[PAUTHABIELF64] Possible typo in R_AARCH64_AUTH_GOT_LO12_NC
#253
Comments
|
I'm on vacation this week. Will get back to you after Easter. It is most likely a typo, my memory is that I wanted to match the existing dynamic relocations. I'll check when I get back to see if there is any other reason. |
smithp35
added a commit
to smithp35/abi-aa
that referenced
this issue
Apr 2, 2024
As pointed out in ARM-software#253 the R_AARCH64_AUTH_GOT_LO12_NC is meant to be the AUTH variant of R_AARCH64_LD64_GOT_LO12_NC. As there is also a R_AARCH64_LD32_GOT_LO12_NC relocation rename the relocation to R_AARCH64_LD64_AUTH_GOT_LO12_NC. These relocations are in the appendix as we are currently expecting the GOT to be RELRO and unsigned in most signing schemas.
|
I agree this is a typo. I've created #255 to fix. To the best of my knowledge the Appendix on GOT generating relocations hasn't been implemented before. At the time of writing the interested parties preferred to have a RELRO unsigned GOT. |
smithp35
added a commit
to smithp35/abi-aa
that referenced
this issue
Apr 3, 2024
There is no equivalent for this relocation in the standard ABI it could be used for runtime code to calculate the address of a GOT slot in a similar way to obtaining the address of a static data item. However there is no way to express the address of a GOT slot in source code, and the compiler can use a load to get the contents of the GOT slot directly without first calculating the address. Part of ARM-software#253
smithp35
added a commit
to smithp35/abi-aa
that referenced
this issue
Apr 15, 2024
There is no equivalent for this relocation in the standard ABI it is used by runtime code to calculate the address of a GOT slot so it can be used as one of the inputs to an authenticate instruction. Add a note that this matches up with the :got_auth_lo12: operator for future reference. Part of ARM-software#253
smithp35
added a commit
that referenced
this issue
Jul 2, 2024
As pointed out in #253 the R_AARCH64_AUTH_GOT_LO12_NC is meant to be the AUTH variant of R_AARCH64_LD64_GOT_LO12_NC. As there is also a R_AARCH64_LD32_GOT_LO12_NC relocation rename the relocation to R_AARCH64_LD64_AUTH_GOT_LO12_NC. These relocations are in the appendix as we are currently expecting the GOT to be RELRO and unsigned in most signing schemas.
smithp35
added a commit
that referenced
this issue
Jul 2, 2024
There is no equivalent for this relocation in the standard ABI it is used by runtime code to calculate the address of a GOT slot so it can be used as one of the inputs to an authenticate instruction. Add a note that this matches up with the :got_auth_lo12: operator for future reference. Part of #253
|
Closing the issue as resolved in #255 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
In https://github.com/ARM-software/abi-aa/blob/main/pauthabielf64/pauthabielf64.rst#auth-variant-got-generating-relocations, AUTH variants of GOT-generating relocations are defined. Their names are in most cases basically the same as corresponding non-AUTH relocs names, but for
R_AARCH64_AUTH_GOT_LO12_NC, it's not true: the non-AUTH variant containsLD64(R_<CLS>_LD64_GOT_LO12_NC), so the AUTH variant should probably beR_AARCH64_AUTH_LD64_GOT_LO12_NC. IfLD64is considered redundant since we don't have 32-bit versions of AUTH relocations, it should be probably deleted from other AUTH relocs names: inR_AARCH64_AUTH_LD64_GOTOFF_LO15andR_AARCH64_AUTH_LD64_GOTPAGE_LO15,LD64is present.Is
LD64omitted inR_AARCH64_AUTH_GOT_LO12_NCintentionally? If yes, it would be nice if you explain rationale of such decision.The text was updated successfully, but these errors were encountered: