[scudo] Fix MallocIterateBoundary test on 32 bit Android.

[cferris]

Summary

On Android, the min alignment is 16 bytes. This test needs
the BlockDelta to match the min alignment, so set this value
differently for Android.

Update the comment in to explain these details.

compiler-rt/lib/scudo/standalone/tests/wrappers_c_test.cpp

 }
 
 // Verify that a block located on an iteration boundary is not mis-accounted.
 // To achieve this, we allocate a chunk for which the backing block will be
 // aligned on a page, then run the malloc_iterate on both the pages that the
 // block is a boundary for. It must only be seen once by the callback function.
 TEST(ScudoWrappersCTest, MallocIterateBoundary) {
   const size_t PageSize = sysconf(_SC_PAGESIZE);
+#if SCUDO_ANDROID
+  // Android uses a 16 byte alignment for both 32 bit and 64 bit.
+  const size_t BlockDelta = 16U;
+#else
   const size_t BlockDelta = FIRST_32_SECOND_64(8U, 16U);
+#endif
   const size_t SpecialSize = PageSize - BlockDelta;
 
   // We aren't guaranteed that any size class is exactly a page wide. So we need
-  // to keep making allocations until we succeed.
+  // to keep making allocations until we get an allocation that starts exactly
+  // on a page boundary. The BlockDelta value is expected to be the number of
+  // bytes to subtract from a returned pointer to get to the actual start of
+  // the pointer in the size class. In practice, this means BlockDelta should
+  // be set to the minimum alignment in bytes for the allocation.
   //
   // With a 16-byte block alignment and 4096-byte page size, each allocation has
   // a probability of (1 - (16/4096)) of failing to meet the alignment
   // requirements, and the probability of failing 65536 times is
   // (1 - (16/4096))^65536 < 10^-112. So if we still haven't succeeded after
   // 65536 tries, give up.
   uintptr_t Block;
   void *P = nullptr;