Regression in split load/store

[utkonos]

Version and Platform (required):

• Binary Ninja Version: 5.3.8622-dev Ultimate, 62a8cd21
• OS: macos
• OS Version: 15.7
• CPU Architecture: arm64

Bug Description:
This assembly is not accurately lifted to IL:

18001b53d  65488b03           mov     rax, qword [gs:rbx]

Here is the resulting LLIL:

  45 @ 18001b53d  rax = [gsbase + rbx {_TEB::ProcessEnvironmentBlock+3.q}].q

And HLIL:

18001b53d        TEB* gsbase
18001b53d        rax_4.b = gsbase->ProcessEnvironmentBlock.b
18001b53d        rax_4:1.b = gsbase->ProcessEnvironmentBlock:1.b
18001b53d        rax_4:2.b = gsbase->ProcessEnvironmentBlock:2.b
18001b53d        rax_4:3.b = gsbase->ProcessEnvironmentBlock:3.b

This causes more problems downstream because rax_4 is not of type PEB*

Steps To Reproduce:
Open the database or binary and look at 0x18001b53d

BNDB: dense atom prunes validly
Binary: dual yard releases fractally

Expected Behavior:
HLIL should be:

rax_4 = gsbase->ProcessEnvironmentBlock

Screenshots:
Setting rax_4 manually to type PEB* fixes most of the downstream problems (remaining problems listed below):

Even after manually setting the type of PEB*, this instruction is missing an annotation. Downstream instructions do get correct annotations, but this one is still missing:

[rssor]

This is fixed on dev, I can pull in the commit/build number later when I'm at a computer again. There's several of these to close, actually.

[rssor]

Fixed in 5.3.8689.

[utkonos]

@rssor This looks good. Thanks. It didn't fix the bug with relative offsets in the loader linked lists in issue #7648 however. I think that's due to an incomplete fix to #6492

[rssor]

Yeah that's something worse, I'm taking a look at it now.

[rssor]

Oh, nope, it's nowhere near as bad, just the offsets being wrong like you said.

[utkonos]

FYI: If you have the above issue, and you also have a version of Binja where all the underlying fixes have been implemented, to get a function looking as expected there are a few steps needed. That's all here:

#7769 (comment)