Don't overestimate proprietary Trusted Execution Environments. We might never know for sure whether they are another sophisticated NSA honeypots or not. Issues have already been found in them; Nvidia is not excluded.
Not that I'm against these technologies, but I'd like to see more solutions based on something open and independently verifiable as well.
If it's not something identical to hardware TEE solutions, then at least there's a not-yet-well-known Linux syscall `memfd_secret`, which is basically a way to allocate private RAM regions that are, to some degree, isolated even from the kernel (pages with these regions won't be swapped to disk, core dumps won't include this memory, etc.).
It's a limited solution. It's not for GPUs. However, it's open and independent from any particular vendor.
Originally it was designed for cryptography. Useful for NIP-46 signers and CPU-only ML models for example.
https://www.man7.org/linux/man-pages/man2/memfd_secret.2.html#NOTES
#privacy #linux #TEE
Not that I'm against these technologies, but I'd like to see more solutions based on something open and independently verifiable as well.
If it's not something identical to hardware TEE solutions, then at least there's a not-yet-well-known Linux syscall `memfd_secret`, which is basically a way to allocate private RAM regions that are, to some degree, isolated even from the kernel (pages with these regions won't be swapped to disk, core dumps won't include this memory, etc.).
It's a limited solution. It's not for GPUs. However, it's open and independent from any particular vendor.
Originally it was designed for cryptography. Useful for NIP-46 signers and CPU-only ML models for example.
https://www.man7.org/linux/man-pages/man2/memfd_secret.2.html#NOTES
#privacy #linux #TEE
