Turn off local APIC timer

This slowed down drawing to video in QEMU/VirtualBox but greatly sped up
doing so on real hardware (no longer reading from video memory).
Despite the speed-up, it is still quite slow though.

README.md

@@ -1,4 +1,4 @@
-= HOS System Requirements
+# HOS System Requirements
 
 HOS requires:
 
@@ -6,15 +6,24 @@ HOS requires:
 * A UEFI system.
 * 128MB of RAM.
 
-= Running HOS
+# Building HOS
 
-== Physical Machine
+From the top level of the repository, run:
+
+    ./rscons
+
+If you get errors about missing packages, install the packages and then re-run
+the build command.
+
+# Running HOS
+
+## Physical Machine
 
 To install HOS to an EFI partition, follow these steps:
 
-1. Copy build/hello/BOOTX64.EFI to EFI partition EFI/HOS.EFI. For example, from Linux:
+1. Copy build/hello/HOS.EFI to EFI partition EFI/HOS.EFI. For example, from Linux:
 
-    sudo cp build/hello/BOOTX64.EFI /boot/efi/EFI/HOS.EFI
+    sudo cp build/hello/HOS.EFI /boot/efi/EFI/HOS.EFI
 
 2. Register EFI loader for HOS (one time only) with command:
 
@@ -40,7 +49,7 @@ with an entry like this:
 Just update the root setting for the correction location to your system's EFI
 partition.
 
-== VirtualBox
+## VirtualBox
 
 To create a VM in VirtualBox:
 

Rsconscript

@@ -58,7 +58,7 @@ class Image < Builder
       system(*%W[mformat -i #{@target} ::])
       system(*%W[mmd -i #{@target} ::/EFI])
       system(*%W[mmd -i #{@target} ::/EFI/BOOT])
-      system(*%W[mcopy -i #{@target} #{@sources.first} ::/EFI/BOOT])
+      system(*%W[mcopy -i #{@target} #{@sources.first} ::/EFI/BOOT/BOOTX64.EFI])
       partition_contents = File.binread(@target)
       disk_image = empty_mb + partition_contents + empty_mb
       File.binwrite(@target, disk_image)
@@ -69,6 +69,18 @@ class Image < Builder
   end
 end
 
+class CheckThreadLocal < Builder
+  def run(options)
+    map_file = File.binread(@sources.first)
+    if map_file =~ /\.(tdata|tbss)\b/
+      $stderr.puts "Error: found thread-local data in #{@sources.first}"
+      false
+    else
+      true
+    end
+  end
+end
+
 class FontGen < Builder
   def run(options)
     if @command
@@ -88,11 +100,39 @@ end
 
 hulk_env = env "hulk", use: %w[ldc2 x86_64-elf-gcc] do |env|
   env.add_builder(FontGen)
+  env.add_builder(CheckThreadLocal)
   env.FontGen("^/src/hulk/kfont.d", "font/Hack-Regular.ttf",
               "fontgen" => fontgen_env.expand("^/fontgen"))
   env["sources"] = glob("src/hulk/**/*.d")
   env["sources"] << "^/src/hulk/kfont.d"
-  env["DFLAGS"] += %w[-mtriple=x86_64-unknown-elf --betterC -release -O3 --wi --enable-cross-module-inlining -code-model=large]
+  cpu_attrs = %w[
+    -avx
+    -avx2
+    -avx512bf16
+    -avx512bitalg
+    -avx512bw
+    -avx512cd
+    -avx512dq
+    -avx512er
+    -avx512f
+    -avx512ifma
+    -avx512pf
+    -avx512vbmi
+    -avx512vbmi2
+    -avx512vl
+    -avx512vnni
+    -avx512vp2intersect
+    -avx512vpopcntdq
+    -sse
+    -sse-unaligned-mem
+    -sse2
+    -sse3
+    -sse4.1
+    -sse4.2
+    -sse4a
+    -ssse3
+  ]
+  env["DFLAGS"] += %W[-mtriple=x86_64-unknown-elf -mattr=#{cpu_attrs.join(",")} --betterC -release -O3 --wi --enable-cross-module-inlining -code-model=large]
   env["D_IMPORT_PATH"] += %w[src]
   env["D_IMPORT_PATH"] << env.expand("^/src")
   env["LD"] = "x86_64-elf-gcc"
@@ -102,6 +142,7 @@ hulk_env = env "hulk", use: %w[ldc2 x86_64-elf-gcc] do |env|
   env["OBJCOPY"] = "x86_64-elf-objcopy"
   env.Program("^/hulk.elf", "${sources}")
   env.produces("^/hulk.elf", "^/hulk.elf.map")
+  env.CheckThreadLocal(:hulk_map_check, "^/hulk.elf.map")
   env.depends("^/hulk.elf", "src/hulk/hulk.ld")
   env["SIZE"] = "x86_64-elf-size"
   env.Size("^/hulk.size", "^/hulk.elf")
@@ -118,6 +159,7 @@ end
 
 hello_env = env "hello", use: %w[ldc2 x86_64-w64-mingw32-gcc] do |env|
   env.add_builder(Image)
+  env.add_builder(CheckThreadLocal)
   env.add_builder(HulkBinObj)
   env["sources"] = glob("src/hello/**/*.d")
   env["sources"] += glob("uefi-d/source/**/*.d")
@@ -131,19 +173,20 @@ hello_env = env "hello", use: %w[ldc2 x86_64-w64-mingw32-gcc] do |env|
   env["LDFLAGS"] += %w[-nostdlib -Wl,-dll -shared -Wl,--subsystem,10 -e efi_main -Wl,-Map,${_TARGET}.map]
   env["LDCMD"] = %w[${LD} -o ${_TARGET} ${LDFLAGS} ${_SOURCES} ${LIBDIRPREFIX}${LIBPATH} ${LIBLINKPREFIX}${LIBS}]
   env["OBJDUMP"] = "x86_64-w64-mingw32-objdump"
-  env.Program("^/BOOTX64.EFI", "${sources}")
+  env.Program("^/HOS.EFI", "${sources}")
-  env.produces("^/BOOTX64.EFI", "^/BOOTX64.EFI.map")
+  env.produces("^/HOS.EFI", "^/HOS.EFI.map")
+  env.CheckThreadLocal(:hos_map_check, "^/HOS.EFI.map")
   env["SIZE"] = "x86_64-w64-mingw32-size"
-  env.Size("^/BOOTX64.size", "^/BOOTX64.EFI")
+  env.Size("^/HOS.size", "^/HOS.EFI")
-  env.Disassemble("^/BOOTX64.txt", "^/BOOTX64.EFI")
+  env.Disassemble("^/HOS.txt", "^/HOS.EFI")
-  env.Image("^/HOS.img", "^/BOOTX64.EFI")
+  env.Image("^/HOS.img", "^/HOS.EFI")
 end
 
 task "run", desc: "Run HOS in QEMU" do
   Dir.mktmpdir do |tmpdir|
     img = hello_env.expand("^/HOS.img")
     FileUtils.cp(img, tmpdir)
-    sh %W[qemu-system-x86_64 -bios OVMF.fd -drive file=#{tmpdir}/HOS.img,format=raw -device qemu-xhci -device usb-tablet]
+    sh %W[qemu-system-x86_64 -cpu max -bios OVMF.fd -drive file=#{tmpdir}/HOS.img,format=raw -device qemu-xhci -device usb-tablet]
   end
 end
 

src/hello/hello.d

@@ -13,7 +13,9 @@ import hulk.memory;
 import ldc.llvmasm;
 
 __gshared EFI_SYSTEM_TABLE * st;
+/** HULK binary start address, 4KB-aligned. */
 extern extern(C) __gshared ubyte _hulk_bin_start;
+/** HULK binary end address, 4KB-aligned. */
 extern extern(C) __gshared ubyte _hulk_bin_end;
 private align(4096) __gshared ubyte[1024 * 1024] scratch_buffer;
 /** Index to the memory map region being used to allocate page tables. */
@@ -39,14 +41,9 @@ private ulong hulk_bin_size()
     return cast(ulong)&_hulk_bin_end - cast(ulong)&_hulk_bin_start;
 }
 
