RFR: 8326306: RISC-V: Re-structure MASM calls and jumps [v8]

[Robbin Ehn]

On Wed, 8 May 2024 10:21:08 GMT, Fei Yang <fyang at openjdk.org> wrote:

>> The JNI_FastGetField::generate_fast_get_XXX_field0 write the code in a CodeBuffer, this where slow_case_addr points to.
>> 
>> As I test with ReservedCodeCacheSize=2047M if we generated a li() relocation would fail. (there is usually around 120 MB between them)
>> 
>> Added your assert, it passes also.
>> 
>>      ExternalAddress target(slow_case_addr);
>> +    assert(CodeCache::contains(slow_case_addr), "Must be");
>>      __ relocate(target.rspec(), [&] {
>> 
>> 
>> 
>> I guess you are running on apple, the code concerning "static_fast_get_field_wrapper" smells.
>> Maybe you found a bug here.
>
> Try this: `make test TEST="runtime/jni/FastGetField/FastGetField.java"`
> The assertion failed on both linux-aarch64 and linux-riscv64 platforms.

If `-UseFastJNIAccessors or +VerifyJNIFields or +CheckJNICalls` we never use the stubs.
In this case the JVM TI agent have callbacks since it have field watches, so it turns the stubs off.

Good catch thanks!

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

PR Review Comment: https://git.openjdk.org/jdk/pull/18942#discussion_r1593956722