Re: [Qemu-devel] [PATCH v2 00/16] tcg mips64 and mips r6 improvements

[Aurelien Jarno <aurelien@aurel32.net>]

On 2016-02-15 14:42, Richard Henderson wrote:
> Changes since v1:
>   * Some bugs pointed out by Mark fixed.
>   * Canonicalize the whole file on tcg_debug_assert.
>   * Switch bswap code to subroutine earlier; the first patch is
>     standalone for mips32, and there is no longer an intermediate
>     patch with inline bswap for mips64.
>   * Use NAL for pre-r6 mips64 loading of the slow path return address.
> 

Thanks a lot for working on that, it's something I have on my TODO list
for months. I have finally found time to have a look and give a try over
the week-end (sorry about the delay).

It seems to work perfectly for 64-bit guests on mips64el but 32-bit
guests end-up quickly in a segmentation fault. It's easily reproducible 
by starting qemu-system-i386 on a mips64el host, it crashes when
executing seabios.

More problematic it seems that the patch "Adjust qemu_ld/st for mips64"
causes a regression on at least a big-endian 32-bit host running
qemu-system-i386. It is reproducible by booting a Debian i386 wheezy
guest on such a system. Unfortunately the week-end was too short for
finding the issue, I'll continue looking in the next days.

I have a few comments on the individual patches, I'll send them asap.
Note that I don't have an R6 machine, so I haven't been able to test
that part.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net