[Qemu-devel] Re: [PATCH] Snapshot block device support

[Fabrice Bellard <fabrice.bellard@free.fr>]

Rusty Russell wrote:
 > In message <3F02AA9D.1030307@free.fr> you write:
 >
 >> I had plans to implement "copy on write" disk images...I find your
 >>  approach original. My initial plan was just to add a second disk
 >> image for each disk containing the changes. It should be easy to
 >> do it from your code (for example by adding options '-hdacow file'
 >> where file is the snapshot file you use).
 >
 >
 > Sure: I used a temporary (unlinked) file because I couldn't see much
 > point in keeping it around.  Since most filesystems do sparse files,
 > it's quite efficient, too.
 >
 > -hdacow with an optional argument would work really well, if that's 
desired
 > behaviour.
 >
 > I'm glad I finally beat you to implementing something: you're
 > usually too quick 8)

Well, I don't have to maintain parts of the Linux kernel at the same 
time :-)

I am going to modify your patch today so that -hdacow is supported.

 >> About IDE, it is working (at least with kernel 2.4.20). Someone
 >> has even compiled a linux kernel inside QEMU (slowdown factor of
 >> about 15), so vl begins to be usable for real tasks. There are
 >> still known problems if several processes use floating point at
 >> the same time.
 >
 >
 > OK, I'm getting errors with IDE on 2.5, but shouldn't be too hard to
 > track down.

I tested with 2.5.74 and fixed the initial IDE error. I also added the
emulation of the PC 'reset' pin so that vl can be shut down as a normal 
Linux system (only hard reset is supported, not power off, as it would 
require a more complicated real mode APM bios emulation).

Now I am looking for potential original uses of vl/QEMU !

Fabrice.