Small SlabAllocator fixes

Summary:
- Pull from back of vector (back() + pop_back()) when reclaiming slabs since we're using a vector
- Remove unused headers (and add necessary ones in other .cpp files)
- Fix pointer provenance lint (verified via [godbolt](https://gcc.godbolt.org/z/YPrccWfvd) that it should produce identical assembly)

Reviewed By: AlnisM

Differential Revision: D99835775

fbshipit-source-id: c46e9ac3c0463bc9ae7f24fe5c0efd6d2d395d1b

Author: rlyerly

cachelib/allocator/memory/MemoryAllocator.cpp

@@ -21,6 +21,8 @@
 #include <algorithm>
 #include <utility>
 
+#include "cachelib/common/Utils.h"
+
 using namespace facebook::cachelib;
 
 namespace {

cachelib/allocator/memory/MemoryPool.cpp

@@ -17,7 +17,7 @@
 #include "cachelib/allocator/memory/MemoryPool.h"
 
 #include <algorithm>
-#include <string>
+#include <numeric>
 #include <utility>
 
 #include "cachelib/allocator/memory/AllocationClass.h"

cachelib/allocator/memory/SlabAllocator.cpp

@@ -400,9 +400,8 @@ Slab* FOLLY_NULLABLE SlabAllocator::reclaimSlab(PoolId id) {
   {
     LockHolder l(lock_);
     if (!advisedSlabs_.empty()) {
-      auto it = advisedSlabs_.begin();
-      slab = *it;
-      advisedSlabs_.erase(it);
+      slab = advisedSlabs_.back();
+      advisedSlabs_.pop_back();
     }
   }
 

cachelib/allocator/memory/SlabAllocator.h

@@ -21,7 +21,6 @@
 #include <atomic>
 #include <mutex>
 #include <thread>
-#include <unordered_map>
 #include <vector>
 
 #pragma GCC diagnostic push
@@ -35,7 +34,6 @@
 
 #include "cachelib/allocator/memory/CompressedPtr.h"
 #include "cachelib/allocator/memory/Slab.h"
-#include "cachelib/common/Utils.h"
 
 namespace facebook {
 namespace cachelib {
@@ -214,8 +212,9 @@ class SlabAllocator {
   FOLLY_ALWAYS_INLINE const Slab* getSlabForMemory(
       const void* memory) const noexcept {
     // returns the closest slab boundary for the memory address.
-    return reinterpret_cast<const Slab*>(reinterpret_cast<uintptr_t>(memory) &
-                                         kAddressMask);
+    auto offset = reinterpret_cast<uintptr_t>(memory) & ~kAddressMask;
+    return reinterpret_cast<const Slab*>(
+        reinterpret_cast<const uint8_t*>(memory) - offset);
   }
 
   using SlabIdx = uint32_t;