syscall: Getpagesize should return actual system value on Solaris

[binarycrusader]

As noted in #12076 Getpagesize doesn't currently return the actual system's default page size, but rather a hard-coded value. Per the discussion with @ianlancetaylor on golang-dev [1], the opinion was that syscall.Getpagesize should return the same value as the POSIX call sysconf(_SC_PAGESIZE).

This change will be necessary for future platform support on Solaris.

While this change could be generally applied (more than likely) to all other POSIX platforms, I'm only planning on making this change for Solaris for now.

This issue is being filed to track the change against golang/go separately from x/sys/unix.

While it may be viewed as related to #10180, I believe it can be resolved independently.

[1] https://groups.google.com/forum/#!topic/golang-dev/ZLzxYzULzU8

[minux]

I think we should just fix #10180 and make syscall.Getpagesize return runtime._PhysPageSize.

[davecheney]

@minux agreed, having syscall and runtime return different values is an accident waiting to happen.

[binarycrusader]

/CC @4ad

[minux]

Your CL is fine for x/sys/unix (#12076). But for the main repo, we need to update runtime and make syscall and runtime using the same page size.

[gopherbot]

CL https://golang.org/cl/14421 mentions this issue.

[binarycrusader]

Your CL is fine for x/sys/unix (#12076). But for the main repo, we need to update runtime and make syscall and runtime using the same page size.

Like I said, I'm going off what Ian told me, so I'll let all of you work things out.

For the record though, if the runtime does not return the same pagesize as what sysconf(_SC_PAGESIZE) returns, that's going to be a problem too.

[4ad]

I agree, this is fine for x/sys/unix, but for syscall is not the right thing. It has to involve the runtime, and it's not just making the runtime call this function either. Unfortunately, the runtime doesn't work well with big page sizes. We should fix #10180, plus any other bugs that linger in the runtime, like #9993 and #11933 (some of them have been fixed, but they're just symptoms of a larger problem).

In particular, we need to make the runtime work with multiple dynamic page sizes.

[binarycrusader]

@4ad on Solaris, the runtime may not have a choice about whether or not big pages are used. In particular, on SPARC systems, much larger page sizes may be used by default as very large SPARC systems are already at a max of 32 terabytes of memory and I imagine that will increase in the future.

If this change should not be made for syscall today, should I close this issue as a duplicate of #10180 then?

[binarycrusader]

I'm going to close this with the assumption that it's a duplicate of #10180.