diff options
author | David Spickett <david.spickett@linaro.org> | 2023-11-02 15:56:46 +0000 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-11-02 15:56:46 +0000 |
commit | 0d0ca51ffe1002cec3b1b7a332e290176b650390 (patch) | |
tree | 2cd363803701799d9a9ec67f9ccd438ee464349e /clang/lib/CodeGen/CodeGenModule.cpp | |
parent | 5e1d81ac680482eef401f944bccb6cf03bc7115f (diff) | |
download | llvm-0d0ca51ffe1002cec3b1b7a332e290176b650390.zip llvm-0d0ca51ffe1002cec3b1b7a332e290176b650390.tar.gz llvm-0d0ca51ffe1002cec3b1b7a332e290176b650390.tar.bz2 |
[lldb][AArch64] Read SME2's ZT0 register from Linux core files (#70934)
The ZT0 register is always 64 bytes in size so it is a lot easier to
handle than ZA which is scalable. In addition, reading an inactive ZT0
via ptrace returns all 0s, unlike ZA which returns no register data.
This means that a corefile from a process where ZA and ZT0 were inactive
still contains an NT_ARM_ZT note and we can simply say that if it's
there, then we should be able to read from it.
Along the way I removed a redundant check on the size of the ZA note. If
that note's size is < the ZA header size, we do not have SME, and
therefore could not have SME2 either.
I have added ZT0 to the existing SME core files tests. This means that
you need an SME2 system to generate them (Arm's FVP at this point). I
think this is a fair tradeoff given that this is all running in
simulation anyway and seperate ZT0 tests would be 99% identical copies
of the ZA only tests.
Diffstat (limited to 'clang/lib/CodeGen/CodeGenModule.cpp')
0 files changed, 0 insertions, 0 deletions