RFR: 8332689: RISC-V: Use load instead of trampolines [v14]

[Robbin Ehn]

On Fri, 21 Jun 2024 07:35:38 GMT, Fei Yang <fyang at openjdk.org> wrote:

>> Robbin Ehn has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains 22 commits:
>> 
>>  - Minor review comments
>>  - Merge branch 'master' into 8332689
>>  - To be pushed
>>  - Merge branch 'master' into 8332689
>>  - Review comments, removed dead code.
>>  - Merge branch 'master' into 8332689
>>  - Merge branch 'master' into 8332689
>>  - Merge branch 'master' into 8332689
>>  - Merge branch 'master' into 8332689
>>  - Merge branch 'master' into 8332689
>>  - ... and 12 more: https://git.openjdk.org/jdk/compare/d7dad50a...e47f2454
>
> src/hotspot/cpu/riscv/riscv.ad line 1244:
> 
>> 1242:     return 1 * NativeInstruction::instruction_size; // jal
>> 1243:   }
>> 1244:   return 3 * NativeInstruction::instruction_size; // auipc + ld + jalr
> 
> Question: As we will only patch the address in the stub, do we still need the handling in compute_padding (`CallStaticJavaDirectNode::compute_padding` & `CallDynamicJavaDirectNode::compute_padding`) when `UseTrampolines` is false?

No, not that I know of. But we need additional fixes, some asserts needs to tweaked and PostCallNop expects aligned calls (as we want to patch nop -> trap not crossing pages) so we instead need padding after.

I think we can either have:
`c.nop, auipc, ld, jalr, nop (cmodx to trap), nop`
`auipc, ld, jalr, c.nop, nop (cmodx to trap), nop`

It seems to me keeping what we have is just simplest.

What you reckon?

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

PR Review Comment: https://git.openjdk.org/jdk/pull/19453#discussion_r1651211204