[libc++abi] Fix test on Android

[maksfb]

Follow up to #72727. The added test could not be executed on Android.

[llvmbot]

@llvm/pr-subscribers-libcxxabi

Author: Maksim Panchenko (maksfb)

Changes

Follow up to #72727. The added test could not be executed on Android.

---

Full diff: https://github.com/llvm/llvm-project/pull/74753.diff

1 Files Affected:

• (modified) libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s (+1-1)

diff --git a/libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s b/libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s
index ea9dd0104c086..249d243d921ea 100644
--- a/libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s
+++ b/libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s
@@ -2,7 +2,7 @@
 # RUN: %t.exe
 
 # REQUIRES: linux && target={{x86_64-.+}}
-# UNSUPPORTED: no-exceptions
+# UNSUPPORTED: no-exceptions || android
 
 ## Check that libc++abi works correctly when LPStart address is explicitly set
 ## to zero.

[maksfb]

Unrelated ibm-libc++-shared.cfg.in test failed on AIX.

[maksfb]

Checked that all test failures are unrelated.

[ldionne]

@rprichard Can you take a look? I suspect the right thing to do here is not to mark the test as UNSUPPORTED, since that would simply reduce your test coverage.

[rprichard]

I think the UNSUPPORTED is correct here.

The RUN: %t.exe line should be RUN: %{exec} %t.exe, but after fixing that, the test will still fail on Android because of the -no-pie compile option in RUN: %{cxx} %{flags} %s %{link_flags} -no-pie -o %t.exe. The Android dynamic loader rejects non-PIE executables:

"./t.tmp.exe": error: Android 5.0 and later only support position-independent executables (-fPIE).

I can't make the executable work as a PIE because the address of the landing pad would require a dynamic relocation, which can't be applied because .gcc_except_table is read-only. Making the section writable instead is rejected by the assembler:

/llvm/libcxxabi/test/native/x86_64/lpstart-zero.pass.sh.s:76:2: error: changed section flags for .gcc_except_table, expected: 0x2
        .section        .gcc_except_table,"aw",@progbits
        ^

Also:

.Lexception0:
	.byte	3                               # @LPStart Encoding = udata4
  .long 0
	.byte	155                             # @TType Encoding = indirect pcrel sdata4
	.uleb128 .Lttbase0-.Lttbaseref0
.Lttbaseref0:
	.byte	11                              # Call site Encoding = udata4

DW_EH_PE_udata4 is 3 but DW_EH_PE_sdata4 is 11. I think the comment on the call side encoding is wrong -- it should be sdata4 instead of udata4.

[ldionne]

Thanks for the analysis @rprichard!

@maksfb Can you please address those comments?

[ldionne]

@rprichard What would be the right target-based annotation here?

// UNSUPPORTED: target={{.+}}-android{{.+}}

?

[rprichard]

I think I would change the suffix to use {{.*}} to also match a triple without the version number suffix, just for completeness:

// UNSUPPORTED: target={{.+}}-android{{.*}}

The two unversioned Android annotations I currently see in libc++ also use a {{.*}} suffix:

// XFAIL: LIBCXX-ANDROID-FIXME && target={{i686|x86_64}}-{{.+}}-android{{.*}}

// UNSUPPORTED: target=i686-{{.+}}-android{{.*}}

[maksfb]

Fixed the execution command and the comment. The test is supposed to be non-PIE, which excludes it from testing on Android.