[Qemu-devel] Re: POLL: Why do you use kqemu?

[Blue Swirl]

On 6/7/09, Jan Kiszka <address@hidden> wrote:
> Blue Swirl wrote:
>  > On 6/7/09, Jan Kiszka <address@hidden> wrote:
>  >> Avi Kivity wrote:
>  >>  > Jan Kiszka wrote:
>  >>  >>> Maybe the backwards compatibility features should be ported to QEMU?
>  >>  >>> For example, is there a workaround for
>  >>  >>> #error Missing KVM capability KVM_CAP_DESTROY_MEMORY_REGION_WORKS
>  >>  >>> ?
>  >>  >>>
>  >>  >>
>  >>  >> Given that we have always-up-to-date kvm-kmod packages with support 
> down
>  >>  >> to reasonable kernel versions, I would prefer to keep upstream clean
>  >>  >> from old workarounds. They should only be needed for issues found very
>  >>  >> recently (KVM_CAP_JOIN_MEMORY_REGIONS_WORKS) or that might be found in
>  >>  >> the future.
>  >>  >>
>  >>  >
>  >>  > Requiring the latest up-to-date modules is pushing the problem to the
>  >>  > users.  Sometimes there is no choice, but when there is, the
>  >>  > implementation that cares about its uses prefer unclean code and
>  >>  > functionality over perfection and brokenness.
>  >>
>  >>
>  >> Let's make it more concrete:
>  >>
>  >>  By the time upstream is as well tested, feature-rich and with comparable
>  >>  performance as qemu-kvm, its current baseline requirement (2.6.29 due to
>  >>  KVM_CAP_DESTROY_MEMORY_REGION_WORKS) will no longer be a problem to most
>  >>  normal users. Until then they are better off with qemu-kvm anyway.
>  >>
>  >>  So all I wanted to express is that I see no point in merging workarounds
>  >>  upstream that hardly anyone will need but that restrict non-kvm code in
>  >>  upstream. Basically I have the current line along
>  >>  KVM_CAP_DESTROY_MEMORY_REGION_WORKS / clean memory slot management in
>  >>  mind. Anything older should be skipped when merging upstream. And unless
>  >>  something more problematic comes along (rather unlikely), 2.6.29 or
>  >>  compatible kvm-kmod is a reasonable minimum requirement for the long 
> term.
>  >
>  > I pulled qemu-kvm and it looks to me that
>  > KVM_CAP_DESTROY_MEMORY_REGION_WORKS and the derived functions
>  > kvm_destroy_memory_region_works() and must_use_aliases_*() are only
>  > used in very few places. Do I miss something, how can this be of any
>  > restriction?
>
>
> Check e.g the diff of hw/vga.c against upstream: All the magic dances
>  there are required as qemu-kvm tracking cpu_register_physical_memory and
>  kvm_log_start cannot cope with all the patterns normal qemu code comes
>  up with. Upstream slot management now provides the same features
>  (including migration) like qemu-kvm, it just does not deal with legacy,
>  thus it does not have to patch qemu code (rather, we were able to remove
>  some already merged hooks - vga_dirty_log_stop).

Still not very restrictive, all this could be encapsulated with for
example macro COMPAT_NO_DMRW which could be removed when we don't care
anymore. Next?