fix the fragment_in_dst_padding_gets_overwritten test on s390x

[fneddy]

on s390x 128bit types have a smaller alignment then on x86¹. This leads to smaller structs (24 instead of 32 bytes) and therefore the write_unaligned will write outside of the structs boundary.

To fix the test, change the trailing u32 into a u8. This will generate 7 trailing padding bytes on s390x and 15 on x86_64. Also change the start of the garbage data so it will span over x,y and padding:

s390x:   XXXXXXXXXXXXXXXXY-------
x86_64:  XXXXXXXXXXXXXXXXY---------------
                      GGGGGGGG

Footnotes

1. s390x ELF ABI Table 1.1, Page 12 https://github.com/IBM/s390x-abi ↩

[rustbot]

r? @madsmtm

rustbot has assigned @madsmtm.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[fneddy]

tagging @RalfJung who introduced the test in #148259

[RalfJung]

Given that what we are testing here is entirely target-independent code (the handling of provenance in the interpreter), I think I'd prefer just skipping the test on s390x over the risk of making the test not test the right thing any more.

[RalfJung]

This leads to smaller structs (24 instead of 32 bytes) and therefore the write_unaligned will write outside of the structs boundary.

If you add align(16) to the struct Pair, does that suffice? That should make it 32 bytes large, right?

[RalfJung]

Actually never mind that, repr(align) would make this not be a ScalarPair any more.