This is an archive of the discontinued LLVM Phabricator instance.

Differential D120102

[HWASAN] Add test for optnone function.
ClosedPublic

Authored by fmayer on Feb 17 2022, 4:52 PM.

Details

Reviewers
eugenis
Commits
rG864820bb3cf5: [HWASAN] Add test for optnone function.

Diff Detail

Event Timeline

fmayer requested review of this revision.Feb 17 2022, 4:52 PM
fmayer created this revision.
Herald added a project: Restricted Project. · View Herald TranscriptFeb 17 2022, 4:52 PM
Herald added a subscriber: llvm-commits. · View Herald Transcript
fmayer added a reviewer: eugenis.Feb 17 2022, 4:53 PM
Harbormaster completed remote builds in B150355: Diff 409821.Feb 17 2022, 5:33 PM
dblaikie added a subscriber: dblaikie.Feb 20 2022, 10:35 PM
eugenis added inline comments.Feb 23 2022, 1:19 PM
llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
113

Should this say [[TMP11]]? It is otherwise unused.

Overall I feel this test is too detailed and potentially brittle. If you want to check that tagging/untagging happens within lifetime bounds, that can be done with just a few lines.

Also, I could be misreading, but does this show __hwasan_tag_memory after lifetime.end?

fmayer added inline comments.Feb 23 2022, 2:00 PM
llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
113

This is copy-pasted from the test just above, just with added _optnull.

fmayer updated this revision to Diff 410948.Feb 23 2022, 2:34 PM

rebase

fmayer added a parent revision: D120437: [HWASAN] erase lifetime intrinsics if tag is outside..Feb 23 2022, 2:34 PM
Harbormaster completed remote builds in B151146: Diff 410948.Feb 23 2022, 3:50 PM
fmayer updated this revision to Diff 411452.Feb 25 2022, 10:25 AM

rebase

Harbormaster completed remote builds in B151503: Diff 411452.Feb 25 2022, 11:36 AM
fmayer added inline comments.Mar 1 2022, 10:46 AM
llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll
113

Rebased this to a CL that fixes this problem. PTAL.

eugenis accepted this revision.Mar 1 2022, 11:19 AM

LGTM

This revision is now accepted and ready to land.Mar 1 2022, 11:19 AM
fmayer updated this revision to Diff 412185.Mar 1 2022, 11:31 AM

rebase

Harbormaster completed remote builds in B152015: Diff 412185.Mar 1 2022, 12:31 PM
Herald added a project: Restricted Project. · View Herald TranscriptMar 1 2022, 12:31 PM
This revision was landed with ongoing or failed builds.Mar 1 2022, 2:49 PM
Closed by commit rG864820bb3cf5: [HWASAN] Add test for optnone function. (authored by fmayer). · Explain Why
This revision was automatically updated to reflect the committed changes.
fmayer added a commit: rG864820bb3cf5: [HWASAN] Add test for optnone function..

