RFR: 8261710: SA DSO objects have sizes that are too large

[Chris Plummer]

On Wed, 17 Feb 2021 06:46:27 GMT, Yasumasa Suenaga <ysuenaga at openjdk.org> wrote:

> Linux kernel seems to separate final segment (0xbc78) into RO and RW segments when it attempts to load shared library. (but I'm not sure)

I know I've seen this brought up before, if not with SA then with hotspot itself. Maybe it was CDS or NMT. Or maybe it has something to do with how hotspot generates core dumps (I seem to recall there are some bits you can set somewhere that determine what does or does not get dumped to the core).

In any case, I think the main issue it causes for you is that your rounding up the size of the last (4th) segment may not enough. I think in most cases you would need to round it up to a page boundary, and then add another page to it. Consider a page as 4k and let's say the segment size is 2k, but it is split into two 1k segments. Each of those segment would take one page, so a total of 8k. But if you round 2k up to a page boundary you only get to 4k. You need to add another page to that. I think the only time you don't want to add an extra page to the size is if one of the segment's size was already page aligned. However, you have know way of knowing if that will be the case, unless you can determine the RO and RW sizes of the segment.

> But they makes big change to SA.
As an option, we can integrate this change at first, then we will refactor them.
What do you think?

Whatever works for you. I'm actually not too concerned about getting this right, because with my PointerFinder workaround I don't think this issue with the map sizes has much impact on SA. Probably the only place it will show up is with SA pmap output.

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

PR: https://git.openjdk.java.net/jdk/pull/2563