Hacker News Reader

Comment by TazeTSchnitzel | Hacker News Reader

The amazing thing about DVI, and by extension HDMI, is that it's just VGA but digital, with all the timing and synchronisation complexity that implies. Recall that DVI can have both an analog (VGA/DVI-A) and digital (DVI-D) signal in the same cable. They aren't independent; they share some pins and have the same timing. You could have made a CRT monitor that used DVI-D just by adding a DAC, though I'm not sure if anyone ever did.

DisplayPort does away with all that legacy. I assume the hardware to implement it is much simpler and more reliable.

View on HN

This also has side-channel implications, and is a reason why eDP is recommended in voting applications. https://www.eerstekamer.nl/nonav/overig/20160428/richtlijnen...

>DisplayPort uses a scrambler as part of its line encoding in order to flatten the Fourier spectrum of its emissions and suppress spectral peaks caused by particular image contents. This reduces the chances of any particular image content causing a problem spectral peak during SDIP-27 and EMI spectrum measurements. According to the standard, the scrambler reduces spectral peaks by about 7 dB. As a side effect, the scrambler also makes it far more difficult, probably even impractical, for an attacker to reconstruct any information about the displayed image from the DisplayPort emissions. [..] DisplayPort uses a small number of fixed bit rates, independent of the video mode used. Unlike with most other digital interfaces, video data is transmitted in data packets with header and padding bytes, and not continuously with a television-like timing. As a result, DisplayPort cables are not a common source of van-Eckstyle video emanations and this again will make it very hard for an eavesdropper to synchronize to the transmitted data.

Speaking of which: how does one force HDCP (say, in Linux)? That would transform this technology from a DRM nuisance (strippers easily available on AliExpress) into a Van Eck-countermeasure.

HDCP can't really be stripped, if it's on then something is an HDCP receiver. For v1.4 signals you can probably get something with the leaked key, but not for v2.2 as yet.

HDCP itself needs to be licensed in order to generate it, and that licence is only available to HDMI Adopter companies. However there is code to get some chipsets to turn it on: https://github.com/intel/hdcp

Grab it now before Intel deletes it completely.

HDCP is really a very ugly protocol designed just for anti-copying, I wouldn't build anything relying on it, and everything is harder with HDCP from an AV integration perspective. If you have long links that you want to secure, use something like SDVoE with encryption and authentication (bits are easily flipped in HDCP).

There are devices that will give you HDCP 2.3->1.4 conversion for compatibility reasons (or maybe wink-wink reasons), and after that you can use a HDCP 1.4 stripper. There are also devices that will strip 2.2 outright (they are marketed as HDMI splitters but, oops, one of the two outputs has unencrypted signal, an honest mistake); not sure about 2.3 though. Completely agree that HDCP makes everything harder, but in this scenario it would also make attacker's job of descrambling the eavesdropped signal harder. And this is after they have overcome packetization and spread spectrum hurdles, which the paper suggests is very challenging to do, so would be a (hacky) defense-in-depth for DisplayPort. Even more important for HDMI/DVI which lacks the aforementioned hurdles.

My understanding is that the downside of those downgrade devices is HDCP 1.4 will leave you stuck without 4k and HDR for content that uses HDCP.

Yes you can extract plaintext 1080p but not 4k HDR.

Voting computers are not recommended anyways.

It's still shocking to me that they're used widely, and also pains me that my 15+ years of railing against digital voting computers is now looked upon the same way as the hard-right-wing's complaints (made up ones, but in a semi-related space to my complaints) about said computers.

It should be a national citizen identity logon system, which a voting web site uses, everyone can vote from home or from a laptop provided by the independent voting facility.

No.

Sounds nice, but computer based voting systems, no matter how secure, clean or simple, are fundamentally opaque to the average citizen. That alone undermines the very democracy it tries to support. Paper based voting systems may be bloody tedious but they’re simple, and anyone can participate.

As a programmer myself, I have to say no to electronic voting systems. We need to keep it low tech.

Also, they may be opaque even to software engineers, it's not like we have magic powers to look at a physical computer and know what code is running.

Even if the code is released:

(1) Who can read and understand tens or hundreds of thousands of lines of probably not very comprehensible code? Not many people can and even fewer will.

(2) Is the code running on the machines the same as the published code?

I don't even know what code is running on my phone or laptop right now.

Even if you work in the space as a software developer, the shear amount of variance in government voting systems from registration, petition verification/validation, and voting itself is really wide. Some states have rules that vary from county to county. Some of these states have lots of counties. Election systems are some of the most byzantine you can imagine.

Disclosure: I used to work for an Election Services company.

The process needs to be understandable to non-programmers. Right now, even those who program it don't know every instruction/interaction/oddity from the processor running the code.

I can explain a voting process with pen and paper to a child.

If a voter is too inept to properly operate cryptography, can it be said that their vote is legitimate? Seems to me that not everyone should vote.

Yeah, I’ve heard variations of that argument spoken non-ironically. Opens a whole can of worm. I’d rather assume, barring special circumstances, that anyone old enough is able to do politics (voting, being elected, being randomly drafted in a commission or jury…). With the right mechanisms groups are smarter than we give them credit for.

This "can of worms" is why democracy will fail and a Caesar will eventually ride into town and promise all kinds of things to the median voter. If it is considered immoral to hold voters to any standards, why would it be immoral for politicians to lie, cheat and steal?

When you vote from home I can show up and put a gun to your kid's head and tell you to vote for Pedro.

When you vote with some sort of association with your name I can look up how you voted after the fact and if you didn't vote for Pedro I can blow up your house.

The fact is that the secret, anonymous ballot is the only valid way of voting.

> When you vote from home I can show up and put a gun to your kid's head and tell you to vote for Pedro

If you could do that at a scale large enough to affect the result, I would think you could also do that at each ballot box and make the vote counters report the results you want them to report.

Also, if you could do that at scale, why would you bother with elections at all?

IMO, the main problem with digital voting is that it makes it possible to change the vote count without needing lots of people to do so.

With thousands of ballot boxes, open counting of votes for each ballot box and the publication of election results per ballot box that’s basically impossible.

In the 2020 election one of the seats for the House of Representatives had an initial margin of victory of six votes. After recounting it was something like twenty. And another had a final margin of victory around one hundred.

I doubt anybody could tell it was going to get that close. The uncertainty likely ranked into the thousands, if not tens of thousands of votes.

I think few fraudsters will be happy with slightly increasing their odds by swinging the vote by a few hundred votes. Instead, they’ll want almost certainty, if not complete certainty.

Same reason I advocate for ranked choice voting instead of other even better methods of choosing a winning candidate. Some of them require matrix multiplication.

(I still support electronic voting though).

You mean election type voting?

Screen scraping is the least of the problems.

Voting machines are simply not trustworthy given these requirements for a free and fair election:

0. Freedom to vote or not

1. Anonymity in that no vote may be associated with the voter who cast it

(These two are most of why e-voting is not trustworthy and are what differentiates banking from voting.)

2. Transparency. The law, means, methods must all be known to the voting public. It is their election. In particular, seeing how a cast vote moves through the entire process and into the final tally is necessary.

3. Oversight. The first two of these foundation requirements necessary for free and fair public elections have an important ratification:

The record of votes cast must be human readable and said record must be used directly to realize the final tally of all votes cast.

There is a chain of trust between voter intent and the vote cast. When one casts a vote on physical media, that is a direct record of voter intent. The voter can be sure their intent adds to the tally because they can directly verify their votes cast.

Electronic votes are a vote by proxy. The direct expression of intent is not kept and exists as a bit of skin residue left on the machine interface. Put simply, we ask people to vote and then we discard their expression!

What gets used instead is whatever the machine thought the voter intent was!

Consider this: You have a machine in front of you with two buttons. One brings the world into harmony, the other casts it into war, grief and darkness.

You can:

A. Submit a paper form you fill out manually indicating your choice

, or

B. Press one of the buttons and get a receipt for your records.

Say this machine has an indicator showing your choice and a third "approval" button you use when you feel good about the machine getting your choice right.

