30 minSeason 2, episode 23

The Intersection of Hardware and Software Security

With Chris Romeo, Izar Tarandach, and Matt Coles

Infrastructure, identity & resilience

In this episode of The Security Table, Chris, Izar, and Matt discuss an article that discusses threat modeling in the context of hardware. They explore the intersection of hardware and software security, the importance of understanding attack surfaces, and the challenges posed by vulnerabilities in hardware components, such as speculative execution faults and the impact of supply chain security.

Episode title image: The Intersection of Hardware and Software Security

Listen to the conversation

About this episode

In this episode of The Security Table, Chris, Izar, and Matt discuss an article that discusses threat modeling in the context of hardware. They explore the intersection of hardware and software security, the importance of understanding attack surfaces, and the challenges posed by vulnerabilities in hardware components, such as speculative execution faults and the impact of supply chain security. Join the conversation as they examine the critical points in the ongoing dialogue around hardware and software security integration.



Follow AI Security Table:

➜ Home: https://securitytable.ai/

➜ X: https://x.com/AISecurityTable

➜ LinkedIn: https://www.linkedin.com/company/ai-security-table/

➜ YouTube: https://www.youtube.com/@AISecurityTable

Transcript · 30 min conversation

0:09Chris Romeo Hey folks, welcome to another episode of The Security Table. My name is Chris Romeo, joined as always by Ezar Tarendosh and Matt Coles. And as we were just chuckling about, it is security week in the desert right now. And as you can see, All three of us are safely stowed away in our bunkers so that there's— what's the concept of the— in the US government, there's always one person left behind in the bunker.

0:41Matt Coles Designated Survivor.

0:42Chris Romeo Designated Survivor. I can never remember.

0:43Izar Tarandach Actually a very good series.

0:45Matt Coles Okay.

0:46Chris Romeo That's— but we are the designated survivors of cybersecurity who are staying out of the desert. Case of some other bigger problems. So yes, it's Black Hat besides Las Vegas and DEF CON week, but we're here.

1:03Matt Coles Besides Las Vegas, where else? Yeah.

1:05Chris Romeo Well, that could be a whole episode, whole different episode on leaving Las Vegas. Um, okay. So, Izar, you got, you found an article to tee us up here. So why don't you, uh, why don't you set the stage, kind of talk about, give Give us the background on why this caught your attention.

1:22Izar Tarandach Okay, so after learning a lot from Matt, from the embedded and hardware side of things, somehow my Google feeds, I don't even know how to call the thing on mobile.

1:36Matt Coles Discover.

1:38Izar Tarandach That, yeah, thank you. It picked up on the fact that I have a passing interest in track modeling and apparently some hardware as well. and it brought me this, this interesting, uh, uh, article from a site called semiengineering.com. Notice that I am being very, very careful with my pronunciation because it's been pointed out to me that my accent may be wrong in some of our past episodes.

2:05Matt Coles Well, wait, may be wrong. Let's, let's, we could unpack that.

2:09Chris Romeo I always thought that's part of your charm.

2:12Matt Coles Right?

2:12Izar Tarandach It's what I'm going for. But no, I have to have somebody with even worse accent than me.

2:18Matt Coles You're the South American and European accents we need. Got it.

2:25Izar Tarandach See? There you go. Somebody valued me finally. Anyways, so there is this semiengineering.com site and they posted an article called Defining Cheap threat models to identify security risk.

2:44Matt Coles And immediately, chip, chip, not cheap.

2:48Izar Tarandach Chip, chip. Yeah, C-H-I-P, like hardware black thingies with a lot of pointy legs on the bottom. Not, not expensive things.

3:00Matt Coles You never met a CPU, did you?

3:04Izar Tarandach We can go there later, but But anyways, so it starts, it begins with experts at the table, right? So it's a group of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 experts that begin by answering the question, how do you define a threat model?

3:34Chris Romeo And that's where the wheels start to slowly fall off.

3:41Izar Tarandach Trust me. The thingy, you know, I have no problem with people having their own definitions. After all, even in the context of the Threat Modeling Manifesto, we agreed on one definition that we said this is not, the definition. This is a definition that we all can work with. But if you look at the many presentations that we have out there and writings, you'll see that each one of us has their own definition, working definition we use every day to do what we do. But, oh my God, I don't, you know, I, I don't want to throw shade on people.

4:24Chris Romeo Well, we're not going to mention any names as to what comments we pulled or whatever, but let me throw this point before we go there. It just hit me. For some reason, the Threat Modeling Manifesto has not drifted into the world of hardware at all. Because nobody used our definition though. At least this, I'm looking at this as a representative sample of academia, government and, um, commercial companies, right?

4:55Matt Coles Yeah.

4:56Chris Romeo And I would've thought one of these definitions would've been like analyzing representations of a system to understand the security and privacy characteristics. Like, I feel like the manifesto, at least in the circles we run in, has changed, has driven people towards more of a common definition. But what I'm not seeing is I'm not seeing that here. In any of the answers. This—

5:21Matt Coles there's a couple that are close, but yeah, you're right. I mean, in looking at the, at the list of folks, um, some of them are at least aware. I mean, we know that they've been privy to the Threat Line Manifesto. I mean, they're part of the same circles, you know, that, that we run in. So, um, well, I guess don't read in too much about it. I mean, they could have paraphrased it. And, and not use the definition. Um, alternatively, they could think there's a better way of describing it, and, and for some of these, that may be the approach they want to take. So let's not, let's not read it too much of it. However, uh, if any of them are listening, they're going to get a copy of the Sharp Modeling Manifesto in their inbox.

6:07Izar Tarandach But Matt, let me ask you something. As, as my, uh, Embedded and Hardware Guru.

6:14Matt Coles You're so, so doomed.

6:15Izar Tarandach Hey, it's good enough for me. So this point that Chris made, which I agree with completely, would it be fair to say that it's not only the manifesto, that there is a disconnect in terms of security between the hardware community and the software community? And just to point out that something, once upon a time in the previous lives, I worked at Intel. In a fab and all that good stuff.

6:44Matt Coles And did not go to TIL. Yeah.

6:47Izar Tarandach And I got to look at the over-the-shoulder of people who actually wrote code that described chips and sent it to the wafer machine to actually make the chips. And it's a very powerful language. It's like, it's a full language, of course.

7:02Matt Coles VHDL you're talking about?

7:04Izar Tarandach I believe so. Don't remember exactly the details, but I could imagine at this point, not then, but at this point in my life, I could imagine some constructs in that language having a security value and you being able to introduce security faults in the hardware itself by means of the language that's used to define the chip and to make a long loop shorter. I'm willing to bet that there are no static analysis tools for those languages that cover security cases, for example, because embedded people still see that stuff as not really where we play.

7:46Chris Romeo Is that true?

7:47Matt Coles It's not entirely true. So there's been work done by various folks. I don't, I'd have to pull the research. I know MathWorks has been doing some work in this area. So I'll have, I'll say, so I used to work at, I worked briefly at Analog Devices, which of course is a semiconductor manufacturer, designer. So the, there is some work being done in that area. It's interesting too because threat modeling started in a cyber-physical world, you know, with the NIST risk risk assessment framework. Ooh, get that out today. Um, where it went through threat modeling of harms, physical harms mostly, and harms to things like national security from physical systems, right? Mostly weapon systems and critical infrastructure and whatnot. Um, so to say that hardware is, is not, uh, up to speed on security is sort of misleading. I will say though, there is a, there's two parts. Software definitely ran really far, really fast when it came to security. But then again, it's sort of like the, oh, it's like sort of like the OS debate. If you, it, or, or when it comes to cybersecurity, at least if you use it, they will come. And so software, so software was laden with holes and needed to, you needed to rapidly advance. And hardware, I, you know, definitely I think got sort of went head in the sand approach, you know, where, you know, hardware, there wasn't a lot of security research into it, at least, you know, not visibly. And it didn't hit, you know, billion-dollar industries as quickly. And then, and now it's starting to, you know, starting to really take on more focus with, you know, a lot of our software being hardware dependent. It always was, right? I mean, running in the cloud, you're running off hardware. But now with specialized chips and GPUs and, and, and other things, you know, running on specialized silicon, uh, you know, they— you have, you have more emphasis now on hardware. You also have side channels, things like that, right? Which have always existed. My— one of my favorite things about hardware security was a project called Tempest, uh, out of the DoD, where, you know, if you, if you had, um, EMF Uh, emanations from your hardware devices. People could, you know, monitor network communications and see what you're reading on screen and, you know, big, you know, big spy stuff, right? Um, and, and so hardware has always been part of security, but it's like I said, software, I think just ran very fast, very quickly because they, they made a mountain out of security vulnerabilities.

