[Qemu-devel] whats up with QEMU and virtual disks?

[Qemu-devel] whats up with QEMU and virtual disks?

[Hetz Ben Hamo]

Hi,

Today I installed NT 4 server and XP Pro.

In both cases I used new virtual disk partitions (4GB, and 5GB respectivly)..

As both OS's have 2 stages install, both run the first stage without
any problem (both of them were booted from their original CDs),
however, upon rebooting to the 2nd stage I had problems.. (on XP I had
the problem only after quitting qemu and restarting it)..

In NT 4 - after starting qemu (after finishing first stage install) it
refused to reboot.
in XP Pro - after restarting qemu (after fully finishing installing XP
Pro) it says that the NTLDR is missing..

in both cases, the only way to bypass this problem is by using the
-hdachs parameter..

Why? what has changed that broke the normal behavior? how can it be fixed?

Thanks,
Hetz

Re: [Qemu-devel] whats up with QEMU and virtual disks?

[ML]

Hi list!

---
Hetz Ben Hamo said the following on 11/15/2004 10:14 PM:

> Hi,
> 
> Today I installed NT 4 server and XP Pro.
> In both cases I used new virtual disk partitions (4GB, and 5GB respectivly)..

Me too, I created a new 2GB virtual disk with qemu-img to make a fresh 
installation of XP Home.

> As both OS's have 2 stages install, both run the first stage without
> any problem (both of them were booted from their original CDs),
> however, upon rebooting to the 2nd stage I had problems.. (on XP I had
> the problem only after quitting qemu and restarting it)..

Yep, same thing to me.

> In NT 4 - after starting qemu (after finishing first stage install) it
> refused to reboot.
> in XP Pro - after restarting qemu (after fully finishing installing XP
> Pro) it says that the NTLDR is missing..

I get a Disk I/O error instead.

> in both cases, the only way to bypass this problem is by using the
> -hdachs parameter..
> Why? what has changed that broke the normal behavior? how can it be fixed?

My quick and dirty fix of commenting out the call to bdrv_set_geometry_hint in 
line 1898 of hw/ide.c doesn't appear to work anymore with the disk image newly 
created with the 0.6.1 version.
However my older disk images, created with previous versions of Qemu, don't 
work with 0.6.1 (it says that NTLDR is missing) but, if I comment out the line 
as per above, everything goes back to normal behavior.

Ciao,
Piero

Re: [Qemu-devel] whats up with QEMU and virtual disks?

[Hetz Ben Hamo]

> > in both cases, the only way to bypass this problem is by using the
> > -hdachs parameter..
> > Why? what has changed that broke the normal behavior? how can it be fixed?
> 
> My quick and dirty fix of commenting out the call to bdrv_set_geometry_hint in
> line 1898 of hw/ide.c doesn't appear to work anymore with the disk image newly
> created with the 0.6.1 version.
> However my older disk images, created with previous versions of Qemu, don't
> work with 0.6.1 (it says that NTLDR is missing) but, if I comment out the line
> as per above, everything goes back to normal behavior.

I see...

Fabrice, since this change breaks compatibility, is there a way to either:

* Add a parameter to load "old" images with the old bdrv_set_geometry_hint 

or

* Provide a simple script to convert from the old to new format?

One more thing Fabrice, regarding the message about the RTC. Could you
detect if the user is running 2.6.x kernel and if so, avoid showing
the RTC message please?

Thanks,
Hetz

Re: [Qemu-devel] whats up with QEMU and virtual disks?

[Fabrice Bellard]

Hetz Ben Hamo wrote:
>>>in both cases, the only way to bypass this problem is by using the
>>>-hdachs parameter..
>>>Why? what has changed that broke the normal behavior? how can it be fixed?
>>
>>My quick and dirty fix of commenting out the call to bdrv_set_geometry_hint in
>>line 1898 of hw/ide.c doesn't appear to work anymore with the disk image newly
>>created with the 0.6.1 version.
>>However my older disk images, created with previous versions of Qemu, don't
>>work with 0.6.1 (it says that NTLDR is missing) but, if I comment out the line
>>as per above, everything goes back to normal behavior.
> 
> 
> I see...
> 
> Fabrice, since this change breaks compatibility, is there a way to either:
> 
> * Add a parameter to load "old" images with the old bdrv_set_geometry_hint 

No one gave precise enough information to fix the problem. The problem 
is related to the default Bochs BIOS tranlation mode, and I would like 
to understand first before doing a fix.

> * Provide a simple script to convert from the old to new format?

If it is the only solution, I can do it, but there may be a better one.

> One more thing Fabrice, regarding the message about the RTC. Could you
> detect if the user is running 2.6.x kernel and if so, avoid showing
> the RTC message please?

I'll try.

Fabrice.

Re: [Qemu-devel] whats up with QEMU and virtual disks?

[Fabrice Bellard]

>> * Provide a simple script to convert from the old to new format?

I repeat: there is _no_ "old" or "new" disk image format. Raw disk 
images are exactly the same in every QEMU version...

Fabrice.

Re: [Qemu-devel] whats up with QEMU and virtual disks?

[Fabrice Bellard]

Fabrice Bellard wrote:
> 
>>> * Provide a simple script to convert from the old to new format?
> 
> 
> I repeat: there is _no_ "old" or "new" disk image format. Raw disk 
> images are exactly the same in every QEMU version...

However, when formatting the disk, Windows or MSDOS may write the 
current BIOS geometry in it, and it is likely to be the problem.

If I disable the BIOS translation (for example by commenting 
'bdrv_set_geometry_hint()' or by removing 'rtc_set_memory(s, 0x39, val)' 
in hw/pc.c), then NT 4.0 won't see disks bigger than 512 MB. So a better 
solution must be found.

Fabrice.