Support multi heap managers (#45)

* add init function in each heap manager

* add ptmalloc glibc version detection

* fix ptmalloc glibc 2.27 + gdb9.2 on Ubuntu 18.04

* move get_glibc_version to heap_ptmalloc files

* rename gHeapInitFuncs to gHeapRegistrationFuncs

* heap.cpp: add macro to check CA_HEAP and enhance its error message

Co-authored-by: Michael Yan <myan@microstrategy.com>
Former-commit-id: ed32610

.devcontainer/Dockerfile

@@ -7,6 +7,7 @@ FROM mcr.microsoft.com/vscode/devcontainers/cpp:0-${VARIANT}
 # [Optional] Uncomment this section to install additional packages.
 RUN apt-get update && export DEBIAN_FRONTEND=noninteractive \
     && apt install -y texinfo \
+    && apt install -y libgmp-dev \
     && apt install -y build-essential \
     && apt install -y wget \
     && apt install -y python-dev \

gdbplus/gdb-12.1/gdb/Makefile.in

@@ -1092,6 +1092,8 @@ COMMON_SFILES = \
 	go-typeprint.c \
 	go-valprint.c \
 	heap.c \
+	heap_ptmalloc_2_27.c \
+	heap_ptmalloc_2_31.c \
 	heap_ptmalloc_2_35.c \
 	heapcmd.c \
 	i386-decode.c \

gdbplus/gdb-12.1/gdb/gdb_dep.c

@@ -48,6 +48,37 @@ inferior_memory_read (address_t addr, void* buffer, size_t sz)
 		return false;
 }
 
+void ca_switch_to_thread(struct thread_info *info)
+{
+    switch_to_thread (info);
+}
+
+int ca_num_fields(struct type *type)
+{
+    return type->num_fields();
+}
+
+const char *ca_field_name(struct type *type, int i)
+{
+    return type->field(i).name();
+}
+
+bool get_gv_value(const char *varname, char *buf, size_t bufsz)
+{
+	struct symbol *sym;
+
+	sym = lookup_static_symbol(varname, VAR_DOMAIN).symbol;
+	if (sym) {
+		struct value *val = value_of_variable(sym, 0);
+		if (bufsz >= TYPE_LENGTH(value_type(val))) {
+			gdb::array_view<const gdb_byte> content = value_contents(val);
+			memcpy(buf, content.data(), TYPE_LENGTH(value_type(val)));
+			return true;
+		}
+	}
+	return false;
+}
+
 static int
 thread_callback (struct thread_info *info, void *data)
 {
@@ -379,8 +410,6 @@ build_segments(void)
 bool
 update_memory_segments_and_heaps(void)
 {
-	bool rc = true;
-
 	/* reset internal quit flag */
 	g_quit = false;
 
@@ -422,9 +451,7 @@ update_memory_segments_and_heaps(void)
 		return false;
 	}
 	/* Probe for heap segments */
-	CA_HEAP->init_heap();
-
-	return rc;
+    return init_heap_managers();
 }
 
 int

gdbplus/gdb-12.1/gdb/heap_ptmalloc_2_27.c

@@ -0,0 +1 @@
+../../../src/heap_ptmalloc_2_27.cpp

gdbplus/gdb-12.1/gdb/heap_ptmalloc_2_31.c

@@ -0,0 +1 @@
+../../../src/heap_ptmalloc_2_31.cpp

gdbplus/gdb-12.1/gdb/heapcmd.c

@@ -265,6 +265,10 @@ display_help_command (const char *arg, int from_tty)
 static void
 switch_heap_command(const char *arg, int from_tty)
 {
+	/* Ensure heap manager is initialized */
+	if (!update_memory_segments_and_heaps())
+		return;
+
 	if (!arg) {
 		auto supported_heaps = get_supported_heaps();
 		CA_PRINT("Please provide the heap manager name, currently supported heap managers: %s.\n", supported_heaps.c_str());
@@ -292,7 +296,6 @@ void _initialize_heapcmd ();
 void
 _initialize_heapcmd ()
 {
-	register_heap_managers(); // todo: find a better place to register heaps. maybe update_memory_segments_and_heaps is better than here.
 	add_cmd("ref", class_info, ref_command, _("Search for references to a given object.\nref <addr_exp>\nref [/thread or /t] <addr_exp> <size> [level]"), &cmdlist);
 	add_cmd("obj", class_info, obj_command, _("Search for object and reference to object of the same type as the input expression\nobj <type|variable>"), &cmdlist);
 	add_cmd("shrobj", class_info, shrobj_command, _("Find objects that currently referenced from multiple threads\nshrobj [tid0] [tid1] [...]"), &cmdlist);

