Realtek RTL836x series of Ethernet switch ICs has a 8051 hidden inside
twitter.com
twitter.com
I'm not seeing anything nefarious here at all, the RTL8367 is designed for 'smart switches', the 8051 is there to run the webUI. In general, 8051s are _everywhere_, it's one of the most used IP cores in the world. I'd bet the average household has a a few dozen of them in various devises.
EDIT: Looking through some older tweets from the same author, I think the claim is based on pinout similarities between the RTL8366 and the RTL8367. My guess would be that these two parts are the same die and the RTL8366 parts are the ones where a defect was found in the CPU core which would be a likely source of manufacturing defects.
I've used similar parts, like bluetooth transceivers, where there are two SKUs: one is a cheaper commodity part and comes with built-in firmware.
The other more expensive part allows for modifiable firmware (and, thus, exposes the processor inside). My cynical POV is that the chipmaker knows they'll be spending more internally on SDK development/support and field rep time. So that needs to be recaptured in the cost of the part.
Likely cheaper to manufacture a single chip and do market segmentation via software than manufacture two if the savings are measly (let’s assume 16nm process) 1/500 mm^2 per chip.
https://investorshub.advfn.com/boards/read_msg.aspx?message_...
But you're right about the rest. The segmentation is achieved with fuses and firmware.
I don't even see the cynicism. But, in any case, most people would probably consider the standard business school price finding advice to be more cynical: costs are a lower limit only. Beyond that, only the customer's willingness to pay is relevant.
Airline fares are good examples, or any of the food products being sold branded as well as unbranded/store-brand/generic/white label versions.
I don't think NCR really recovered till the NCR 7878 came out, setting the bar for a robust scale/scanner
My old employer is prepping for the shift with their software vendor currently, should allow for a much longer period of support & remote debugging...
On the other hand, if everything came out good, then the manufacturer sells the part as the high-end product.
AMD CPUs, especially the 3-core parts, were often unlockable by re-enabling the disabled cores in BIOS. Lucky buyers could get 4 cores for the price of 3.
However, there's another view of fairness (or maybe just a desirable feature of the markey). It should be possible for a customer to purchase only the features required to meet their requirements, and that they should be able to spend less money on less features.
Those two views of fairness can be in contradiction with each other.
The ability of a company to be able to provide a family of products at a variety of feature and price points cheaply is aided by their ability to do software based control of features. Use of a single physical platform reduces cost (which aids the final customer). If a company had to price everything at a single feature/price point, the average customer will see an increase in price to pay for features they do not want.
Anyhow, that's the argument. Still feels icky though.
The alternative is to design a car with a smaller battery pack. This would mean additional design work, and less efficiencies of scale. This would make the cost of both tiers of car go up.
The way this is phrased it's just differing margins for the same physical good and no additional cost for the manufacturer to justify a higher price. (Not sure if this is an accurate assessment, but that is what I get from the comment.)
Suppose that once r&d is considered, two chips (single/dual core) cost roughly the same to manufacture.
I could sell everyone the 2 core processor and say sorry to customers needing less, you have to pay full price. Or I could maximize the market I can sell to, and offer the same product with a burned out pin that locks out the second core. Or even better, sell everyone a soft upgradable 1 core, so you buy the entry level product, and when/if you need it, you can buy into the upmarket model.
But the important part here is: ignoring r&d. Yes, the hardware I'm selling you could do more, but you're subsidizing fewer engineers, so you only get access to their efforts, not every effort my company has ever produced.
This part:
> Or even better, sell everyone a soft upgradable 1 core, so you buy the entry level product, and when/if you need it, you can buy into the upmarket model
... is still one that I would have trouble with, were I the manufacturer. I would be inclined to throw away profits on high end sales and price them all low for these scenarios, based on my desire to be fair to the customer and not run the thing like a casino with physically dubious upsells for what they already purchased. Yes you read that right. This is one reason why I am not in a position to be running Intel.
Are software companies allowed to charge a price for their free-to-produce products at all?
If they are, how would you justify that?
And why would that justification not apply to CPU vendors, if they’d label their hardware as freebies where you’re just paying for a perpetual license?
> Also, as you seem to acknowledge, when you purchase software, you are technically acquiring a license.
That's really just semantics.
When you buy a mainframe with some CPUs disabled, you're not only buying the hardware, you're also buying a license for the OS, which includes the ability to use a certain number of CPUs of the machine.
You're commenting earlier that you'd price the higher SKU the same as the lower SKU, but I think it's just as likely that you'd simply not make a higher SKU because there'd be no extra profit in doing so.
It may seem perverse that hardware is being sold with capabilities deliberately crippled, but doing so if often one of the only ways to justify developing these higher SKU capabilities to begin with. This kind of product differentiation makes it possible for customers to buy a higher SKUs that otherwise simply wouldn't exist.
This, I think, is the crux of the matter.
Yes, they could just offer one model, at a price point somewhere between the two different models. However, now some customers who were happy with the lower-end model are priced entirely out of the market, which is bad both for those customers and the seller. Balancing that, customers for the higher-end version have paid less (good for them) and some new customers will have entered the market because the new features/price-point is attractive to them (good for them and the seller).
It's not obvious which of these will dominate (so whether everyone is net better off or not), but in the absence of any hard data relying on the self-interest of the seller to maximise their own benefit seems reasonable here.
A strange accusation without proof.
Fair warning I can't open the Tweet, Twitter is blocking me... or maybe I uBlock them? so I'm making a statement in similar unsupported ignorance.
>whitequark
>nonconsensually submitted to the orange website
I could understand if your personal site gets "HN'd", but a Twitter user doesn't pay for bandwidth.
I'm not talking about myself. Never even remotely implied it.
HN is filled with the 0.1% that truly are experts on individual topics. You'll find people that wrote papers on quantum mechanics chiming in on QM discussions, people with decades of RF HW experience correcting those 5Ghz conspiracies, start-up founders jumping in on discussions, and (in this case) people that can read spec sheets and understand how systems like these are built and programmed.
No one's an expert on much, but plenty are experts on something. Twitter's often too shallow to even entertain the possibility of being wrong.
I think it's in fact super healthy for people to set clear boundaries about how they engage with this place, and "this Twitter thread was not posted as a solicitation for an HN thread, or for HN people to jump onto my TL" is a very sane boundary indeed.
That people read this kind of boundary drawing as "hatred" is itself a good illustration of the issue!
http://arny.tjps.eu/OpenWrt/EasyBox904xDSL/oem-firmware-info...
And
https://github.com/uwehermann/easybox-904-xdsl-firmware/blob...
>SMI SLAVE FOR EXTERNAL CPU
your links talk about external programming of smart switches, you could do this 10 years ago with RTL3866
https://hackaday.com/2010/05/26/unlocking-the-crippled-poten...
http://spritesmods.com/?art=rtl8366sb
or 15 years ago with RTL8309sb
https://web.archive.org/web/20061118184321/http://pupa.da.ru...
I remember in the times of region locked DVD players, some device hacks targeted the same processor inside the DVD player unit (I'm talking about the IDE/SATA drive, not the whole player)
You would want to leave the circuitry in place for production because it's largely unnoticeable and changing it could introduce bugs.
>so apparently most unmanaged switches have a 8051 connected to the switch fabric inside. would it occur to you that pretty much any switch on your network could be monitoring or injecting traffic? sure would not to me
https://twitter.com/whitequark/status/1175692567150653440
>i wasn't previously aware that it's possible (it does makes sense i guess? doing switching entirely in gateware is more risky), and it's an absolutely incredible placement for a near-undetectable implant
Why are we still using old 80xx chips in embedded applications? Expired patents? Existing toolchain? Inertia?
It’s hard to believe we couldn’t do better, cheaper and more power efficient solutions with all the tricks we’ve learned in those almost 40 years.
Well known cores that are proven reliable over several decades, very well documented and seemingly any bugs are known already. Also cheap
I imagine it has to deal with encrypted traffic, IPv6 these days, which doesn’t sound fun on 8bit chips.
It seems weird to me, but again, it’s not my field.
Does any one know what is the market share for 6502, 8051 and ARM?
They sometimes might even exist along side of a fully 32bit ARM micro.
You'd be amazed at the amount of chips that have an 8051 in them...
It's the instruction set that has been retained, not the silicon design. The variants these days are more power-efficient and powerful in terms of MIPS and peripherals, and have indeed benefited from years of R&D.
And if the silicon design is new, we are not benefiting all the much from decades of battle testing, right?
I can’t imagine how a clean, embedded first 32bit ISA design wouldn’t be more appropriate.
But 8051 was designed for embedded:
> The Intel MCS-51 (commonly termed 8051) is a single chip microcontroller (MCU) series developed by Intel in 1980 for use in embedded systems
(wikipedia)
> I can’t imagine how a clean, embedded first 32bit ISA design wouldn’t be more appropriate
I guess we'll see how riscv will develop.
(disclaimer: I sell an 8051 based product, have sold them in the past - never again)
Even if it was the exact same circuit design, that's only a small part of what makes a processor and negates all of the advancements in chemistry, lithography, engineering, and everything thing else that goes into manufacturing a chip.
It's like saying "why are we still using scissors|staplers|brushed electric motors|internal combustion engines?"
Speaking as a member of a 2 electric car family, I can tell you the current EVs are not 100% replacements for ICE vehicles.
200+ mile EVs can easily supplant ICE vehicles in almost all use cases but they're still more expensive than the average selling price of a new car in the US. That says nothing about the used car market.
It's an embedded application - if the core does the job exactly what it is intended to be, why upgrade? What do you expect more from an eth chip when it already provide full 1gbps capabilities? In other words, there is no room to be better. It could be more power efficient but it is alway efficient enough. Also changing the core will cost a huge amount of money as it needs to change the production line.
Using more recent tech such as ARM isn't always cheaper or better.
It's sort of like how MIPS never really got a lot of traction outside of 5ish stage, in order, optional FPU designs. That's ultimately the gate count niche where they're still competitive.
Did you know the processor in the b43 network cards use this really weird architecture developed by the company that made the backplane? It also doesn’t actually handle the radio baseband, that’s done entirely in hardware (If my reading of the firmware is correct.) Which makes me wonder why having more open source WiFi card firmware is rare, I guess it still handles calibration and some other things that you could use to intentionally make the card misbehave but it’s a lot more limited than I expected.
Does realtek publish their documentation on products publicly? I found it interesting that the author of the tweet has, for example, a pinout.
I suspect it's more of a "we're not interested in answering your questions, which you will certainly have if we give you a datasheet directly" than any real concern over IP.
Tweets make bad HN posts; not only because they disappear, but they're too short to have all the details, they don't capture a whole conversation, when they do it's disjointed and in pieces, people jump to conclusions, etc.
The paranoid would say "it's a hidden backdoor for running shady code". However, anyone with even a passing interest in embedded computing would recognize that it's designed to store and run official firmware; nothing shady or underhanded about it.
So in short, in this particular instance it's much ado about nothing.
You can kill someone with a hammer, but that doesn't mean the hammer was designed for killing.
And this is why people avoid this site....