Re: [hatari-devel] Hatari on constrained devices

[Thomas Huth <th.huth@xxxxxxxxx>]

Am Thu, 6 Jan 2022 13:25:46 +0200
schrieb Eero Tamminen <oak@xxxxxxxxxxxxxx>:

> Hi Thomas,
> 
> I noticed that you removed WinCE support.
> 
> I removed the Hatari v1.x config examples for other constrained
> devices, as they're not supporting SDL2 either.

Good idea, thanks!

> Should the SMALL_MEM option be removed too?

I think we could remove it nowadays, yes - but actually I rather tend
to make SMALL_MEM=1 the default nowadays. Rationale:

- The huge STRam[16MiB] array is kind of ugly

- If we ever have a bug in one of the IO memory or ROM access
  functions, it might be possible for a malicious Atari program to
  access other variables in the BSS of Hatari (since they are in
  well-known locations relatively to STRam[]) - if we malloc the memory
  instead, the chances are way smaller that something ugly could happen
  here beside crashing Hatari.

- I think the speed optimization that we get via the huge static
  STRam[] array is completely neglectable on modern CPUs. Actually, I
  just also did a completely unrepresentative benchmark by running
  two Hatari instances in parallel and comparing their CPU usage with
  "top", and the build with --enable-small-mem was even 0.3 % up to
  1.0% better than the one without --enable-small-mem! Might be just
  noise, of course, but at least there is no big difference between the
  two modes.

So unless someone has some good objections, I think I'm simply going to
switch the code to use the SMALL_MEM behavior by default.

 Thomas