Keyless maps default to BPF_MAP_TYPE_ARRAY for some types (#3300)

delphix · Jul 22, 2024 · b4a70de · b4a70de
1 parent 44019ba
commit b4a70de
Show file tree

Hide file tree

Showing 127 changed files with 4,545 additions and 5,498 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -28,6 +28,8 @@ and this project adheres to
   - [#3203](https://github.com/bpftrace/bpftrace/pull/3203)
 - Remove length limitations for the `buf` builtin function
   - [#3249](https://github.com/bpftrace/bpftrace/pull/3249)
+- Faster map access for keyless maps by using BPF_MAP_TYPE_ARRAY
+  - [#3300](https://github.com/bpftrace/bpftrace/pull/3300)
 #### Deprecated
 #### Removed
 - Remove the `-dd` CLI option

diff --git a/src/ast/dibuilderbpf.cpp b/src/ast/dibuilderbpf.cpp
@@ -1,5 +1,6 @@
 #include "dibuilderbpf.h"
 
+#include "libbpf/bpf.h"
 #include "log.h"
 #include "struct.h"
 #include "utils.h"
@@ -204,12 +205,17 @@ DIType *DIBuilderBPF::GetType(const SizedType &stype)
 }
 
 DIType *DIBuilderBPF::GetMapKeyType(const MapKey &key,
-                                    const SizedType &value_type)
+                                    const SizedType &value_type,
+                                    libbpf::bpf_map_type map_type)
 {
   // No-key maps use '0' as the key.
-  // No-key count() maps use BPF_MAP_TYPE_PERCPU_ARRAY which needs 4-byte key.
+  // - BPF requires 4-byte keys for array maps
+  // - bpftrace uses 8 bytes for the implicit '0' key in hash maps
   if (key.args_.size() == 0)
-    return value_type.IsCountTy() ? getInt32Ty() : getInt64Ty();
+    return (map_type == libbpf::BPF_MAP_TYPE_PERCPU_ARRAY ||
+            map_type == libbpf::BPF_MAP_TYPE_ARRAY)
+               ? getInt32Ty()
+               : getInt64Ty();
 
   // Some map types need an extra 8-byte key.
   uint64_t extra_arg_size = 0;
@@ -261,7 +267,9 @@ DIGlobalVariableExpression *DIBuilderBPF::createMapEntry(
   uint64_t size = 128;
   if (!value_type.IsNoneTy()) {
     fields.push_back(createPointerMemberType(
-        "key", size, createPointerType(GetMapKeyType(key, value_type), 64)));
+        "key",
+        size,
+        createPointerType(GetMapKeyType(key, value_type, map_type), 64)));
     fields.push_back(createPointerMemberType(
         "value", size + 64, createPointerType(GetType(value_type), 64)));
     size += 128;

diff --git a/src/ast/dibuilderbpf.h b/src/ast/dibuilderbpf.h
@@ -38,7 +38,9 @@ class DIBuilderBPF : public DIBuilder {
   DIType *createPointerMemberType(const std::string &name,
                                   uint64_t offset,
                                   DIType *type);
-  DIType *GetMapKeyType(const MapKey &key, const SizedType &value_type);
+  DIType *GetMapKeyType(const MapKey &key,
+                        const SizedType &value_type,
+                        libbpf::bpf_map_type map_type);
   DIType *GetMapFieldInt(int value);
   DIGlobalVariableExpression *createMapEntry(const std::string &name,
                                              libbpf::bpf_map_type map_type,

