[Qemu-devel] Everything OK with kqemu vs. new SUNWqemu-0.8.2, REV=2006.10.18-sol10-i386-opt.pkg.gz for x64/x86

[Martin Bochnig]

It was a misunderstanding.
You know, I'm new to x86 HOSTs running qemu  (doing most things on sparc 
normally).

I therefore didn't have prior kqemu experience.


Eric Lowe wrote:

>>OH, oh, -kernel-kqemu does NOT work with Solaris guests. 
>>
>  
>

Your binary 32bit version actually does in 70% of tested cases (Sol10_U2
boot DVD and marTux_0.2_x64__snv46) !
It does work with  -kernel-kqemu and boots into marTux_0.2 x86 OpenSolaris 
32bit mode!
This is what confused me, and which made me believe, we had been talking about 

A)
(appropriate kqemu loaded)  &&  (-kernel-kqemu)

rather than

B)
(appropriate kqemu loaded)  &&  !(-no-kqemu) 

all the time.
A total misunderstanding which made me wonder, how to see the light.


>>You can only use kqemu without any kqemu parameters to QEMU so it does the 
>>default of virtualizing only the user-mode bits and emulating CPL=0.
>>  
>>
>  
>

Ahh, cool.
Then I simply misunderstood the help-comment/option-description of 
-kernel-kqemu, when I believed, "user-mode bits" only applies to the
linux_X_linux usermode case (not softmmu full system emulation).
Well, then it works ok what I get.
I need to do a performace analysis covering module_loaded versus
driver_removed (or -no-kqemu).


>>-kernel-kqemu does appear to work OK with Linux and Windows guests.
>>  
>>
>  
>

Sun_JDS1-Linux yes.
Dos/Win9x no. (also a no-go with /opt/SUNWqemu/32/bin/qemu from your
previous binary tar archive version, whereas OpenSolaris guests do work with 
your 20061002 version)
Apparrently NT based Win would potentially work with -kernel-kqemu, when you 
say that you could sucessfully boot into XP.

Do Linux-HOSTS running qemu/kqemu have the same incompatibility problems with 
certain guests (i.e. OpenSolaris or Win9x), or rather the BSD
kqemu wrapper which you have ported to Solaris?


>>After discussing with Fabrice the ins and outs of -kernel-kqemu I built a 
>>kernel which fixes some of the issues (Solaris has some naughty code which 
>>clears IF using POPF instead of using the CLI instruction) but there is one 
>>problem I could not solve kernel-side; kqemu does not do full page table 
>>shadowing for the kernel but instead updates the page tables of the guest 
>>in-situ with the translated pfns. Solaris gets very cranky about this since 
>>it assumes that the PTE.pfn == page_t.p_pagenum when doing the reverse 
>>translation. In your case I think you probably got bitten instead by an 
>>interrupt coming into the guest when it expected interrupts to be off -- the 
>>dispatcher does some twiddling with interrupts in some of its code which 
>>invoke the bad macro that uses PUSH $2 followed by POPF to clear IF.
>>  
>>
>  
>

Interesting.
I need to play around with that.


>>Testing without -kernel-kqemu should not have any problems.. if you do 
>>encounter any problems let me know. 
>>
>  
>

The only problem is, that the Sol10_U2 DVD stops booting the installer
after Configuring Devices.
^C brings me to the # prompt  (letting it running over night doesn't
have it make progress).
Not sure why that happens.
May not have to do with kqemu.


--->>
Thanks for introducing me to the kqemu(-wrapper) matter.
The more respect do I have for your wrapper-port.


-Martin

Eric Lowe wrote:

>kqemu should work with OpenSolaris guests provided you don't use 
>-kernel-kqemu. -kernel-kqemu is incompatible with the OpenSolaris kernel for a 
>couple of reasons. First, it appears kqemu will need a change in its page 
>table shadowing if we want to support OpenSolaris guests with full 
>virtualization, due to the way the HAT layer manages its page tables which is 
>very different from Linux and Windows; second, there is a bug (well, an RFE) 
>in OpenSolaris itself which would also need to be addressed.
>
>I have booted Windows XP and Linux using -kernel-kqemu with OpenSolaris as the 
>host using the binary bits mentioned below and they worked fine for me. I also 
>had no difficulty booting OpenSolaris guests on these bits in 32-bit and 
>64-bit mode w/ kqemu. I will install the latest packages on my home system 
>tonight and retry my tests with the latest kernel accelerator (my machine is 
>an Athlon X2 so I can test both 32-bit and 64-bit).
>
>- Eric
>
>----- Original Message ----
>From: Martin Bochnig <address@hidden>
>Cc: Open Solaris <address@hidden>; address@hidden; address@hidden
>Sent: Thursday, October 19, 2006 2:50:37 AM
>Subject: err  Re: [qemu-discuss] New SUNWqemu-0.8.2, 
>REV=2006.10.18-sol10-i386-opt.pkg.gz for x64/x86 ready for public testing
>
>Sorry, the new version doesn't work with kqemu very well.
>This affects the old and the new src equally.
>The compiler or build env seems to be the problem.
>
>Please continue to use the previous version that can still be obtained
>from
>http://opensolaris.org/os/project/qemu/downloads/qemu-osol-0.8.2-bin-x86.tar.gz
>
>Sorry for any inconveniences.
>
>
>Martin Bochnig wrote:
>
>  
>
>>http://opensolaris.org/os/project/qemu/downloads/
>>
>>Latest patched QEMU sources 20061013
>><http://opensolaris.org/os/project/qemu/downloads/qemu-0.8.2-solaris_src_20061013fri.tar.bz2>QEMU
>>0.8.2 + patches snapshot October 13th 2006
>>Previous patched QEMU sources 20061002
>><http://www.opensolaris.org/os/project/qemu/downloads/qemu-0.8.2-solaris.tar.gz>QEMU
>>0.8.2 + patches snapshot October 2nd 2006
>>Pre-built x86 binaries_pkg
>><http://opensolaris.org/os/project/qemu/downloads/SUNWqemu-0.8.2_REV_2006.10.18-sol10-i386-opt.pkg.bz2>
>>32-bit and 64-bit x86/x64 pre-compiled QEMU 0.8.2 binaries for Solaris
>>10 / 11 / OpenSolaris
>>
>>Should run on Sol10-x86 FCS aka 03/05 or higher (has been built under
>>Solaris 10 03/05 FCS).
>> 
>>
>>    
>>
>
>_______________________________________________
>qemu-discuss mailing list
>address@hidden
>http://opensolaris.org/mailman/listinfo/qemu-discuss
>
>
>
>
>
>
>
>  
>