[kernel] Improve VMA lifecycle

The vm_area objects had a number of issues I have been running into when
working on srv.init:

- It was impossible to map a VMA, fill it, unmap it, and hand it to
  another process. Unmapping the VMA in this process would cause all the
  pages to be freed, since it was removed from its last mapping.
- If a VMA was marked with vm_flag::zero, it would be zeroed out _every
  time_ it was mapped into a vm_space.
- The vm_area_open class was leaking its page_tree nodes.

In order to fix these issues, the different VMA types all work slightly
differently now:

- Physical pages allocated for a VMA are now freed when the VMA is
  deleted, not when it is unmapped.
- A knock-on effect from the first point is that vm_area_guarded is now
  based on vm_area_open, instead of vm_area_untracked. An untracked area
  cannot free its pages, since it does not track them.
- The vm_area_open type now deletes its root page_tree node. And
  page_tree nodes will delete child nodes or free physical pages in
  their dtors.
- vm_flag::zero has been removed; pages will need to be zeroed out
  further at a higher level.
- vm_area also no longer deletes itself only on losing its last handle -
  it will only self-delete when all handles _and_ mappings are gone.

src/include/j6/flags.h

@@ -6,5 +6,5 @@ enum j6_vm_flags {
 #define VM_FLAG(name, v) j6_vm_flag_ ## name = v,
 #include "j6/tables/vm_flags.inc"
 #undef VM_FLAG
-    j6_vm_flags_MAX
+    j6_vm_flag_MAX
 };

src/include/j6/tables/vm_flags.inc

@@ -1,7 +1,6 @@
 VM_FLAG( none,            0x00000000)
 VM_FLAG( write,           0x00000001)
 VM_FLAG( exec,            0x00000002)
-VM_FLAG( zero,            0x00000010)
 VM_FLAG( contiguous,      0x00000020)
 VM_FLAG( large_pages,     0x00000100)
 VM_FLAG( huge_pages,      0x00000200)

src/include/kernel_memory.h

@@ -10,6 +10,9 @@ namespace memory {
     /// Size of a single page frame.
     constexpr size_t frame_size = 0x1000;
 
+    /// Number of bits of addressing within a page
+    constexpr size_t frame_bits = 12;
+
     /// Start of kernel memory.
     constexpr uintptr_t kernel_offset = 0xffff800000000000ull;
 

src/kernel/objects/vm_area.cpp

@@ -27,19 +27,20 @@ vm_area::add_to(vm_space *space)
     return true;
 }
 
-bool
+void
 vm_area::remove_from(vm_space *space)
 {
     m_spaces.remove_swap(space);
-    return
+    if (!m_spaces.count() &&
-        !m_spaces.count() &&
+        check_signal(j6_signal_no_handles))
-        !(m_flags && vm_flags::mmio);
+        delete this;
 }
 
 void
 vm_area::on_no_handles()
 {
     kobject::on_no_handles();
+    if (!m_spaces.count())
         delete this;
 }
 
@@ -66,15 +67,24 @@ vm_area_fixed::vm_area_fixed(uintptr_t start, size_t size, vm_flags flags) :
 {
 }
 
-vm_area_fixed::~vm_area_fixed() {}
+vm_area_fixed::~vm_area_fixed()
+{
+    if (m_flags && vm_flags::mmio)
+        return;
 
-size_t vm_area_fixed::resize(size_t size)
+    size_t pages = memory::page_count(m_size);
+    frame_allocator::get().free(m_start, pages);
+}
+
+size_t
+vm_area_fixed::resize(size_t size)
 {
     // Not resizable
     return m_size;
 }
 
-bool vm_area_fixed::get_page(uintptr_t offset, uintptr_t &phys)
+bool
+vm_area_fixed::get_page(uintptr_t offset, uintptr_t &phys)
 {
     if (offset > m_size)
         return false;
@@ -83,13 +93,15 @@ bool vm_area_fixed::get_page(uintptr_t offset, uintptr_t &phys)
     return true;
 }
 
-
 vm_area_untracked::vm_area_untracked(size_t size, vm_flags flags) :
     vm_area {size, flags}
 {
 }
 
-vm_area_untracked::~vm_area_untracked() {}
+vm_area_untracked::~vm_area_untracked()
+{
+    kassert(false, "An untracked VMA's pages cannot be reclaimed, leaking memory");
+}
 
 bool
 vm_area_untracked::get_page(uintptr_t offset, uintptr_t &phys)
@@ -115,7 +127,11 @@ vm_area_open::vm_area_open(size_t size, vm_flags flags) :
 {
 }
 
-vm_area_open::~vm_area_open() {}
+vm_area_open::~vm_area_open()
+{
+    // the page_tree will free its pages when deleted
+    delete m_mapped;
+}
 
 bool
 vm_area_open::get_page(uintptr_t offset, uintptr_t &phys)
@@ -128,7 +144,7 @@ vm_area_guarded::vm_area_guarded(uintptr_t start, size_t buf_pages, size_t size,
     m_start {start},
     m_pages {buf_pages},
     m_next {memory::frame_size},
-    vm_area_untracked {size, flags}
+    vm_area_open {size, flags}
 {
 }
 
