JIT: Remove dataSection flexible array (#124373)

jakobbotsch · web-flow · commit 3249bf9c00a2 · 2026-02-16T13:00:35.000+01:00
Storing pointers and `emitLocation` instances in this `BYTE` flexible
array is problematic due to alignment. We could use `alignas`, but the
flexible array here is a micro optimization so just go with a simpler
representation without the footguns.
diff --git a/src/coreclr/jit/codegencommon.cpp b/src/coreclr/jit/codegencommon.cpp
@@ -6815,7 +6815,7 @@ void CodeGen::genReportAsyncDebugInfo()
         uint32_t diagNativeOffset = 0;
         if (genAsyncResumeInfoTable != nullptr)
         {
-            emitLocation& emitLoc = ((emitLocation*)genAsyncResumeInfoTable->dsCont)[i];
+            emitLocation& emitLoc = genAsyncResumeInfoTable->Locations()[i];
             if (emitLoc.Valid())
             {
                 diagNativeOffset = emitLoc.CodeOffset(GetEmitter());
diff --git a/src/coreclr/jit/codegenlinear.cpp b/src/coreclr/jit/codegenlinear.cpp
@@ -961,7 +961,7 @@ void CodeGen::genRecordAsyncResume(GenTreeVal* asyncResume)
     emitter::dataSection* asyncResumeInfo;
     genEmitAsyncResumeInfoTable(&asyncResumeInfo);
 
-    ((emitLocation*)asyncResumeInfo->dsCont)[index] = emitLocation(GetEmitter());
+    asyncResumeInfo->Locations()[index] = emitLocation(GetEmitter());
 }
 
 /*
diff --git a/src/coreclr/jit/emit.cpp b/src/coreclr/jit/emit.cpp
@@ -7811,14 +7811,14 @@ UNATIVE_OFFSET emitter::emitDataGenBeg(unsigned size, unsigned alignment, var_ty
 
     /* Allocate a data section descriptor and add it to the list */
 
-    dataSection* secDesc = emitDataSecCur = (dataSection*)emitGetMem(roundUp(sizeof(*secDesc) + size));
+    dataSection* secDesc = emitDataSecCur = (dataSection*)emitGetMem(sizeof(*secDesc));
+    secDesc->dsType                       = dataSection::data;
+    secDesc->Data()                       = (BYTE*)emitGetMem(size);
 
     secDesc->dsSize = size;
 
     secDesc->dsAlignment = alignment;
 
-    secDesc->dsType = dataSection::data;
-
     secDesc->dsDataType = dataType;
 
     secDesc->dsNext = nullptr;
@@ -7864,14 +7864,14 @@ UNATIVE_OFFSET emitter::emitBBTableDataGenBeg(unsigned numEntries, bool relative
 
     /* Allocate a data section descriptor and add it to the list */
 
-    secDesc = emitDataSecCur = (dataSection*)emitGetMem(roundUp(sizeof(*secDesc) + numEntries * sizeof(BasicBlock*)));
+    secDesc = emitDataSecCur = (dataSection*)emitGetMem(sizeof(*secDesc));
+    secDesc->dsType          = relativeAddr ? dataSection::blockRelative32 : dataSection::blockAbsoluteAddr;
+    secDesc->Blocks()        = (BasicBlock**)emitGetMem(numEntries * sizeof(BasicBlock*));
 
     secDesc->dsSize = emittedSize;
 
     secDesc->dsAlignment = elemSize;
 
-    secDesc->dsType = relativeAddr ? dataSection::blockRelative32 : dataSection::blockAbsoluteAddr;
-
     secDesc->dsDataType = TYP_UNKNOWN;
 
     secDesc->dsNext = nullptr;
@@ -7905,14 +7905,15 @@ void emitter::emitAsyncResumeTable(unsigned numEntries, UNATIVE_OFFSET* dataSecO
     unsigned       emittedSize = sizeof(CORINFO_AsyncResumeInfo) * numEntries;
     emitConsDsc.dsdOffs += emittedSize;
 
