evp_setpeer: fix -k all

[esyr]

As max_time was initialised only once, the runs after the first one terminated immediately. Move max_time inside the key type loop.

Fixes: 7388d62 "Set a constant amount of runtime on more perf tests"

[jogme]

At many (if not all) places the max time is not set directly before calling run_tests. I'm not sure if that's a feature or a bug, but if the latter it would be worth fixing through the codebase

[esyr]

Well, the only other place where it induced a bug has been already fixed in e5cd2d4, and touching it just to get the proximity seems superficial to me; if anything, the whole test harness needs to be factored out, to avoid such bugs altogether.

[jogme]

If this specifically solves a bug then MB.

I was nitpicking about the fact, that we are doing operations such as memory allocation in the measured section, which if not explicitly intended then it's not correct IMO. Also it's probably not such an overload, but it's picking my OCD

[esyr]

There's a lot of that in terms of measurement: the fact that threads hammer the same counts array and constantly invalidate the cache line, the ossl_time_now() call on each iteration, inclusion of thread creation in the measured time, the list goes on; but, well, one thing at a time.

[jogme]

thanks!

[andrewkdinh]

Ah, my bad.

@quarckster fyi this might affect evp_setpeer perf test

Edit: Oh actually, since this is just for -k all then it probably won't