How do you the voter know the receipt and or indicators match what the proxy machine says your choice was.

What happens when we have a dispute? We know the machine can't be trusted to match receipts. With the direct expression of intent lost, what do we bring to court?

We can and have brought paper ballots into court to resolve and election facing genuine ambiguity.

I submit to you we can't really be sure who wins an electronic election, unless we also associate your identity with the vote.

Say a bunch of people get to choose the fate of the world this way. What is your confidence level?

How do we know the machines tallied up the winning choices correctly?

What do the tests for that look like and how do they get around the forced trust problem we have with all electronic inputs today?

There is a lot more trouble to discuss. What I put here is the core trouble and computers have always had this problem too.

It just does not present any real difficulty, until elections are brought into the discussion.

Display trust issues are well above the core tech trust issues we face today.

I was guessing that the oddly beautiful 17” Apple CRT Studio Display [1] from 2000 might have been a CRT that used a digital signal cable because it had the short-lived Apple Display Connector, but apparently ADC carried analog too.

[1] https://everymac.com/monitors/apple/studio_cinema/specs/appl...

The ADC carried power, too, which is why Apple went with it.

One (extreeeemely pedantic) difference is that (I think) HDMI begins and ends its vsync pulses when the preceding hblank begins, whereas VGA begins and ends them slightly later when the preceding hsync pulse begins (I documented this at https://nyanpasu64.gitlab.io/blog/crt-modeline-cvt-interlaci...).

I also think that DP rouuuughly reflects CRT timings complete with active vs. blanking intervals, but doesn't actually have a fixed pixel clock (perhaps it does? I didn't quite figure out synchronous/asynchronous pixel transmission when reading the spec) and doesn't transmit hsync pulses once per scanline.

> You could have made a CRT monitor that used DVI-D just by adding a DAC, though I'm not sure if anyone ever did.

From memory IBM offered a crt with DVI.

Plenty did but they were usually DVI-A or DVI-I

>DisplayPort does away with all that legacy.

You wish. DP sends exactly same bytes DVI does (blanking and all), just broken up into packets.

The important difference is that (like VGA) DVI and HDMI still dedicate particular physical wires to separate red, green, and blue channels. Display port does not shard individual pixels across multiple channels: all bits (all colours) of a pixel are sent on the same diffpair. If a DP link has multiple lanes then the bits of different successive pixels are sent on each lane.

This is great, but then they drop the ball by forcing garbage (“interspersed stuffing symbols”) between the packets instead of letting you use that BW.

I think that you can actually transmit audio, or other displays' signals (DisplayPort Multi-Stream Transport), in place of stuffing symbols.

Afaik its a no-no zone. You can only use blanking periods.

You may be correct (and I was mistaken in my previous post) for audio. The leaked (DP isn't really an open standard) DP-1.2.pdf on glenwing's site says that "The dummy stuffing data symbols during the video blanking periods (both vertical and horizontal) may be substituted either with main stream attributes data or a secondary-data packet", and the public 1.1a PDF indicates similarly.

MST is quite complicated and appears to interleave bytes from different video streams within 64-byte packets (rather than solely during blanking):

> The MTP (Multi-stream Transport Packet) is 64 link-symbol (1 byte of video data or special symbols) cycles (that is, 64 time slots) long, starting with MTP Header in the first time slot (or Time Slot 0), and is constantly transported regardless of the presence/absence of streams.

> The Payload Bandwidth Manager of each uPacket TX in the path from a DP Source to a target Sink device allocates time slots within the MTP to a VC (Virtual Channel) Payload to establish the virtual channel for transporting a stream.

You are not giving the full picture in this trade. The people making the spec knew the trade well. You add a lot of complexity to a receiver by forcing it to generate its own timing information from whatever video signal is sent. That work has to be done somewhere and it is most robust when done at the source.

That optional infrastructure on the receiving end is already provisioned in the standard in form of eDP PSR (self refresh).

Panel Self-Refesh is an optional feature, is it not?

Why isnt there an optional matching DP mode where you could just transfer screen data without timing? Why are you forced to sent fake blanking to PSR displays?

Reply on news.ycombinator.com