[Qemu-devel] reverting commit a53ae8e934cd54686875b5bcfc2f434244ee55d6 Re: [edk2] [PATCH 0/2] Re: exec: fix regression by making system-memory region UINT64_MAX size

[Paolo Bonzini]

Il 07/11/2013 23:23, Laszlo Ersek ha scritto:
> On 11/07/13 22:24, Marcel Apfelbaum wrote:
>> Why pci-hole and system.flash collide? IMHO we should not play with
>> priorities here, better solve the collision.
> 
> What about this "beautiful" series? It produces
> 
> memory
> 0000000000000000-000fffffffffffff (prio 0, RW): system
>   [...]
>   0000000060000000-00000000ffdfffff (prio 0, RW): alias pci-hole @pci
>   0000000060000000-00000000ffdfffff
>   [...]
>   00000000ffe00000-00000000ffffffff (prio 0, R-): system.flash
> 
> and I can run OVMF with it. It also stays within i386/pc.

This definitely works, and make sense.

But I think for 1.7 we should just revert the commit.  It can be done
later in 1.8.  And it should have a qtest testcase that shows the effect
of the patch.

Other patches can still be applied to 1.7, but the change definitely had
much bigger ramifications than anticipated.  The patches are

[PATCH for-1.7 v2 5/8] pci: fix address space size for bridge
[PATCH for-1.7 v2 7/8] pc: s/INT64_MAX/UINT64_MAX/
[PATCH for-1.7 v2 8/8] spapr_pci: s/INT64_MAX/UINT64_MAX/

There's also

[PATCH 1/2] split definitions for exec.c and translate-all.c radix trees
[PATCH 2/2] exec: make address spaces 64-bit wide

which however has a 2% perf hit for TCG.  In 1.8 we can apply it and
recover the hit with other optimizations.

Paolo