Re: [hatari-devel] Issues after recent WinUAE CPU core changes

[Toni Wilen <twilen@xxxxxxxxxx>]

If you enable MMU logging (mmu_common.h and cpummu030.h) and check the
log, do you see bus error from address that ends to FFC before it
crashes? (la=c0007ffc etc, it also probably have "ffff ILLEGAL._" at the
end of line which means prefetch bus error)

If yes, below patch improved prefetch handling after branch to address
space that generates bus error. Previously it partially reloaded already
prefetched data after RTE, now it does not but it seems there is some
bug or MMUSR flags are not right in this situation.

I tried few different Amiga Linux variants (some ancient Debian) and one
of them does not crash but generates unexpected line-F because of above
prefetch fault.

> Hi,
> 
> Commit regressing this is:
> -------------------------
> commit 759d60f9dcf3f2971aabd905adb3d7404e8d3b19 (HEAD, refs/bisect/bad)
> Author: Nicolas Pomarede <npomarede@xxxxxxxxxxxx>
> Date:   Thu May 21 22:10:15 2020 +0200
> 
>     68030 MMU fixes. Software fixed pipeline stage fixes and more
> compatible (prefetch) locked rmw and mmufixup fix (from WinUAE 4.4.0
> beta3+ 2020/05/21)
> -------------------------
> 
> See:
> https://git.tuxfamily.org/hatari/hatari.git/commit/?id=759d60f9dcf3f2971aabd905adb3d7404e8d3b19
> 
> 
>     - Eero
> 
> 
> On 6/2/20 5:37 PM, Eero Tamminen wrote:
>> On 6/1/20 7:49 PM, Toni Wilen wrote:
>>>> Issue goes away only when I disable *both* 030 cache emulation and
>>>> prefetch with:
>>>> --cpu-exact off --compatible off
>>>>
>>>> Disabling just one of them isn't enough.
>>>
>>> 68030 + MMU + prefetch has been broken for long time
>>
>> MMU support is not that old, 030 MMU + cache
>> support was added some time in 2017, and many bugs
>> were still fixed in 2018.
>>
>> Do you remember how long ago the above combo was
>> fully working (in winAUE CPU core version)?
>>
>>
>>  > (It might have accidentally worked in some
>>  > situations)
>>
>> There were quite a lot of mystery crashes on user-
>> space entry / exit also earlier, so I'm not too
>> surprised about that. :-)
>>
>>
>>> It was fixed (at least Amix boots again) only 1-2 weeks ago.
>>
>> I was testing Hatari version that had Nicolas' CPU core updates from
>> yesterday.
>>
>> I'll hope to have time to bisect it this evening.
>> Happily it's fully reproducible.
>