[DWARF] Add option to add linkage_names to call_origin declaration refs #89640
Conversation
If -mllvm -add-linkage-names-to-external-call-origins is true then add
DW_AT_linkage_name attributes to DW_TAG_subprogram DIEs referenced by
DW_AT_call_origin attributes that would otherwise be omitted.
A debugger may use DW_TAG_call_origin attributes to determine whether any
frames in a callstack are missing due to optimisations (e.g. tail calls).
For example, say a() calls b() tail-calls c(), and you stop in your
debugger in c():
The callstack looks like this:
c()
a()
Looking "up" from c(), call site information can be found in a(). This
includes a DW_AT_call_origin referencing b()'s subprogram DIE, which means
the call at this call site was to b(), not c() where we are currently stopped.
This indicates b()'s frame has been lost due to optimisation (tail call or
ICF).
This patch makes it easier for a debugger to check whether the referenced DIE
describes the target function or not, for example by comparing the referenced
function name to the current frame.
|
@llvm/pr-subscribers-debuginfo Author: Orlando Cazalet-Hyams (OCHyams) ChangesIf A debugger may use For example, say The callstack looks like this: Looking "up" from This patch makes it easier for a debugger to check whether the referenced There's already an option to apply DW_AT_linkage_name in a targeted manner: The new flag shouldn't change anything upstream whether it is enabled or not because the non-SCE-tuned behaviour upstream is to always add linkage names to subprogram DIEs, AFAICT. Enabling this feature alongside Full diff: https://github.com/llvm/llvm-project/pull/89640.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp b/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
index 14f2a363f9be61..49954bf10abd5b 100644
--- a/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
+++ b/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
@@ -35,6 +35,7 @@
#include "llvm/Target/TargetLoweringObjectFile.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
+#include "llvm/Support/CommandLine.h"
#include <iterator>
#include <optional>
#include <string>
@@ -42,6 +43,20 @@
using namespace llvm;
+/// Query value using AddLinkageNamesToDeclCallOriginsForTuning.
+cl::opt<cl::boolOrDefault> AddLinkageNamesToDeclCallOrigins(
+ "add-linkage-names-to-declaration-call-origins", cl::Hidden,
+ cl::desc("Add DW_AT_linkage_name to function declaration DIEs "
+ "referenced by DW_AT_call_origin attributes. Enabled by default "
+ "for -gsce debugger tuning."));
+
+static bool AddLinkageNamesToDeclCallOriginsForTuning(const DwarfDebug *DD) {
+ bool EnabledByDefault = DD->tuneForSCE();
+ if (EnabledByDefault)
+ return AddLinkageNamesToDeclCallOrigins != cl::boolOrDefault::BOU_FALSE;
+ return AddLinkageNamesToDeclCallOrigins == cl::boolOrDefault::BOU_TRUE;
+}
+
static dwarf::Tag GetCompileUnitType(UnitKind Kind, DwarfDebug *DW) {
// According to DWARF Debugging Information Format Version 5,
@@ -1260,6 +1275,12 @@ DIE &DwarfCompileUnit::constructCallSiteEntryDIE(DIE &ScopeDIE,
} else {
DIE *CalleeDIE = getOrCreateSubprogramDIE(CalleeSP);
assert(CalleeDIE && "Could not create DIE for call site entry origin");
+ if (AddLinkageNamesToDeclCallOriginsForTuning(DD) &&
+ !CalleeSP->isDefinition() &&
+ !CalleeDIE->findAttribute(dwarf::DW_AT_linkage_name)) {
+ addLinkageName(*CalleeDIE, CalleeSP->getLinkageName());
+ }
+
addDIEEntry(CallSiteDIE, getDwarf5OrGNUAttr(dwarf::DW_AT_call_origin),
*CalleeDIE);
}
diff --git a/llvm/test/DebugInfo/X86/call-origin-linkage-names.ll b/llvm/test/DebugInfo/X86/call-origin-linkage-names.ll
new file mode 100644
index 00000000000000..c8ca3075c0ff7a
--- /dev/null
+++ b/llvm/test/DebugInfo/X86/call-origin-linkage-names.ll
@@ -0,0 +1,96 @@
+; RUN: llc %s --filetype=obj -o - -dwarf-linkage-names=Abstract -add-linkage-names-to-declaration-call-origins=false \
+; RUN: | llvm-dwarfdump - | FileCheck %s --check-prefixes=COMMON,DISABLE --implicit-check-not=DW_AT_linkage_name
+; RUN: llc %s --filetype=obj -o - -dwarf-linkage-names=Abstract -add-linkage-names-to-declaration-call-origins=true \
+; RUN: | llvm-dwarfdump - | FileCheck %s --check-prefixes=COMMON,ENABLE --implicit-check-not=DW_AT_linkage_name
+
+;; Check that -add-linkage-names-to-declaration-call-origins controls whether
+;; linkage names are added to declarations referenced by DW_AT_call_origin
+;; attributes.
+;;
+;; $ cat test.cpp
+;; void a();
+;; __attribute__((optnone))
+;; void b() {}
+;; void c();
+;; extern "C" {
+;; void d();
+;; }
+;;
+;; void e() {
+;; a(); //< Reference declaration DIE (add linkage name).
+;; b(); //< Reference definition DIE (don't add linkage name).
+;; c(); //< Reference definition DIE (don't add linkage name).
+;; d(); //< Reference declaration DIE, but there's no linkage name.
+;; }
+;;
+;; __attribute__((optnone))
+;; void c() {}
+;; $ clang++ -emit-llvm -S -O1 -g
+
+; COMMON: DW_TAG_call_site
+; ENABLE-NEXT: DW_AT_call_origin (0x[[a:[a-z0-9]+]] "_Z1av")
+; DISABLE-NEXT: DW_AT_call_origin (0x[[a:[a-z0-9]+]] "a")
+; COMMON: DW_TAG_call_site
+; COMMON-NEXT: DW_AT_call_origin (0x[[b:[a-z0-9]+]] "b")
+; COMMON: DW_TAG_call_site
+; COMMON-NEXT: DW_AT_call_origin (0x[[c:[a-z0-9]+]] "c")
+; COMMON: DW_TAG_call_site
+; COMMON-NEXT: DW_AT_call_origin (0x[[d:[a-z0-9]+]] "d")
+
+; COMMON: 0x[[a]]: DW_TAG_subprogram
+; COMMON: DW_AT_name ("a")
+; ENABLE: DW_AT_linkage_name ("_Z1av")
+; COMMON: 0x[[b]]: DW_TAG_subprogram
+; COMMON: DW_AT_name ("b")
+; COMMON: 0x[[c]]: DW_TAG_subprogram
+; COMMON: DW_AT_name ("c")
+; COMMON: 0x[[d]]: DW_TAG_subprogram
+; COMMON: DW_AT_name ("d")
+
+target triple = "x86_64-unknown-linux-gnu"
+
+define dso_local void @_Z1ev() local_unnamed_addr !dbg !13 {
+entry:
+ tail call void @_Z1av(), !dbg !14
+ tail call void @_Z1bv(), !dbg !15
+ tail call void @_Z1cv(), !dbg !16
+ tail call void @d(), !dbg !17
+ ret void, !dbg !18
+}
+
+define dso_local void @_Z1bv() local_unnamed_addr !dbg !9 {
+entry:
+ ret void, !dbg !12
+}
+
+declare !dbg !19 void @_Z1av() local_unnamed_addr
+
+define dso_local void @_Z1cv() local_unnamed_addr !dbg !20 {
+entry:
+ ret void, !dbg !21
+}
+declare !dbg !22 void @d() local_unnamed_addr
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+!llvm.ident = !{!8}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 19.0.0git", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "test.cpp", directory: "/")
+!2 = !{i32 7, !"Dwarf Version", i32 5}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!8 = !{!"clang version 19.0.0"}
+!9 = distinct !DISubprogram(name: "b", linkageName: "_Z1bv", scope: !1, file: !1, line: 3, type: !10, scopeLine: 3, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!10 = !DISubroutineType(types: !11)
+!11 = !{null}
+!12 = !DILocation(line: 3, column: 11, scope: !9)
+!13 = distinct !DISubprogram(name: "e", linkageName: "_Z1ev", scope: !1, file: !1, line: 9, type: !10, scopeLine: 9, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!14 = !DILocation(line: 10, column: 3, scope: !13)
+!15 = !DILocation(line: 11, column: 3, scope: !13)
+!16 = !DILocation(line: 12, column: 3, scope: !13)
+!17 = !DILocation(line: 13, column: 3, scope: !13)
+!18 = !DILocation(line: 14, column: 1, scope: !13)
+!19 = !DISubprogram(name: "a", linkageName: "_Z1av", scope: !1, file: !1, line: 1, type: !10, flags: DIFlagPrototyped, spFlags: DISPFlagOptimized)
+!20 = distinct !DISubprogram(name: "c", linkageName: "_Z1cv", scope: !1, file: !1, line: 17, type: !10, scopeLine: 17, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0)
+!21 = !DILocation(line: 17, column: 11, scope: !20)
+!22 = !DISubprogram(name: "d", scope: !1, file: !1, line: 6, type: !10, flags: DIFlagPrototyped, spFlags: DISPFlagOptimized)
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
| @@ -35,13 +35,28 @@ | |||
| #include "llvm/Target/TargetLoweringObjectFile.h" | |||
| #include "llvm/Target/TargetMachine.h" | |||
| #include "llvm/Target/TargetOptions.h" | |||
| #include "llvm/Support/CommandLine.h" | |||
There was a problem hiding this comment.
clang-format reminds me this is not in correct order.
| ; RUN: | llvm-dwarfdump - | FileCheck %s --check-prefixes=COMMON,DISABLE --implicit-check-not=DW_AT_linkage_name | ||
| ; RUN: llc %s --filetype=obj -o - -dwarf-linkage-names=Abstract -add-linkage-names-to-declaration-call-origins=true \ | ||
| ; RUN: | llvm-dwarfdump - | FileCheck %s --check-prefixes=COMMON,ENABLE --implicit-check-not=DW_AT_linkage_name | ||
|
|
There was a problem hiding this comment.