-private ulong hulk_total_size()
-{
-    return cast(ulong)hulk_header().total_size;
-}
-
 private ulong hulk_bss_size()
 {
-    return hulk_total_size() - hulk_bin_size();
+    return cast(ulong)hulk_header().hulk_bss_size;
 }
 
 private ulong hulk_stack_size()
@@ -146,7 +143,7 @@ private bool set_graphics_mode()
 /**
  * Walk the EFI memory map and translate it to the HULK bootinfo format.
  */
-private void get_memory_map(ulong * bss_phys, ulong * stack_phys, ulong * physical_address_limit, UINTN * memory_map_key)
+private void get_memory_map(ulong * physical_address_limit, UINTN * memory_map_key)
 {
     immutable static ubyte[] efi_to_hulk_memory_map_type = [
         BootInfo.MemoryRegion.Type.Reserved, // EfiReservedMemoryType
@@ -184,6 +181,8 @@ private void get_memory_map(ulong * bss_phys, ulong * stack_phys, ulong * physic
     size_t count;
     bool found_bss;
     bool found_stack;
+    bool found_fb_buffer1;
+    const(size_t) fb_size = (bootinfo().fb.stride * bootinfo().fb.height * 4u + PAGE_SIZE - 1u) & ~(PAGE_SIZE - 1u);
     for (count = 0u; count < n_entries; count++)
     {
         if (count > bootinfo().memory_map.length)
@@ -215,7 +214,7 @@ private void get_memory_map(ulong * bss_phys, ulong * stack_phys, ulong * physic
             (descriptor.Type == EfiConventionalMemory) &&
             (bootinfo().memory_map[count].size >= hulk_bss_size()))
         {
-            *bss_phys = bootinfo().memory_map[count].base;
+            bootinfo().bss_phys = bootinfo().memory_map[count].base;
             bootinfo().memory_map[count].base += hulk_bss_size();
             bootinfo().memory_map[count].size -= hulk_bss_size();
             found_bss = true;
@@ -224,14 +223,23 @@ private void get_memory_map(ulong * bss_phys, ulong * stack_phys, ulong * physic
             (descriptor.Type == EfiConventionalMemory) &&
             (bootinfo().memory_map[count].size >= hulk_stack_size()))
         {
-            *stack_phys = bootinfo().memory_map[count].base;
+            bootinfo().stack_phys = bootinfo().memory_map[count].base;
             bootinfo().memory_map[count].base += hulk_stack_size();
             bootinfo().memory_map[count].size -= hulk_stack_size();
             found_stack = true;
         }
+        if ((!found_fb_buffer1) &&
+            (descriptor.Type == EfiConventionalMemory) &&
+            (bootinfo().memory_map[count].size >= fb_size))
+        {
+            bootinfo().fb_buffer1_phys = bootinfo().memory_map[count].base;
+            bootinfo().memory_map[count].base += fb_size;
+            bootinfo().memory_map[count].size -= fb_size;
+            found_fb_buffer1 = true;
+        }
     }
     bootinfo().memory_map_count = count;
-    if ((!found_bss) || (!found_stack))
+    if ((!found_bss) || (!found_stack) || (!found_fb_buffer1))
     {
         for (;;) {}
     }
@@ -382,7 +390,7 @@ private bool map2m(ulong source_page, ulong dest_page, PageTable * pt_base)
  *
  * @param pt_base Page table base address.
  */
-private bool map_hulk(PageTable * pt_base, ulong bss_phys, ulong stack_phys)
+private bool map_hulk(PageTable * pt_base)
 {
     /* Map HULK bin region. */
     ulong virt = hulk_virt_base_address();
@@ -392,13 +400,13 @@ private bool map_hulk(PageTable * pt_base, ulong bss_phys, ulong stack_phys)
     }
     /* Map HULK bss region. */
     virt += hulk_bin_size();
-    if (!map4kregion(virt, bss_phys, hulk_bss_size(), pt_base))
+    if (!map4kregion(virt, bootinfo().bss_phys, hulk_bss_size(), pt_base))
     {
         return false;
     }
     /* Map HULK stack. */
     virt = hulk_virt_stack_top() - hulk_stack_size();
-    if (!map4kregion(virt, stack_phys, hulk_stack_size(), pt_base))
+    if (!map4kregion(virt, bootinfo().stack_phys, hulk_stack_size(), pt_base))
     {
         return false;
     }
@@ -410,7 +418,7 @@ private bool map_hulk(PageTable * pt_base, ulong bss_phys, ulong stack_phys)
  *
  * @param physical_address_limit Maximum physical address to identity map.
  */
-private bool build_page_tables(ulong physical_address_limit, ulong bss_phys, ulong stack_phys)
+private bool build_page_tables(ulong physical_address_limit)
 {
     PageTable * pt_base = new_page_table();
     if (pt_base == null)
@@ -446,7 +454,7 @@ private bool build_page_tables(ulong physical_address_limit, ulong bss_phys, ulo
         return false;
     }
     /* Map HULK regions. */
-    if (!map_hulk(pt_base, bss_phys, stack_phys))
+    if (!map_hulk(pt_base))
     {
         return false;
     }
@@ -521,13 +529,11 @@ extern (C) EFI_STATUS efi_main(EFI_HANDLE image_handle, EFI_SYSTEM_TABLE * st)
         return EFI_SUCCESS;
     }
 
-    ulong bss_phys;
-    ulong stack_phys;
     ulong physical_address_limit;
     UINTN memory_map_key;
     for (;;)
     {
-        get_memory_map(&bss_phys, &stack_phys, &physical_address_limit, &memory_map_key);
+        get_memory_map(&physical_address_limit, &memory_map_key);
 
         EFI_STATUS status = st.BootServices.ExitBootServices(image_handle, memory_map_key);
         if (status == EFI_INVALID_PARAMETER)
@@ -543,14 +549,12 @@ extern (C) EFI_STATUS efi_main(EFI_HANDLE image_handle, EFI_SYSTEM_TABLE * st)
         return EFI_SUCCESS;
     }
 
-    if (!build_page_tables(physical_address_limit, bss_phys, stack_phys))
+    if (!build_page_tables(physical_address_limit))
     {
         return EFI_SUCCESS;
     }
 
     bootinfo().hulk_phys = hulk_bin_phys();
-    bootinfo().bss_phys = bss_phys;
-    bootinfo().stack_phys = stack_phys;
 
     jump_to_hulk();
 

src/hulk/acpi.d

@@ -12,7 +12,7 @@ struct acpi
     enum uint APIC_SIGNATURE = 0x43495041u;
     enum uint XSDT_SIGNATURE = 0x54445358u;
 
-    struct XsdtHeader
+    static struct XsdtHeader
     {
         uint signature;
         uint length;
@@ -23,16 +23,16 @@ struct acpi
         uint oem_revision;
         uint creator_id;
         uint creator_revision;
-    };
+    }
 
-    struct Xsdt
+    static struct Xsdt
     {
         XsdtHeader header;
         uint _pad;
         ulong[0] tables;
     }
 
-    struct MadtHeader
+    static struct MadtHeader
     {
         uint signature;
         uint length;
@@ -47,7 +47,7 @@ struct acpi
         uint flags;
     }
 
-    struct MadtEntry
+    static struct MadtEntry
     {
         ubyte type;
         ubyte length;

src/hulk/apic.d

@@ -6,25 +6,20 @@ module hulk.apic;
 import hulk.klog;
 import hulk.hurl;
 import hulk.acpi;
+import hulk.idt;
 
 struct apic
 {
+    private enum uint PERIODIC_MODE = 0x2_0000u;
+
     static struct ApicRegister
     {
-        private uint register;
+        public uint value;
+        alias value this;
+
         private ubyte[12] _padding;
-
-        public uint read() const
-        {
-            return register;
     }
 
-        public void write(uint value)
-        {
-            register = value;
-        }
-    };
-
     static struct ApicRegisters
     {
         ApicRegister[2] _reserved0;
@@ -55,15 +50,55 @@ struct apic
         ApicRegister current_count;
         ApicRegister[4] _reserved3;
         ApicRegister divide_configuration;
-    };
+    }
+
+    static struct IoApicRegisters
+    {
+        ApicRegister address;
+        ApicRegister data;
+    }
+
+    private static __gshared ApicRegisters * apic_registers;
+    private static __gshared IoApicRegisters * io_apic_registers;
 