gdbplus/gdb-9.2/gdb/Makefile.in

@@ -1063,7 +1063,9 @@ COMMON_SFILES = \
 	go-typeprint.c \
 	go-valprint.c \
 	heap.c \
-	heap_ptmalloc.c \
+	heap_ptmalloc_2_27.c \
+	heap_ptmalloc_2_31.c \
+	heap_ptmalloc_2_35.c \
 	heap_tcmalloc.c \
 	heapcmd.c \
 	i386-decode.c \

gdbplus/gdb-9.2/gdb/gdb_dep.c

@@ -48,6 +48,37 @@ inferior_memory_read (address_t addr, void* buffer, size_t sz)
 		return false;
 }
 
+void ca_switch_to_thread(struct thread_info *info)
+{
+    switch_to_thread (info->ptid);
+}
+
+int ca_num_fields(struct type *type)
+{
+    return TYPE_NFIELDS (type);
+}
+
+const char *ca_field_name(struct type *type, int i)
+{
+    return TYPE_FIELD_NAME (type, i);
+}
+
+bool get_gv_value(const char *varname, char *buf, size_t bufsz)
+{
+	struct symbol *sym;
+
+	sym = lookup_static_symbol(varname, VAR_DOMAIN).symbol;
+	if (sym) {
+		struct value *val = value_of_variable(sym, 0);
+		if (bufsz >= TYPE_LENGTH(value_type(val))) {
+			const gdb_byte *content = value_contents(val);
+			memcpy(buf, content, TYPE_LENGTH(value_type(val)));
+			return true;
+		}
+	}
+	return false;
+}
+
 static int
 thread_callback (struct thread_info *info, void *data)
 {
@@ -382,8 +413,6 @@ build_segments(void)
 bool
 update_memory_segments_and_heaps(void)
 {
-	bool rc = true;
-
 	/* reset internal quit flag */
 	g_quit = false;
 
@@ -425,9 +454,7 @@ update_memory_segments_and_heaps(void)
 		return false;
 	}
 	/* Probe for heap segments */
-	CA_HEAP->init_heap();
-
-	return rc;
+    return init_heap_managers();
 }
 
 int

gdbplus/gdb-9.2/gdb/heap_ptmalloc_2_27.c

@@ -0,0 +1 @@
+../../../src/heap_ptmalloc_2_27.cpp

gdbplus/gdb-9.2/gdb/heap_ptmalloc_2_31.c

@@ -0,0 +1 @@
+../../../src/heap_ptmalloc_2_31.cpp

gdbplus/gdb-9.2/gdb/heap_ptmalloc_2_35.c

@@ -0,0 +1 @@
+../../../src/heap_ptmalloc_2_35.cpp

gdbplus/gdb-9.2/gdb/heapcmd.c

@@ -289,7 +289,6 @@ switch_heap_command(const char *arg, int from_tty)
 void
 _initialize_heapcmd (void)
 {
-	register_heap_managers(); // todo: find a better place to register heaps. maybe update_memory_segments_and_heaps is better than here.
 	add_cmd("ref", class_info, ref_command, _("Search for references to a given object.\nref <addr_exp>\nref [/thread or /t] <addr_exp> <size> [level]"), &cmdlist);
 	add_cmd("obj", class_info, obj_command, _("Search for object and reference to object of the same type as the input expression\nobj <type|variable>"), &cmdlist);
 	add_cmd("shrobj", class_info, shrobj_command, _("Find objects that currently referenced from multiple threads\nshrobj [tid0] [tid1] [...]"), &cmdlist);

gdbplus/gdb-9.2/gdb/i386-decode.c.REMOVED.git-id

@@ -1 +1 @@
-d2f46fb1f1e55c960eedabbc1b9150bf1fae008b
+8a5aa1178b538f607392a2d584f5c82acf226932

