[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: etch-m68k in a linux container



Hi Thorsten,

Le jeudi 06 décembre 2012 à 20:56 +0000, Thorsten Glaser a écrit :
> Laurent Vivier dixit:
> 
> >I think it should be a good idea to play with the linux-user mode of
> >qemu.
> 
> Does that emulate an MMU? The system mode doesn’t, IIRC.

MMU is useless in usermode emulation.

> >qemu linux-user mode traps guest (m68k) syscalls to translate them to
> >native ones. It's not perfect.
> 
> Well right. It also doesn’t catch genuine kernel bugs, and
> probably a lot less other bugs (I remember having FPU issues
> which turned out to be ARAnyM bugs, but still).

indeed

> 
> >really usable for a debian buildd, but it can be used to develop and
> 
> Hum, but right now, all we need is buildds.

I agree

> >correct build issues. For instance, you can build glibc, kernel and gcc
> >concurrently on one machine in less than a day ;-)
> 
> I’m down to about 1.5 days on gcc. With your bogomips from below,
> I believe you can’t get any faster, unless you go parallel building,
> which won’t work because, right now, m68k is not SMP.

A full system simulator is slow mainly because of I/O. User-mode
emulation has native I/O performance.

> (I also don’t think it gets the new cmpxchg syscall right ;)
> 
> >I have tested it on an ubuntu 12.10 x86_64 system, and it seems to work
> 
> Urgh… I’d not trust *buntu to get anything right…
> 
> >Note3: and for those that want to know the real power of their new m68k
> >machine, I've added in attachement bogomips.c to compute the bogomips
> >and the equivalent 040 cpu frequency. Mine (on a Q6600 a 2.4 Ghz) is:
> >$ ./bogomips 
> >Clocking:      132
> >BogoMips:      104.00
> >Calibration:   524288
> 
> I get this on ARAnyM, on a 3.2 GHz host box:
> 
> Clocking:       265.269
> BogoMips:       209.00
> Calibration:    1048576
> 
> So I believe it’s not slower ;-) but more reliable.

On a full system emulation, like ARAnyM or even KVM, clock (needed to
compute instructions per second) is not reliable. So you can't
compare ... 

> But still, thanks. I guess your method makes for a quicker start,
> although I believe that, for most people, Aranym/Quick from the
> Debian wiki would be enough (and possibly, as you already said
> it’s not complete, needed) to work on build issues.
> 
> (Actually, the general consensus is that we need *less* emulation
> and more bare iron right now.)

As I said, it is just to play or to help to solve build issue and saving
time. Don't play, don't have fun ;-)

Regards,
Laurent



Reply to: