Page MenuHomePhabricator
Feed Advanced Search

Wed, May 5

hctim updated the diff for D101865: [scudo] [GWP-ASan] Add GWP-ASan variant of scudo benchmarks..

Forgot to initialize the global for the _loop variant.

Wed, May 5, 3:05 PM · Restricted Project
hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Unfortunately, the small code size increase does hurt performance a little, 2.3%-ish.

Wed, May 5, 2:59 PM · Restricted Project
hctim accepted D101936: [hwasan] Stress test for thread creation..

LGTM w/ nits

Wed, May 5, 1:17 PM · Restricted Project
hctim updated the diff for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Add locks to the global stats, and call the locks around GWP-ASan allocations that aren't part of already mutually exclusive initialization.

Wed, May 5, 11:32 AM · Restricted Project
hctim accepted D101874: [scudo] Align objects with alignas.

Shouldn't we just mark the individual test rather than bumping the whole allocator?

Why? No one complained when you asked for entire compiler-rt.

Wed, May 5, 10:56 AM · Restricted Project
hctim accepted D101881: [hwasan] Fix missing synchronization in AllocThread..
Wed, May 5, 10:29 AM · Restricted Project

Tue, May 4

hctim added a comment to D101874: [scudo] Align objects with alignas.

I'm not sure that we can do this until all of our minimum compiler versions support C++17. E.g. minimum gcc version is 5.1.0 which was released in 2015, so it presumably doesn't have C++17 support.

Tue, May 4, 6:04 PM · Restricted Project
hctim updated the diff for D101865: [scudo] [GWP-ASan] Add GWP-ASan variant of scudo benchmarks..

Forgot to update the once-init nonce. It needs to be moved to a member, rather than a global.

Tue, May 4, 3:28 PM · Restricted Project
hctim requested review of D101865: [scudo] [GWP-ASan] Add GWP-ASan variant of scudo benchmarks..
Tue, May 4, 3:13 PM · Restricted Project

Mon, May 3

hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Update to provide correct statistics for live/free allocations in GWP-ASan. Use the slot size in the calculation rather than actual size, as there's no way to pre-calculate the size of the allocations that can land in the guarded pool.

Mon, May 3, 1:28 PM · Restricted Project
hctim updated the diff for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Update to provide correct statistics for live/free allocations in GWP-ASan. Use the slot size in the calculation rather than actual size, as there's no way to pre-calculate the size of the allocations that can land in the guarded pool.

Mon, May 3, 1:23 PM · Restricted Project
hctim committed rGe8f7241e0b48: [scudo] Don't track free/use stats for transfer batches. (authored by hctim).
[scudo] Don't track free/use stats for transfer batches.
Mon, May 3, 11:50 AM
hctim closed D101653: [scudo] Don't track free/use stats for transfer batches..
Mon, May 3, 11:50 AM · Restricted Project
hctim updated the diff for D101653: [scudo] Don't track free/use stats for transfer batches..

Integrate Kostya's excellent suggestion.

Mon, May 3, 11:15 AM · Restricted Project

Fri, Apr 30

hctim updated the diff for D101653: [scudo] Don't track free/use stats for transfer batches..

Looks much better now:

Fri, Apr 30, 5:54 PM · Restricted Project
hctim updated the diff for D101653: [scudo] Don't track free/use stats for transfer batches..

Instead of introducing a branch to the fast path, just undo the stats changes in the slow path. allocate()/deallocate() are inlined everywhere by the compiler (which is a correct optimisation, no inlining causes a 10% performance penalty). By adding extra instructions, we drastically diminish icache performance by a few percent.

Fri, Apr 30, 5:40 PM · Restricted Project
hctim added a comment to D101653: [scudo] Don't track free/use stats for transfer batches..

I originally thought the +X wasn't a %% difference. I looked into one of the worst-looking benchmarks that showed a 3% increase.

Fri, Apr 30, 4:47 PM · Restricted Project
hctim accepted D101514: [scudo] Use require_constant_initialization.

LGTM! Thank you!

Fri, Apr 30, 1:56 PM · Restricted Project
hctim added a comment to D101653: [scudo] Don't track free/use stats for transfer batches..

Right, it makes sense to not account for the batches. I don't think this should have a meaningful impact, but have you noticed any perf changes adding this since it's in the fastpath?

Fri, Apr 30, 11:20 AM · Restricted Project
hctim requested review of D101653: [scudo] Don't track free/use stats for transfer batches..
Fri, Apr 30, 11:01 AM · Restricted Project

Thu, Apr 29

hctim added inline comments to D101514: [scudo] Use require_constant_initialization.
Thu, Apr 29, 5:07 PM · Restricted Project
hctim added inline comments to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..
Thu, Apr 29, 3:51 PM · Restricted Project
hctim updated the diff for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Moved GWP-ASan allocator back to a member, now depends on D101514 where
Scudo enforces static initialization.

Thu, Apr 29, 3:50 PM · Restricted Project
hctim added inline comments to D101514: [scudo] Use require_constant_initialization.
Thu, Apr 29, 3:40 PM · Restricted Project

Wed, Apr 28

hctim edited reviewers for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks., added: vitalybuka; removed: pcc.
Wed, Apr 28, 7:32 PM · Restricted Project
hctim updated the diff for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Make GWP-ASan's allocator a static data member and make the state have a
constexpr constructor. This prevents dynamic initializers. This may need
some attention if we end up having multiple initialized Scudo instances
in a program, but I don't know of anywhere where this is the case right
now.

Wed, Apr 28, 7:31 PM · Restricted Project
hctim added a comment to D101473: [compiler-rt] Move to cxx17..

I'd slightly prefer setting CXX_STANDARD 17 on individual targets (you might also need CXX_STANDARD_REQUIRED YES if you want to rely on C++17 features) which is a more idiomatic CMake, but either way it's fine with us.

Wed, Apr 28, 4:02 PM · Restricted Project
hctim added a comment to D101473: [compiler-rt] Move to cxx17..

@srhines, @mcgrathr, @phosek, @echristo - please raise your hand if you foresee any problems with bumping compiler-rt to c++17. The rest of LLVM seems to use it just fine.

Wed, Apr 28, 11:11 AM · Restricted Project
hctim added reviewers for D101473: [compiler-rt] Move to cxx17.: srhines, mcgrathr, phosek, echristo.
Wed, Apr 28, 11:10 AM · Restricted Project
hctim requested review of D101473: [compiler-rt] Move to cxx17..
Wed, Apr 28, 10:56 AM · Restricted Project

Tue, Apr 27

hctim accepted D101407: [gwp_asan] Use __sanitizer_fast_backtrace on Fuchsia.

LGTM with Kostya's nit. Glad to see that the modularity works for Fuchsia out of the box :).

Tue, Apr 27, 4:13 PM · Restricted Project

Fri, Apr 23

hctim reopened D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..
Fri, Apr 23, 5:18 PM · Restricted Project
hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Curious, I can't find dynamic initializers (or even a constructor) for gwp_asan::GuardedPoolAllocator and scudo::Allocator<> in clang's binary, but they exist in the gcc binary. Maybe gcc really wants to give them a dynamic initializer?

Fri, Apr 23, 5:13 PM · Restricted Project
hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Where GPAForSignalHandler gets zero-initialized:

Fri, Apr 23, 4:48 PM · Restricted Project
hctim added a reverting change for rGa683abe5c026: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks.: rG643ccf6e4b85: Revert "[Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks.".
Fri, Apr 23, 3:41 PM
hctim committed rG643ccf6e4b85: Revert "[Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks." (authored by hctim).
Revert "[Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks."
Fri, Apr 23, 3:41 PM
hctim added a reverting change for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks.: rG643ccf6e4b85: Revert "[Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks.".
Fri, Apr 23, 3:41 PM · Restricted Project
hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Broke the gcc bots. Crashes in gtest shutdown. Looks like a GWP-ASan allocation is getting freed by scudo.

Fri, Apr 23, 3:37 PM · Restricted Project
hctim added a comment to D100783: [HWASan] Use memory addresses in tag dump, not shadow..

why is this necessary? there should be a memory address in the report header I think

Fri, Apr 23, 1:02 PM · Restricted Project
hctim committed rGf1a47181f5c6: [hwasan] Remove untagging of kernel-consumed memory (authored by hctim).
[hwasan] Remove untagging of kernel-consumed memory
Fri, Apr 23, 11:05 AM
hctim closed D101121: [hwasan] Remove untagging of kernel-consumed memory.
Fri, Apr 23, 11:05 AM · Restricted Project
hctim added a reverting change for rG90118563ad0f: [X86][AMX] Try to hoist AMX shapes' def: rGcaea37b37e6a: Revert "[X86][AMX] Try to hoist AMX shapes' def".
Fri, Apr 23, 10:43 AM
hctim committed rGcaea37b37e6a: Revert "[X86][AMX] Try to hoist AMX shapes' def" (authored by hctim).
Revert "[X86][AMX] Try to hoist AMX shapes' def"
Fri, Apr 23, 10:43 AM
hctim added a reverting change for D101067: [X86][AMX] Try to hoist AMX shapes' def: rGcaea37b37e6a: Revert "[X86][AMX] Try to hoist AMX shapes' def".
Fri, Apr 23, 10:43 AM · Restricted Project
hctim added a comment to D101067: [X86][AMX] Try to hoist AMX shapes' def.

Looks like this patch broke the MSan buildbots: https://lab.llvm.org/buildbot/#/builders/5/builds/6967/steps/9/logs/stdio

Fri, Apr 23, 10:42 AM · Restricted Project
hctim committed rGa683abe5c026: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks. (authored by hctim).
[Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks.
Fri, Apr 23, 10:08 AM
hctim closed D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..
Fri, Apr 23, 10:08 AM · Restricted Project
hctim updated the diff for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

Rebase.

Fri, Apr 23, 10:03 AM · Restricted Project

Thu, Apr 22

hctim updated the diff for D100783: [HWASan] Use memory addresses in tag dump, not shadow..
  • Remove dependency.
Thu, Apr 22, 5:52 PM · Restricted Project
hctim updated the diff for D100783: [HWASan] Use memory addresses in tag dump, not shadow..
  • Remove dependency.
Thu, Apr 22, 5:50 PM · Restricted Project
hctim abandoned D101120: [hwasan] Add __hwasan_untag_pointer.
Thu, Apr 22, 5:49 PM · Restricted Project
hctim updated the diff for D100783: [HWASan] Use memory addresses in tag dump, not shadow..

Add test. Depends on D10112.

Thu, Apr 22, 5:47 PM · Restricted Project
hctim retitled D101121: [hwasan] Remove untagging of kernel-consumed memory from Summary: Now that page aliasing for x64 has landed, we don't need to worry about passing tagged pointers to libc, and thus D98875 removed it. Unfortunately, we still test on aarch64 devices that don't have the kernel tagged address ABI (https... to [hwasan] Remove untagging of kernel-consumed memory.
Thu, Apr 22, 5:38 PM · Restricted Project
hctim requested review of D101121: [hwasan] Remove untagging of kernel-consumed memory.
Thu, Apr 22, 5:37 PM · Restricted Project
hctim requested review of D101120: [hwasan] Add __hwasan_untag_pointer.
Thu, Apr 22, 5:30 PM · Restricted Project

Mon, Apr 19

hctim requested review of D100783: [HWASan] Use memory addresses in tag dump, not shadow..
Mon, Apr 19, 12:13 PM · Restricted Project

Mon, Apr 12

hctim added inline comments to D100220: [ASan] Allow new/delete replacement by making interceptors weak.
Mon, Apr 12, 10:06 AM · Restricted Project
hctim committed rG15689f3af018: [asan] Replaceable new/delete is unsupported in Windows. (authored by hctim).
[asan] Replaceable new/delete is unsupported in Windows.
Mon, Apr 12, 10:04 AM

Fri, Apr 9

hctim added a reverting change for rG5a0117b2d0ea: [AMDGPU] Remove MachineDCE after SIFoldOperands: rG092f288d3644: Revert "[AMDGPU] Remove MachineDCE after SIFoldOperands".
Fri, Apr 9, 3:48 PM
hctim committed rG092f288d3644: Revert "[AMDGPU] Remove MachineDCE after SIFoldOperands" (authored by hctim).
Revert "[AMDGPU] Remove MachineDCE after SIFoldOperands"
Fri, Apr 9, 3:48 PM
hctim added a reverting change for rGd19a42eba98f: [AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs: rG3d4730a73fd1: Revert "[AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs".
Fri, Apr 9, 3:48 PM
hctim added a reverting change for D100189: [AMDGPU] Remove MachineDCE after SIFoldOperands: rG092f288d3644: Revert "[AMDGPU] Remove MachineDCE after SIFoldOperands".
Fri, Apr 9, 3:48 PM · Restricted Project
hctim added a reverting change for D100188: [AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs: rG3d4730a73fd1: Revert "[AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs".
Fri, Apr 9, 3:48 PM · Restricted Project
hctim committed rG3d4730a73fd1: Revert "[AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs" (authored by hctim).
Revert "[AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs"
Fri, Apr 9, 3:48 PM
hctim accepted D100230: [docs][scudo] Update Scudo documentation.

It would be nice to include some details about MTE support (like the header size is 16 bytes rather than 8 bytes, etc). But we can make those changes afterwards.

Fri, Apr 9, 3:26 PM · Restricted Project
hctim added a comment to D100188: [AMDGPU] SIFoldOperands: eagerly erase dead REG_SEQUENCEs.

Looks like this patch might have broken the ASan buildbots (https://lab.llvm.org/buildbot/#/builders/99/builds/2766):

Fri, Apr 9, 3:05 PM · Restricted Project
hctim committed rG7df30e77d499: [ASan] Allow new/delete replacement by making interceptors weak (authored by hctim).
[ASan] Allow new/delete replacement by making interceptors weak
Fri, Apr 9, 2:20 PM
hctim closed D100220: [ASan] Allow new/delete replacement by making interceptors weak.
Fri, Apr 9, 2:20 PM · Restricted Project
hctim added a reverting change for rG16fe741c6942: [PowerPC] Add ROP Protection Instructions for PowerPC: rG1a2756b777b6: Revert "[PowerPC] Add ROP Protection Instructions for PowerPC".
Fri, Apr 9, 1:37 PM
hctim committed rG1a2756b777b6: Revert "[PowerPC] Add ROP Protection Instructions for PowerPC" (authored by hctim).
Revert "[PowerPC] Add ROP Protection Instructions for PowerPC"
Fri, Apr 9, 1:37 PM
hctim added a reverting change for D99375: [PowerPC] Add ROP Protection Instructions for PowerPC: rG1a2756b777b6: Revert "[PowerPC] Add ROP Protection Instructions for PowerPC".
Fri, Apr 9, 1:37 PM · Restricted Project
hctim updated the diff for D100220: [ASan] Allow new/delete replacement by making interceptors weak.

wrong diffbase messed up phabricator rendering

Fri, Apr 9, 1:02 PM · Restricted Project
hctim updated the diff for D100220: [ASan] Allow new/delete replacement by making interceptors weak.

Add it to CXX_OPERATOR_ATTRIBUTE, on non-mac non-windows?

Fri, Apr 9, 1:01 PM · Restricted Project
hctim added a comment to D99375: [PowerPC] Add ROP Protection Instructions for PowerPC.

Looks like this patch broke the UBSan buildbot (https://lab.llvm.org/buildbot/#/builders/5/builds/6546), PTAL.

Fri, Apr 9, 12:13 PM · Restricted Project
hctim requested review of D100220: [ASan] Allow new/delete replacement by making interceptors weak.
Fri, Apr 9, 11:37 AM · Restricted Project

Apr 7 2021

hctim committed rZORG43077c5c525b: Unconditionally pre-define CMAKE_ARGS in sanitizer bot. (authored by hctim).
Unconditionally pre-define CMAKE_ARGS in sanitizer bot.
Apr 7 2021, 12:27 PM

Mar 16 2021

hctim accepted D98732: scudo: Allow TBI to be disabled on Linux with a macro..
Mar 16 2021, 12:49 PM · Restricted Project

Mar 11 2021

hctim accepted D98456: [scudo][standalone] Add shared library to makefile.
Mar 11 2021, 4:21 PM · Restricted Project
hctim added a comment to D98456: [scudo][standalone] Add shared library to makefile.

LGTM w/ nit

Mar 11 2021, 3:19 PM · Restricted Project

Mar 9 2021

hctim added a comment to D96120: [scudo] Port scudo sanitizer to Windows.
In D96120#2597877, @kcc wrote:

I am afraid we will have to delete the older (non-standalone) variant entirely.
(And the sooner the better)

Okay, think we found a few things out by doing this before porting standalone.

