3 ms·

Boy oh boy do I have bad news for you about every other computer you're using today. The ccNUMA model they teach in school is well and truly dead even on deskto

Boy oh boy do I have bad news for you about every other computer you're using today. The ccNUMA model they teach in school is well and truly dead even on desktops. There is no modern computer that just has the CPU in charge of memory. I don't think there are any general purposes devices you or I can purchase that have actual memory segmentation (either true separation or via SMMUs/IOMMUs. On a modern desktop, you have dozens of random cores sitting on the memory bus that have full access to physical memory, and there's nothing you can do about it because those cores run their own proprietary (often baked in) software. A Linux-first phone does absolutely nothing to fix this. Even if you're running your own operating system that you wrote yourself, your GPU or hell even the power management controller might reach around and steal all of your stuff. The issue lies in the way we build computers and the rapid proliferation of cores without any serious thought as to how we're supposed to manage and secure the enormous fleet of heterogeneous hardware that lives on a modern SoC.

mindslight 5y ago
First, my main desktop is a librebooted KGPE-D16. Second, it matters if those possibly-hostile cores have network access. Technically they could subvert the software on the main CPUs to communicate with command/control, but that seems like raising the bar to such an attack. Whereas with the standard mobile architecture, I can totally see some phone manufacturer getting the "bright idea" to have the baseband processor collect statistics on the application processor's software for market research. But I agree with your general point. Still, I think moving in the direction of Linux phones gives us a starting point to do something about this insecurity - proving the market allows there to be devices that truly separate out the cell modem.
- Sirened 5y ago
  They wouldn't need to even talk to the AP to get out on to the network though :) If you have physical write access, you can talk to the network just by writing data in the network device's mailbox using the exactly same MMIO techniques that the AP uses. Its one of the reasons I am so disenchanted with current rootkit detection tech—if malware can get physical memory write, they can corrupt other cores memory persist into other SoC elements, well and truly out of view from the AP while still maintaining the exact same amount of control. I would not be surprised in the _least_ if the US three letter agencies have at least one implant that does this because its not exactly difficult but is so insanely hard to fix or detect that it makes it ideal for covert monitoring. That being said, I think there's a fundamental disconnect between how SoCs work and how people actually think they work. Linux lives in just one tiny little corner of the machine and the hardware does so much to coddle Linux into thinking it's in control. The AP is just a core on the memory bus, it typically has no greater privileges (it often has _less_ because hardware vendors hide ""implementation details"" away from OS developers!) than any other core. I don't know that this is so much a radical take in the OS community though it might be one here: Linux _is_ part of the problem. Linux makes several assumptions about how hardware works. Chiefly, it assumes that it is in the driver's seat and it makes cooperating with other cores running their own copies of Linux (let alone other OSs) damn near impossible. I think the next step in computer evolution is going to come from software engineers rediscovering hardware and building code that not only runs on the AP but across all cores—from the power management controller to the AP--in an SoC. This may come from radically changing Linux or it may very well come from a new OS that treats modern hardware platforms as they are: platforms.
  - pdoege 5y ago
    The baseband/PMIC is such a huge threat and it is entirely in the hands of the carriers that desperately want user data. Coupled with very tightly held IP the problem seems to be insurmountable. When I see "removing the Google ecosystem" I have a hard time taking the rest of statement seriously.

mindslight 5y ago

First, my main desktop is a librebooted KGPE-D16. Second, it matters if those possibly-hostile cores have network access. Technically they could subvert the software on the main CPUs to communicate with command/control, but that seems like raising the bar to such an attack. Whereas with the standard mobile architecture, I can totally see some phone manufacturer getting the "bright idea" to have the baseband processor collect statistics on the application processor's software for market research. But I agree with your general point. Still, I think moving in the direction of Linux phones gives us a starting point to do something about this insecurity - proving the market allows there to be devices that truly separate out the cell modem.

Sirened 5y ago
They wouldn't need to even talk to the AP to get out on to the network though :) If you have physical write access, you can talk to the network just by writing data in the network device's mailbox using the exactly same MMIO techniques that the AP uses. Its one of the reasons I am so disenchanted with current rootkit detection tech—if malware can get physical memory write, they can corrupt other cores memory persist into other SoC elements, well and truly out of view from the AP while still maintaining the exact same amount of control. I would not be surprised in the _least_ if the US three letter agencies have at least one implant that does this because its not exactly difficult but is so insanely hard to fix or detect that it makes it ideal for covert monitoring. That being said, I think there's a fundamental disconnect between how SoCs work and how people actually think they work. Linux lives in just one tiny little corner of the machine and the hardware does so much to coddle Linux into thinking it's in control. The AP is just a core on the memory bus, it typically has no greater privileges (it often has _less_ because hardware vendors hide ""implementation details"" away from OS developers!) than any other core. I don't know that this is so much a radical take in the OS community though it might be one here: Linux _is_ part of the problem. Linux makes several assumptions about how hardware works. Chiefly, it assumes that it is in the driver's seat and it makes cooperating with other cores running their own copies of Linux (let alone other OSs) damn near impossible. I think the next step in computer evolution is going to come from software engineers rediscovering hardware and building code that not only runs on the AP but across all cores—from the power management controller to the AP--in an SoC. This may come from radically changing Linux or it may very well come from a new OS that treats modern hardware platforms as they are: platforms.
- pdoege 5y ago
  The baseband/PMIC is such a huge threat and it is entirely in the hands of the carriers that desperately want user data. Coupled with very tightly held IP the problem seems to be insurmountable. When I see "removing the Google ecosystem" I have a hard time taking the rest of statement seriously.