[lworld] RFR: 8372806: [lworld] x64: save bad values instead of rfp and lr above the extension space
Tobias Hartmann
thartmann at openjdk.org
Fri Dec 19 09:44:53 UTC 2025
On Fri, 19 Dec 2025 07:56:27 GMT, Marc Chevalier <mchevalier at openjdk.org> wrote:
> That's the x86 counterpart of [JDK-8371993](https://bugs.openjdk.org/browse/JDK-8371993).
>
> I've added some comments, using the offsets given in `frame_x86.hpp` to make sure to put the frame start at the right place. In particular, the frame start is 2 pointer sizes under the sender's sp:
>
> https://github.com/openjdk/valhalla/blob/3c41c2aa442076827cb4480373a1d481e481cdf1/src/hotspot/cpu/x86/frame_x86.hpp#L61-L62
>
> Unlike aarch64, with x64 we have only one copy of rbp. As with aarch64, I had to get rid of an assert that can't be checked anymore. A small price to pay.
>
> Now, in debug, instead of
>
> pop r13
> sub rsp,0x20
> push r13
>
> we have
>
> pop r13
> sub rsp,0x20
> mov DWORD PTR [rsp-0x4],0xdeadda7a
> mov DWORD PTR [rsp-0x8],0xdeadda7a
> sub rsp,0x8
>
> I've kept the `pop r13` to limit the difference of behavior between debug and product builds: both will overwrite `r13` with the return address, whether it's a good idea or not.
>
> And at runtime, on my favorite `compiler/valhalla/inlinetypes/CorrectlyRestoreRfp.java` example, instead of the stack:
>
> 0x7fd7345fe660: 0x000000042724f5d0 0x00007fd7345fe750 <-- rsp
> 0x7fd7345fe670: 0xffffffffffffffff 0x00007fd740acc0f7
> 0x7fd7345fe680: 0x00007fd700000000 0x000000042724f5d0
> 0x7fd7345fe690: 0x000000056f517f28 0x00007fd7390002a6
> 0x7fd7345fe6a0: 0x00007fd72c90c8b3 0x0000000000000078 <-- # | sp_inc
> 0x7fd7345fe6b0: 0x00007fd7345fe750 0x00007fd740541306 <-- rsp - 0x50 # rbp | return address
> 0x7fd7345fe6c0: 0x000000056f49f4e0 0x000000042724f5d0 <-- # String (arg2) | Object (arg3)
> 0x7fd7345fe6d0: 0x0000000000000000 0x00007fd740541306 <-- rsp - 0x70 # boolean (arg4) | return address
> 0x7fd7345fe6e0: 0x00007fd740541306 0x0000000000000000
> 0x7fd7345fe6f0: 0x000000042724f5d0 0x000000056f49f4e0
>
> we have
>
> 0x7f93cfdfe660: 0x000000042724f5d0 0x00007f93cfdfe750 <-- rsp
> 0x7f93cfdfe670: 0xffffffffffffffff 0x00007f93e4acc107
> 0x7f93cfdfe680: 0x00007f9300000000 0x000000042724f5d0
> 0x7f93cfdfe690: 0x000000056f517fb0 0x00007f93dd0002a6
> 0x7f93cfdfe6a0: 0x00007f93d090c8b3 0x0000000000000078 <-- # | sp_inc
> 0x7f93cfdfe6b0: 0x00007f93cfdfe750 0xdeadda7adeadda7a <-- rsp - 0x50 # rbp | >>>> bad word <<<<
> 0x7f93cfdfe6c0: 0x000000056f49f540 0x000000042724f5d0 <-- # String (arg2) | Object (arg3)
> 0x7f93cfdfe6d0: 0x0000000000000000 0x00007f93e4541306 <-- rsp - 0x70 # boolean (arg4) | return address
> 0x7f93cfdfe6e0: 0x00007f93e4541306 0x00000...
Nice comments! The changes look good to me.
-------------
Marked as reviewed by thartmann (Committer).
PR Review: https://git.openjdk.org/valhalla/pull/1839#pullrequestreview-3597794053
More information about the valhalla-dev
mailing list