-    dataSection* secDesc = (dataSection*)emitGetMem(roundUp(sizeof(dataSection) + numEntries * sizeof(emitLocation)));
+    dataSection* secDesc = (dataSection*)emitGetMem(sizeof(dataSection));
+    secDesc->dsType      = dataSection::asyncResumeInfo;
+    secDesc->Locations() = (emitLocation*)emitGetMem(numEntries * sizeof(emitLocation));
 
     for (unsigned i = 0; i < numEntries; i++)
-        new (secDesc->dsCont + i * sizeof(emitLocation), jitstd::placement_t()) emitLocation();
+        new (&secDesc->Locations()[i], jitstd::placement_t()) emitLocation();
 
     secDesc->dsSize      = emittedSize;
     secDesc->dsAlignment = TARGET_POINTER_SIZE;
-    secDesc->dsType      = dataSection::asyncResumeInfo;
     secDesc->dsDataType  = TYP_UNKNOWN;
     secDesc->dsNext      = nullptr;
 
@@ -7949,7 +7950,7 @@ void emitter::emitDataGenData(unsigned offs, const void* data, UNATIVE_OFFSET si
 
     assert(emitDataSecCur->dsType == dataSection::data);
 
-    memcpy(emitDataSecCur->dsCont + offs, data, size);
+    memcpy(emitDataSecCur->Data() + offs, data, size);
 }
 
 /*****************************************************************************
@@ -7967,7 +7968,7 @@ void emitter::emitDataGenData(unsigned index, BasicBlock* label)
 
     assert(emitDataSecCur->dsSize >= emittedElemSize * (index + 1));
 
-    ((BasicBlock**)(emitDataSecCur->dsCont))[index] = label;
+    emitDataSecCur->Blocks()[index] = label;
 }
 
 /*****************************************************************************
@@ -8011,7 +8012,7 @@ UNATIVE_OFFSET emitter::emitDataGenFind(const void* cnsAddr, unsigned cnsSize, u
         //
         if ((secDesc->dsType == dataSection::data) && (secDesc->dsSize >= cnsSize) && ((curOffs % alignment) == 0))
         {
-            if (memcmp(cnsAddr, secDesc->dsCont, cnsSize) == 0)
+            if (memcmp(cnsAddr, secDesc->Data(), cnsSize) == 0)
             {
                 cnum = curOffs;
 
@@ -8393,7 +8394,7 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
             target_size_t* bDstRW   = (target_size_t*)dstRW;
             for (unsigned i = 0; i < numElems; i++)
             {
-                BasicBlock* block = ((BasicBlock**)dsc->dsCont)[i];
+                BasicBlock* block = dsc->Blocks()[i];
 
                 // Convert the BasicBlock* value to an IG address
                 insGroup* lab = (insGroup*)emitCodeGetCookie(block);
@@ -8424,7 +8425,7 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
 
             for (unsigned i = 0; i < numElems; i++)
             {
-                BasicBlock* block = ((BasicBlock**)dsc->dsCont)[i];
+                BasicBlock* block = dsc->Blocks()[i];
 
                 // Convert the BasicBlock* value to an IG address
                 insGroup* lab = (insGroup*)emitCodeGetCookie(block);
@@ -8444,7 +8445,7 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
             CORINFO_AsyncResumeInfo* aDstRW = (CORINFO_AsyncResumeInfo*)dstRW;
             for (size_t i = 0; i < numElems; i++)
             {
-                emitLocation* emitLoc = &((emitLocation*)dsc->dsCont)[i];
+                emitLocation* emitLoc = &dsc->Locations()[i];
 
                 // Async call may have been removed very late, after we have introduced suspension/resumption.
                 // In those cases just encode null.
@@ -8468,7 +8469,7 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
             // Simple binary data: copy the bytes to the target
             assert(dsc->dsType == dataSection::data);
 
-            memcpy(dstRW, dsc->dsCont, dscSize);
+            memcpy(dstRW, dsc->Data(), dscSize);
 