diff --git a/src/ast/passes/codegen_llvm.cpp b/src/ast/passes/codegen_llvm.cpp
@@ -3519,11 +3519,10 @@ libbpf::bpf_map_type CodegenLLVM::get_map_type(const SizedType &val_type,
   if (val_type.IsCountTy() && key.args_.empty()) {
     return libbpf::BPF_MAP_TYPE_PERCPU_ARRAY;
   } else if (bpftrace_.feature_->has_map_percpu_hash() &&
-             (val_type.IsHistTy() || val_type.IsLhistTy() ||
-              val_type.IsCountTy() || val_type.IsSumTy() ||
-              val_type.IsMinTy() || val_type.IsMaxTy() || val_type.IsAvgTy() ||
-              val_type.IsStatsTy())) {
+             val_type.NeedsPercpuMap()) {
     return libbpf::BPF_MAP_TYPE_PERCPU_HASH;
+  } else if (!val_type.NeedsPercpuMap() && key.args_.empty()) {
+    return libbpf::BPF_MAP_TYPE_ARRAY;
   } else {
     return libbpf::BPF_MAP_TYPE_HASH;
   }
@@ -3563,7 +3562,12 @@ void CodegenLLVM::generate_maps(const RequiredResources &resources)
 
     auto max_entries = bpftrace_.config_.get(ConfigKeyInt::max_map_keys);
     auto map_type = get_map_type(val_type, key);
-    if (val_type.IsCountTy() && key.args_.empty()) {
+    // This is special casing for keyless maps eg: @x or @++. For a subset of
+    // keyless maps we can special case them to BPF_MAP_TYPE_ARRAY instead of
+    // HASH for efficiency
+    if ((map_type == libbpf::BPF_MAP_TYPE_PERCPU_ARRAY ||
+         map_type == libbpf::BPF_MAP_TYPE_ARRAY) &&
+        key.args_.empty()) {
       max_entries = 1;
     }
 

diff --git a/src/types.cpp b/src/types.cpp
@@ -568,6 +568,11 @@ bool SizedType::FitsInto(const SizedType &t) const
   return IsEqual(t);
 }
 
+bool SizedType::NeedsPercpuMap() const
+{
+  return IsHistTy() || IsLhistTy() || IsCountTy() || IsSumTy() || IsMinTy() ||
+         IsMaxTy() || IsAvgTy() || IsStatsTy();
+}
 } // namespace bpftrace
 
 namespace std {

diff --git a/src/types.h b/src/types.h
@@ -453,6 +453,8 @@ class SizedType {
              type_ == Type::lhist || type_ == Type::stats);
   }
 
+  bool NeedsPercpuMap() const;
+
   friend std::ostream &operator<<(std::ostream &, const SizedType &);
   friend std::ostream &operator<<(std::ostream &, Type);
 

diff --git a/tests/codegen/llvm/basic_while_loop.ll b/tests/codegen/llvm/basic_while_loop.ll
@@ -9,13 +9,13 @@ target triple = "bpf-pc-linux"
 
 @LICENSE = global [4 x i8] c"GPL\00", section "license"
 @AT_ = dso_local global %"struct map_t" zeroinitializer, section ".maps", !dbg !0
-@ringbuf = dso_local global %"struct map_t.0" zeroinitializer, section ".maps", !dbg !20
-@event_loss_counter = dso_local global %"struct map_t.1" zeroinitializer, section ".maps", !dbg !34
+@ringbuf = dso_local global %"struct map_t.0" zeroinitializer, section ".maps", !dbg !22
+@event_loss_counter = dso_local global %"struct map_t.1" zeroinitializer, section ".maps", !dbg !36
 
 ; Function Attrs: nounwind
 declare i64 @llvm.bpf.pseudo(i64 %0, i64 %1) #0
 
-define i64 @interval_s_1_1(ptr %0) section "s_interval_s_1_1" !dbg !50 {
+define i64 @interval_s_1_1(ptr %0) section "s_interval_s_1_1" !dbg !41 {
 entry:
   %"@_val" = alloca i64, align 8
   %"@_key" = alloca i64, align 8
@@ -30,7 +30,7 @@ while_cond:                                       ; preds = %while_body, %entry
   %2 = icmp sle i64 %1, 150
   %3 = zext i1 %2 to i64
   %true_cond = icmp ne i64 %3, 0
-  br i1 %true_cond, label %while_body, label %while_end, !llvm.loop !57
+  br i1 %true_cond, label %while_body, label %while_end, !llvm.loop !48
 
 while_body:                                       ; preds = %while_cond
   %4 = load i64, ptr %"$a", align 8
@@ -58,8 +58,8 @@ declare void @llvm.lifetime.end.p0(i64 immarg %0, ptr nocapture %1) #1
 attributes #0 = { nounwind }
 attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
 