     public static void initialize()
     {
-        ApicRegisters * apic_registers =
+        apic_registers = cast(ApicRegisters *)acpi.apic_address;
-            cast(ApicRegisters *)acpi.apic_address;
+        io_apic_registers = cast(IoApicRegisters *)0xFEC0_0000u;
         hurl.map(cast(ulong)apic_registers, cast(ulong)apic_registers,
                 PT_WRITABLE | PT_WRITE_THROUGH | PT_DISABLE_CACHE | PT_NO_EXECUTE);
-        klog.writefln("LAPIC ID: 0x%08x", apic_registers.lapic_id.read());
+        hurl.map(cast(ulong)io_apic_registers, cast(ulong)io_apic_registers,
-        klog.writefln("LAPIC version: 0x%08x", apic_registers.lapic_version.read());
+                PT_WRITABLE | PT_WRITE_THROUGH | PT_DISABLE_CACHE | PT_NO_EXECUTE);
+        klog.writefln("LAPIC ID: 0x%08x", apic_registers.lapic_id.value);
+        klog.writefln("LAPIC version: 0x%08x", apic_registers.lapic_version.value);
+        /* Enable local APIC to receive interrupts and set spurious interrupt
+         * vector to 0xFF. */
+        apic_registers.spurious_interrupt_vector.value = 0x1FFu;
+//        apic_registers.lvt_timer.value = idt.INT_LAPIC_TIMER | PERIODIC_MODE;
+        apic_registers.lvt_timer.value = idt.INT_LAPIC_TIMER;
+        apic_registers.lvt_lint[0].value = idt.INT_LAPIC_LINT0;
+        apic_registers.lvt_lint[1].value = idt.INT_LAPIC_LINT1;
+//        apic_registers.divide_configuration.value = 3u;
+        configure_io_apic_irq(8u, 0x48u);
+    }
+
+    private static void configure_io_apic_irq(size_t io_apic_irq, size_t interrupt_id)
+    {
+        ulong entry = interrupt_id;
+        io_apic_registers.address.value = cast(uint)(0x10u + io_apic_irq * 2u);
+        io_apic_registers.data.value = entry & 0xFFFF_FFFFu;
+        io_apic_registers.address.value = cast(uint)(0x10u + io_apic_irq * 2u + 1u);
+        io_apic_registers.data.value = entry >> 32u;
+    }
+
+    public static void eoi()
+    {
+        apic_registers.eoi.value = 0u;
+    }
+
+    public static void isr(ulong vector)
+    {
+        klog.writefln("APIC ISR 0x%x", vector);
+        eoi();
     }
 }

src/hulk/bootinfo.d

@@ -69,6 +69,9 @@ struct BootInfo
     /* Physical address of stack while jumping to HULK. */
     ulong stack_phys;
 
+    /* Physical address of framebuffer buffer1. */
+    ulong fb_buffer1_phys;
+
     /* Physical address of ACPI XSDT table. */
     ulong acpi_xsdt_phys;
 }

src/hulk/cpu.d

@@ -5,6 +5,57 @@ module hulk.cpu;
 
 import ldc.llvmasm;
 
+/** CR0 bits. @{ */
+enum ulong CR0_PE = 0x1u;
+enum ulong CR0_MP = 0x2u;
+enum ulong CR0_EM = 0x4u;
+enum ulong CR0_TS = 0x8u;
+enum ulong CR0_ET = 0x10u;
+enum ulong CR0_NE = 0x20u;
+enum ulong CR0_WP = 0x1_0000u;
+enum ulong CR0_AM = 0x4_0000u;
+enum ulong CR0_NW = 0x2000_0000u;
+enum ulong CR0_CD = 0x4000_0000u;
+enum ulong CR0_PG = 0x8000_0000u;
+/** @} */
+
+/** CR4 bits. @{ */
+enum ulong CR4_VME = 0x1u;
+enum ulong CR4_PVI = 0x2u;
+enum ulong CR4_TSD = 0x4u;
+enum ulong CR4_DE = 0x8u;
+enum ulong CR4_PSE = 0x10u;
+enum ulong CR4_PAE = 0x20u;
+enum ulong CR4_MCE = 0x40u;
+enum ulong CR4_PGE = 0x80u;
+enum ulong CR4_PCE = 0x100u;
+enum ulong CR4_OSFXSR = 0x200u;
+enum ulong CR4_OSXMMEXCPT = 0x400u;
+enum ulong CR4_UMIP = 0x800u;
+enum ulong CR4_VMXE = 0x2000u;
+enum ulong CR4_SMXE = 0x4000u;
+enum ulong CR4_FSGSBASE = 0x1_0000u;
+enum ulong CR4_PCIDE = 0x2_0000u;
+enum ulong CR4_OSXSAVE = 0x4_0000u;
+enum ulong CR4_SMEP = 0x10_0000u;
+enum ulong CR4_SMAP = 0x20_0000u;
+enum ulong CR4_PKE = 0x40_0000u;
+enum ulong CR4_CET = 0x80_0000u;
+enum ulong CR4_PKS = 0x100_0000u;
+/** @} */
+
+/** XCR0 bits. @{ */
+enum ulong XCR0_X87 = 0x1u;
+enum ulong XCR0_SSE = 0x2u;
+enum ulong XCR0_AVX = 0x4u;
+enum ulong XCR0_BNDREG = 0x8u;
+enum ulong XCR0_BNDSCR = 0x10u;
+enum ulong XCR0_OPMASK = 0x20u;
+enum ulong XCR0_ZMM_HI256 = 0x40u;
+enum ulong XCR0_HI16_ZMM = 0x80u;
+enum ulong XCR0_PKRU = 0x100u;
+/** @} */
+
 /** MSR register numbers. @{ */
 enum uint MSR_EFER = 0xC000_0080u;
 /** @} */
@@ -20,6 +71,70 @@ enum uint EFER_FFXSR = 0x4000u;
 enum uint EFER_TCE = 0x8000u;
 /** @} */
 