 #ifdef DEBUG
             if (EMITVERBOSE)
@@ -8477,7 +8478,7 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
 
                 for (size_t i = 0; i < dscSize; i++)
                 {
-                    printf("%02x ", dsc->dsCont[i]);
+                    printf("%02x ", dsc->Data()[i]);
                     if ((((i + 1) % 16) == 0) && (i + 1 != dscSize))
                     {
                         printf("\n\t\t\t\t\t");
@@ -8487,10 +8488,10 @@ void emitter::emitOutputDataSec(dataSecDsc* sec, AllocMemChunk* chunks)
                 {
                     case TYP_FLOAT:
                         printf(" ; float  %9.6g",
-                               FloatingPointUtils::convertToDouble(*reinterpret_cast<float*>(&dsc->dsCont)));
+                               FloatingPointUtils::convertToDouble(*reinterpret_cast<float*>(dsc->Data())));
                         break;
                     case TYP_DOUBLE:
-                        printf(" ; double %12.9g", *reinterpret_cast<double*>(&dsc->dsCont));
+                        printf(" ; double %12.9g", *reinterpret_cast<double*>(dsc->Data()));
                         break;
                     default:
                         break;
@@ -8551,7 +8552,7 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                     printf(labelFormat, "");
                 }
 
-                BasicBlock* block = reinterpret_cast<BasicBlock**>(data->dsCont)[i];
+                BasicBlock* block = data->Blocks()[i];
                 insGroup*   ig    = static_cast<insGroup*>(emitCodeGetCookie(block));
 
                 const char* blockLabel = emitLabelString(ig);
@@ -8617,7 +8618,7 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                     printf(labelFormat, label);
                 }
 
-                emitLocation* emitLoc = &((emitLocation*)data->dsCont)[i];
+                emitLocation* emitLoc = &data->Locations()[i];
 
                 printf("\tdq\t%s\n", resumeStubName);
 
@@ -8668,9 +8669,9 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                         {
                             printf("\t<Unexpected data size %d (expected >= 4)\n", data->dsSize);
                         }
-                        printf("\tdd\t%08llXh\t", (UINT64) * reinterpret_cast<uint32_t*>(&data->dsCont[i]));
+                        printf("\tdd\t%08llXh\t", (UINT64) * reinterpret_cast<uint32_t*>(&data->Data()[i]));
                         printf("\t; %9.6g",
-                               FloatingPointUtils::convertToDouble(*reinterpret_cast<float*>(&data->dsCont[i])));
+                               FloatingPointUtils::convertToDouble(*reinterpret_cast<float*>(&data->Data()[i])));
                         i += 4;
                         break;
 
@@ -8679,21 +8680,21 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                         {
                             printf("\t<Unexpected data size %d (expected >= 8)\n", data->dsSize);
                         }
-                        printf("\tdq\t%016llXh", *reinterpret_cast<uint64_t*>(&data->dsCont[i]));
-                        printf("\t; %12.9g", *reinterpret_cast<double*>(&data->dsCont[i]));
+                        printf("\tdq\t%016llXh", *reinterpret_cast<uint64_t*>(&data->Data()[i]));
+                        printf("\t; %12.9g", *reinterpret_cast<double*>(&data->Data()[i]));
                         i += 8;
                         break;
 
