RISC-V from scratch 2: Hardware layouts, linker scripts, and C runtimes

https://twilco.github.io/riscv-from-scratch/2019/04/27/riscv-from-scratch-2.html

by pplonski86 • 7 years ago

170 20 7 years ago

twilco.github.io

tindleaj• 7 years ago

Exciting writeup! Looking forward to reading this later.

PaulHoule6• 7 years ago

RISC-V, System/360 for the 2020s!

microcolonel3• 7 years ago

Indeed, though I think people here think you're saying that as an insult rather than a compliment.

PaulHoule6• 7 years ago

Yep.

The real genius behind the 360 was scalablity. Not only could you buy a wide range of different sized machines, but IBM has kept making better 360s to this day.

Intel has accomplished the same with the 8086 (by accident more than design) and so has ARM.

Once you get past the x86 and ARM the price/performance ratio gets a lot worse. Thus you end up with expensive WiFi routers that allegedly can work as a file server but can't really handle it, or things like the Samsung Smartthings hub that has to offload even simple things to the AWS cloud.

RISC-V could make the market for embedded chips competitive again. RISC-V scales even more than the 360 did since you can put in just the features you want, thus you might be able to build something which is high performance but doesn't waste a transistor on features you don't want.

bogomipz3• 7 years ago

Could you elaborate? What is the connection between IBM System/360 and RISC-V?

chubot• 7 years ago

I definitely wasn't around then, but Fred Brooks talks about it in some of his writing, and the System/360 work is cited by his Turing award.

My understanding is that before System/360, software was tied to particular machines, which were naturally of a particular speed and capacity (and such resources were a very big deal back then).

When institutions bought new machines, they would MANUALLY REWRITE their software in the assembly language of the new machine. In other words, they were like manual compilers.

I think a lot of the compilers were actually specific to the machine as well! The hardware manufacturers invented their own programming language for their own machines. Portability wasn't a "thing". Many institutions probably only owned a 1 or 3 computers from the same manufacturer anyway.

So basically System/360 was a single instruction set for a lot of different machines, with differing speed and capacity. So you didn't have to do this thing that seems crazy to us today -- you could run the same software on may machines.

Software back then was really just an extension of hardware... The views have changed a lot [1]

Also from what I'm reading, the notion of an "abstract" machine was also not really a thing. There's a conceptual leap from a physical machine on your department's floor to some binary interface you can program against abstractly (the ISA), which we take for granted today.

https://amturing.acm.org/award_winners/brooks_1002187.cfm

Bob Evans promoted IBM’s vision to develop a single product line of general purpose computers with a common instruction set that permitted customers to preserve their investment in software as the moved from slower machines to faster ones. Evans assigned Brooks to lead the team to design this product line, called the System/360, which was announced in 1964. Brooks coined the term “computer architecture” to mean the structure and behavior of computer processors and associated devices, as separate from the details of any particular hardware implementation.

The importance of the System /360 cannot be understated: it was a widely successful project that transformed the face of business computing and reshaped the landscape of the computer companies throughout the world. Among many important contributions to the design of the System/360, Brooks was particularly proud of the 8-bit byte, which permitted the use of uppercase and lowercase alphabets and expanded the role of computers in text processing.

While the hardware architecture for the System/360 was well underway, it was clear that there was considerable risk in delivering the operating system for the new series of machines. Brooks was assigned to lead the software team in building what was perhaps the largest operating system project of its time. Brooks describes the lessons he learned in his classic text on software engineering, The Mythical Man-Month. It is from that experience that Brooks proposed “Brook’s Law”: that “adding manpower to a late software project makes it later."

[1] For another example of how wildly views have changed, I recall that early networking assumed that if you were admin on ANY MACHINE on the network, you had admin rights on this machine. There was almost no notion of network security.

RISC-V from scratch 2: Hardware layouts, linker scripts, and C runtimes

Related Stories