8364269: Simplify code cache API by storing adapter entry offsets in blob

[adinn]

The AOT stub blob save and restore API handles Adapter blobs differently to other single-stub blobs by passing the associated entries in a separate auxiliary array. Storing the entry offsets in the blob, as happens with other generated blobs, simplifies the current save/restore API and implementation. It also makes it easier to define and implement an API extension supporting save and restore of multi-stub (StubGen) blobs.

---

Progress

• Change must be properly reviewed (1 review required, with at least 1 Reviewer)
• Change must not contain extraneous whitespace
• Commit message must refer to an issue

Issue

• JDK-8364269: Simplify code cache API by storing adapter entry offsets in blob (Enhancement - P4)

Reviewers

• Vladimir Kozlov (@vnkozlov - Reviewer)
• Aleksey Shipilev (@shipilev - Reviewer)
• Ashutosh Mehra (@ashu-mehra - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/26532/head:pull/26532
$ git checkout pull/26532

Update a local copy of the PR:
$ git checkout pull/26532
$ git pull https://git.openjdk.org/jdk.git pull/26532/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 26532

View PR using the GUI difftool:
$ git pr show -t 26532

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/26532.diff

Using Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back adinn! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

@adinn This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8364269: Simplify code cache API by storing adapter entry offsets in blob

Reviewed-by: kvn, shade, asmehra

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 122 new commits pushed to the master branch:

• c71be80: 8361748: Enforce limits on the size of an XBM image
• b8acbc3: 8364315: Remove unused xml files from test/jaxp/javax/xml/jaxp/functional/javax/xml/transform/xmlfiles
• 244e629: 8364984: Many jpackage tests are failing on Linux after JDK-8334238
• ... and 119 more: https://git.openjdk.org/jdk/compare/0226c0298f5398c185db3df30ad35ee6022aab1b...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

[openjdk]

@adinn The following label will be automatically applied to this pull request:

• hotspot

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

[mlbridge]

Webrevs

• 02: Full - Incremental (a7925825)
• 01: Full - Incremental (ad4a1a27)
• 00: Full (452f2d86)

[shipilev]

Cross-compile GHA jobs would be fixed once you pull from recent master.

[adinn]

Testing: tier1 + runtime/cds/appcds

[adinn]

@ashu-mehra @vnkozlov could you review this cleanup?

[adinn]

Ah, sorry, the merge appears to have caused some issues in debug builds. Will let you know when this is ready again.

[adinn]

@vnkozlov @ashu-mehra @shipilev

Ok, problem fixed, I believe, but I'm still gob-smacked that this built, ran and passed tier1 plus cds/appcds tests on Linux/aarch64 given the failure to run the exploded image on Linux/x86 and MacOS/aarch64.

The problem was with the blob layout. Adding extra fields to AdapterBlob requires propagating the adjusted blob size (i.e. header_size) up the Blob constructor chain, through BufferBlob into RuntimeBlob. Which I have now done -- previously, all subclasses were sized using sizeof(BufferBlob) as none of them had extra fields.

without that fix the content_start address for the Adpater blob was being reported in the tail of the blob proper which meant that in the broken builds data fields were being treated as code and we end up in hyperspace when we first use an i2c adapter to jump to Object.init(). Yet not on Linux/aarch64? Go figure.

Should be ready to review now.

[ashu-mehra]

If the entry offsets are stored in AdapterBlob, AdapterHandlerEntry can then just store a pointer to the blob and get rid of the offsets. SharedRuntime::generate_i2c2i_adapters can be updated to return the AdapterBlob. And when storing the AdapterHandlerEntry, pointer to the blob can be cleared and restored in AdapterHandlerLibrary::lookup_aot_cache.
This assumes the AdapterBlob won't be moved around in the code cache.
@adinn does this make sense?

[adinn]

@ashu-mehra

If the entry offsets are stored in AdapterBlob, AdapterHandlerEntry can then just store a pointer to the blob and get rid of the offsets.

I planned that originally but it doesn't work because of one special case. Method AdapterHandlerLibrary::create_abstract_method_handler does this:

  address wrong_method_abstract = SharedRuntime::get_handle_wrong_method_abstract_stub();
  _abstract_method_handler = AdapterHandlerLibrary::new_entry(AdapterFingerPrint::allocate(0, nullptr));
  _abstract_method_handler->set_entry_points(SharedRuntime::throw_AbstractMethodError_entry(),
                                             wrong_method_abstract,
                                             wrong_method_abstract,
                                             nullptr);

So, in this one case we have a Frankenstein handler assembled from 3 disparate entry points (+ nullptr) which are not derived from an associated adapter blob.

[ashu-mehra]

So, in this one case we have a Frankenstein handler assembled from 3 disparate entry points (+ nullptr) which are not derived from an associated adapter blob.

That's true. I missed this special adapter handler. That special handler is a pain to accommodate.
I looked at how these entry points are accessed and it seems they are mostly accessed by the routines in Method. So it should be feasible to handle the case of abstract methods there and get rid of this special adapter completely. @adinn wdyt?

[adinn]

I looked at how these entry points are accessed and it seems they are mostly accessed by the routines in Method. So it should be feasible to handle the case of abstract methods there and get rid of this special adapter completely. @adinn wdyt?

I think maybe in follow-up RFE ;-)

[ashu-mehra]

think maybe in follow-up RFE ;-)

Yeah, I can take a stab at that.
Meanwhile this PR is still in draft. Do you have any other changes to be made? Seems ready to me.

[vnkozlov]

Looks good. Let me test it.

[shipilev]

This looks fine to me. API got indeed simpler.

[shipilev]

I was today years old when I realized you can pass const-sized arrays as arguments.

[ashu-mehra]

lgtm

[vnkozlov]

My testing passed.

[adinn]

Thanks all for the reviews. Integrating now.

[adinn]

/integrate

[openjdk]

Going to push as commit 241808e.
Since your change was applied there have been 129 commits pushed to the master branch:

• 1b3e231: 8360048: NMT crash in gtest/NMTGtests.java: fatal error: NMT corruption: Block at 0x0000017748307120: header canary broken
• a26a6f3: 8364649: G1: Move collection set related full gc reset code into abandon_collection_set() method
• 47017e3: 8364760: G1: Remove obsolete code in G1MergeCardSetClosure
• ... and 126 more: https://git.openjdk.org/jdk/compare/0226c0298f5398c185db3df30ad35ee6022aab1b...master

Your commit was automatically rebased without conflicts.

[openjdk]

@adinn Pushed as commit 241808e.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

[shqking]

Hi @adinn,

We encountered runtime error on Arm32 after this patch. I filed in https://bugs.openjdk.org/browse/JDK-8365229.
I was wondering if you have spare time to take a look at this issue. Thanks.