Following issue #109529 and PR
#109715, this PR removes the
`std::variant` in rtsan's diagnostics code, in favour of a solution by
`enum` without the C++ runtime.
With ASan testing enabled on SPARC as per PR #107405, the
```
AddressSanitizer-sparc-sunos-dynamic :: TestCases/Posix/coverage-fork.cpp
```
test occasionally `FAIL`s on Solaris/sparcv9:
```
compiler-rt/test/asan/TestCases/Posix/coverage-fork.cpp:46:15: error: CHECK-DAG: expected string not found in input
// CHECK-DAG: Parent PID: [[ParentPID:[0-9]+]]
^
```
It turns out that the output for parent and child processes is
interleaved like
```
Parent PID: Child PID: 27426
27425
```
Checking with `truss` shows that the `fprintf`s are implemented as 3
separate `write`s:
```
28489: write(2, " P a r e n t P I D : ", 12) = 12
28489: write(2, " 2 8 4 8 9", 5) = 5
28489: write(2, "\n", 1) = 1
```
To avoid this, this patch switches the test to use `snprintf`/`write` to
guarantee the output is atomic.
Tested on `sparcv9-sun-solaris2.11`, `amd64-pc-solaris2.11`, and
`x86_64-pc-linux-gnu`.
When ASan testing is enabled on SPARC as per PR #107405, the
```
AddressSanitizer-sparc-linux :: TestCases/Posix/print_cmdline.cpp
```
test `FAIL`s. Either `ASAN_OPTIONS=print_cmdline=true` yielded binary
garbage in the `Command:` output or just an empty string.
It turns out one needs to apply an offset to `__libc_stack_end` to get
at the actual `argc`/`argv`, as described in `glibc`'s
`sysdeps/sparc/sparc{32,64}/dl-machine.h` (`DL_STACK_END`).
This patch does this, fixing the test.
Tested on `sparc64-unknown-linux-gnu`.
…ceptors
When ASan testing is enabled on SPARC as per PR #107405, a couple of
tests `FAIL` on Linux/sparc64:
```
AddressSanitizer-sparc-linux :: TestCases/printf-2.c
AddressSanitizer-sparc-linux :: TestCases/printf-3.c
AddressSanitizer-sparc-linux :: TestCases/printf-4.c
AddressSanitizer-sparc-linux :: TestCases/printf-5.c
SanitizerCommon-asan-sparc-Linux :: Linux/unexpected_format_specifier_test.cpp
```
It turns out the interceptors aren't used since on Linux/sparc64
`double` and `long double` are the same, and a couple of `stdio`
functions are prefixed with `__nldbl_` (no long double) accordingly.
This patch handles this.
Tested on `sparc64-unknown-linux-gnu`.
When ASan testing is enabled on SPARC as per PR #107405, the
```
AddressSanitizer-sparc-sunos :: TestCases/Posix/stack-overflow.cpp
```
test `FAIL`s:
```
compiler-rt/test/asan/TestCases/Posix/stack-overflow.cpp:80:12: error: CHECK: expected string not found in input
// CHECK: {{stack-overflow on address 0x.* \(pc 0x.* bp 0x.* sp 0x.* T.*\)}}
^
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer:DEADLYSIGNAL
=================================================================
==11358==ERROR: AddressSanitizer: SEGV on unknown address 0xff3fff90 (pc 0x000db0c0 bp 0xfeed59f8 sp 0xfeed5978 T0)
==11358==The signal is caused by a READ memory access.
AddressSanitizer:DEADLYSIGNAL
AddressSanitizer: nested bug in the same thread, aborting.
```
It turns out that `sanitizer_linux.cpp` (`GetPcSpBp`) tries to
dereference the stack pointer to get at the saved frame pointer, which
cannot work since `sp` has been invalidated by the stack overflow in the
test. The access attempt thus leads to a second `SEGV`.
Solaris `walkcontext(3C)` doesn't have that problem: in the original
OpenSolaris sources (`$SRC/lib/libc/port/gen/walkstack.c`) they used
`/proc/self/as` to avoid the fault, which is quite heavy-handed. Solaris
11.4 uses a non-faulting load instead (`load_no_fault_uint32`, which
just uses the `lduwa` insn).
This patch follows this lead, returning a `NULL` `bp` in the failure
case. Unfortunately, this leads to `SEGV`s in the depth of the unwinder,
so this patch avoids printing a stack trace in this case.
Tested on `sparcv9-sun-solaris2.11` and `sparc64-unknown-linux-gnu`.
When two threads dlopen a shared library, one instance of the library
is loaded. Hence, code inside the library needs to be thread safe,
so use atomic updates for profile counters.
