commit 0e96d95d13d9f7b2a96bcaa569ce0a0181a6c7f3
Author: Jeremy Morse <jeremy.morse@sony.com>
Date:   Fri Mar 4 16:39:22 2022 +0000

    [DebugInfo][InstrRef] Accept register-reads after isel in any block
    
    When lowering LLVM-IR to instruction referencing stuff, if a value is
    defined by a COPY, we try and follow the register definitions back to where
    the value was defined, and build an instruction reference to that
    instruction. In a few scenarios (such as arguments), this isn't possible.
    I added some assertions to catch cases that weren't explicitly whitelisted.
    
    Over the course of a few months, several more scenarios have cropped up,
    the lastest is the llvm.read_register intrinsic, which lets LLVM-IR read an
    arbitary register at any point. In the face of this, there's little point
    in validating whether debug-info reads a register in an expected scenario.
    Thus: this patch just deletes those assertions, and adds a regression test
    to check that something is done with the llvm.read_register intrinsic.
    
    Fixes #54190
    
    Differential Revision: https://reviews.llvm.org/D121001

diff --git a/llvm/lib/CodeGen/MachineFunction.cpp b/llvm/lib/CodeGen/MachineFunction.cpp
index fd5ea5cad072..884f609c1e75 100644
--- a/llvm/lib/CodeGen/MachineFunction.cpp
+++ b/llvm/lib/CodeGen/MachineFunction.cpp
@@ -1141,26 +1141,13 @@ auto MachineFunction::salvageCopySSA(MachineInstr &MI)
   MachineBasicBlock &InsertBB = *CurInst->getParent();
 
   // We reached the start of the block before finding a defining instruction.
-  // It could be from a constant register, otherwise it must be an argument.
-  if (TRI.isConstantPhysReg(State.first)) {
-    // We can produce a DBG_PHI that identifies the constant physreg. Doesn't
-    // matter where we put it, as it's constant valued.
-    assert(CurInst->isCopy());
-  } else if (State.first == TRI.getFrameRegister(*this)) {
-    // LLVM IR is allowed to read the framepointer by calling a
-    // llvm.frameaddress.* intrinsic. We can support this by emitting a
-    // DBG_PHI $fp. This isn't ideal, because it extends the behaviours /
-    // position that DBG_PHIs appear at, limiting what can be done later.
-    // TODO: see if there's a better way of expressing these variable
-    // locations.
-    ;
-  } else {
-    // Assert that this is the entry block, or an EH pad. If it isn't, then
-    // there is some code construct we don't recognise that deals with physregs
-    // across blocks.
-    assert(!State.first.isVirtual());
-    assert(&*InsertBB.getParent()->begin() == &InsertBB || InsertBB.isEHPad());
-  }
+  // There are numerous scenarios where this can happen:
+  // * Constant physical registers,
+  // * Several intrinsics that allow LLVM-IR to read arbitary registers,
+  // * Arguments in the entry block,
+  // * Exception handling landing pads.
+  // Validating all of them is too difficult, so just insert a DBG_PHI reading
+  // the variable value at this position, rather than checking it makes sense.
 