-!llvm.dbg.cu = !{!47}
-!llvm.module.flags = !{!49}
+!llvm.dbg.cu = !{!38}
+!llvm.module.flags = !{!40}
 
 !0 = !DIGlobalVariableExpression(var: !1, expr: !DIExpression())
 !1 = distinct !DIGlobalVariable(name: "AT_", linkageName: "global", scope: !2, file: !2, type: !3, isLocal: false, isDefinition: true)
@@ -68,55 +68,46 @@ attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: re
 !4 = !{!5, !11, !16, !19}
 !5 = !DIDerivedType(tag: DW_TAG_member, name: "type", scope: !2, file: !2, baseType: !6, size: 64)
 !6 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !7, size: 64)
-!7 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 32, elements: !9)
+!7 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 64, elements: !9)
 !8 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
 !9 = !{!10}
-!10 = !DISubrange(count: 1, lowerBound: 0)
+!10 = !DISubrange(count: 2, lowerBound: 0)
 !11 = !DIDerivedType(tag: DW_TAG_member, name: "max_entries", scope: !2, file: !2, baseType: !12, size: 64, offset: 64)
 !12 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !13, size: 64)
-!13 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 131072, elements: !14)
+!13 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 32, elements: !14)
 !14 = !{!15}
-!15 = !DISubrange(count: 4096, lowerBound: 0)
+!15 = !DISubrange(count: 1, lowerBound: 0)
 !16 = !DIDerivedType(tag: DW_TAG_member, name: "key", scope: !2, file: !2, baseType: !17, size: 64, offset: 128)
 !17 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !18, size: 64)
-!18 = !DIBasicType(name: "int64", size: 64, encoding: DW_ATE_signed)
-!19 = !DIDerivedType(tag: DW_TAG_member, name: "value", scope: !2, file: !2, baseType: !17, size: 64, offset: 192)
-!20 = !DIGlobalVariableExpression(var: !21, expr: !DIExpression())
-!21 = distinct !DIGlobalVariable(name: "ringbuf", linkageName: "global", scope: !2, file: !2, type: !22, isLocal: false, isDefinition: true)
-!22 = !DICompositeType(tag: DW_TAG_structure_type, scope: !2, file: !2, size: 128, elements: !23)
-!23 = !{!24, !29}
-!24 = !DIDerivedType(tag: DW_TAG_member, name: "type", scope: !2, file: !2, baseType: !25, size: 64)
-!25 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !26, size: 64)
-!26 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 864, elements: !27)
-!27 = !{!28}
-!28 = !DISubrange(count: 27, lowerBound: 0)
-!29 = !DIDerivedType(tag: DW_TAG_member, name: "max_entries", scope: !2, file: !2, baseType: !30, size: 64, offset: 64)
-!30 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !31, size: 64)
-!31 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 8388608, elements: !32)
-!32 = !{!33}
-!33 = !DISubrange(count: 262144, lowerBound: 0)
-!34 = !DIGlobalVariableExpression(var: !35, expr: !DIExpression())
-!35 = distinct !DIGlobalVariable(name: "event_loss_counter", linkageName: "global", scope: !2, file: !2, type: !36, isLocal: false, isDefinition: true)
-!36 = !DICompositeType(tag: DW_TAG_structure_type, scope: !2, file: !2, size: 256, elements: !37)
-!37 = !{!38, !43, !44, !19}
-!38 = !DIDerivedType(tag: DW_TAG_member, name: "type", scope: !2, file: !2, baseType: !39, size: 64)
-!39 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !40, size: 64)
-!40 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 64, elements: !41)
-!41 = !{!42}
-!42 = !DISubrange(count: 2, lowerBound: 0)
-!43 = !DIDerivedType(tag: DW_TAG_member, name: "max_entries", scope: !2, file: !2, baseType: !6, size: 64, offset: 64)
-!44 = !DIDerivedType(tag: DW_TAG_member, name: "key", scope: !2, file: !2, baseType: !45, size: 64, offset: 128)
-!45 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !46, size: 64)
-!46 = !DIBasicType(name: "int32", size: 32, encoding: DW_ATE_signed)
-!47 = distinct !DICompileUnit(language: DW_LANG_C, file: !2, producer: "bpftrace", isOptimized: false, runtimeVersion: 0, emissionKind: LineTablesOnly, globals: !48)
-!48 = !{!0, !20, !34}
-!49 = !{i32 2, !"Debug Info Version", i32 3}
-!50 = distinct !DISubprogram(name: "interval_s_1_1", linkageName: "interval_s_1_1", scope: !2, file: !2, type: !51, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !47, retainedNodes: !55)
-!51 = !DISubroutineType(types: !52)
-!52 = !{!18, !53}
-!53 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !54, size: 64)
-!54 = !DIBasicType(name: "int8", size: 8, encoding: DW_ATE_signed)
-!55 = !{!56}
-!56 = !DILocalVariable(name: "ctx", arg: 1, scope: !50, file: !2, type: !53)
-!57 = distinct !{!57, !58}
-!58 = !{!"llvm.loop.unroll.disable"}
+!18 = !DIBasicType(name: "int32", size: 32, encoding: DW_ATE_signed)
+!19 = !DIDerivedType(tag: DW_TAG_member, name: "value", scope: !2, file: !2, baseType: !20, size: 64, offset: 192)
+!20 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !21, size: 64)
+!21 = !DIBasicType(name: "int64", size: 64, encoding: DW_ATE_signed)
+!22 = !DIGlobalVariableExpression(var: !23, expr: !DIExpression())
+!23 = distinct !DIGlobalVariable(name: "ringbuf", linkageName: "global", scope: !2, file: !2, type: !24, isLocal: false, isDefinition: true)
+!24 = !DICompositeType(tag: DW_TAG_structure_type, scope: !2, file: !2, size: 128, elements: !25)
+!25 = !{!26, !31}
+!26 = !DIDerivedType(tag: DW_TAG_member, name: "type", scope: !2, file: !2, baseType: !27, size: 64)
+!27 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !28, size: 64)
+!28 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 864, elements: !29)
+!29 = !{!30}
+!30 = !DISubrange(count: 27, lowerBound: 0)
+!31 = !DIDerivedType(tag: DW_TAG_member, name: "max_entries", scope: !2, file: !2, baseType: !32, size: 64, offset: 64)
+!32 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !33, size: 64)
+!33 = !DICompositeType(tag: DW_TAG_array_type, baseType: !8, size: 8388608, elements: !34)
+!34 = !{!35}
+!35 = !DISubrange(count: 262144, lowerBound: 0)
+!36 = !DIGlobalVariableExpression(var: !37, expr: !DIExpression())
+!37 = distinct !DIGlobalVariable(name: "event_loss_counter", linkageName: "global", scope: !2, file: !2, type: !3, isLocal: false, isDefinition: true)
+!38 = distinct !DICompileUnit(language: DW_LANG_C, file: !2, producer: "bpftrace", isOptimized: false, runtimeVersion: 0, emissionKind: LineTablesOnly, globals: !39)
+!39 = !{!0, !22, !36}
+!40 = !{i32 2, !"Debug Info Version", i32 3}
+!41 = distinct !DISubprogram(name: "interval_s_1_1", linkageName: "interval_s_1_1", scope: !2, file: !2, type: !42, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !38, retainedNodes: !46)
+!42 = !DISubroutineType(types: !43)
+!43 = !{!21, !44}
+!44 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !45, size: 64)
+!45 = !DIBasicType(name: "int8", size: 8, encoding: DW_ATE_signed)
+!46 = !{!47}
+!47 = !DILocalVariable(name: "ctx", arg: 1, scope: !41, file: !2, type: !44)
+!48 = distinct !{!48, !49}
+!49 = !{!"llvm.loop.unroll.disable"}