[PATCH 12/42] libcrypt/crypt-md5.c: Fix overflow issues

[Peter Dufault]

> On Mar 26, 2015, at 17:26 , Joel Sherrill <Joel.Sherrill at oarcorp.com> wrote:
> 
>> I think if 16 bit Harvard architecture (64K instruction / 64K data) targets are no longer supportable then 16 bit should be deprecated and then abandoned.  If you can still do a lot with RTEMS in 128K then that useful subset of the code should be identified and kept 16 bit clean, that would be a good requirement on developers and that part of the code base.
> 
>> That is, if anyone wants to do that, I currently use 4MB instruction / 4MB data as my minimal targets that can be comfortably extended during the support life time (I want TCP/IP and NFS as part of my minimum, your mileage will definitely vary).
>> 
> The original target platform for RTEMS was 1MB RAM that was used for 
> everything.
> I still think this is a very reasonable memory profile for most 
> applications.
> 
> I am not sure if the 64K instruction/data is that useful but 16-bit 
> architectures are
> not necessarily limited to 64K address spaces. We all remember the 8086 and
> variants. The m32c has 24 bit address space.
> 
> I don't know the requirements for what I used to frequently call 
> Tiny/RTEMS.
> If you have a 20+ bit address space, then there isn't much of a code 
> size issue.
> There will be combinations of code that just won't fit. I expect the new 
> TCP/IP
> stack to be a casualty there. But the old TCP/IP stack worked quite well in
> a 1MB environment and I would expect LWIP to do even better.
> 
> So my focus is just on being 16-bit integer clean. I don't think we will 
> shrink
> into the smallest AVR CPU models but something like an 8086 in large memory
> model or an m32c shouldn't be an issue for most of RTEMS.
> 
> But yes.. we should have some insight into which features are hopeless in
> a 16-bit integer environment and maybe some idea of what is really too 
> small.
> 

We're talking about different things.  I looked quickly at the M32C and see it has 24 bit address registers, and the 8086 is a nightmare.  I think of 16 bit architectures as having 16 bit address and data, and not a mixed architecture with larger sized pointers (clean, un-aliased, pointers).  I think you want to keep RTEMS 16 bit data clean, but not 16 bit address clean.

Peter
-----------------
Peter Dufault
HD Associates, Inc.      Software and System Engineering