RFR: JDK-8307356: Metaspace: simplify BinList handling [v2]

[Thomas Stuefe]

On Thu, 4 May 2023 12:50:34 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>> In preparation for Lilliput, I'd like to simplify BinList handling a bit.
>> 
>> BinList are a data structure that stores small blocks that had been prematurely deallocated for re-use. Due to the way BinList is implemented, it cannot store blocks smaller than two words. That causes complexities in metaspace and makes estimating memory usage awkward (important for testing). We can eliminate this complexity by shrinking a Block to just one word. 
>> 
>> ----
>> 
>> Details:
>> 
>> A BinList block before: 
>> ``` Block { Block* next; size_t size; }```
>> A BinList block now: 
>> ``` Block { Block* next; }```
>> 
>> We don't need to store the block size, since all blocks in a single sub-list have the same size. In fact, we used the block size only for verification.
>> 
>> With this patch, gross- and net-allocation sizes in Metaspace are the same for 64-bit (see `get_raw_word_size_for_requested_word_size`), which will make writing tests that try to predict metaspace usage based on allocation pattern a lot less onerous to write. For 32-bit, unfortunately, we still need to align allocation sizes to 2 words since allocation addresses need to be aligned to 64-bit.
>
> Thomas Stuefe has updated the pull request incrementally with one additional commit since the last revision:
> 
>   feedback roman

test/hotspot/jtreg/TEST.groups line 381:

> 379:  -runtime/Metaspace/FragmentMetaspaceSimple.java \
> 380:  -runtime/Metaspace/elastic/TestMetaspaceAllocationMT1.java \
> 381:  -runtime/Metaspace/elastic/TestMetaspaceAllocationMT2.java \

Reviewer hint: I need these tests back in GHA tier 1 for the moment (also because more Metaspace changes are coming for Lilliput)

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

PR Review Comment: https://git.openjdk.org/jdk/pull/13776#discussion_r1185693658