Re: [hatari-devel] Hatari/EmuTOS Conflict

[Vincent Rivière <vincent.riviere@xxxxxxxxxxx>]

On 11/06/2013 01:03, Nicolas Pomarède wrote:
> I don't see any problem in this part, acia reset is correctly detected and
> emutos-cvs-20130604 boots with functional mouse/keyboard, even without the
> patch I added to force acia at 9600 bauds on cold start.
>
> Here're the traces :
>
> acia ikbd write fffc00 cr=0x03 video_cyc=59888 452@117 pc=e06b1a instr_cycle 16
> acia ikbd write cr data=0x03 VBL=53 HBL=117
> acia ikbd master reset VBL=53 HBL=117
> -> reset is made
>
> acia ikbd write fffc00 cr=0x96 video_cyc=59912 476@117 pc=e06b20 instr_cycle 16
> acia ikbd write cr data=0x96 VBL=53 HBL=117
> acia ikbd start timer divider=64 cpu_cycles=1024 VBL=53 HBL=117
> -> acia is initialized at 9600 bauds

Even if it works, this looks like a bug in EmuTOS.
The ACIA speed should be initialized before sending bytes, isn't it?

> But there's a problem in Emutos : we can see it still sends 0x1a and 0x12
> *during* the reset, which is not correct, it should send them after getting
> a 0xf0 or 0xf1 from the ikbd.

I wrote ikbd_readb() especially to avoid this situation.

> static UBYTE ikbd_readb(void)
> {
> #if CONF_WITH_IKBD_ACIA
>     unsigned int i;
> zz
>     /* We have to use a timeout to avoid waiting forever
>      * if the keyboard is unplugged. The standard value is 300 ms.
>      */
>     for (i = 0; i < 300; i++)
>     {
>         if (ikbd_acia.ctrl & ACIA_RDRF)
>             return ikbd_acia.data;
>
>         delay_loop(loopcount_1_msec);
>     }
>
>     return 0; /* bogus value when timeout */
> #else
>     return 0; /* bogus value */
> #endif
> }

If it returns before the byte is received, either the test of the ACIA 
status bit is wrong, or the function timeouts. Maybe the timeout value is 
not adjusted correctly. Also as I told you, in those early stages 
delay_loop(loopcount_1_msec) is not accurate and will not wait enough time 
if he CPU is faster than expected.

By looking at the disassembly, the bit 0 of 0xfffffc00 is correctly tested 
in the loop.

> Is this version of Emutos CVS really compiled with CONF_WITH_IKBD_ACIA ?

Yes, proven by the disassembly.

> Maybe this line should be enabled :
>
> 560    #if DBG_KBD
> 561        kprintf("ikbd_version = 0x%02x\n", ikbd_version);
> 562    #endif
>
> To see if emutos really received 0xf1 or 0xf0, or bypassed the test.

I will setup a test environment (latest Hatari, EmuTOS, MIDI debugging...), 
but I'm currently very busy. Anyway, I will need to find the time to look at 
that before the next Hatari/EmuTOS release.

--
Vincent Rivière