design flaw: operating system page size is not always comptime known, but std lib assumes it is

[daurnimator]

There is no longer just one page size; "huge pages" are well supported by modern processors and operating systems.

Originally posted by @daurnimator in #2527

[shawnl]

Yes, but many systems make them transparent so they don't break things.

[LemonBoy]

Yes, but many systems make them transparent so they don't break things.

It's not really transparent if you consider other architectures.

[andrewrk]

user space can only know it at execution time from the aux vector

I was afraid of this being the case. This is so inconvenient that I might consider making the "page size" part of the target. I don't know yet. This will be an important decision for the limitations & capabilities of the standard library.

[shawnl]

@LemonBoy I just said that cause i'm using ppc64le and the large page sizes are automatic, but 4096 still works. That is the default config.

user space can only know it at execution time from the aux vector

This could be used with symbol multi-versioning, but it definitely is a pain.

[daurnimator]

Maybe all we need is the minimum page size?
That would suffice for e.g. mmap alignment

[andrewrk]

Solution: std.mem.page_size is the comptime-known minimum page size. Whether or not to rename it to "minimum_page_size" or not is a separate question.