Mistakes Microsoft made in the Xbox security system (2005)

The fundamental problem was that x86 had no mechanism for verifying first instruction at the time (Boot Guard and Platform Secure Boot provide that now), and the only way to try to deal with this was by adding immutable storage - but given where they put it, that was expensive, so small. And that led to making poor tradeoffs, influenced by having what was clearly not a great level of adversarial security analysis, but even implementing that perfectly they'd still have been fucked by the gate A20 thing which is maybe the absolute funniest legacy design failure that perpetuated well into the 21st century.

(The Intel/AMD difference on IP rollover is also funny but given the number of other ways to circumvent things...)

I actually use this as a teaching example - it's a great way to talk about how CPUs actually work and interact with other hardware, and a good understanding of this gives a lot of insight into low level platform design

Microsoft clearly learned from their Xbox and Xbox 360 mistakes, leading to unhacked (?) Xbox One and Xbox Series X consoles: https://www.platformsecuritysummit.com/2019/speaker/chen/

I do take issue with this fluffy statement at the top:

> If the reader finds the mistakes in the design, this proves that Microsoft has weak developers.

(The article even goes on later to say basically "don't attribute all of these to stupid engineers", and the explicit 17 mistakes are almost entirely not related to the technical content of the security breakages, so there's a sleight of hand being performed here already between "mistakes in the design" and "mistakes in organisational software engineering practice"!)

While I have no love of Microsoft software, and clearly the Xbox was woefully insecure, the statement ignores the fact that knowing there is something to find is often enough to find it. I am failing to find the definitive article about this, but there's something about this by Michael Nielsen or Andy Matuschak or similar. One of its examples is a quote by Kasparov or Magnus Carlsen or similar, to the effect that the single word "now" at the right time would be enough to win a game, because it would announce that there was a discovery to be found. This article is entitled "17 Mistakes…", and it also presents the relevant details of the design rather than all the details of the design, so the problem of finding the mistakes given the description is much, much easier than the problem of reviewing a complete design spec.

>Keeping this in mind, the attack on the Xbox is pretty straightforward: If we connect the CPU's A20# pin to GND, the Xbox will not start from FFFFFFF0, but from FFEFFFF0 - this is not covered by the secret ROM, but is ordinary flash memory, because flash is mirrored over the upper 16 MB. So by only connecting a single pin, the secret ROM is completely bypassed

This is interesting to me because it sounds remarkably similar to my admittedly very naive understanding of how jailbreaking the Nintendo Switch (original console, not sure about the recent Switch 2) works, just inverted; there's a single pin under where the right joycon slots in that seems to be responsible for ensuring that the OS (or firmware? my understanding of what step this is in the boot process is pretty fuzzy) is properly loaded, so shorting it is enough to allow injection of pretty much whatever you want to boot instead. The entirety of what you need to do from a hardware perspective even as a complete novice is access to a desktop/laptop, the USB-C cable that comes with the console, and a plastic nub people 3D print and sell on Amazon for like $8 to slide into the joycon slot, and you're done.

All this makes me wonder how hard it should actually be for console makers to catch these sorts of things ahead of time. Shouldn't be be fairly straightforward to just look at every single pin and consider what happens if it's turned on or off? Assuming the others are in the default state would be sufficient to have noticed both the issue with the Xbox and the one I describe in the Switch, so it's not like there's some insanely high number of combinations you need to consider to at least make it twice as hard to hack. It's understandable that in 2005 maybe this wasn't something that Microsoft would have thought of in their first time making a console, but the Switch came out over a decade after that, and Nintendo has been pretty vigorously going after "piracy" since before then, so it's hard to imagine that this was due to indifference.

Discussed once (and I do mean once):

17 Mistakes Microsoft Made in the Xbox Security System - https://news.ycombinator.com/item?id=781036 - Aug 2009 (1 comment)

Alternatively: Paths to Freedom.

RSA doesn’t take that much code. The Atari lynx would initialize all the hardware, perform RSA, and load the next step from the cart in 512 bytes.

All of this nonsense because they underpriced the console to overcharge on games.

This is from 2005.