Crash on macosx-aarch64 when merging Leyden premain with mainline

[Vladimir Kozlov]

Briefly looking through 8361376 aarch64 changes which uses 
relocInfo::entry_guard_type to set _guard_addr.

We don't update it when loading AOT code. May be this is the issue.

Vladimir K

On 9/18/25 2:04 PM, Dean Long wrote:
> 8367131 is a test-only change.  What is the code doing at the crash 
> site?  It looks like the crash is because of an unaligned memory 
> address, but the reported address 0x000000012a8c7108 looks aligned 
> unless the code is doing a 16-byte memory access.  It could possibly be 
> related to my recent change 8361376.
> 
> dl
> 
> On 9/18/25 1:37 PM, ioi.lam at oracle.com wrote:
>> I got a crash when building the JDK after merging the latest premain 
>> with the following changeset in the mainline
>>
>>
>> commit fdc11a1569248c9b671b66d547b4616aeb953ecf
>> Author: Sergey Bylokhov <serb at openjdk.org>
>> Date:   Wed Sep 10 11:41:42 2025
>>
>>     8367131: Test com/sun/jdi/ThreadMemoryLeakTest.java fails on 32 bits
>>
>> I had to do "make jdk-image LOG=debug" to find out the exact command 
>> that crashed and ran it by hand to get hs_err, which shows:
>>
>>
>> V  [libjvm.dylib+0x2c49c8] 
>> BarrierSetNMethod::nmethod_stub_entry_barrier(unsigned char**)+0x140
>> v  ~StubRoutines::Stub Generator method_entry_barrier_stub 
>> 0x0000000131de4938
>> siginfo: si_signo: 10 (SIGBUS), si_code: 1 (BUS_ADRALN), si_addr: 
>> 0x000000012a8c7108
>>
>>
>> Could anyone familiar with the code help trouble shooting this? (You 
>> don't want to merge beyond the above changeset as you'd run into some 
>> CDS conflicts, which I plan to address above this crash is fixed).
>>
>> This happens on my laptop (M3 and Sequoia 15.6.1) and our CI as well. 
>> For some reason, the crash didn't happen with GitHub action, so it may 
>> be specific to macos or CPU versions.
>>
>> Thanks
>>
>> - Ioi
>>
>>
>>