[IRGen] Upstream pointer auth for value witness tables #74625
Conversation
|
@swift-ci please test |
egorzhdan
force-pushed
the
egorzhdan/upstream-ptrauth-vwt
branch
from
b16d6d5 to
162ff58
Compare
|
@swift-ci please test |
|
@swift-ci please test Windows |
lib/IRGen/GenMeta.cpp
Outdated
|
|
||
| const auto &ptrAuthOpts = IGM.getOptions().PointerAuth; | ||
| if (ptrAuthOpts.FunctionPointers && ptrAuthOpts.FunctionPointers.getKind() == | ||
| PointerAuthSchema::Kind::ARM8_3) { |
There was a problem hiding this comment.
This test should really just be if (auto schema = IGM.getOptions().PointerAuth.ValueWitnessTable) {. We don't need to test function pointers or the kind of ptrauth we're doing in general; knowing that we have a request to sign vwtables is enough.
There was a problem hiding this comment.
Right, this check is already performed in IRGen's setPointerAuthOptions. Removed it here.
lib/IRGen/GenMeta.cpp
Outdated
| auto *Clang = | ||
| static_cast<ClangImporter *>(IGM.Context.getClangModuleLoader()); | ||
| auto &targetInfo = Clang->getTargetInfo(); | ||
| auto intSize = Size::forBits(targetInfo.getIntWidth()); |
There was a problem hiding this comment.
This part of the patch seems to be trying to set a more accurate bound on the dereferenceable range of the vwtable. That's fine, but it should be a separate patch (and probably extracted into a nice helper function).
There was a problem hiding this comment.
Sounds good, I removed this change from the patch.
egorzhdan
force-pushed
the
egorzhdan/upstream-ptrauth-vwt
branch
from
162ff58 to
30cd90c
Compare
|
@swift-ci please test |
This makes Swift emit a signed pointer to the value witness table in type metadata. The original change was done by Varun Gandhi.
egorzhdan
force-pushed
the
egorzhdan/upstream-ptrauth-vwt
branch
from
30cd90c to
b22057e
Compare
|
@swift-ci please test |
This makes Swift emit a signed pointer to the value witness table in type metadata.
The original change was done by Varun Gandhi.