10:39Chris Romeo I mean, think about, think about the, avenue, the vector of attack. Like when we think about what are the biggest hardware issues that we've seen in the last 20 years, they've been Intel chip derived due to, there was the Ghost vulnerability. Um, what was the last one?

11:01Matt Coles You had the, uh, you had the Pentium divide by zero.

11:04Chris Romeo Yep. Okay.

11:05Izar Tarandach Yeah. Oh yeah. I remember that one.

11:07Chris Romeo Forgot about that one. Okay. But then there was another—

11:09Matt Coles so that wasn't technically a security problem, right? That was, that was a chip flaw that, that resulted in denial, you know, mass denial of service for, for devices.

11:16Chris Romeo What was the one, what was the big one before Ghost? Ghost wasn't as big as the—

11:22Matt Coles we had Rowhammer. You had Rowhammer. Oh, you're thinking about Spectre and Meltdown.

11:26Chris Romeo Spectre and Meltdown.

11:27Matt Coles Spectre and Meltdown. Just for those who aren't familiar, Spectre and Meltdown is about, um, uh, uh, oh my God. Just, left my head.

11:34Izar Tarandach Isn't it the parallel lines of execution?

11:37Matt Coles Yeah, it's speculative execution faults, right? And the ability for somebody to both influence and guess the outcomes. And then you have RowHammer, which is as chips, as when chips, when hardware devices are manufactured, they're manufactured in layers. There's power and processing, you know, circuitry that are put in the layers. Those layers are getting thinner and thinner and thinner. You may hear them as your, your nano, nanometer technologies. You know, I think we're at 3 nanometers now. Uh, and as they get closer and closer together, it's possible for one layer to influence the other layer. And so in, uh, in RowHammer, you have, uh, circuitry that is able to flip bits in, in me either in memory or in, in, um, well, either mem like memory or in storage, depending on what you're using for storage. And then do things like, uh, gain kernel access, hit the receipt, or, or become administrator because you can flip that little bit that says, hey, I'm an admin. Uh, same as thin. And, and so, yeah, now you can do that through a webpage.

12:42Chris Romeo So is that a fair assessment then to say, do you agree with my statement that the primary hardware attack vectors that we've seen resulting in attacks in the last 20 years have been around the Intel chips?

12:57Matt Coles No, not strictly the Intel chips. So Rowhammer is not an Intel chip problem. It may be an x86. So let's be careful of architecture versus manufacturer, right? So x86 is a processor technology. Intel makes x86 chips, but so does AMD and others, right? So the processor technology may be at fault, but the chips themselves either can have additional flaws or are impacted like a protocol flaw, right? Heartbleed was a TLS protocol flaw, not OpenSSL, not an OpenSSL flaw.

13:31Chris Romeo Okay. So then if we, if we use something like Spectre and Meltdown, what cat, what box does that go in? Is that an x86 problem or a specific processor?

13:43Matt Coles I think those were, if anything, anything that has speculative execution. So I think that was x86 and that, you you know, x64. Um, uh, I don't recall. I have to go back and check. I don't recall if ARM had, uh, was impacted by it or if they were impacted the same way, but it's certainly x86. And, and again, 32 and 64-bit.

14:03Chris Romeo So are there other hardware attacks, other hardware vulnerabilities— let's say vulnerabilities instead— there are in other things that just don't get any, like, they just don't get airtime? Is that like— because I feel like I don't know of anything outside of—

14:16Matt Coles they do, they do. So there's There's a whole class of issues around misconfigurations. So a couple of the common ones is, so remember with, so just back up for just a sec. So remember with hardware especially, things get designed, like hardware is the ultimate waterfall model. Things get designed, they get prototyped, they get built as a test, you know, test system, and then you mass produce them. And once you start mass producing them, whatever you baked in is baked in until you replace them. So if you have, if you misconfigure something like doing things like not baking in a root key to do Secure Boot, right? So that hardware won't be able to support Secure Boot at all, unless it's reprogrammable. But, and, Okay, I do. Sorry.

