memory: let the debug allocator mimic the standard allocator more clo…

…sely

The standard allocator returns page-aligned addresses for large allocations.
Some osv code incorrectly relies on this.

While we should fix the incorrect code, for now, adjust the debug allocator
to return aligned addresses.

The debug allocator now uses the following layout:

  [header page][guard page][user data][pattern tail][guard page]

core/mempool.cc

@@ -670,7 +670,7 @@ struct header {
     char fence[16];
     size_t size2;
 };
-static const size_t pad_before = mmu::page_size;
+static const size_t pad_before = 2 * mmu::page_size;
 static const size_t pad_after = mmu::page_size;
 
 void* malloc(size_t size)
@@ -679,30 +679,30 @@ void* malloc(size_t size)
         return std_malloc(size);
     }
 
-    auto hsize = size + sizeof(header);
-    auto asize = align_up(hsize, mmu::page_size);
+    auto asize = align_up(size, mmu::page_size);
     auto padded_size = pad_before + asize + pad_after;
     void* v = free_area.fetch_add(padded_size, std::memory_order_relaxed);
+    mmu::vpopulate(v, mmu::page_size);
+    new (v) header(size);
     v += pad_before;
     mmu::vpopulate(v, asize);
-    auto h = new (v) header(size);
-    memset(v + hsize, '$', asize - hsize);
-    return h + 1;
+    memset(v + size, '$', asize - size);
+    return v;
 }
 
 void free(void* v)
 {
     if (v < debug_base) {
         return std_free(v);
     }
-    auto h = static_cast<header*>(v) - 1;
+    auto h = static_cast<header*>(v - pad_before);
     auto size = h->size;
-    auto hsize = size + sizeof(header);
-    auto asize = align_up(hsize, mmu::page_size);
-    char* vv = reinterpret_cast<char*>(h);
-    assert(std::all_of(vv + hsize, vv  + asize, [=](char c) { return c == '$'; }));
+    auto asize = align_up(size, mmu::page_size);
+    char* vv = reinterpret_cast<char*>(v);
+    assert(std::all_of(vv + size, vv + asize, [=](char c) { return c == '$'; }));
     h->~header();
-    mmu::vdepopulate(h, asize);
+    mmu::vdepopulate(h, mmu::page_size);
+    mmu::vdepopulate(v, asize);
 }
 
 void* realloc(void* v, size_t size)