4 ms·

No. Do not want. Rowhammer is a hardware problem ---- defective RAM --- not a software one. The sooner everyone starts returning defective RAM and putting pr

No. Do not want. Rowhammer is a hardware problem ---- defective RAM --- not a software one. The sooner everyone starts returning defective RAM and putting pressure on the hardware manufacturers to maintain correctness, the sooner we can stop this descent into insanity. "They can always work around it in software" is the attitude that let Rowhammer exist, and continuing to fulfill that expectation will only make things worse.

mikewarot 3y ago
Came here to say the same thing... take my up-vote. ;-)
bogantech 3y ago
> Rowhammer is a hardware problem ---- defective RAM --- not a software one. It always amazes me how people can be so confident yet so wrong. It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM and that is going to be extremely expensive when talking 16GB and not nearly dense enough. It's not some conspiracy by "Big RAM"
- oasisaimlessly 3y ago
  > there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM Why wouldn't storing a cryptographically-secure checksum on every RAM row work?
- theamk 3y ago
  There are plenty of problems of physics in the RAM design, and it's hardware designers jobs to find the operating regime where those problems do not matter. I mean, if you had a DRAM labeled "DDR4-3200" but it can only work at a much lower speed (say DDR4-2400), it's clearly a problem of physics - the gate capacitance is too high, the driver transistors are not strong enough. And yet my reaction would be to take that RAM back to store and get my money back, not to defend manufacturers which claim false things about their chips.
- rstuart4133 3y ago
  > It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM and that is going to be extremely expensive when talking 16GB and not nearly dense enough. And yet, when the bit flip problem caused by physics got so bad in DDR5 it couldn't be ignored they did fix it - by adding error correction codes. It wasn't that expensive to do so. Notice that HDD's hit the same problem as the got denser, and solved it in the same way (ie, by throwing lots of ECC at it). I agree with the original poster. It's a hardware problem, caused by the manufacturers pushing the limits. And it's their problem to fix, which they can do. DRAM that doesn't corrupt itself isn't a big ask.
- userbinator 3y ago
  "It always amazes me how people can be so confident yet so wrong" could apply to your post too. Yes it's a problem of physics and it's because they are trying to make DRAM too dense. It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM It's not some conspiracy by "Big RAM" Look at the evidence. This didn't start showing up until around 2009-2010, and the industry managed to convince authors of widely-used memory testing tools to downplay the severity and/or not enable RH tests by default, because they didn't want the truth to be known that almost all RAM is defective. It might not be a conspiracy, but it sure is corporate greed. Would you rather pay a little more for RAM that will work correctly under all access patterns, or RAM that is certain to produce bit errors under some conditions that can be encountered in practice? Unfortunately, with newer DDR3 and later generations, it seems you don't get a choice.
  - DANmode 3y ago
    You get a "choice". The choice just boils down to "buy pricey server boards/chipsets that have ECC RAM available" or "get bent".
    - bogantech 3y ago
      Apparently ECC does not prevent this https://www.vusec.net/projects/eccploit/ https://www.vusec.net/projects/eccploit/
      - DANmode 3y ago
        Awesome link - the ECC side channel seems obvious in retrospect. ECC still superior by a longshot, this information notwithstanding.
xgk 3y ago
This is misleading. I recommend [1] as an introduction to the semiconductor physics behind the Rowhammer problem. Rowhammer is an instance of the "weird machine" problem behind many security problems, i.e. a mismatch between two abstractions: the abstraction we pretend describes the system, vs the reality of the system. In the case of Rowhammer, that is the abstraction of memory as a digital device, against the reality of storing bits with capacitors and wires, ie. analog devices. Clearly a leaky abstraction. The denser you pack those capacitors and wires, the more leaky. [1] A. J. Walker, S. Lee, D. Beery, On DRAM Rowhammer and the Physics of Insecurity. https://ieeexplore.ieee.org/document/9366976 https://ieeexplore.ieee.org/document/9366976
- theamk 3y ago
  I think it's important to differentiate "a mismatch between two abstractions" and "hardware bug". Because you can frame any sort of hardware (or even software!) problem like this: "Capacitor plague of 2000 was a mismatch between two abstractions: the abstraction that capacitor actually provides datasheet-described amount of capacitance vs the reality of the system" "Toyota unintended acceleration was a mismatch between two abstractions: the abstraction that ECU properly responds to accelerator pedal release vs the reality of the system" Yes, digital systems are made of analog parts, but that's not a reason to accept systems behaving out of spec. For the last 50 years, the specifications for RAM have been pretty clear: as long as all datasheet requirements are obeyed, the only way to change stored data in one location should be to do a write to that location. If a memory chip does not act according to its own datasheet, it's not a "leaky abstraction", it's a hardware bug. (Now, can this be fixed economically? I don't know, I could believe the answer is "no". However, the solution in this case is not software workarounds, but rather to make a new spec: "RH-RAM is like regular RAM but cannot tolerate certain access pattern")
  - xgk 3y ago
    I think if you (row)hammer hard enough, every DRAM will eventually flip a bit.
    - 3y ago

