How a months-old AMD microcode bug destroyed my weekend [UPDATED]

[jim bell]

 On Tuesday, October 29, 2019, 11:42:42 AM PDT, coderman <coderman at protonmail.com> wrote:
 
 >What’s an RDRAND?>The microcode bug in question is a faulty response to the RDRAND instruction. Modern x86_64 CPUs—beginning with Intel's Broadwell andAMD's Zen architectures—are supposed to have high-quality onboard random number generators (RNGs), which use thermal "noise" to very rapidlyoffer high-entropy pseudorandom numbers to anybody with kernel-levelaccess who wants it. RDRAND is, in turn, the instruction that provides these random numbers.>All of this is supposed to be fairly failsafe. There's a CPUID function call that checks for the availability of RDRAND, and there's also a "carry bit" in the return value from a call to RDRAND that's supposed to let the calling application know if the CPU's RNGwas unable to generate a sufficiently random number. Unfortunately,unpatched Ryzen 3000 says "yes" to the CPUID 01H call, sets the carry bit indicating it has successfully created the mostartisanal, organic high-quality random number possible... and gives you a 0xFFFFFFFF for the "random" number, every single time.


I think that any microprocessor which purports to be able to internally-generate "random numbers" should also be equipped with an input (possibly a single line) which is intended to be connected to an external source of random numbers, intended to be mixed with the internal random source, for example:
http://www.fdk.com/cyber-e/pi_ic_rpg100.html       or  https://www.idquantique.com/random-number-generation/products/quantis-qrng-chip/?gclid=CjwKCAjwxt_tBRAXEiwAENY8hU5d5R0aujGaQjDCHS6Ej_gwwLk8Sz9Z6XG71zec2o9HlcTyjHPaIhoC7HMQAvD_BwE
   orhttps://ieeexplore.ieee.org/document/847868


This should minimize the possibility that defects in one source can affect the randomness of the ultimately-used data stream,
            Jim Bell
  
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/html
Size: 3783 bytes
Desc: not available
URL: <https://lists.cpunks.org/pipermail/cypherpunks/attachments/20191029/acc561e8/attachment.txt>