src/core_elf.cpp

@@ -793,7 +793,6 @@ static bool InitLinkMap_32(MmapFile& irExec, MmapFile& irCore)
 //////////////////////////////////////////////////////////////
 bool InitCoreAnalyzer(MmapFile& irExec, MmapFile& irCore)
 {
-	register_heap_managers();
 	int ptr_bit = g_ptr_bit;
 	bool rc;
 	if (ptr_bit == 64)
@@ -812,7 +811,7 @@ bool InitCoreAnalyzer(MmapFile& irExec, MmapFile& irCore)
 	else
 		rc = InitLinkMap_32 (irExec, irCore);
 
-	return rc;
+	return init_heap_managers();
 }
 
 static bool VerifyELFHeader(Elf64_Ehdr* elfhdr )

src/core_pe_x86.cpp

@@ -688,10 +688,9 @@ extern bool g_dbgheap;
 
 bool InitCoreAnalyzer(MmapFile& irExec, MmapFile& irCore)
 {
-	register_heap_managers();
 	bool rc = BuildSegments(irCore) /*&& test_segments(true)*/;
 
-	return rc;
+	return init_heap_managers();
 }
 
 static struct ca_segment*

src/heap.cpp

@@ -12,32 +12,63 @@
 #include "search.h"
 
 
+CoreAnalyzerHeapInterface* gCAHeap;
+
+#define ENSURE_CA_HEAP()							\
+	do {											\
+		if (!CA_HEAP) {								\
+			CA_PRINT("No heap manager is detedted or selected.\n");	\
+			return false;							\
+		}											\
+	} while (0)
+
 std::map<std::string, CoreAnalyzerHeapInterface*> gCoreAnalyzerHeaps;
 
-CoreAnalyzerHeapInterface* gCAHeap;
-void register_heap_managers() {
+static std::vector<void(*)()> gHeapRegistrationFuncs = {
 	#ifdef WIN32
-	gCoreAnalyzerHeaps[HeapManagerMscrtMalloc] = get_mscrt_malloc_heap_manager();
-	gCAHeap = get_mscrt_malloc_heap_manager();
+    register_mscrt_malloc,
 	#else
-	// heap.cpp is used by ref.dll, and we only support switching heap in Linux currently.
-	//gCoreAnalyzerHeaps.emplace(std::make_pair<std::string, CoreAnalyzerHeapInterface*>("pt", get_pt_malloc_heap_manager()));
-	//gCoreAnalyzerHeaps.emplace(std::make_pair<std::string, CoreAnalyzerHeapInterface*>("tc",get_tc_malloc_heap_manager()));
-	gCoreAnalyzerHeaps.emplace(std::make_pair<std::string, CoreAnalyzerHeapInterface*>("pt 2.35", get_pt_malloc_2_35_heap_manager()));
-	// default is pt malloc heap, in the future we will auto detect the heap.
-	gCAHeap = get_pt_malloc_2_35_heap_manager();
-	#endif
+    register_pt_malloc_2_27,
+    register_pt_malloc_2_31,
+    register_pt_malloc_2_35,
+    //register_tc_malloc,
+    #endif
+};
+
+bool init_heap_managers() {
+    gCoreAnalyzerHeaps.clear();
+    gCAHeap = nullptr;
+
+    for (auto f: gHeapRegistrationFuncs)
+        f();
+
+    if (gCAHeap) {
+        CA_HEAP->init_heap();
+        return true;
+    }
+	CA_PRINT("failed to parse heap data\n");
+    return false;
+}
+
+void register_heap_manager(std::string name, CoreAnalyzerHeapInterface* heapIntf, bool detected) {
+    gCoreAnalyzerHeaps[name] = heapIntf;
+    if (detected) {
+        /* TODO we need to resolve the scenario that multi heap managers are present */
+        gCAHeap = heapIntf;
+    }
 }
 
 std::string get_supported_heaps() {
 	std::string lSupportedHeaps;
 	bool first_entry = true;
-	for (const auto&[k, v] : gCoreAnalyzerHeaps) {
+	for (const auto &itr : gCoreAnalyzerHeaps) {
 		if (!first_entry)
 		{
 			lSupportedHeaps += ", ";
 		}
-		lSupportedHeaps +=k;
+		if (itr.second == gCAHeap)
+			lSupportedHeaps += "(current)";
+		lSupportedHeaps += itr.first;
 		first_entry = false;
 	}
 	return lSupportedHeaps;
@@ -128,6 +159,8 @@ char ca_help_msg[] = "Commands of core_analyzer " CA_VERSION_STRING "\n"
  */
 bool heap_command_impl(const char* args)
 {
+	ENSURE_CA_HEAP();
+
 	bool rc = true;
 
 	address_t addr = 0;
@@ -320,6 +353,8 @@ bool heap_command_impl(const char* args)
  */
 bool ref_command_impl(const char* args)
 {
+	ENSURE_CA_HEAP();
+
 	int rc;
 	bool threadref = false;
 	address_t addr = 0;
@@ -542,6 +577,8 @@ bool segment_command_impl(const char* args)
  */
 bool pattern_command_impl(const char* args)
 {
+	ENSURE_CA_HEAP();
+
 	address_t lo = 0, hi = 0;
 	// Parse user input options
 	// argument is in the form of <start> <end>
@@ -1293,7 +1330,6 @@ bool display_heap_leak_candidates(void)
  */
 void display_mem_histogram(const char* prefix)
 {
-
 	if (!g_mem_hist.num_buckets || !g_mem_hist.bucket_sizes
 		|| !g_mem_hist.inuse_cnt || !g_mem_hist.inuse_bytes
 		|| !g_mem_hist.free_cnt || !g_mem_hist.free_bytes)

src/heap.h

@@ -64,12 +64,30 @@ extern std::map<std::string, CoreAnalyzerHeapInterface*> gCoreAnalyzerHeaps;
 
 extern CoreAnalyzerHeapInterface* gCAHeap;
 #define CA_HEAP gCAHeap
-extern void register_heap_managers();
 
-extern CoreAnalyzerHeapInterface* get_pt_malloc_heap_manager();
-extern CoreAnalyzerHeapInterface* get_pt_malloc_2_35_heap_manager();
-extern CoreAnalyzerHeapInterface* get_tc_malloc_heap_manager();
-extern CoreAnalyzerHeapInterface* get_mscrt_malloc_heap_manager();
+/*
+* This function is called at bootstrap or when target is changed
+* and after target memory layout is scanned.
+*/
+extern bool init_heap_managers();
+
+/*
+* Individual heap manager calls this function in its init function
+* to declare its name, heap interface, and whether it detects its heap data in the target.
+*/
+extern void register_heap_manager(std::string, CoreAnalyzerHeapInterface*, bool);
+
+/*
+* Each heap manager implements an init function
+* Maybe we don't need to explicitly expose them. We may scape these functions at compile time
+* the same way gdb commands initializers are collected.
+*/
+extern void register_pt_malloc_2_27();
+extern void register_pt_malloc_2_31();
+extern void register_pt_malloc_2_35();
+extern void register_tc_malloc();
+extern void register_mscrt_malloc();
+
 extern std::string get_supported_heaps();
 
 extern struct inuse_block* build_inuse_heap_blocks(unsigned long*);

src/heap_mscrt.cpp

@@ -237,8 +237,8 @@ CoreAnalyzerHeapInterface sMscrtMallocHeap = {
    get_biggest_blocks,
    walk_inuse_blocks,
 };
-CoreAnalyzerHeapInterface* get_mscrt_malloc_heap_manager() {
-	return &sMscrtMallocHeap;
+void register_mscrt_malloc() {
+    register_heap_manager(HeapManagerMscrtMalloc, &sMscrtMallocHeap, true);
 }
 /////////////////////////////////////////////////////
 // Implementation of heap walk

src/heap_ptmalloc.cpp

@@ -665,8 +665,8 @@ static CoreAnalyzerHeapInterface sPtMallHeapManager = {
    walk_inuse_blocks,
 };
 
-CoreAnalyzerHeapInterface* get_pt_malloc_heap_manager() {
-	return &sPtMallHeapManager;
+void register_pt_malloc() {
+    return register_heap_manager("pt", &sPtMallHeapManager, true);
 }
 /***************************************************************************
 * Ptmalloc Helper Functions