This is an archive of the discontinued LLVM Phabricator instance.

Differential D51364

[msan] Tentative fix for failing aarch64 test
AbandonedPublic

Authored by cryptoad on Aug 28 2018, 9:12 AM.

Details

Reviewers
eugenis
vitalybuka
yroux
javed.absar
Summary

Follow up to a discussion in post D50940 commit.

http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/5703/steps/ninja%20check%202/logs/FAIL%3A%20MemorySanitizer-AARCH64%3A%3A%20mmap.cc

0xf00000000 appears to be a valid mapping on aarch64 now. Add it to the list of
app address mappings in the mmap.cc test. Reformat the list a bit as well so
that all the addresses are easier to read.

I can't test this, for all I know lower mappings might also work now and that
might require a couple of iterations. If someone can verify that this solves the
issue at hand, please go ahead!

Diff Detail

Event Timeline

cryptoad created this revision.Aug 28 2018, 9:12 AM
Herald added a reviewer: javed.absar. · View Herald TranscriptAug 28 2018, 9:12 AM
Herald added subscribers: Restricted Project, delcypher, kristof.beyls. · View Herald Transcript
Harbormaster completed remote builds in B22008: Diff 162888.Aug 28 2018, 9:12 AM
cryptoad added a comment.Aug 29 2018, 11:21 AM

Could I please get a LGTM and/or other options so that I can try and fix the aarch64 bots?

eugenis added a comment.Aug 29 2018, 11:33 AM

Looking at msan.h, 0x000F00000000 is not a valid mapping address:

{0x00000000000ULL, 0x01000000000ULL, MappingDesc::INVALID, "invalid"},

Something's broken.

cryptoad added a comment.Aug 29 2018, 11:36 AM
In D51364#1217985, @eugenis wrote:

Something's broken.

Interesting. Do you see anything in https://reviews.llvm.org/D50940 that could have lead to this?

eugenis added a comment.Aug 29 2018, 11:48 AM

I wonder if ProtectMemoryRange in msan_linux.cc needs updating?

cryptoad added a comment.Aug 29 2018, 12:24 PM
In D51364#1218027, @eugenis wrote:

I wonder if ProtectMemoryRange in msan_linux.cc needs updating?

Very good point! I guess now that it returns null on error, differentiating with mapping 0 success isn't working out.

cryptoad abandoned this revision.EditedAug 29 2018, 12:35 PM

Reverting the initial patch with D51451.
Abandoning this.

Revision Contents

PathSize
test/
msan/
31 lines

Diff 162888

test/msan/mmap.cc

Show All 24 Lines
#elif defined(__powerpc64__) #elif defined(__powerpc64__)
return addr < 0x000100000000ULL || addr >= 0x300000000000ULL; return addr < 0x000100000000ULL || addr >= 0x300000000000ULL;
#elif defined(__aarch64__) #elif defined(__aarch64__)
struct AddrMapping { struct AddrMapping {
uintptr_t start; uintptr_t start;
uintptr_t end; uintptr_t end;
} mappings[] = { } mappings[] = {
{0x05000000000ULL, 0x06000000000ULL}, {0x000F00000000ULL, 0x001000000000ULL},
{0x07000000000ULL, 0x08000000000ULL}, {0x005000000000ULL, 0x006000000000ULL},
{0x0F000000000ULL, 0x10000000000ULL}, {0x007000000000ULL, 0x008000000000ULL},
{0x11000000000ULL, 0x12000000000ULL}, {0x00F000000000ULL, 0x010000000000ULL},
{0x20000000000ULL, 0x21000000000ULL}, {0x011000000000ULL, 0x012000000000ULL},
{0x2A000000000ULL, 0x2B000000000ULL}, {0x020000000000ULL, 0x021000000000ULL},
{0x2E000000000ULL, 0x2F000000000ULL}, {0x02A000000000ULL, 0x02B000000000ULL},
{0x3B000000000ULL, 0x3C000000000ULL}, {0x02E000000000ULL, 0x02F000000000ULL},
{0x3F000000000ULL, 0x40000000000ULL}, {0x03B000000000ULL, 0x03C000000000ULL},
{0x0041000000000ULL, 0x0042000000000ULL}, {0x03F000000000ULL, 0x040000000000ULL},
{0x0050000000000ULL, 0x0051000000000ULL}, {0x041000000000ULL, 0x042000000000ULL},
{0x0058000000000ULL, 0x0059000000000ULL}, {0x050000000000ULL, 0x051000000000ULL},
{0x0061000000000ULL, 0x0062000000000ULL}, {0x058000000000ULL, 0x059000000000ULL},
{0x0AAAAA0000000ULL, 0x0AAAB00000000ULL}, {0x061000000000ULL, 0x062000000000ULL},
{0x0FFFF00000000ULL, 0x1000000000000ULL}, {0xAAAAA0000000ULL, 0xAAAB00000000ULL},
{0xFFFF00000000ULL, 0x1000000000000ULL},
}; };
const size_t mappingsSize = sizeof (mappings) / sizeof (mappings[0]); const size_t mappingsSize = sizeof (mappings) / sizeof (mappings[0]);
for (int i=0; i<mappingsSize; ++i) for (int i=0; i<mappingsSize; ++i)
if (addr >= mappings[i].start && addr < mappings[i].end) if (addr >= mappings[i].start && addr < mappings[i].end)
return true; return true;
return false; return false;
#endif #endif
Show All 28 Lines