Revision Contents

PathSize
llvm/
test/
Instrumentation/
HWAddressSanitizer/
61 lines

Diff 412263

llvm/test/Instrumentation/HWAddressSanitizer/use-after-scope.ll

Show First 20 Lines • Show All 62 Lines • ▼ Show 20 Lines; We should tag the memory before the next br (before the jump back).
call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull %1) call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull %1)
br i1 %3, label %4, label %2 br i1 %3, label %4, label %2
4: ; preds = %2 4: ; preds = %2
call void @use(i8* nonnull %1) #2 call void @use(i8* nonnull %1) #2
ret i32 0 ret i32 0
} }
define dso_local i32 @standard_lifetime_optnone() local_unnamed_addr optnone noinline sanitize_hwaddress {
; SCOPE-LABEL: @standard_lifetime_optnone(
; SCOPE-NEXT: [[DOTHWASAN_SHADOW:%.*]] = call i8* asm "", "=r,0"(i8* null)
; SCOPE-NEXT: [[TMP1:%.*]] = alloca { i8, [15 x i8] }, align 16
; SCOPE-NEXT: [[TMP2:%.*]] = bitcast { i8, [15 x i8] }* [[TMP1]] to i8*
; SCOPE-NEXT: [[TMP3:%.*]] = call i8 @__hwasan_generate_tag()
; SCOPE-NEXT: [[TMP4:%.*]] = zext i8 [[TMP3]] to i64
; SCOPE-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64
; SCOPE-NEXT: [[TMP6:%.*]] = shl i64 [[TMP4]], 57
; SCOPE-NEXT: [[TMP7:%.*]] = or i64 [[TMP5]], [[TMP6]]
; SCOPE-NEXT: [[ALLOCA_0_HWASAN:%.*]] = inttoptr i64 [[TMP7]] to i8*
; SCOPE-NEXT: br label [[TMP8:%.*]]
; SCOPE: 8:
; SCOPE-NEXT: call void @llvm.lifetime.start.p0i8(i64 1, i8* nonnull [[ALLOCA_0_HWASAN]])
; SCOPE-NEXT: [[TMP9:%.*]] = trunc i64 [[TMP4]] to i8
; SCOPE-NEXT: call void @__hwasan_tag_memory(i8* [[TMP2]], i8 [[TMP9]], i64 16)
; SCOPE-NEXT: [[TMP10:%.*]] = tail call i1 (...) @cond()
; SCOPE-NEXT: call void @__hwasan_tag_memory(i8* [[TMP2]], i8 0, i64 16)
; SCOPE-NEXT: call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull [[ALLOCA_0_HWASAN]])
; SCOPE-NEXT: br i1 [[TMP10]], label [[TMP11:%.*]], label [[TMP8]]
; SCOPE: 11:
; SCOPE-NEXT: call void @use(i8* nonnull [[ALLOCA_0_HWASAN]])
; SCOPE-NEXT: ret i32 0
;
; NOSCOPE-LABEL: @standard_lifetime_optnone(
; NOSCOPE-NEXT: [[DOTHWASAN_SHADOW:%.*]] = call i8* asm "", "=r,0"(i8* null)
; NOSCOPE-NEXT: [[TMP1:%.*]] = alloca { i8, [15 x i8] }, align 16
; NOSCOPE-NEXT: [[TMP2:%.*]] = bitcast { i8, [15 x i8] }* [[TMP1]] to i8*
; NOSCOPE-NEXT: [[TMP3:%.*]] = call i8 @__hwasan_generate_tag()
; NOSCOPE-NEXT: [[TMP4:%.*]] = zext i8 [[TMP3]] to i64
; NOSCOPE-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64
; NOSCOPE-NEXT: [[TMP6:%.*]] = shl i64 [[TMP4]], 57
; NOSCOPE-NEXT: [[TMP7:%.*]] = or i64 [[TMP5]], [[TMP6]]
; NOSCOPE-NEXT: [[ALLOCA_0_HWASAN:%.*]] = inttoptr i64 [[TMP7]] to i8*
; NOSCOPE-NEXT: [[TMP8:%.*]] = trunc i64 [[TMP4]] to i8
; NOSCOPE-NEXT: call void @__hwasan_tag_memory(i8* [[TMP2]], i8 [[TMP8]], i64 16)
; NOSCOPE-NEXT: br label [[TMP9:%.*]]
; NOSCOPE: 9:
; NOSCOPE-NEXT: [[TMP10:%.*]] = tail call i1 (...) @cond()
; NOSCOPE-NEXT: br i1 [[TMP10]], label [[TMP11:%.*]], label [[TMP9]]
; NOSCOPE: 11:
; NOSCOPE-NEXT: call void @use(i8* nonnull [[ALLOCA_0_HWASAN]])
; NOSCOPE-NEXT: call void @__hwasan_tag_memory(i8* [[TMP2]], i8 0, i64 16)
eugenisUnsubmitted
Not Done
ReplyInline Actions

Should this say [[TMP11]]? It is otherwise unused.

Overall I feel this test is too detailed and potentially brittle. If you want to check that tagging/untagging happens within lifetime bounds, that can be done with just a few lines.

Also, I could be misreading, but does this show __hwasan_tag_memory after lifetime.end?

eugenis: Should this say `[[TMP11]]`? It is otherwise unused. Overall I feel this test is too detailed…
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

This is copy-pasted from the test just above, just with added _optnull.

fmayer: This is copy-pasted from the test just above, just with added _optnull.
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

Rebased this to a CL that fixes this problem. PTAL.

fmayer: Rebased this to a CL that fixes this problem. PTAL.
; NOSCOPE-NEXT: ret i32 0
;
%1 = alloca i8, align 1
br label %2
2: ; preds = %2, %0
; We should tag the memory after the br (in the loop).
call void @llvm.lifetime.start.p0i8(i64 1, i8* nonnull %1)
%3 = tail call i1 (...) @cond() #2
; We should tag the memory before the next br (before the jump back).
call void @llvm.lifetime.end.p0i8(i64 1, i8* nonnull %1)
br i1 %3, label %4, label %2
4: ; preds = %2
call void @use(i8* nonnull %1) #2
ret i32 0
}
define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress { define dso_local i32 @multiple_lifetimes() local_unnamed_addr sanitize_hwaddress {
; SCOPE-LABEL: @multiple_lifetimes( ; SCOPE-LABEL: @multiple_lifetimes(
; SCOPE-NEXT: [[DOTHWASAN_SHADOW:%.*]] = call i8* asm "", "=r,0"(i8* null) ; SCOPE-NEXT: [[DOTHWASAN_SHADOW:%.*]] = call i8* asm "", "=r,0"(i8* null)
; SCOPE-NEXT: [[TMP1:%.*]] = alloca { i8, [15 x i8] }, align 16 ; SCOPE-NEXT: [[TMP1:%.*]] = alloca { i8, [15 x i8] }, align 16
; SCOPE-NEXT: [[TMP2:%.*]] = bitcast { i8, [15 x i8] }* [[TMP1]] to i8* ; SCOPE-NEXT: [[TMP2:%.*]] = bitcast { i8, [15 x i8] }* [[TMP1]] to i8*
; SCOPE-NEXT: [[TMP3:%.*]] = call i8 @__hwasan_generate_tag() ; SCOPE-NEXT: [[TMP3:%.*]] = call i8 @__hwasan_generate_tag()
; SCOPE-NEXT: [[TMP4:%.*]] = zext i8 [[TMP3]] to i64 ; SCOPE-NEXT: [[TMP4:%.*]] = zext i8 [[TMP3]] to i64
; SCOPE-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64 ; SCOPE-NEXT: [[TMP5:%.*]] = ptrtoint i8* [[TMP2]] to i64
▲ Show 20 Lines • Show All 108 Lines • Show Last 20 Lines