If I find time I’ll have a go at removing the sanitizer version.

The scudo documentation (https://llvm.org/docs/ScudoHardenedAllocator.html) still seems to refer to the sanitizer version so will need updating.

Mar 9 2021, 2:46 PM · Restricted Project, Restricted Project

Mar 5 2021

hctim committed rGe58d68fcd06d: Revert "[AMDGPU] Restore the s_memtime instruction in gfx1030" (authored by hctim).
Revert "[AMDGPU] Restore the s_memtime instruction in gfx1030"
Mar 5 2021, 6:25 PM
hctim added a reverting change for rGfc28f600e558: [AMDGPU] Restore the s_memtime instruction in gfx1030: rGe58d68fcd06d: Revert "[AMDGPU] Restore the s_memtime instruction in gfx1030".
Mar 5 2021, 6:25 PM
hctim added a reverting change for D97928: [AMDGPU] Restore the s_memtime instruction in gfx1030: rGe58d68fcd06d: Revert "[AMDGPU] Restore the s_memtime instruction in gfx1030".
Mar 5 2021, 6:25 PM · Restricted Project, Restricted Project
hctim added a comment to D97928: [AMDGPU] Restore the s_memtime instruction in gfx1030.

Hello, it looks like this change broke the MSan and ASan buildbots.

Mar 5 2021, 4:48 PM · Restricted Project, Restricted Project

Mar 4 2021

hctim committed rG1be97975ccc6: Change instrprof LLVM_VP_MAX_NUM_VALS_PER_SITE threshold. (authored by hctim).
Change instrprof LLVM_VP_MAX_NUM_VALS_PER_SITE threshold.
Mar 4 2021, 12:14 PM

Feb 23 2021

hctim added a comment to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..

@pcc - friendly ping for a final pass when you have some spare time. Could you take a look at the comment in lit.site.cfg.py.in and see if you have any suggestions?

Feb 23 2021, 1:33 PM · Restricted Project
hctim accepted D96435: [scudo][standalone] Compact pointers for Caches/Batches.

LGTM

Feb 23 2021, 11:05 AM · Restricted Project
hctim accepted D96435: [scudo][standalone] Compact pointers for Caches/Batches.

LGTM w/ one nit.

Feb 23 2021, 9:44 AM · Restricted Project

Feb 22 2021

hctim added inline comments to D96435: [scudo][standalone] Compact pointers for Caches/Batches.
Feb 22 2021, 1:27 PM · Restricted Project

Feb 10 2021

hctim committed rGb93786907c54: [GWP-ASan] Add back some headers removed by IWYU. (authored by hctim).
[GWP-ASan] Add back some headers removed by IWYU.
Feb 10 2021, 9:14 AM
hctim closed D96374: [GWP-ASan] Add back some headers removed by IWYU..
Feb 10 2021, 9:14 AM · Restricted Project

Feb 9 2021

hctim updated the diff for D96374: [GWP-ASan] Add back some headers removed by IWYU..

Inline the definitions required only on Android.

Feb 9 2021, 2:57 PM · Restricted Project
hctim requested review of D96374: [GWP-ASan] Add back some headers removed by IWYU..
Feb 9 2021, 2:55 PM · Restricted Project

Feb 8 2021

hctim added inline comments to D93731: scudo: Support memory tagging in the secondary allocator..
Feb 8 2021, 3:50 PM · Restricted Project
hctim added inline comments to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..
Feb 8 2021, 12:37 PM · Restricted Project
hctim added inline comments to D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks..
Feb 8 2021, 11:25 AM · Restricted Project
hctim edited reviewers for D95884: [Scudo] Use GWP-ASan's aligned allocations and fixup postalloc hooks., added: pcc; removed: eugenis.
Feb 8 2021, 11:24 AM · Restricted Project
hctim committed rG3d8823b8e48a: [GWP-ASan] Add aligned allocations. (authored by hctim).
[GWP-ASan] Add aligned allocations.
Feb 8 2021, 11:23 AM
hctim closed D94830: [GWP-ASan] Add aligned allocations..
Feb 8 2021, 11:23 AM · Restricted Project
hctim updated the diff for D94830: [GWP-ASan] Add aligned allocations..

Address Evgenii's comments.

Feb 8 2021, 10:17 AM · Restricted Project