15:13Izar Tarandach No, no, no, no. You finished the thought. I'm just getting— okay. Yeah. Sorry.

15:17Matt Coles So, so, so then you also have something called, you have some common interfaces that are used to communicate to the processors, to the chips, things like SPI and, and JTAG, which is a debug interface, uh, programming interface. And often those are left open or they're left accessible, right? And therefore people can do things like, um, you know, flash new firmware.

15:39Chris Romeo Uh, you can, you know, there's other vulnerabilities, but those require physical access though, right?

15:44Matt Coles Like it generally requires physical access, but if you can trick somebody into flashing firmware, you can, you can potentially gain access. It depends on how it's, how it's connected. Usually that's physical. Uh, more recently there was the issue that was reported around, um, uh, somebody, a company's using test boot keys. So they would use, they would use secure boot keys that were not intended for public use.

16:12Izar Tarandach Yeah, we spoke about it last week.

16:14Matt Coles Yeah. And then, and then lastly, you have, where was I gonna go? Well, yeah, so there's other vulnerabilities. Many of them are not, many of them are public. They often happen in consumer devices where no, uh, general consumers aren't gonna, aren't gonna know, and maybe not even care that their JTAG is exposed because that JTAG is on a chip that's embedded into a, say a, you know, a, a, you know, smart speaker or router or something like that. Uh, I'm gonna look at that device as a, as a regular consumer. I'm gonna look at the device and go, yeah, nobody's getting into that. Well, there's a whole supply chain discussion we could talk about because there are—

16:59Izar Tarandach exactly, we spoke about that back on when we talked about SBOM, that nobody's going to look at it and say, oh, this has XML too, I'm not going to use it. But one thing that you said in there that, uh, and actually the, the article mentions it as well, is the, the, the curve of cost in hardware. So yeah, true. We, we got to a point where hardware is getting cheaper by the day, at least at some, some scale, right? You can go into Amazon, get a thing like this for like bucks, uh, and do whatever you want with it. But even so, a lot of the hardware that comes out does have the capability of flashing new firmware. And I think that my point is that even though there might be hardware security failures. There is a, um, there are two things that I see happening. First, more and more of the functionality is actually regulated by the firmware that it's using. So in other words, if something gets bad, you could potentially reflash the thing and close that that issue. Sometimes even at the point of, uh, what's it called, uh, micro— like the, the code in the chip itself.

18:22Matt Coles It's microcontroller. Yeah, microcode.

18:25Izar Tarandach Microcode. But, uh, the, the other point is that hardware is getting so cheap that if you're a consumer and you don't want to deal with the security issue that has been found, It's cheap enough for you to unplug the thing and plug the new version.

18:42Chris Romeo Why don't we just replace the device?

18:44Matt Coles Or just replace the device.

18:45Chris Romeo I mean, yes, but why don't we update? I don't ever update firmware. Am I running ancient firmware?

18:51Matt Coles Oh, we'll be coming to your house next.

18:53Izar Tarandach Missing. No, no, no, no, you're so missing. Like, it's, it's the best thing to get like your heart pumping if you don't want to do aerobics. Flashing firmware is Amazing.

19:04Chris Romeo I remember needing to flash firmware in the old days, but is it, maybe it's just happening behind the scenes. I don't, I don't even pay attention anymore.

19:15Matt Coles You don't use an Android phone, do you?

19:17Chris Romeo Oh no, no, I wouldn't.

19:18Izar Tarandach Oh, I haven't rooted one of those in years. Oh my God. But flashing firmware, for example, on your router, right? Or on a Workday. Until you see that red lights turn green. It's like the biggest suspense in your life. You, you, you have never experienced the joy of seeing red become green until you, you flashed the only router that you have working.

19:42Matt Coles Yeah. And until you, until you don't and you see the red stay red.

19:46Chris Romeo Yeah. Yeah.

19:46Izar Tarandach Yeah.

19:46Chris Romeo And then fuck, you're troubleshooting on Sunday night knowing everybody's gonna come in the office on Monday morning and nothing's gonna work.

