For use only on NTSC Genesis Systems.
Avatar image by CyanSorcery!


Tumblr (inactive)
tumblokami.tumblr.com/
Twitter (inactive)
twitter.com/Techokami

summonercat
@summonercat

I am beginning to see why ARM is looked at as the future of computing. Also, I'm still waiting on ASUS to deploy the update for the motherboard! How do I light their arse on fire via Ethernet to get it done?!

I might be a little grumpy because the hernia's flaring up again. Goodie.


summonercat
@summonercat

...Are you fucking kidding me?! Oh.. fucking hell!


techokami
@techokami

And I didn't even touch on how downright cagey the manufacturers of ARM CPUs are. I dare you to try and find an actual datasheet for Qualcomm's latest ARM offerings! Now compare that to uh... Intel. Who doesn't even ask for an email address to download both books for the 13th/14th gen Core CPUs.


techokami
@techokami

It's in the official datasheet, Errata: RP2350-E9, p.1340
The "fix" is either to do hacky software workarounds or "don't do this". Damned shame, this was looking to be an interesting and useful part! Though, most of them got used up for DEF CON:


You must log in to comment.

in reply to @techokami's post:

in reply to @techokami's post:

we did fwiw

The 68LC040 is a low cost version of the Motorola 68040 microprocessor with no FPU. This makes it less expensive and it draws less power. Although the CPU now fits into a feature chart more like the Motorola 68030, it continues to include the 68040's caches and pipeline and is thus significantly faster than the 68030.

Some mask revisions of the 68LC040 contained a bug that prevents the chip from operating correctly when a software FPU emulator is used. According to Motorola's errata, any chip with a mask set 2E71M or later does not contain the bug. This new mask was introduced in mid-1995 and converted the 68LC040 chip to MC status.

The buggy revisions are typically found in 68LC040-based Apple Macintosh computers. Chips with mask set 2E23G (as used in the LC 475) have been confirmed to be faulty. The fault relates to pending writes being lost when the F-line exception is triggered. The 68040 cannot update its microcode in the manner of modern x86 chips. This means that the only way to use software that requires floating-point functionality is to replace the buggy 68LC040 with a later revision, or a full 68040.

in reply to @techokami's post:

in reply to @techokami's post:

in reply to @techokami's post:

Qualcomm is definitely one of the worst.

They don't want to talk to anyone if it's not going to shift at least a couple hundred thousand SoCs, and even then, you're going to be passed off to one of Qualcomm's design partners, and only receive redacted documentation for the parts.

The part that always amused me was that certain vendors have inconsistent redaction. So a feature that is redacted in one SoC's product manual might have all the details described in another that is available without NDA.

I worked at a contract engineering firm that was a design partner for most of the major vendors, and we worked with all of them except AMD at least once during my time there.

So I have a good understanding of what's available publicly, what's available under a standard customer NDA, what's available under a design partner NDA, and what's "Hell no, we're not ever giving anyone that." for almost all of them.

I see. I've not been able to go that deep before because I don't work in that field (I'm just a hobbyist EE) but I just find it so strange that you can get so much info from Intel without needing to give anything in return, yet you need to give up all your PII to, say, NXP to get any datasheets about their current products, which are like toys compared to Qualcomm's stuff which told me I needed to sign partnership contracts to see anything. Not like the days of the 68K, where you could just write to Motorola and they'd send you a box of documentation and samples. (I actually obtained one of these boxes off eBay, it's so awesome)

A lot of the documentation that Intel and Arm Technology gives away is specifically of interest to software developers, who these days would be unlikely to be a direct sales lead that Intel or Arm Holdings would want to pursue.

They however contribute to indirect sales, through Arm's and Intel's partners. So there is value to having the information out there.

Vendors like TI and NXP, which are some of the easier ones to get documentation out of, haven't actually changed as much as you would have thought.

Pre-late 90s, early 2000s, whether it was gathered by you calling up a sales office, writing in, visiting a vendor's booth at a trade fair, or sending in the forms that used to commonly be included as the last pages of their printed documentation, the information being collected to ship you the printed material became a sales lead, which someone from the nearest regional sales office would determine if it was worth following up on. (Nowadays, followup is mostly automated email spam, but sales people still call or make site visits to more lucrative leads.)

Since engineers and corporate buyers would fill out that material with office phone numbers and addresses, it wasn't really considered PII at the time, and they wouldn't normally bother retaining information from a hobbyist or student because it wouldn't become a sale.

Chip samples and free or at printing cost documentation being sent to hobbyist was mostly a goodwill thing, since positive experiences made it more likely you would pick them if you went out into industry.

Ironically, that sort of thing started dying out in the late 2000s, early 2010s, because the growth in popularity of hobby electronics happened around the same time vendors started doing major cuts in budgets for their sales department. (It became pretty common that the regional sales office would be closed, and the vendor would have just one sales rep for an entire region. There is also a whole digression into the Arduino-chasing low cost devkits trend that happened at the same time, I could get into.)

Another factor that lead to material being placed behind free account walls, is the US State Department enforcement of export restrictions on dual-use technologies. Vendors (and their sales channels) have legal obligations to ensure that the people receiving access to technical information or dual-use technology aren't on the export restrictions Entity List, or fronting for a person, organization, or country that is on it.

Now companies like Qualcomm are another barrel of fish. They deliberately do not want to deal with customers that are doing low-volume production. By focusing on high-volume OEMs, they can operate with fewer sales/application engineering staff than vendors that support lower volume customers. Modern ICs are not trivial to design into products, and it doesn't take many hours of sales support to make small-volume sales unprofitable for a vendor. (Support for a hobbyist integrating a sampled part is a net negative.) Issues arising from mistakes or oversights during integration by OEMs sometimes get blamed by the public and reviewers on the SoC too, so poor integration can cause PR problems that impact sales. There are also regulatory requirements that come into play, because they specifically focus on the telecoms segment. Code running on or integrating with baseband processors needs to be certified. (Network operators also often require their own interop certification before a device is allowed to be connected to their network.)

Another issue that impacts almost all vendors is that SoCs often contain a lot of licensed IP. PowerVR GPUs were integrated into a lot of SoCs, for example. Encode/decode codecs are often under restrictive IP licenses, and the hardware details of the specialized coprocessors that run them can reveal a lot about implemented specification. Complex peripherals like USB controllers and even interconnects like an internal network on chip fabric are other common licensed third party IP.

Then vendors also still try to do security by obscurity. The ME in Intel processors is one of those classic examples of that.

Another factor is vendor's trying to preserve product segmentation. The use of common chip dies across a product family, and then binning or fusing a particular die for a particular product segment is another reason vendors like to make documentation hard to get. It is sometimes effective in preventing customers from trying to use the questionably good undocumented functionality on cheaper parts instead of buying the more expensive part. It also obfuscates how a particular generic IP block might be getting reused across market segments with specialized firmware blobs to have it function as a dedicated and specialized coprocessor for that market.

In an ideal world, a lot of this stuff wouldn't be an issue, and I'm also frustrated by a lot of this stuff when it comes to my own personal projects too, especially since I'm no longer in the know on the latest stuff, but it is understandable why things are the way they are, even if I find a lot of it very silly.