Page 2 of 2

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Sat Mar 01, 2025 8:37 am
by calxalot
libfah is private source because it has confidential stuff in it. Possibly the private key used to sign cores.

I think the core wrappers are private because they use libfah and developer doesn’t have time to audit them.

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Sun Mar 02, 2025 12:42 pm
by arisu
I assumed libfah just had some sort of proprietary anti-cheat sanity checks. I hope it doesn't have the private key in it. If libfah is statically included inside core then that means that the private key could be pulled out from the binaries.

Anyway I will keep working on core22 to see if there is a relatively simple fix.

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Sun Mar 02, 2025 10:50 pm
by calxalot
libfah is also used by private tools.
I don’t expect crypto keys get compiled into cores.

To be clear, confidential stuff may be in the repository, not the compiled lib.

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Sun Mar 02, 2025 10:54 pm
by calxalot
If you become aware of any security issues, please report it to email on page
https://foldingathome.org/contact-us/

Please do not announce publicly here or on github.

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Tue Mar 04, 2025 3:14 am
by arisu
Of course! I didn't mean to suggest that I know of any security issues.

Back to the core22 issue, what exactly is it about newer glibc that is causing it to be unable to find the GPU resource? Is it even known?

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Tue Mar 04, 2025 3:32 am
by calxalot
I believe they are unrelated issues.
A version mismatch with the expected glibc causes the core to crash at launch.

Re: Core 0x22 is unable to detect OpenCL device (bad work unit)

Posted: Tue Mar 04, 2025 4:08 am
by arisu
It didn't crash like that when I tested it (using strace). It initialized OpenCL and found only two compute platforms (Reference and CPU) and not the third (OpenCL). So it exited and the client dumped the WU thinking that it was bad (even though it's the core that is "bad").

Unless that's what you mean by it crashing? But it isn't like it failed to execute or load libraries like would be expected for a symbol mismatch.