack right now - most excellent.
Top Most Bizarre/Disturbing Error Messages
It'll be interesting to see how fast x86-64 GCC takes to be stable enough for production use.
But this is REALLY cool stuff - don't take the above as criticism, its not - just curious.
Top Most Bizarre/Disturbing Error Messages
Must resist .. Imagine a Beow .. no .. no .. clust .. not gonna ... say ....it!
Whew - made it!
Top Most Bizarre/Disturbing Error Messages
Actually I replaced Linux with NetBSD on the machine that AMD sent me, and ran the simulator under emulation ;-) Sorry..
aww... all you have to do is fork some children processes off and manage some domain sockets or pipes, and voila, you've got more than 2GB addressable on ia32! speed? who needs speed with datasets that large? it's not like you're going to be doing anything more than playing large mpegs, right? i mean, why else would you need more memory?
of course, having more than 2GB of memory on x86 can only be bad news... software is already too bloated as it is.
/me pants anxiously at the prospect of buying a Real Computer someday... too bad x86 will probably always be faster.
It's quite common for system software bring-up to happen on simulators before the chips come out. In a lot of ways, its a much more desirable approach than *having* to deal with buggy, early-rev. hardware.
-- Jason R. Thorpe, NetBSD and FreSSH developer
Why? Because it's there. People are going to want to run software for their shiny new expensive processors in 64-bit mode, and since other companies will not provide the service for a while, you've got to run something. Besides, there were Linux distros that ran on IA-64 processors a full year before their release. It gives people a chance to clean up their code and make it 64-bit clean (there are some differences) before the release of the processor.
If god had intended you to be naked, you would have been born that way.
you know...the Society for the Prevention of Incessantly Repeating Annoying Crap.
--
Given that Gates was operating without modern compilers and debuggers, his achievment is indisputably more impressive, but kudos to the BSD team for being almost as good as Bill Gates.
Way to go guys!
"I don't know that atheists should be considered citizens, nor should they be considered patriots." - George Bush