> There is a reason so much effort is put in TPM and remote attestation and so on

If you trust TPM not to be backdoored... come on, you don't think the NSA or who else has put effort in getting a backdoor inside? They even tried to put one in Linux and it's documented, never the less in anything proprietary...

> It can just read the generated seed directly from user space without the program ever knowing about it.

Not that simple: it has to know exactly where in memory it's stored, and that requires understanding of the source code of the program that is encrypting data. That is not of course a simple task if someone wants to write a malware that just "steals" encrypted data from any software just by looking at the network traffic, like you would do if you compromise the RNG of the OS.

> clock_gettime() just reads a value that the kernel has set, so that's not particularly difficult to fake.

You can sample the call millions of time and understand if the value is truly random or there is a pattern. It's something detectable. Software like GPG that doesn't trust what the OS gives you already do that (as well as combining multiple entropy sources).

> It's fine if you use it as a strictly additional source of entropy, but then the whole argument that it is superior because it avoids syscalls goes out of the window, because you're doing strictly _more_ work.

Avoiding the syscall could have other benefits, not only performance. For example: a program making that syscall may be flagged by a possible backdoor as a process with something interesting in it, and thus a potential spyware may be interested in take, for example, the memory image of that program and send it to a remote system for it to be analyzed. The fact that the reading of the current time doesn't pass from a system calls means that it's not possible to identify that process as "some process that uses cryptography and thus has something interesting in it to hide".