Brendan wrote:
Yes, all of this awful old junk had to stay because BIOS requires it; and making it possible to get rid of all the legacy hardware is the reason Intel wants to get rid of BIOS. I'm thinking it'll probably take less than 6 months between "old chipset without BIOS" and "new chipset without BIOS, A20, PS/2, PIC chips, PIT, legacy ROM areas, legacy VGA area, ..."); and by 2021 we'll finally end up with lean and mean PCs without the unnecessary baggage that no major operating system (Windows, OSX, Linux) has required for at least a decade.
More than likely, they will simply eliminate real mode entirely at the 2020 mark, making the A20 line and the BIOS routines moot and making UEFI booting the only option.
Several of the other things you've mentioned are ones which that were imposed on Intel by IBM's designs for the PC, PC/AT, and PS/2 product lines, and are, for the most part, only emulated even now, IIUC. For example, the
Local APIC, inside of x86 CPUs as far back as the Pentium 'refresh' of 1994, bears little if any actual resemblance to the 8253 PIT even though the former was made by Intel originally - and IIUC, the legacy behavior is in the CPU's internal firmware, not hardware.
I wouldn't be surprised if they eliminated booting into 32-bit p-mode, for that matter, and made 32-bit code a legacy support feature to be phased out over a few years, as well. Even supporting just two different operating modes is getting burdensome, and they have already stopped making CPUs without 64-bit modes for mainstream use, so this would simplify a lot.
This would be a lot less painful than the shift away from real mode has been, simply because (unlike with real mode software) most of the programs were written in high-level languages, and with less direct operations on the hardware, and most of the source code is still around and can be re-compiled with some tweaks. For that matter, the ubiquity of code compiled to .NET and JVM, or written for interpreted languages such as Python or Ruby, means most won't even need recompiling.
I'd finish by adding that recent developments - in particular, the problems Intel is having with implementing their CPUs on a 10nm process - raise the hope that we will finally be able to put that horrible x86 garbage out of our misery for good soon. I'm not sure, as we've thought that revenant was down for the count before, but maybe.