Kernel: Free allocated regions when destroying threads

CHANGELOG.md

@@ -12,5 +12,4 @@
 
   - Deallocate previous stack when loading executable
 
-  - The `PageAllocator` would split blocks and them put the other half back at the incorrect size,
-    leaking an enourmous amount of memory.
+  - Free allocated regions when destroying threads

Kernel/Loader.cpp

@@ -207,9 +207,11 @@ namespace Kernel
 
             // We allocated a stack for this thread in the Scheduler, free it here
             VERIFY(Scheduler::the().active_thread().m_owned_ranges.size() == 1);
-            for (auto range : Scheduler::the().active_thread().m_owned_ranges.iter()) {
-                PageAllocator::the().deallocate(range);
-            }
+            Scheduler::the().active_thread().free_owned_ranges();
+
+            // FIXME: This is a bit ugly, RAII for the PageRanges?
+            // Take ownership of the regions of this executable
+            Scheduler::the().active_thread().m_owned_ranges.append(PageRange{ power_of_two(executable.m_writable_size), executable.m_writable_base });
 
             dbgln("[hand_over_to_loaded_executable] Droping privileges");
 

Kernel/Scheduler.cpp

@@ -110,7 +110,7 @@ namespace Kernel
         thread.m_context = context;
 
         if (thread.m_die_at_next_opportunity) {
-            // FIXME: Free stack here
+            thread.free_owned_ranges();
         } else {
             m_threads.enqueue(move(thread));
         }
@@ -152,6 +152,8 @@ namespace Kernel
 
     void Scheduler::terminate_active_thread()
     {
+        dbgln("[Scheduler::terminate_active_thread]");
+
         active_thread().m_die_at_next_opportunity = true;
         donate_my_remaining_cpu_slice();
     }

Kernel/Scheduler.hpp

@@ -88,6 +88,17 @@ namespace Kernel
         {
         }
 
+        // FIXME: RAII
+        void free_owned_ranges()
+        {
+            dbgln("[Thread::free_owned_ranges] name={}", m_name);
+
+            for (auto range : m_owned_ranges.iter()) {
+                PageAllocator::the().deallocate(range);
+            }
+            m_owned_ranges.clear();
+        }
+
         Vector<PageRange> m_owned_ranges;
 
         String m_name;

README.md

@@ -1,8 +1,9 @@
 ### TODO
 
-#### Next Version
+#### Bugs
 
-  - Free allocated segments when destroying threads
+  - We sometimes crash in `Process::create` when computing the new region.
+    This appears to be a race condition so the Scheduler is involved?
 
 #### Future features
 
@@ -16,10 +17,11 @@
 
 #### Future tweaks (Kernel)
 
-  - Free resources when threads and process are destroyed
+  - Use RAII to manage `PageRange`s
 
-  - We were allocating the stack twice; is this still accurate with
-    `posix_spawn`?
+  - Group `PageRange`s together in `PageAllocator::deallocate`.
+
+  - Provide a shortcut for `Scheduler::the().active_thread()` similar: `Process::current()`.
 
   - Lookup devices via `devno`