Fix Clang vs CONFIG_FORTIFY_SOURCE

[kees]

This is the KSPP bug to track this same issue from ClangBuiltLinux:
ClangBuiltLinux#1002

Basically, CONFIG_FORTIFY_SOURCE does nothing under Clang.

• Handle __builtin_constant_p(end[] - start[]) https://bugs.llvm.org/show_bug.cgi?id=41459
• Do not collapse callers that look like __builtin_mem*() and __builtin_str*(): bug and possible fix?
• __builtin_object_size() does not reliably work within inlined functions: bug

These bugs can be worked around by reorganizing the fortify macros in the kernel, which is underway.

[kees]

Maybe use PASS_ARGUMENT_SIZE, but doesn't seem to help with collapsing: https://godbolt.org/z/_2Fj3z

[kees]

Using external __*_chk() functions can convince Clang to DTRT, but code generation doesn't look right, and this fails in the face of -ffreestanding:
ClangBuiltLinux@41a9c9c

[kees]

Clang and GCC differ in their beliefs about flexible-array address calculations: https://godbolt.org/z/PWjEoo

[kees]

This still could use some improvement in Clang, but overall this is fixed in the kernel now with the various commits leading up to and include commit 281d0c9 (v5.18).