With ASan testing enabled on SPARC as per PR #107405, the
```
AddressSanitizer-sparc-linux-dynamic :: TestCases/Linux/preinstalled_signal.cpp
```
test `FAIL`s on Linux/sparc64. See Issue #109573 for all the details,
but the core is that `syscall(__NR_rt_sigaction)` cannot be used because
it takes an additional argument that isn't accessible outside of `libc`,
while switching to `sigaction` instead changes the order of
`AsanInitInternal` and `Init`, breaking the test.
Therefore this patch `XFAIL`s the test.
Tested on `sparc64-unknown-linux-gnu` and `x86_64-pc-linux-gnu`.
This fixes two problems with asan's interception of `strtol` on Windows:
1. In the dynamic runtime, the `strtol` interceptor calls out to ntdll's
`strtol` to perform the string conversion. Unfortunately, that function
doesn't set `errno`. This has been a long-standing problem (#34485), but
it was not an issue when using the static runtime. After the static
runtime was removed recently (#107899), the problem became more urgent.
2. A module linked against the static CRT will have a different instance
of `errno` than the ASan runtime, since that's now always linked against
the dynamic CRT. That means even if the ASan runtime sets `errno`
correctly, the calling module will not see it.
This patch fixes the first problem by making the `strtol` interceptor
call out to `strtoll` instead, and do 32-bit range checks on the result.
I can't think of any reasonable way to fix the second problem, so we
should stop intercepting `strtol` in the static runtime thunk. I checked
the list of functions in the thunk, and `strtol` and `strtoll` are the
only ones that set `errno`. (`strtoll` was already missing, probably by
mistake.)
# Why?
In llvm, we need to add a call to `__rtsan_notify_blocking_call()` when
a function is marked `[[clang::blocking]]`. This will produce a
different error message than a call to an unsafe malloc etc
Disentangles (and simplifies) integration-like tests for
`__rtsan::ExpectNotRealtime` and `__rtsan::Context` into simpler unit
tests. None of the tests are new, but their assertions have changed to
reflect the more direct testing strategy.
In preparation for providing more information to rtsan's diagnostics
output (via `__rtsan_expect_not_realtime`), this PR separates out all
logic for i) making rtsan's assertions about real-time context state and
ii) displaying diagnostics to the user - disentangling them both from
the rtsan `Context`.
We'll follow up this PR with a simplification to the unit tests that
reflect this new separation.
Originally I tried spliting these features in the compiler with
https://github.com/llvm/llvm-project/pull/101712, but we decided to lump
those features in the ACLE specification (see
https://github.com/ARM-software/acle/pull/346). Since there are no
hardware implementations out there which implement ls64 without ls64_v
or ls64_accdata, this shouldn't be a regression for feature detection.
Since 2c69a09bee94acca859a1adf5b04d01dc13f7295, the Solaris build is
broken like
```
Undefined first referenced
symbol in file
_ZN11__sanitizer20ThreadDescriptorSizeEv projects/compiler-rt/lib/sanitizer_common/CMakeFiles/RTSanitizerCommonLibc.i386.dir/sanitizer_linux_libcdep.cpp.o
```
The `ThreadDescriptorSize` reference is from
`sanitizer_linux_libcdep.cpp` (`GetTls`), l.590. This isn't actually
needed on non-glibc targets AFAICS, so this patch provides a dummy to
restore the build.
Tested on `sparcv9-sun-solaris2.11`, `amd64-pc-solaris2.11`, and
`x86_64-pc-linux-gnu`.
This patch fixes:
compiler-rt/lib/builtins/truncxfbf2.c:16:7: error: ISO C requires a
translation unit to contain at least one declaration
[-Werror,-Wempty-translation-unit]
MSVC can sometimes generate instructions in function prologues that asan
previously didn't know the size of. This teaches asan those sizes. This isn't
super useful for using ASAN with non-msvc compilers, but it does stand alone.
From https://reviews.llvm.org/D151008
When ASan testing is enabled on SPARC as per PR #107405, the
```
AddressSanitizer-sparc-linux :: TestCases/Linux/odr_c_test.c
```
test `FAIL`s on Linux/sparc64:
```
+ projects/compiler-rt/test/asan/SPARCLinuxConfig/TestCases/Linux/Output/odr_c_test.c.tmp
+ count 0
Expected 0 lines, got 13.
AddressSanitizer:DEADLYSIGNAL
=================================================================
==4165420==ERROR: AddressSanitizer: BUS on unknown address (pc 0x7012d5b4 bp 0xffa3b938 sp 0xffa3b8d0 T0)
==4165420==The signal is caused by a READ memory access.
==4165420==Hint: this fault was caused by a dereference of a high value address (see register values below). Disassemble the provided pc to learn which register was used.
```
The test relies on an unaligned access, which cannot work on a
strict-alignment target like SPARC.
Thus this patch skips the test.
Tested on `sparc64-unknown-linux-gnu`.
This fixes#108936, but the calling convention doesn't match with GCC. I
doubt we have such a lib function for now, so leave the calling
convention as is.
