RFR: JDK-8326586: Improve Speed of System.map

[Thomas Stuefe]

On Tue, 27 Feb 2024 06:08:42 GMT, David Holmes <dholmes at openjdk.org> wrote:

>>> In other parts of HotSpot we have been changed usages of `uintx` to instead use `size_t`.
>> 
>> Why? size_t implies memory size to me. This is not a memory size, its a counter.
>> 
>> (If reviewers prefer uintx, I can change it; I just want to understand the reasoning behind choosing size_t)
>
> Why `uintx` rather than just `uint`? Is there a reason to have a variable-sized counter for this? Does it even need to be unsigned?

Hmm... I guess that would work.

Its number of NMT ranges. 

My original thought was: we have a 64-bit address space, usually 128TB, which would give us 32 billion 4K pages, and with region folding (since you need regions separated by a page) this comes to max. 16 billion 4K regions mapped, which would be more than a 32-bit value can represent.

OTOH maybe that's overthinking. unsigned gives us 4 billion max, and the largest processes I have ever seen had  ~20 million mappings. I think a lot of other things start failing before we reach 4 billion.

Okay, I make it an unsigned then.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/17984#discussion_r1503740264