[gold-devel] PlayStation3 target
ADAM David Alan Martin
adamlsd at gmail.com
Thu Apr 12 04:40:47 EDT 2007
James,
Replying to both your emails here.
> I agree we should possibly think of a new name for the OS (other than
> gold)... but PreciOS is not it. ;)
Sorry to hear you dislike it. I know I'm most keen on it. Justin
seemed ambivalent. I suggest, at least until a better proposal comes
along, that we call it PreciOS, just to avoid confusion with whether
we talk about Gold as an OS or just a userspace library. (But given
that temporary names tend to become permanent over time, maybe it's
best to wait until we all agree on a better name.)
> I still kinda like the name of my (defunct) "Terra OS". Unfortunately
> it clashes a bit with "Terra Soft". But I'm not going to push for that
> name anyway, just throwing ideas around.
Too many overlaps to existing things, and defunct projects. Name is
not too important, but it's nice to have a "name" for the finished
project. My biggest preferences here are that the name resemble some
word, that the name be tied to the "gold" theme, and that the name
end in "OS" to specify that it's an operating system, and not a
blender or a toaster. (Justin already is considering products like
the "Gold Nugget" as a PDA, so we should stay within the Gold theme.
PreciOS, as a first approximation fit that bill. And the Golem jokes
are just a bonus! :-)
In the end, I don't much care. But I hate the GNU route: LSD/Gold/
Golem/... And Gold being both the whole OS and the userland is a
problem. (Apple calls their base OS Darwin, but their product MacOS,
and the kernel XNU. Such name variance is not uncommon.)
On Mar 29, 2007, at 5:02 AM, James Lee wrote:
> I think we should evaluate the PS3 as a common target (I know you guys
> are targeting mac/etc, but I mean for the future)
At this point, since we're all time constrained, ESPECIALLY me, I
think it's best we focus on ONE hardware platform. I won't eradicate
x86 support, and I'll try not to break what I have. I'll even fix
what I might break if I can. But now that PPC/Mac is approved, I
would like to develop all new SYSTEM primitives for this. We can
then port it back to x86, or forward to ARM, SPARC, or whatever. (I
suggest we abandon x86 completely, regardless.)
PS3 is a viable Machine target in the PPC family, but I want to
avoid 64-bit arches for the time being. Also we've just changed
development platforms. I don't want to change it AGAIN! We'll be
stuck in development far longer than the 3 to 7 years we've been in
development (depending where you start counting.) If I approve
another hardware platform, or change platforms again, even if we do
not do any more, we just leave ourselves open to more such fiascos
later. PS3 is an intriguing possibility, which I would like to
explore in the future, but I have to be firm here. If we flail about
changing platforms, nothing will get done.
> PS3 pros:
> * Mass produced 64bit PowerPC platform
So is PPC/Mac
> * Boots any OS out-of-the-box (without hacks)
So does PPC/Mac, with OpenFirmware. (No hacks. Just put image as
elf in an HFS or ext2 partition.)
> * Large install base
PPC/Mac is larger. And now obsolete, making future platform changes
not an issue.
> * Hypervisor hardware abstraction (drivers are super easy to write)
Useful, but not mandatory.
> * Standard hardware platform (don't need a million drivers)
Apple's Mac w/ OpenFirmware uses ATA, USB, FireWire OHCI and more
standard hardware too. Some system hardware like OpenPIC needs
support for specific machine, but this is to be expected with any
machine variant.
> * USB 2.0/SATA/blu-ray/HDMI/SD/CF/gbit ethernet... so it's pretty
> modern standard interfaces generally
Mac in last revisions had: USB2.0, Firewire, SATA, HDMI video cards,
SD and CF via extensions, and gbit ethernet too. Not much change there.
> * SPU processors are perfect for gold
This seems to be PS3's greatest argument. But let's get a working
multitasking PPC kernel up. Porting to Cell can be done later.
> * Complete cell simulator available for free from IBM
Useful. But not needed. I have PowerPC simulators, but more
important is hardware testing of the whole platform.
> * Super fast RAM/busses/etc
So?
> * Might be able to port open firmware easily, in which case it will
> act extremly similar to a mac.
When/if we ever port to PS3 we could do this.
> * Jbit has alot of experience with this platform.
How could you? It's not even a year old? You've played with it.
Mac is sufficiently old, with enough example OS code out there, that
it makes writing an OS from scratch a far easier task. (Most from-
scratch OSes are started on stock, older hardware for this reason.)
> PS3 cons:
> * only 256MBytes of RAM (and 256MBytes of graphics RAM)
Expandability is necessary, I would wager.
> * No hardware drawing access (yet?)
That is a problem.
> * Slow (in-order-execution) CPU
This I do not like. What does this mean for Imprecise and Precise
interrupts in PPC's ABI on the Cell? System programming for this CPU
may be trickier than PPC.
> * Can't upgrade RAM or CPU
Expandability again...
> * Pretty expensive at the moment
I don't have one. Justin doesn't have one. Do you? Neither he nor
I can spare the ~600USD to buy one either. I might eventually save
up for it, but Gold alone doesn't justify such an attempt, nor am I
much of a games fanatic. (I'll play older stuff, if anything.)
> * Hypervisor hardware abstraction (no direct access to most hardware)
This is a serious problem for LSD's design. Emulated hardware or
emulated interfaces changes timing from native systems, and requires
a serious rethink of the optimization concepts in LSD -- things like
Event delivery, context switching, and piggybacking.
> * No easy debug solution on consumer units
Terrible. This makes development essentially impossible for me.
> * Jbit might be limited to what he can release/work on.
We don't want development tied to a platform I've never seen nor
touched. PPC/Mac is sufficiently available to justify using it. You
have one. I have several. Justin has one. Many on the channel have
one. They're also far cheaper for simple models than a PS3. PearPC
emulation is a possibility for some, if it gets OpenHackware replaced
with OpenBIOS.
> Let me know your thoughts on this.
> I might start looking into porting open firmware or another
> standardish firmware to "otherOS".
Meh. No need for that.
In other news, I understand you'll be leaving us for a while to
focus on your job, and on the PS development scene? We'll miss your
assistance. Hopefully you'll return to us in time to get Golem
working on top of Gold. A year off might be in order in your case.
Good luck, and enjoy working on the PS scene.
Regards,
--
ADAM David Alan Martin
FreeBSD Hacker, TCSH Hacker, general UNIX coding
Author of AutoFS for FreeBSD 6.x
LSD Kernel Lead Developer
Filesystem and Storage Lab, SUNY Stony Brook
More information about the gold-devel
mailing list