Do we want more RUN lines showing default varies with tuning? I'm inclined to think it's not super necessary but curious what other reviewers think.
| /// Query value using AddLinkageNamesToDeclCallOriginsForTuning. | ||
| cl::opt<cl::boolOrDefault> AddLinkageNamesToDeclCallOrigins( | ||
| "add-linkage-names-to-declaration-call-origins", cl::Hidden, | ||
| cl::desc("Add DW_AT_linkage_name to function declaration DIEs " | ||
| "referenced by DW_AT_call_origin attributes. Enabled by default " | ||
| "for -gsce debugger tuning.")); | ||
|
|
There was a problem hiding this comment.
I think we /generally/ try not to use/add raw cl::opts, and instead pass things through the driver/frontend flag handling? (I might be wrong, if there's substantial precedent, let me know)
& not sure how this sits with @pogo59's general premise that debugger tuning shouldn't be reachable only through the tuning flags, but should be independently accessible (does independently accessible mean supported in the driver flags, or is -Xclang/cc1/mllvm flags sufficient to meet that requirement?)
There was a problem hiding this comment.
FWIW -dwarf-linkage-names is an -mllvm flag too (happy to move the flag around as needed though - I don't have a strong opinion on it)
By "upstream" you mean "anyone not using And any chance you've measured the impact of not always adding linkage names to subprogram DIEs for SCE lately? Is that optimization still pulling its weight for you, or would it be possible to converge on the non-sce behavior and remove some special cases here, rather than adding a new one? |
There was a problem hiding this comment.
Thanks both for the reviews.
By "upstream" you mean "anyone not using -gsce"?
Ah, yes indeed. I'll edit the summary so that it makes more sense.
And any chance you've measured the impact of not always adding linkage names to subprogram DIEs for SCE lately? Is that optimization still pulling its weight for you, or would it be possible to converge on the non-sce behavior and remove some special cases here, rather than adding a new one?
I have - across some internal codebases and configurations I measured an increase of total linked binary size including debug info of between ~4% (-O0 builds) to as much as ~19% (thinLTO -O3 builds) when using -dwarf-linkage-names=All instead of =Abstract.
| /// Query value using AddLinkageNamesToDeclCallOriginsForTuning. | ||
| cl::opt<cl::boolOrDefault> AddLinkageNamesToDeclCallOrigins( | ||
| "add-linkage-names-to-declaration-call-origins", cl::Hidden, | ||
| cl::desc("Add DW_AT_linkage_name to function declaration DIEs " | ||
| "referenced by DW_AT_call_origin attributes. Enabled by default " | ||
| "for -gsce debugger tuning.")); | ||
|
|
There was a problem hiding this comment.
FWIW -dwarf-linkage-names is an -mllvm flag too (happy to move the flag around as needed though - I don't have a strong opinion on it)
What we don't want is Tuning sets the defaults for a bunch of behaviors, but each of the behaviors can be individually controlled outside of the tuning parameter. That is, we could decide to abandon the tuning idea entirely, but still get exactly the same behavior for each target, if we were willing to pass the extra umpteen flags. Whether these are driver, cc1, -mllvm, or whatever, doesn't matter. |
If
-mllvm -add-linkage-names-to-external-call-originsis true then addDW_AT_linkage_nameattributes toDW_TAG_subprogramDIEs referenced byDW_AT_call_originattributes that would otherwise be omitted.A debugger may use
DW_TAG_call_originattributes to determine whether anyframes in a call stack are missing due to optimisations (e.g. tail calls).
For example, say
a()callsb()tail-callsc(), and you stop in your debuggerin
c():The callstack looks like this:
Looking "up" from
c(), call site information can be found ina(). This includesa DW_AT_call_origin referencing
b()'s subprogram DIE, which means the call atthis call site was to
b(), notc()where we are currently stopped. Thisindicates
b()'s frame has been lost due to optimisation (or is misleading dueto ICF).
This patch makes it easier for a debugger to check whether the referenced
DIE describes the target function or not, for example by comparing the referenced
function name to the current frame.
There's already an option to apply DW_AT_linkage_name in a targeted manner:
-dwarf-linkage-names=Abstract, which limits adding DW_AT_linkage_names to abstract subprogram DIEs (this is default for SCE tuning).The new flag shouldn't affect non-SCE-tuned behaviour whether it is enabled or not because the non-SCE-tuned behaviour is to always add linkage names to subprogram DIEs, AFAICT.
Enabling this feature alongside
-dwarf-linkage-names=Abstracthas minimal impact on size of CTMark RelWithDebInfo binaries. There's a geomean increase in size by < 0.1%.