mikewarot 3y ago

Came here to say the same thing... take my up-vote. ;-)

bogantech 3y ago

> Rowhammer is a hardware problem ---- defective RAM --- not a software one. It always amazes me how people can be so confident yet so wrong. It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM and that is going to be extremely expensive when talking 16GB and not nearly dense enough. It's not some conspiracy by "Big RAM"

oasisaimlessly 3y ago
> there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM Why wouldn't storing a cryptographically-secure checksum on every RAM row work?
theamk 3y ago
There are plenty of problems of physics in the RAM design, and it's hardware designers jobs to find the operating regime where those problems do not matter. I mean, if you had a DRAM labeled "DDR4-3200" but it can only work at a much lower speed (say DDR4-2400), it's clearly a problem of physics - the gate capacitance is too high, the driver transistors are not strong enough. And yet my reaction would be to take that RAM back to store and get my money back, not to defend manufacturers which claim false things about their chips.
rstuart4133 3y ago
> It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM and that is going to be extremely expensive when talking 16GB and not nearly dense enough. And yet, when the bit flip problem caused by physics got so bad in DDR5 it couldn't be ignored they did fix it - by adding error correction codes. It wasn't that expensive to do so. Notice that HDD's hit the same problem as the got denser, and solved it in the same way (ie, by throwing lots of ECC at it). I agree with the original poster. It's a hardware problem, caused by the manufacturers pushing the limits. And it's their problem to fix, which they can do. DRAM that doesn't corrupt itself isn't a big ask.
userbinator 3y ago
"It always amazes me how people can be so confident yet so wrong" could apply to your post too. Yes it's a problem of physics and it's because they are trying to make DRAM too dense. It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM It's not some conspiracy by "Big RAM" Look at the evidence. This didn't start showing up until around 2009-2010, and the industry managed to convince authors of widely-used memory testing tools to downplay the severity and/or not enable RH tests by default, because they didn't want the truth to be known that almost all RAM is defective. It might not be a conspiracy, but it sure is corporate greed. Would you rather pay a little more for RAM that will work correctly under all access patterns, or RAM that is certain to produce bit errors under some conditions that can be encountered in practice? Unfortunately, with newer DDR3 and later generations, it seems you don't get a choice.
- DANmode 3y ago
  You get a "choice". The choice just boils down to "buy pricey server boards/chipsets that have ECC RAM available" or "get bent".
  - bogantech 3y ago
    Apparently ECC does not prevent this https://www.vusec.net/projects/eccploit/ https://www.vusec.net/projects/eccploit/
    - DANmode 3y ago
      Awesome link - the ECC side channel seems obvious in retrospect. ECC still superior by a longshot, this information notwithstanding.

xgk 3y ago

This is misleading. I recommend [1] as an introduction to the semiconductor physics behind the Rowhammer problem. Rowhammer is an instance of the "weird machine" problem behind many security problems, i.e. a mismatch between two abstractions: the abstraction we pretend describes the system, vs the reality of the system. In the case of Rowhammer, that is the abstraction of memory as a digital device, against the reality of storing bits with capacitors and wires, ie. analog devices. Clearly a leaky abstraction. The denser you pack those capacitors and wires, the more leaky. [1] A. J. Walker, S. Lee, D. Beery, On DRAM Rowhammer and the Physics of Insecurity. https://ieeexplore.ieee.org/document/9366976 https://ieeexplore.ieee.org/document/9366976

theamk 3y ago
I think it's important to differentiate "a mismatch between two abstractions" and "hardware bug". Because you can frame any sort of hardware (or even software!) problem like this: "Capacitor plague of 2000 was a mismatch between two abstractions: the abstraction that capacitor actually provides datasheet-described amount of capacitance vs the reality of the system" "Toyota unintended acceleration was a mismatch between two abstractions: the abstraction that ECU properly responds to accelerator pedal release vs the reality of the system" Yes, digital systems are made of analog parts, but that's not a reason to accept systems behaving out of spec. For the last 50 years, the specifications for RAM have been pretty clear: as long as all datasheet requirements are obeyed, the only way to change stored data in one location should be to do a write to that location. If a memory chip does not act according to its own datasheet, it's not a "leaky abstraction", it's a hardware bug. (Now, can this be fixed economically? I don't know, I could believe the answer is "no". However, the solution in this case is not software workarounds, but rather to make a new spec: "RH-RAM is like regular RAM but cannot tolerate certain access pattern")
- xgk 3y ago
  I think if you (row)hammer hard enough, every DRAM will eventually flip a bit.
  - 3y ago