Eric Lowe wrote:
> Roland Mainz wrote:
> ...
> >> As far as the install times, I think you'll see it is a lot better with
> >> the kernel accelerator module. I just installed Ubuntu Dapper Drake
> >> under QEMU on my AMD64 machine and it took about 20 minutes. I'm hoping
> >> we can get the OpenSolaris accelerator into the next kqemu release.
> >
> > What about security in this case ? Does the kernel accelerator module
> > execute any native code within the kernel context ?
> 
> kqemu has two modes. In the first (and default) mode, all privileged
> code is simulated and the user code runs natively. In the second mode,
> the privileged code runs as a user process and the user code runs
> natively. In either mode, all guest code executes in ring 3, and only
> the kqemu module runs in ring 0.
> 
> All kernel memory which is allocated by the kqemu module is zero'd for
> safety. This is required by the kqemu API.
> 
> Currently the OpenSolaris port of kqemu requires only permissions to
> access the device node. Least privilege is currently not supported, and
> if it was, only locked memory would need to be checked (I opted not to
> add this support so that the 32-bit module can work on older kernels
> like Solaris 8 without recompilation). In the future, we could add RBAC
> support, so users can be given a kqemu role, and use pfexec to invoke
> qemu... but I don't know if it's worth the bother, since kqemu should be
> safe (with the exception of possibly causing realtime scheduling jitter).

Erm, I think the RBAC role may be better here, however the default
should be a mode which is 100% safe.

> If you want to debate this further I would suggest getting on the qemu
> project mailing list when it is set up, I'm open to suggestions.

Ok (the project isn't open yet, right ?) ...

[snip]
> > I've seen the 40h+ install time because I was using QEMU/AMD64 on a
> > 32bit machine in the normal, "unaccerlated" interpreter mode AND QEmu
> > suffers extensively from bugs in the ATA driver, kernel and other parts
> > of Solaris. Examples include busy-loops after starting DMA, that the
> > kernel doesn't call the x86 "HTL" instruction (which can damage emulator
> > performance a lot - normal hardware just stops any execution until the
> > next interrupt is hit, but an emulator than stops interpreting of the
> > native instructions and does other stuff (including looking at I/O
> > registers MUCH earlier (this is also a problem on laptops because it
> > wastes energy))) very often and that Solaris has insane memory
> > consumtion (or better: The "working set" during several 5-15min samples
> > showed that the working set is sometimes beyond 100MB and that more or
> > less kills emulator performance, too (unless the underlying host system
> > has enougth spare bandwidth)) during install time (who had the idea to
> > run JAVA duing installation ?!) ...
> 
> Yep overall it isn't really usable without the accelerator. QEMU ATAPI
> DMA support will help a lot, too, though it still doesn't work for a
> Solaris guest.

Note that it would be nice to ship all "qemu" variants (even those for
SPARC (yes I know that they are not really usefull for end-users right
now)) and not only the x86 ones.

----

Bye,
Roland

-- 
  __ .  . __
 (o.\ \/ /.o) [EMAIL PROTECTED]
  \__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
  /O /==\ O\  TEL +49 641 7950090
 (;O/ \/ \O;)
_______________________________________________
opensolaris-code mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code

Reply via email to