[msan] Increase kNumStackOriginDescrs constant

[thurstond]

This increases the constant size of kNumStackOriginDescrs to 4M (64GB of BSS across two arrays), which ought to be enough for anybody.

This is the easier alternative suggested by eugenis@ in #92826.

[llvmbot]

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Thurston Dang (thurstond)

Changes

This increases the constant size of kNumStackOriginDescrs to 64M (1GB of BSS across two arrays), which ought to be enough for anybody.

This is the easier alternative suggested by eugenis@ in #92826.

---

Full diff: https://github.com/llvm/llvm-project/pull/92838.diff

1 Files Affected:

• (modified) compiler-rt/lib/msan/msan.cpp (+7-1)

diff --git a/compiler-rt/lib/msan/msan.cpp b/compiler-rt/lib/msan/msan.cpp
index a2fc27de1901b..8bef6812f21c2 100644
--- a/compiler-rt/lib/msan/msan.cpp
+++ b/compiler-rt/lib/msan/msan.cpp
@@ -100,7 +100,13 @@ int msan_report_count = 0;
 
 // Array of stack origins.
 // FIXME: make it resizable.
-static const uptr kNumStackOriginDescrs = 1024 * 1024;
+// Although BSS memory doesn't cost anything until used, it is limited to 2GB
+// in some configurations (e.g., relocation R_X86_64_PC32 out of range:
+// 8600110908 is not in [-2147483648, 2147483647]; references section '.bss')
+// hence kNumStackOriginDescrs is limited to roughly 2GB / sizeof(uptr) / 2
+// == 128M per array (StackOriginDescr, StackOriginPC). We set it at 64M each
+// in case other parts of MSan want more BSS space in the future.
+static const uptr kNumStackOriginDescrs = 64 * 1024 * 1024;
 static const char *StackOriginDescr[kNumStackOriginDescrs];
 static uptr StackOriginPC[kNumStackOriginDescrs];
 static atomic_uint32_t NumStackOriginDescrs;

[eugenis]

Hmm I did not consider the relocation offset. This can realistically push some other .bss variable out of the 2Gb range if the linker is not smart enough to place the largest object last. The 2Gb is not just MSan rtl budget, but for the whole program.

Lets bring this down to 4M origins, that should be enough for the foreseeable future.

[thurstond]

Hmm I did not consider the relocation offset. This can realistically push some other .bss variable out of the 2Gb range if the linker is not smart enough to place the largest object last. The 2Gb is not just MSan rtl budget, but for the whole program.

Lets bring this down to 4M origins, that should be enough for the foreseeable future.

Done!

[eugenis]

Thank you.