@@ -164,6 +180,5 @@ vm_area_guarded::get_page(uintptr_t offset, uintptr_t &phys)
     if ((offset >> 12) % (m_pages+1) == 0)
         return false;
 
-    return vm_area_untracked::get_page(offset, phys);
+    return vm_area_open::get_page(offset, phys);
 }
-

src/kernel/objects/vm_area.h

@@ -52,9 +52,7 @@ public:
 
     /// Track that this area was removed frm a vm_space
     /// \arg space  The space that is removing this area
-    /// \returns    True if the removing space should free the pages
+    virtual void remove_from(vm_space *space);
-    ///             mapped for this area
-    virtual bool remove_from(vm_space *space);
 
     /// Change the virtual size of the memory area. This may cause
     /// deallocation if the new size is smaller than the current size.
@@ -141,9 +139,8 @@ public:
 
 
 /// Area split into standard-sized segments, separated by guard pages.
-/// Based on vm_area_untracked, can not be shared.
 class vm_area_guarded :
-    public vm_area_untracked
+    public vm_area_open
 {
 public:
     /// Constructor.

src/kernel/page_tree.cpp

@@ -4,10 +4,10 @@
 #include "kernel_memory.h"
 #include "page_tree.h"
 
-// Page tree levels map the following parts of a pagewise offset. Note the xxx
+// Page tree levels map the following parts of an offset. Note the xxx part of
-// are not part of the offset but represent the bits added for the actual virtual
+// the offset but represent the bits of the actual sub-page virtual address.
-// address. (Also note that level 0's entries are physical page addrs, the rest
+// (Also note that level 0's entries are physical page addrs, the rest map
-// map other page_tree nodes)
+// other page_tree nodes)
 //
 // Level 0: 0000 0000 0003 fxxx    64 pages / 256 KiB
 // Level 1: 0000 0000 00fc 0xxx    4K pages /  16 MiB -- 24-bit addressing
@@ -36,6 +36,20 @@ page_tree::page_tree(uint64_t base, uint8_t level) :
     kutil::memset(m_entries, 0, sizeof(m_entries));
 }
 
+page_tree::~page_tree()
+{
+    if (m_level) {
+        for (auto &e : m_entries)
+            delete e.child;
+    } else {
+        auto &fa = frame_allocator::get();
+        for (auto &e : m_entries) {
+            if (e.entry & 1)
+                fa.free(e.entry & ~0xfffull, 1);
+        }
+    }
+}
+
 bool
 page_tree::contains(uint64_t offset, uint8_t &index) const
 {

src/kernel/page_tree.h

@@ -23,6 +23,8 @@ public:
     /// \returns     True if a page was found
     static bool find_or_add(page_tree * &root, uint64_t offset, uintptr_t &page);
 
+    ~page_tree();
+
 private:
     page_tree(uint64_t base, uint8_t level);
 

src/kernel/vm_space.cpp

@@ -45,11 +45,8 @@ vm_space::vm_space() :
 
 vm_space::~vm_space()
 {
-    for (auto &a : m_areas) {
+    for (auto &a : m_areas)
-        bool free = a.area->remove_from(this);
+        remove_area(a.area);
-        clear(*a.area, 0, memory::page_count(a.area->size()), free);
-        a.area->handle_release();
-    }
 
     kassert(!is_kernel(), "Kernel vm_space destructor!");
     if (active())
@@ -76,15 +73,21 @@ vm_space::add(uintptr_t base, vm_area *area)
     return true;
 }
 
+void
+vm_space::remove_area(vm_area *area)
+{
+    area->remove_from(this);
+    clear(*area, 0, memory::page_count(area->size()));
+    area->handle_release();
+}
+
 bool
 vm_space::remove(vm_area *area)
 {
     for (auto &a : m_areas) {
         if (a.area == area) {
-            bool free = area->remove_from(this);
+            remove_area(area);
-            clear(*area, 0, memory::page_count(area->size()), free);
             m_areas.remove(a);
-            area->handle_release();
             return true;
         }
     }
@@ -277,10 +280,6 @@ vm_space::handle_fault(uintptr_t addr, fault_type fault)
     if (!area->get_page(offset, phys_page))
         return false;
 
-    void *mem = memory::to_virtual<void>(phys_page);
-    if (area->flags() && vm_flags::zero)
-        kutil::memset(mem, 0, memory::frame_size);
-
     page_in(*area, offset, phys_page, 1);
     return true;
 }

src/kernel/vm_space.h

@@ -107,7 +107,6 @@ public:
 
 private:
     friend class vm_area;
-    friend class vm_mapper_multi;
 
     /// Find a given VMA in this address space
     bool find_vma(const vm_area &vma, uintptr_t &base) const;
@@ -118,6 +117,9 @@ private:
     /// Copy a range of mappings from the given address space 
     void copy_from(const vm_space &source, const vm_area &vma);
 
+    /// Remove an area's mappings from this space
+    void remove_area(vm_area *area);
+
     bool m_kernel;
     page_table *m_pml4;