   // Create DBG_PHI for specified physreg.
   auto Builder = BuildMI(InsertBB, InsertBB.getFirstNonPHI(), DebugLoc(),
diff --git a/llvm/test/DebugInfo/X86/instr-ref-ir-reg-read.ll b/llvm/test/DebugInfo/X86/instr-ref-ir-reg-read.ll
new file mode 100644
index 000000000000..ff1ccd939d71
--- /dev/null
+++ b/llvm/test/DebugInfo/X86/instr-ref-ir-reg-read.ll
@@ -0,0 +1,85 @@
+; RUN: llc %s -start-after=codegenprepare -stop-before=finalize-isel \
+; RUN:    -experimental-debug-variable-locations=true  -o - \
+; RUN: | FileCheck %s
+
+; Test that the given input doesn't crash with instrruction referencing variable
+; locations. The use of llvm.read_register allows the IR to access any register
+; at any point, which is unfortunate, but a use case that needs to be supported.
+;
+; Just examine to see that we read something from $rsp.
+; CHECK-LABEL: bb.1.if.then:
+; CHECK:       DBG_PHI $rsp, 1
+; CHECK:       DBG_INSTR_REF 1, 0
+
+source_filename = "tlb-9e7172.c"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-cros-linux-gnu"
+
+@c = dso_local local_unnamed_addr global i32 0, align 4, !dbg !0
+
+; Function Attrs: noredzone nounwind null_pointer_is_valid optsize sspstrong
+define dso_local void @switch_mm_irqs_off() local_unnamed_addr #0 !dbg !16 {
+entry:
+  %0 = load i32, i32* @c, align 4, !dbg !24
+  %tobool.not = icmp eq i32 %0, 0, !dbg !24
+  br i1 %tobool.not, label %if.end, label %if.then, !dbg !25
+
+if.then:                                          ; preds = %entry
+  call void @llvm.dbg.value(metadata i64 0, metadata !20, metadata !DIExpression()), !dbg !26
+  %1 = tail call i64 @llvm.read_register.i64(metadata !7), !dbg !27
+  call void @llvm.dbg.value(metadata i64 %1, metadata !20, metadata !DIExpression()), !dbg !26
+  %call = tail call i32 @b(i64 noundef %1) #4, !dbg !28
+  ret void, !dbg !29
+
+if.end:                                           ; preds = %entry
+  ret void, !dbg !29
+}
+
+; Function Attrs: nofree nounwind readonly
+declare i64 @llvm.read_register.i64(metadata) #1
+
+; Function Attrs: noredzone null_pointer_is_valid optsize
+declare !dbg !30 dso_local i32 @b(i64 noundef) local_unnamed_addr #2
+
+; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
+declare void @llvm.dbg.value(metadata, metadata, metadata) #3
+
+!llvm.dbg.cu = !{!2}
+!llvm.named.register.rsp = !{!7}
+!llvm.module.flags = !{!8, !9, !10, !11, !12, !13, !14}
+!llvm.ident = !{!15}
+
+!0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
+!1 = distinct !DIGlobalVariable(name: "c", scope: !2, file: !5, line: 2, type: !6, isLocal: false, isDefinition: true)
+!2 = distinct !DICompileUnit(language: DW_LANG_C89, file: !3, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, globals: !4, splitDebugInlining: false, debugInfoForProfiling: true, nameTableKind: None)
+!3 = !DIFile(filename: "tlb.c", directory: ".")
+!4 = !{!0}
+!5 = !DIFile(filename: "tlb-9e7172.c", directory: ".")
+!6 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
+!7 = !{!"rsp"}
+!8 = !{i32 7, !"Dwarf Version", i32 4}
+!9 = !{i32 2, !"Debug Info Version", i32 3}
+!10 = !{i32 1, !"wchar_size", i32 2}
+!11 = !{i32 1, !"Code Model", i32 2}
+!12 = !{i32 7, !"frame-pointer", i32 2}
+!13 = !{i32 1, !"override-stack-alignment", i32 8}
+!14 = !{i32 4, !"SkipRaxSetup", i32 1}
+!15 = !{!"clang version 15.0.0 (git@github.com:llvm/llvm-project ab49dce01f211fd80f76f449035d771f5e2720b9)"}
+!16 = distinct !DISubprogram(name: "switch_mm_irqs_off", scope: !5, file: !5, line: 4, type: !17, scopeLine: 4, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !2, retainedNodes: !19)
+!17 = !DISubroutineType(types: !18)
+!18 = !{null}
+!19 = !{!20}
+!20 = !DILocalVariable(name: "d", scope: !21, file: !5, line: 6, type: !23)
+!21 = distinct !DILexicalBlock(scope: !22, file: !5, line: 5, column: 10)
+!22 = distinct !DILexicalBlock(scope: !16, file: !5, line: 5, column: 7)
+!23 = !DIBasicType(name: "long", size: 64, encoding: DW_ATE_signed)
+!24 = !DILocation(line: 5, column: 7, scope: !22)
+!25 = !DILocation(line: 5, column: 7, scope: !16)
+!26 = !DILocation(line: 0, scope: !21)
+!27 = !DILocation(line: 6, column: 14, scope: !21)
+!28 = !DILocation(line: 7, column: 5, scope: !21)
+!29 = !DILocation(line: 9, column: 1, scope: !16)
+!30 = !DISubprogram(name: "b", scope: !5, file: !5, line: 3, type: !31, flags: DIFlagPrototyped, spFlags: DISPFlagOptimized, retainedNodes: !33)
+!31 = !DISubroutineType(types: !32)
+!32 = !{!6, !23}
+!33 = !{}