                     default:
                         switch (elemSize)
                         {
                             case 1:
-                                printf("\tdb\t%02Xh", *reinterpret_cast<uint8_t*>(&data->dsCont[i]));
+                                printf("\tdb\t%02Xh", *reinterpret_cast<uint8_t*>(&data->Data()[i]));
                                 for (j = 1; j < 16; j++)
                                 {
                                     if (i + j >= data->dsSize)
                                         break;
-                                    printf(", %02Xh", *reinterpret_cast<uint8_t*>(&data->dsCont[i + j]));
+                                    printf(", %02Xh", *reinterpret_cast<uint8_t*>(&data->Data()[i + j]));
                                 }
                                 i += j;
                                 break;
@@ -8703,12 +8704,12 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                                 {
                                     printf("\t<Unexpected data size %d (expected size%%2 == 0)\n", data->dsSize);
                                 }
-                                printf("\tdw\t%04Xh", *reinterpret_cast<uint16_t*>(&data->dsCont[i]));
+                                printf("\tdw\t%04Xh", *reinterpret_cast<uint16_t*>(&data->Data()[i]));
                                 for (j = 2; j < 24; j += 2)
                                 {
                                     if (i + j >= data->dsSize)
                                         break;
-                                    printf(", %04Xh", *reinterpret_cast<uint16_t*>(&data->dsCont[i + j]));
+                                    printf(", %04Xh", *reinterpret_cast<uint16_t*>(&data->Data()[i + j]));
                                 }
                                 i += j;
                                 break;
@@ -8719,12 +8720,12 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                                 {
                                     printf("\t<Unexpected data size %d (expected size%%4 == 0)\n", data->dsSize);
                                 }
-                                printf("\tdd\t%08Xh", *reinterpret_cast<uint32_t*>(&data->dsCont[i]));
+                                printf("\tdd\t%08Xh", *reinterpret_cast<uint32_t*>(&data->Data()[i]));
                                 for (j = 4; j < 24; j += 4)
                                 {
                                     if (i + j >= data->dsSize)
                                         break;
-                                    printf(", %08Xh", *reinterpret_cast<uint32_t*>(&data->dsCont[i + j]));
+                                    printf(", %08Xh", *reinterpret_cast<uint32_t*>(&data->Data()[i + j]));
                                 }
                                 i += j;
                                 break;
@@ -8737,12 +8738,12 @@ void emitter::emitDispDataSec(dataSecDsc* section, AllocMemChunk* dataChunks)
                                 {
                                     printf("\t<Unexpected data size %d (expected size%%8 == 0)\n", data->dsSize);
                                 }
-                                printf("\tdq\t%016llXh", *reinterpret_cast<uint64_t*>(&data->dsCont[i]));
+                                printf("\tdq\t%016llXh", *reinterpret_cast<uint64_t*>(&data->Data()[i]));
                                 for (j = 8; j < 64; j += 8)
                                 {
                                     if (i + j >= data->dsSize)
                                         break;
-                                    printf(", %016llXh", *reinterpret_cast<uint64_t*>(&data->dsCont[i + j]));
+                                    printf(", %016llXh", *reinterpret_cast<uint64_t*>(&data->Data()[i + j]));
                                 }
                                 i += j;
                                 break;
diff --git a/src/coreclr/jit/emit.h b/src/coreclr/jit/emit.h
@@ -3555,10 +3555,32 @@ class emitter
         sectionType    dsType;
         var_types      dsDataType;
 
-        // variable-sized array used to store the constant data, BasicBlock*
-        // array in the block cases, or emitLocation for the asyncResumeInfo
-        // case.
-        BYTE dsCont[0];
+    private:
+        union
+        {
+            BYTE*         dsData;      // for data blobs
+            BasicBlock**  dsBlocks;    // for block-based sections
+            emitLocation* dsLocations; // for async resume info
+        };
+
+    public:
+        BYTE*& Data()
+        {
+            assert(dsType == sectionType::data);
+            return dsData;
+        }
+
+        BasicBlock**& Blocks()
+        {
+            assert((dsType == sectionType::blockAbsoluteAddr) || (dsType == sectionType::blockRelative32));
+            return dsBlocks;
+        }
+
+        emitLocation*& Locations()
+        {
+            assert(dsType == sectionType::asyncResumeInfo);
+            return dsLocations;
+        }
     };
 
