Re: Address space limit?

On Jun 7, 10:04am, [email protected] (Chris Hanson) wrote:
-- Subject: Re: Address space limit?

| > On Jun 7, 2019, at 9:28 AM, John Klos <[email protected]> wrote:
| >>=20
| >> Going forward with 9.0, what should we do with 4K versus 8K platforms? T=
| wo sets of binaries seems like a bad idea.
| >=20
| > Can jemalloc be fixed to just obtain the page size of the currently runni=
| ng system? It doesn=E2=80=99t seem like having it be a compile time constan=
| t would really be that advantageous.

It does. Compares the one it gets from the running system to the built-in
one. If the one from the running system is larger, it refuses to run (because
it needs the finer grain alignment).

| It looks like this is also a jemalloc issue being discussed on port-sandpoi=
| nt; sandpoint uses a couple specific PowerPC CPUs with fixed page sizes, bu=
| t other PowerPC-based systems may use a different page size.

Same for sparc; it is all fixed now. Using the MAX_PAGE_SIZE when building
binaries for a platform makes binaries work across all page sizes.

christos