19:54Izar Tarandach No, 'cause you're on a, you're on a drive to, you're on a drive to Best Buy to buy the Or the best thing is when you're like, when you're following a how-to and it says now flash the firmware and then two lines under and says, oh, by the way, you should have downloaded this other tool that you need after you've— and now you can't because you just flash the firmware without the tool.

20:14Matt Coles So yeah, that's why you read instructions first all the way to the end and then you start again.

20:21Izar Tarandach You were asking too much already from me.

20:23Matt Coles To read the first time, but as I go along, please, the directions, the second time, instructions, RTFM, at least once, at least once.

20:34Chris Romeo Instructions slash directions are the manufacturer's suggested way of doing what the operation is. It's not binding. You didn't sign a contract.

20:43Matt Coles No, it's not. It says I'm going to follow the directions. You're free to play, you're free to play above the country.

20:46Chris Romeo Yeah. Let's come back to the article here real quick because I think one of the answers about how to define threat modeling has a brilliant nugget within it that I found very interesting. And that was, and I quote, we ended up defining a model with 4 attack surfaces ranging from side channel to supply chain to reverse engineering to malicious hardware.

21:11Matt Coles Yes, this one caught my eye.

21:13Chris Romeo That's a heck of a categorization of hardware issues that you would want to be exploring in a threat modeling context for a hardware device, from side channel to, you know, supply chain issues with components coming in to how easy is it for people to reverse engineer this thing I've built to somebody trying to inject malicious hardware, which is where we kind of touched on that a little bit.

21:43Matt Coles But by the way, I would extend this. A lot of that applies to software too. Yep. Supply chain attacks, of course.

21:52Chris Romeo So when we think about side channel, is there a side channel attack in software?

21:57Matt Coles There are, there are covert channels in software. There are side channels in effect. Um, you can have, uh, you can have covert channels in software. So, uh, and we know, and we know this because covert channels and side channels do exist in, in the kind of the taxonomy of CWEs for software.

22:15Chris Romeo Sure.

22:16Matt Coles And so, yeah, that was actually really important, I think, because a lot of time when we talk to people about attack surface, they think about the inputs to a system, right? They look at what's coming in, right? User input in particular. Often what gets glossed over even in the software world is process-to-process or machine-to-process communication. So if you take the hardware, uh, hardware as a con— has hardware concepts around supply chain and attack surface and whatnot, and, and apply that to software or to mixed hardware and software systems, even better, uh, your, your attack surface grows. And, and, and if you look at attack surface as what you care about, then, then the, then the security posture of your system And the scope of that, uh, of that exercise grows. That isn't insurmountable. It just is a little more complex.

23:14Izar Tarandach So I have a question on the, the, the supply chain, uh, uh, yeah, vector. So we had that whole thing a couple of years ago with, uh, I don't know how to say the name, the, the Korean router making thing. Huawei. Huawei. That one.

23:34Matt Coles Huawei. Huawei.

23:38Chris Romeo Huawei is a Chinese. Huawei is a Chinese company.

23:41Izar Tarandach Yeah. H-U-A-W-E-Y.

23:44Matt Coles Yeah. They make 5G and other telecommunications equipment.

23:48Izar Tarandach Oh, they make everything. But the point is that when that happened, I could see the problem because it was like this big one new thing with a lot of hardware inside. Inside a lot of software and fine. It came into my head and sat there nicely. But then that moved to cell phones and the actual chipset in there. And then we had, for example, I had this one thing, I needed an Arduino board for something and I bought something off of Amazon. And Turns out that it was fake. And the way that I found it was when I was trying to program it, the memory space was completely different. Nothing was happening. So, but it was stamped Arduino and it looked nice and everything, but it just wasn't. So my question here is, with hardware security issues, How much leeway is there for a bad thing to happen inside a given chip or chipset? In other words, if I have a problem in the modem in my cell phone, how bad is that actually? What can happen?

25:14Chris Romeo Really?

25:14Izar Tarandach No, seriously, because my thing is—

25:17Matt Coles Let's talk about some of the impacts.

25:19Izar Tarandach No, no, no, no, no. Wait, wait, wait. The impacts are clear, but I'm looking at how, how, how bad it is in terms of the probability of something happening because in the chipsets that, that the fault is—

25:31Chris Romeo You mean how widely?

