Skip to content

[tsan] Increase size of shadow mappings for C/C++ on linux/x86_64 #70517

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Nov 2, 2023

Conversation

thurstond
Copy link
Contributor

The current TSan mappings for C/C++ on linux/x86_64 have 0.5TB
for low app mem, 1.5TB for mid app mem and 1.5TB for high app mem.
This can get a bit cramped if the apps are huge, and/or (in the
case of mid/high app mem) with significant ASLR entropy.

This patch increases the mapping sizes to 2TB, 5TB, and 4TB for
the low, mid and high app regions respectively. It is difficult
to make the mappings any larger, given the 44-bit pointer compression.

It also moves the heap region to avoid HeapEnd() overlapping with
the newly enlarged high app region.

For convenience, we now use kShadowAdd instead of kShadowXor for
this set of mappings. This should be roughly equivalent in
runtime performance.

The current TSan mappings for C/C++ on linux/x86_64 have 0.5TB
for low app mem, 1.5TB for mid app mem and 1.5TB for high app mem.
This can get a bit cramped if the apps are huge, and/or (in the
case of mid/high app mem) with significant ASLR entropy.

This patch increases the mapping sizes to 2TB, 5TB, and 4TB for
the low, mid and high app regions respectively. It is difficult
to make the mappings any larger, given the 44-bit pointer compression.

It also moves the heap region to avoid HeapEnd() overlapping with
the newly enlarged high app region.

For convenience, we now use kShadowAdd instead of kShadowXor for
this set of mappings. This should be roughly equivalent in
runtime performance.
@llvmbot
Copy link
Member

llvmbot commented Oct 27, 2023

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

Author: Thurston Dang (thurstond)

Changes

The current TSan mappings for C/C++ on linux/x86_64 have 0.5TB
for low app mem, 1.5TB for mid app mem and 1.5TB for high app mem.
This can get a bit cramped if the apps are huge, and/or (in the
case of mid/high app mem) with significant ASLR entropy.

This patch increases the mapping sizes to 2TB, 5TB, and 4TB for
the low, mid and high app regions respectively. It is difficult
to make the mappings any larger, given the 44-bit pointer compression.

It also moves the heap region to avoid HeapEnd() overlapping with
the newly enlarged high app region.

For convenience, we now use kShadowAdd instead of kShadowXor for
this set of mappings. This should be roughly equivalent in
runtime performance.


Full diff: https://github.com/llvm/llvm-project/pull/70517.diff

1 Files Affected:

  • (modified) compiler-rt/lib/tsan/rtl/tsan_platform.h (+21-22)
diff --git a/compiler-rt/lib/tsan/rtl/tsan_platform.h b/compiler-rt/lib/tsan/rtl/tsan_platform.h
index cfbb57d1d8d8d9b..6a191652ff1ed73 100644
--- a/compiler-rt/lib/tsan/rtl/tsan_platform.h
+++ b/compiler-rt/lib/tsan/rtl/tsan_platform.h
@@ -46,17 +46,16 @@ enum {
 
 /*
 C/C++ on linux/x86_64 and freebsd/x86_64
-0000 0000 1000 - 0080 0000 0000: main binary and/or MAP_32BIT mappings (512GB)
-0040 0000 0000 - 0100 0000 0000: -
-0100 0000 0000 - 1000 0000 0000: shadow
-1000 0000 0000 - 3000 0000 0000: -
-3000 0000 0000 - 3400 0000 0000: metainfo (memory blocks and sync objects)
-3400 0000 0000 - 5500 0000 0000: -
-5500 0000 0000 - 5680 0000 0000: pie binaries without ASLR or on 4.1+ kernels
-5680 0000 0000 - 7d00 0000 0000: -
-7b00 0000 0000 - 7c00 0000 0000: heap
-7c00 0000 0000 - 7e80 0000 0000: -
-7e80 0000 0000 - 8000 0000 0000: modules and main thread stack
+0000 0000 1000 - 0200 0000 0000: main binary and/or MAP_32BIT mappings (2TB)
+0200 0000 0000 - 1000 0000 0000: -
+1000 0000 0000 - 3000 0000 0000: shadow (32TB)
+3000 0000 0000 - 3800 0000 0000: metainfo (memory blocks and sync objects; 8TB)
+3800 0000 0000 - 5500 0000 0000: -
+5500 0000 0000 - 5a00 0000 0000: pie binaries without ASLR or on 4.1+ kernels
+5a00 0000 0000 - 7b00 0000 0000: -
+7a00 0000 0000 - 7b00 0000 0000: heap (1TB)
+7b00 0000 0000 - 7c00 0000 0000: -
+7c00 0000 0000 - 8000 0000 0000: modules and main thread stack (4TB)
 
 C/C++ on netbsd/amd64 can reuse the same mapping:
  * The address space starts from 0x1000 (option with 0x0) and ends with
@@ -72,20 +71,20 @@ C/C++ on netbsd/amd64 can reuse the same mapping:
 */
 struct Mapping48AddressSpace {
   static const uptr kMetaShadowBeg = 0x300000000000ull;
-  static const uptr kMetaShadowEnd = 0x340000000000ull;
-  static const uptr kShadowBeg     = 0x010000000000ull;
-  static const uptr kShadowEnd = 0x100000000000ull;
-  static const uptr kHeapMemBeg    = 0x7b0000000000ull;
-  static const uptr kHeapMemEnd    = 0x7c0000000000ull;
+  static const uptr kMetaShadowEnd = 0x380000000000ull;
+  static const uptr kShadowBeg = 0x100000000000ull;
+  static const uptr kShadowEnd = 0x300000000000ull;
+  static const uptr kHeapMemBeg = 0x7a0000000000ull;
+  static const uptr kHeapMemEnd = 0x7b0000000000ull;
   static const uptr kLoAppMemBeg   = 0x000000001000ull;
-  static const uptr kLoAppMemEnd   = 0x008000000000ull;
+  static const uptr kLoAppMemEnd = 0x020000000000ull;
   static const uptr kMidAppMemBeg  = 0x550000000000ull;
-  static const uptr kMidAppMemEnd  = 0x568000000000ull;
-  static const uptr kHiAppMemBeg   = 0x7e8000000000ull;
+  static const uptr kMidAppMemEnd = 0x5a0000000000ull;
+  static const uptr kHiAppMemBeg = 0x7c0000000000ull;
   static const uptr kHiAppMemEnd   = 0x800000000000ull;
-  static const uptr kShadowMsk = 0x780000000000ull;
-  static const uptr kShadowXor = 0x040000000000ull;
-  static const uptr kShadowAdd = 0x000000000000ull;
+  static const uptr kShadowMsk = 0x700000000000ull;
+  static const uptr kShadowXor = 0x000000000000ull;
+  static const uptr kShadowAdd = 0x100000000000ull;
   static const uptr kVdsoBeg       = 0xf000000000000000ull;
 };
 

This is more useful because the mid and high regions are subject to
ASLR but the low region is not. This configuration allows up to
30-bits of ASLR (assuming 12-bit pages, and that the binaries/libraries
fit in under 1TB).
Copy link
Collaborator

@dvyukov dvyukov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice!
If tsan_shadow_test.cpp passes, then we should be good.

@thurstond thurstond merged commit 7d039ef into llvm:main Nov 2, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants