[v7,0/2] arm64 tagged address ABI

Message

Catalin Marinas Aug. 7, 2019, 3:53 p.m. UTC

Hi,

Thanks for the feedback so far. This is an updated series documenting
the AArch64 Tagged Address ABI as implemented by these patches:

http://lkml.kernel.org/r/[e-mail]

Version 6 of the documentation series is available here:

http://lkml.kernel.org/r/[e-mail]

Changes in v7:

- Dropped the MAP_PRIVATE requirements for tagged pointers for both
  anonymous and file mappings. One reason is that we can't enforce such
  restriction anyway. The other reason is that a future series
  implementing support for the hardware MTE will detect
  incompatibilities of the new PROT_MTE flag with various mmap()
  options.

- As a consequence of the above, I removed Szabolcs ack as I'm not sure
  he's ok with the change.

- Clarified the sysctl and prctl() interaction and reordered the
  descriptions.

- Reworded the prctl(PR_SET_MM) restrictions.

- Removed the description of the tag preservation from the first patch
  as it didn't really make sense (the syscall ABI has always preserved
  all registers other than x0 on return to user).

- s/ARM64/AArch64/ for consistency with the tagged-pointers.rst
  document.

- Other minor rewordings.

Vincenzo Frascino (2):
  arm64: Define Documentation/arm64/tagged-address-abi.rst
  arm64: Relax Documentation/arm64/tagged-pointers.rst

 Documentation/arm64/tagged-address-abi.rst | 151 +++++++++++++++++++++
 Documentation/arm64/tagged-pointers.rst    |  23 +++-
 2 files changed, 167 insertions(+), 7 deletions(-)
 create mode 100644 Documentation/arm64/tagged-address-abi.rst

Comments

Szabolcs Nagy Aug. 8, 2019, 9:32 a.m. UTC | #1

On 07/08/2019 16:53, Catalin Marinas wrote:
> Hi,
> 
> Thanks for the feedback so far. This is an updated series documenting
> the AArch64 Tagged Address ABI as implemented by these patches:
> 
> http://lkml.kernel.org/r/[e-mail]
> 
> Version 6 of the documentation series is available here:
> 
> http://lkml.kernel.org/r/[e-mail]
> 
> Changes in v7:
> 
> - Dropped the MAP_PRIVATE requirements for tagged pointers for both
>   anonymous and file mappings. One reason is that we can't enforce such
>   restriction anyway. The other reason is that a future series
>   implementing support for the hardware MTE will detect
>   incompatibilities of the new PROT_MTE flag with various mmap()
>   options.

OK.

> - As a consequence of the above, I removed Szabolcs ack as I'm not sure
>   he's ok with the change.
> 
> - Clarified the sysctl and prctl() interaction and reordered the
>   descriptions.
> 
> - Reworded the prctl(PR_SET_MM) restrictions.
> 
> - Removed the description of the tag preservation from the first patch
>   as it didn't really make sense (the syscall ABI has always preserved
>   all registers other than x0 on return to user).

preservation is more interesting when a user pointer
is passed to the kernel and later it is passed back
to user space (e.g. set/get_robust_list, or sigaction
where old handler pointer is returned), then the
kernel may want to drop the tag to do something with
the pointer, but user space may want it to be preserved.

in principle segfault si_addr is a similar case when
memory access via tagged pointer faults: currently
the kernel does not preserve the tag.

so i think it's interesting to know when exactly the
kernel preserves the tags, but it may not be easy to
document in a generic way.

> 
> - s/ARM64/AArch64/ for consistency with the tagged-pointers.rst
>   document.
> 
> - Other minor rewordings.
> 
> Vincenzo Frascino (2):
>   arm64: Define Documentation/arm64/tagged-address-abi.rst
>   arm64: Relax Documentation/arm64/tagged-pointers.rst
> 
>  Documentation/arm64/tagged-address-abi.rst | 151 +++++++++++++++++++++
>  Documentation/arm64/tagged-pointers.rst    |  23 +++-
>  2 files changed, 167 insertions(+), 7 deletions(-)
>  create mode 100644 Documentation/arm64/tagged-address-abi.rst
>