tsan: include MBlock/SyncObj stats into mem profile

Include info about MBlock/SyncObj memory consumption in the memory profile.

Depends on D110148.

Reviewed By: melver, vitalybuka

Differential Revision: https://reviews.llvm.org/D110149

compiler-rt/lib/tsan/rtl/tsan_dense_alloc.h

@@ -49,11 +49,7 @@ class DenseSlabAlloc {
   static_assert(sizeof(T) > sizeof(IndexT),
                 "it doesn't make sense to use dense alloc");
 
-  explicit DenseSlabAlloc(LinkerInitialized, const char *name) {
+  DenseSlabAlloc(LinkerInitialized, const char *name) : name_(name) {}
-    freelist_ = 0;
-    fillpos_ = 0;
-    name_ = name;
-  }
 
   explicit DenseSlabAlloc(const char *name)
       : DenseSlabAlloc(LINKER_INITIALIZED, name) {
@@ -89,6 +85,8 @@ class DenseSlabAlloc {
   }
 
   void FlushCache(Cache *c) {
+    if (!c->pos)
+      return;
     SpinMutexLock lock(&mtx_);
     while (c->pos) {
       IndexT idx = c->cache[--c->pos];
@@ -102,33 +100,39 @@ class DenseSlabAlloc {
     internal_memset(c->cache, 0, sizeof(c->cache));
   }
 
+  uptr AllocatedMemory() const {
+    return atomic_load_relaxed(&fillpos_) * kL2Size * sizeof(T);
+  }
+
  private:
   T *map_[kL1Size];
   SpinMutex mtx_;
-  IndexT freelist_;
+  IndexT freelist_ = {0};
-  uptr fillpos_;
+  atomic_uintptr_t fillpos_ = {0};
-  const char *name_;
+  const char *const name_;
 
   void Refill(Cache *c) {
     SpinMutexLock lock(&mtx_);
     if (freelist_ == 0) {
-      if (fillpos_ == kL1Size) {
+      uptr fillpos = atomic_load_relaxed(&fillpos_);
+      if (fillpos == kL1Size) {
         Printf("ThreadSanitizer: %s overflow (%zu*%zu). Dying.\n",
             name_, kL1Size, kL2Size);
         Die();
       }
-      VPrintf(2, "ThreadSanitizer: growing %s: %zu out of %zu*%zu\n",
+      VPrintf(2, "ThreadSanitizer: growing %s: %zu out of %zu*%zu\n", name_,
-          name_, fillpos_, kL1Size, kL2Size);
+              fillpos, kL1Size, kL2Size);
       T *batch = (T*)MmapOrDie(kL2Size * sizeof(T), name_);
       // Reserve 0 as invalid index.
-      IndexT start = fillpos_ == 0 ? 1 : 0;
+      IndexT start = fillpos == 0 ? 1 : 0;
       for (IndexT i = start; i < kL2Size; i++) {
         new(batch + i) T;
-        *(IndexT*)(batch + i) = i + 1 + fillpos_ * kL2Size;
+        *(IndexT *)(batch + i) = i + 1 + fillpos * kL2Size;
       }
       *(IndexT*)(batch + kL2Size - 1) = 0;
-      freelist_ = fillpos_ * kL2Size + start;
+      freelist_ = fillpos * kL2Size + start;
-      map_[fillpos_++] = batch;
+      map_[fillpos] = batch;
+      atomic_store_relaxed(&fillpos_, fillpos + 1);
     }
     for (uptr i = 0; i < Cache::kSize / 2 && freelist_ != 0; i++) {
       IndexT idx = freelist_;

compiler-rt/lib/tsan/rtl/tsan_platform_linux.cpp

@@ -121,17 +121,24 @@ void FillProfileCallback(uptr p, uptr rss, bool file,
 
 void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
   uptr mem[MemCount];
-  internal_memset(mem, 0, sizeof(mem[0]) * MemCount);
+  internal_memset(mem, 0, sizeof(mem));
   __sanitizer::GetMemoryProfile(FillProfileCallback, mem, 7);
+  auto meta = ctx->metamap.GetMemoryStats();
   StackDepotStats *stacks = StackDepotGetStats();
+  // All these are allocated from the common mmap region.
+  mem[MemMmap] -= meta.mem_block + meta.sync_obj + stacks->allocated;
+  if (s64(mem[MemMmap]) < 0)
+    mem[MemMmap] = 0;
   internal_snprintf(buf, buf_size,
-      "RSS %zd MB: shadow:%zd meta:%zd file:%zd mmap:%zd"
+                    "RSS %zd MB: shadow:%zd meta:%zd file:%zd mmap:%zd"
-      " trace:%zd heap:%zd other:%zd stacks=%zd[%zd] nthr=%zd/%zd\n",
+                    " trace:%zd heap:%zd other:%zd memblocks:%zd syncobj:%zu"
-      mem[MemTotal] >> 20, mem[MemShadow] >> 20, mem[MemMeta] >> 20,
+                    " stacks=%zd[%zd] nthr=%zd/%zd\n",
-      mem[MemFile] >> 20, mem[MemMmap] >> 20, mem[MemTrace] >> 20,
+                    mem[MemTotal] >> 20, mem[MemShadow] >> 20,
-      mem[MemHeap] >> 20, mem[MemOther] >> 20,
+                    mem[MemMeta] >> 20, mem[MemFile] >> 20, mem[MemMmap] >> 20,
-      stacks->allocated >> 20, stacks->n_uniq_ids,
+                    mem[MemTrace] >> 20, mem[MemHeap] >> 20,
-      nlive, nthread);
+                    mem[MemOther] >> 20, meta.mem_block >> 20,
+                    meta.sync_obj >> 20, stacks->allocated >> 20,
+                    stacks->n_uniq_ids, nlive, nthread);
 }
 
 #if SANITIZER_LINUX

compiler-rt/lib/tsan/rtl/tsan_sync.cpp

@@ -269,4 +269,11 @@ void MetaMap::OnProcIdle(Processor *proc) {
   sync_alloc_.FlushCache(&proc->sync_cache);
 }
 
+MetaMap::MemoryStats MetaMap::GetMemoryStats() const {
+  MemoryStats stats;
+  stats.mem_block = block_alloc_.AllocatedMemory();
+  stats.sync_obj = sync_alloc_.AllocatedMemory();
+  return stats;
+}
+
 }  // namespace __tsan

compiler-rt/lib/tsan/rtl/tsan_sync.h

@@ -127,6 +127,13 @@ class MetaMap {
 
   void OnProcIdle(Processor *proc);
 
+  struct MemoryStats {
+    uptr mem_block;
+    uptr sync_obj;
+  };
+
+  MemoryStats GetMemoryStats() const;
+
  private:
   static const u32 kFlagMask  = 3u << 30;
   static const u32 kFlagBlock = 1u << 30;