Merge tcmalloc r111 (perftools v. 1.8) with the chromium/ branch.

third_party/tcmalloc/chromium/src/central_freelist.cc

 // Copyright (c) 2008, Google Inc.
 // All rights reserved.
 //
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 // notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
@@ skipping 13 matching lines @@
 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 // ---
 // Author: Sanjay Ghemawat <opensource@google.com>
 
 #include "config.h"
+#include <algorithm>
 #include "central_freelist.h"
-
 #include "internal_logging.h"  // for ASSERT, MESSAGE
 #include "linked_list.h"       // for SLL_Next, SLL_Push, etc
 #include "page_heap.h"         // for PageHeap
 #include "static_vars.h"       // for Static
 
+using std::min;
+using std::max;
+
 namespace tcmalloc {
 
 void CentralFreeList::Init(size_t cl) {
   size_class_ = cl;
   tcmalloc::DLL_Init(&empty_);
   tcmalloc::DLL_Init(&nonempty_);
+  num_spans_ = 0;
   counter_ = 0;
 
+  max_cache_size_ = kMaxNumTransferEntries;
 #ifdef TCMALLOC_SMALL_BUT_SLOW
   // Disable the transfer cache for the small footprint case.
   cache_size_ = 0;
 #else
   cache_size_ = 16;
 #endif
+  if (cl > 0) {
+    // Limit the maximum size of the cache based on the size class.  If this
+    // is not done, large size class objects will consume a lot of memory if
+    // they just sit in the transfer cache.
+    int32_t bytes = Static::sizemap()->ByteSizeForClass(cl);
+    int32_t objs_to_move = Static::sizemap()->num_objects_to_move(cl);
+
+    ASSERT(objs_to_move > 0 && bytes > 0);
+    // Limit each size class cache to at most 1MB of objects or one entry,
+    // whichever is greater. Total transfer cache memory used across all
+    // size classes then can't be greater than approximately
+    // 1MB * kMaxNumTransferEntries.
+    // min and max are in parens to avoid macro-expansion on windows.
+    max_cache_size_ = (min)(max_cache_size_,
+                          (max)(1, (1024 * 1024) / (bytes * objs_to_move)));
+    cache_size_ = (min)(cache_size_, max_cache_size_);
+  }
   used_slots_ = 0;
-  ASSERT(cache_size_ <= kNumTransferEntries);
+  ASSERT(cache_size_ <= max_cache_size_);
 }
 
 void CentralFreeList::ReleaseListToSpans(void* start) {
   while (start) {
     void *next = SLL_Next(start);
     ReleaseToSpans(start);
     start = next;
   }
 }
 
@@ skipping 36 matching lines @@
            Static::sizemap()->ByteSizeForClass(span->sizeclass));
   }
 
   counter_++;
   span->refcount--;
   if (span->refcount == 0) {
     Event(span, '#', 0);
     counter_ -= ((span->length<<kPageShift) /
                  Static::sizemap()->ByteSizeForClass(span->sizeclass));
     tcmalloc::DLL_Remove(span);
+    --num_spans_;
 
     // Release central list lock while operating on pageheap
     lock_.Unlock();
     {
       SpinLockHolder h(Static::pageheap_lock());
       Static::pageheap()->Delete(span);
     }
     lock_.Lock();
   } else {
     *(reinterpret_cast<void**>(object)) = span->objects;
@@ skipping 14 matching lines @@
   ASSERT(t >= 0);
   ASSERT(t < kNumClasses);
   if (t == locked_size_class) return false;
   return Static::central_cache()[t].ShrinkCache(locked_size_class, force);
 }
 
 bool CentralFreeList::MakeCacheSpace() {
   // Is there room in the cache?
   if (used_slots_ < cache_size_) return true;
   // Check if we can expand this cache?
-  if (cache_size_ == kNumTransferEntries) return false;
+  if (cache_size_ == max_cache_size_) return false;
   // Ok, we'll try to grab an entry from some other size class.
   if (EvictRandomSizeClass(size_class_, false) ||
       EvictRandomSizeClass(size_class_, true)) {
     // Succeeded in evicting, we're going to make our cache larger.
     // However, we may have dropped and re-acquired the lock in
     // EvictRandomSizeClass (via ShrinkCache and the LockInverter), so the
     // cache_size may have changed.  Therefore, check and verify that it is
     // still OK to increase the cache_size.
-    if (cache_size_ < kNumTransferEntries) {
+    if (cache_size_ < max_cache_size_) {
       cache_size_++;
       return true;
     }
   }
   return false;
 }
 
 
 namespace {
 class LockInverter {
@@ skipping 36 matching lines @@
   cache_size_--;
   return true;
 }
 
 void CentralFreeList::InsertRange(void *start, void *end, int N) {
   SpinLockHolder h(&lock_);
   if (N == Static::sizemap()->num_objects_to_move(size_class_) &&
     MakeCacheSpace()) {
     int slot = used_slots_++;
     ASSERT(slot >=0);
-    ASSERT(slot < kNumTransferEntries);
+    ASSERT(slot < max_cache_size_);
     TCEntry *entry = &tc_slots_[slot];
     entry->head = start;
     entry->tail = end;
     return;
   }
   ReleaseListToSpans(start);
 }
 
 int CentralFreeList::RemoveRange(void **start, void **end, int N) {
   ASSERT(N > 0);
@@ skipping 97 matching lines @@
     ptr += size;
     num++;
   }
   ASSERT(ptr <= limit);
   *tail = NULL;
   span->refcount = 0; // No sub-object in use yet
 
   // Add span to list of non-empty spans
   lock_.Lock();
   tcmalloc::DLL_Prepend(&nonempty_, span);
+  ++num_spans_;
   counter_ += num;
 }
 
 int CentralFreeList::tc_length() {
   SpinLockHolder h(&lock_);
   return used_slots_ * Static::sizemap()->num_objects_to_move(size_class_);
 }
 
+size_t CentralFreeList::OverheadBytes() {
+  SpinLockHolder h(&lock_);
+  if (size_class_ == 0) {  // 0 holds the 0-sized allocations
+    return 0;
+  }
+  const size_t pages_per_span = Static::sizemap()->class_to_pages(size_class_);
+  const size_t object_size = Static::sizemap()->class_to_size(size_class_);
+  ASSERT(object_size > 0);
+  const size_t overhead_per_span = (pages_per_span * kPageSize) % object_size;
+  return num_spans_ * overhead_per_span;
+}
+
 }  // namespace tcmalloc