[hatari-devel] Hatari startup freeze regression with 030 MMU + EmuTOS + non-VDI screen

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

Everything was fine with TOS boot tester, until I started testing EmuTOS 
with 030 MMU.  That seems currently completely broken, Hatari freezes at 
startup with 100% CPU usage.

There's no freeze with TOS v3 or v4, but with latest EmuTOS 1024k and 
512k versions, Hatari freezes to black screen.

(With 0.9.x EmuTOS versions screen is white and freezes happens a bit 
later.)

It does not matter whether TT or Falcon is emulated, or at what clock, 
or what other options are used.

As long as 030 + MMU is enabled in cycle-exact mode, and something else 
than VDI mode [1] is used for screen, Hatari freezes.

For example:
$ hatari --tos etos512k.img --machine falcon --dsp none --memsize 4 
--cpu-exact on --mmu on -m

Perf shows time being spent in:
-------------------------------
 15.11%  Video_InterruptHandler_EndLine [1]
 13.23%  CycInt_AcknowledgeInterrupt
 12.28%  Video_ConvertPosition
 12.15%  CycInt_AddRelativeInterruptWithOffset
  9.36%  CycInt_InsertInt
  5.68%  Video_GetPosition
  4.36%  Cycles_UpdateCounters
  4.18%  CycInt_CallActiveHandler
  3.26%  Cycles_GetClockCounterImmediate
  2.72%  Cycles_GetCounter
  2.66%  Video_TimerB_GetDefaultPos
  2.19%  m68k_run_mmu030
-------------------------------

(rest is spent on libs and non-Hatari processes.)

[1] maybe video interrupt emulation is now broken  with cycle-accurate 
030 MMU emulation?


	- Eero