Re: [hatari-devel] Cycle exact setting affects nf_scsidrv

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 10/12/18 12:25 AM, Uwe Seimet wrote:
> > Is st_buffer data something that could be program code read from
> > the SCSI disk?  If yes, all caches would need to be flushed for that.
>
> Yes, you are probably right. Just checked that I'm also flushing both in
> HDDRIVER. One more updated patch attached. The sense data buffer and
> the buffer for the bus name do not need flushing the instruction cache
> because you cannot transfer program code this way.

Thanks, I pushed you patch:
https://hg.tuxfamily.org/mercurialroot/hatari/hatari/rev/d6355bb2dc25

When checking that, I noticed that nf_scsidrv.c doesn't validate
that the memory addresses specified through the NF API are valid,
like GEMDOS HD driver does:
-----------------------------
        /* Check that read is to valid memory area */
        if ( !STMemory_CheckAreaType ( Addr, Size, ABFLAG_RAM ) )
        {
                Log_Printf(LOG_WARN, "GEMDOS Fread() failed due to 
invalid RAM range 0x%x+%i\n", Addr, Size);
                Regs[REG_D0] = GEMDOS_ERANGE;
                return true;
        }
-----------------------------

Could you send also a (tested) patch that adds those checks?


	- Eero