25:31Izar Tarandach No, no, no, no. Wait, wait, wait. The fault is in the chipset. So immediately we jump and say, oh, it's in the modem. Of course it can copy all the traffic and send it to wherever.

25:41Matt Coles Yeah.

25:41Izar Tarandach Sending it to wherever.

25:43Chris Romeo This is the same debate we're having in software right now about reachability.

25:47Izar Tarandach No, no, but it's different. Wait, wait, wait, wait, wait, wait. Having that small piece of code inside a chip where everything is limited, where basically you can't just put an IP in there pointing to some S3 bucket and say, write everything over there because it's not how the modem itself works. It's on a lower level in the stack.

26:06Matt Coles Yeah.

26:06Izar Tarandach So, yeah.

26:07Matt Coles So how risky is that? How risky is that vulnerability compared to the thing, the thing that it's in? And the, then this is a, this is the risk management discussion. This is the attack chain, you know, kill chain and all this, all the stuff we love to talk about. This is exactly what you're asking, right? That chip has a flaw. How bad is it really? So it depends on the use of that chip. So if that chip is being used as your, as your radio, sure, it can intercept data. But if it doesn't have access to a processor that can then manage an IP stack, et cetera, then so it's just capturing data. So what? Right. And, and that's, that's problematic, but not necessarily catastrophic. If, however, that chip is your radio and it's used for authentication to the, to the cellular network, and you can use that and you can potentially, you know, if it has a flaw where you can potentially clone identities. You now have the ability to flood a, flood a cellular network, put compromised or rogue devices on the network and gain significant level of access that you wouldn't have before. And on top of that, there is the whole counterfeiting thing you're talking about, right? So if somebody has the ability to clone an identity and put it on a, on an arbitrary network, there's a lot of money to be made in that. So individual vulnerabilities may or may not, depending on what level of capability those devices have or that they are within an environment that gives them capability. But that's true if it's software or hardware. This is not a hardware-specific problem. This is a hardware— it's important to differentiate though a vulnerability that affects a single device versus a vulnerability that affects like a protocol or chipset because that's how many millions of devices does it impact because hardware doesn't affect, you Usually doesn't affect just one.

28:05Izar Tarandach Shit.

28:06Chris Romeo You know, just build one. Come on.

28:07Izar Tarandach But, but my question, my question stays even at that level of, of scale. If you, if you have millions or even billions of cell phones out there and you have a vulnerability at the modem level, in order to exploit that, that thing, it's almost a nation state thing, right? Unless you're going to the middle of the desert and getting a femtocell and whatever.

28:27Matt Coles Well, it, it depends. I mean, it depends on, is it, it doesn't require an active attack. Or is it a firmware problem? Like, in other words, is it, is it a chipset problem that is exacerbated by firmware that gets loaded on every device? Right? It, there's a number of ways that a, that an attack could occur. Those that, you know, require some interaction with the outside in order to trigger an exploit, then yeah, you're probably, you're probably talking either a well-placed actor, right? Insider collusion. Or a nation state or organized crime or, or a ransomware group or malware group that, you know, is building targeted device, targeted software destined for these devices, you know, that, that hacked version of Candy Crush that everyone loves to put on their phone and that, that gains access, right? So this is, yeah, I mean, it depends on, on the nature of the, of the issue. As to how easy it is to exploit, and then what does it take as in, in terms of number of resources, and then whether you do it at scale or you're targeting a singular device, right? Because if you're going to go after, if we're going to go after, you know, the president or vice president's phone because they're the high-value target versus I want to go after as many consumers as I can get in my net to create a botnet of of cellular phones with IP addresses. That's different, different scale and different, uh, well, we're out of time for today.

29:59Chris Romeo So, uh, we'll have to continue this conversation in the hardware space going forward. I don't think this is the last time we'll talk about hardware. Now we've kind of, it seems like we've kind of opened up, uh, uh, something that could be, we could dive a lot deeper into. So let's plan to do that for the future. So folks, thanks for listening to another episode of Security Table.

Transcript supplied by AssemblyAI. It may contain transcription errors.

Originally published as The Security Table. Part of the AI Security Table archive.