     /* These describe the entire initialized/uninitialized data sections */
diff --git a/src/coreclr/jit/emitarm.cpp b/src/coreclr/jit/emitarm.cpp
@@ -7326,7 +7326,7 @@ void emitter::emitDispInsHelp(
                 if (jdsc != nullptr && id->idIns() == INS_movt)
                 {
                     unsigned     cnt = jdsc->dsSize / TARGET_POINTER_SIZE;
-                    BasicBlock** bbp = (BasicBlock**)jdsc->dsCont;
+                    BasicBlock** bbp = jdsc->Blocks();
 
                     bool isBound = (emitCodeGetCookie(*bbp) != nullptr);
 
diff --git a/src/coreclr/jit/emitxarch.cpp b/src/coreclr/jit/emitxarch.cpp
@@ -12347,7 +12347,7 @@ void emitter::emitDispAddrMode(instrDesc* id, bool noDetail) const
     if (jdsc && !noDetail)
     {
         unsigned     cnt = (jdsc->dsSize - 1) / TARGET_POINTER_SIZE;
-        BasicBlock** bbp = (BasicBlock**)jdsc->dsCont;
+        BasicBlock** bbp = jdsc->Blocks();
 
 #ifdef TARGET_AMD64
 #define SIZE_LETTER "Q"

Original file line number	Diff line number	Diff line change
`@@ -6815,7 +6815,7 @@ void CodeGen::genReportAsyncDebugInfo()`
`6815`	`6815`	`uint32_t diagNativeOffset = 0;`
`6816`	`6816`	`if (genAsyncResumeInfoTable != nullptr)`
`6817`	`6817`	`{`
`6818`		`- emitLocation& emitLoc = ((emitLocation*)genAsyncResumeInfoTable->dsCont)[i];`
	`6818`	`+ emitLocation& emitLoc = genAsyncResumeInfoTable->Locations()[i];`
`6819`	`6819`	`if (emitLoc.Valid())`
`6820`	`6820`	`{`
`6821`	`6821`	`diagNativeOffset = emitLoc.CodeOffset(GetEmitter());`
Original file line number	Diff line number	Diff line change
`@@ -961,7 +961,7 @@ void CodeGen::genRecordAsyncResume(GenTreeVal* asyncResume)`
`961`	`961`	`emitter::dataSection* asyncResumeInfo;`
`962`	`962`	`genEmitAsyncResumeInfoTable(&asyncResumeInfo);`
`963`	`963`
`964`		`- ((emitLocation*)asyncResumeInfo->dsCont)[index] = emitLocation(GetEmitter());`
	`964`	`+ asyncResumeInfo->Locations()[index] = emitLocation(GetEmitter());`
`965`	`965`	`}`
`966`	`966`
`967`	`967`	`/*`
Original file line number	Diff line number	Diff line change
`@@ -7326,7 +7326,7 @@ void emitter::emitDispInsHelp(`
`7326`	`7326`	`if (jdsc != nullptr && id->idIns() == INS_movt)`
`7327`	`7327`	`{`
`7328`	`7328`	`unsigned cnt = jdsc->dsSize / TARGET_POINTER_SIZE;`
`7329`		`- BasicBlock bbp = (BasicBlock)jdsc->dsCont;`
	`7329`	`+ BasicBlock** bbp = jdsc->Blocks();`
`7330`	`7330`
`7331`	`7331`	`bool isBound = (emitCodeGetCookie(*bbp) != nullptr);`
`7332`	`7332`
Original file line number	Diff line number	Diff line change
`@@ -12347,7 +12347,7 @@ void emitter::emitDispAddrMode(instrDesc* id, bool noDetail) const`
`12347`	`12347`	`if (jdsc && !noDetail)`
`12348`	`12348`	`{`
`12349`	`12349`	`unsigned cnt = (jdsc->dsSize - 1) / TARGET_POINTER_SIZE;`
`12350`		`- BasicBlock bbp = (BasicBlock)jdsc->dsCont;`
	`12350`	`+ BasicBlock** bbp = jdsc->Blocks();`
`12351`	`12351`
`12352`	`12352`	`#ifdef TARGET_AMD64`
`12353`	`12353`	`#define SIZE_LETTER "Q"`