Yes, that's weird. The instruction that's loading the G pointer from the TLS area is faulting, and on the address of the instruction itself, not the address it's loading from.
Are you using any unsafe? cgo?
I can't imagine it would fail this way unless you're calling syscall.Mmap to unmap the code segment.
Any chance you could give us a program with which we could reproduce this ourselves? Otherwise I don't see any way to figure out what is going on.
Unfortunately, I don't think we can hand over a binary for this particular service.
Are there any debug steps that we can take on our end to gather more data for you? As I mentioned in the initial post, this service has been running for quite a long time and has only crashed twice with this error, I'm not quite sure how we could reproduce it.