+/** CPUID 1 bits. @{ */
+enum uint CPUID_1_EDX_FPU = 0x1u;
+enum uint CPUID_1_EDX_VME = 0x2u;
+enum uint CPUID_1_EDX_DE = 0x4u;
+enum uint CPUID_1_EDX_PSE = 0x8u;
+enum uint CPUID_1_EDX_TSC = 0x10u;
+enum uint CPUID_1_EDX_MSR = 0x20u;
+enum uint CPUID_1_EDX_PAE = 0x40u;
+enum uint CPUID_1_EDX_MCE = 0x80u;
+enum uint CPUID_1_EDX_CX8 = 0x100u;
+enum uint CPUID_1_EDX_APIC = 0x200u;
+enum uint CPUID_1_EDX_SEP = 0x800u;
+enum uint CPUID_1_EDX_MTRR = 0x1000u;
+enum uint CPUID_1_EDX_PGE = 0x2000u;
+enum uint CPUID_1_EDX_MCA = 0x4000u;
+enum uint CPUID_1_EDX_CMOV = 0x8000u;
+enum uint CPUID_1_EDX_PAT = 0x1_0000u;
+enum uint CPUID_1_EDX_PSE36 = 0x2_0000u;
+enum uint CPUID_1_EDX_PSN = 0x4_0000u;
+enum uint CPUID_1_EDX_CLFSH = 0x8_0000u;
+enum uint CPUID_1_EDX_DS = 0x20_0000u;
+enum uint CPUID_1_EDX_ACPI = 0x40_0000u;
+enum uint CPUID_1_EDX_MMX = 0x80_0000u;
+enum uint CPUID_1_EDX_FXSR = 0x100_0000u;
+enum uint CPUID_1_EDX_SSE = 0x200_0000u;
+enum uint CPUID_1_EDX_SSE2 = 0x400_0000u;
+enum uint CPUID_1_EDX_SS = 0x800_0000u;
+enum uint CPUID_1_EDX_HTT = 0x1000_0000u;
+enum uint CPUID_1_EDX_TM = 0x2000_0000u;
+enum uint CPUID_1_EDX_IA64 = 0x4000_0000u;
+enum uint CPUID_1_EDX_PBE = 0x8000_0000u;
+enum uint CPUID_1_ECX_SSE3 = 0x1u;
+enum uint CPUID_1_ECX_PCLMULQDQ = 0x2u;
+enum uint CPUID_1_ECX_DTES64 = 0x4u;
+enum uint CPUID_1_ECX_MONITOR = 0x8u;
+enum uint CPUID_1_ECX_DSCPL = 0x10u;
+enum uint CPUID_1_ECX_VMX = 0x20u;
+enum uint CPUID_1_ECX_SMX = 0x40u;
+enum uint CPUID_1_ECX_EST = 0x80u;
+enum uint CPUID_1_ECX_TM2 = 0x100u;
+enum uint CPUID_1_ECX_SSSE3 = 0x200u;
+enum uint CPUID_1_ECX_CNXTID = 0x400u;
+enum uint CPUID_1_ECX_SDBG = 0x800u;
+enum uint CPUID_1_ECX_FMA = 0x1000u;
+enum uint CPUID_1_ECX_CX16 = 0x2000u;
+enum uint CPUID_1_ECX_XTPR = 0x4000u;
+enum uint CPUID_1_ECX_PDCM = 0x8000u;
+enum uint CPUID_1_ECX_PCID = 0x2_0000u;
+enum uint CPUID_1_ECX_DCA = 0x4_0000u;
+enum uint CPUID_1_ECX_SSE41 = 0x8_0000u;
+enum uint CPUID_1_ECX_SSE42 = 0x10_0000u;
+enum uint CPUID_1_ECX_X2APIC = 0x20_0000u;
+enum uint CPUID_1_ECX_MOVBE = 0x40_0000u;
+enum uint CPUID_1_ECX_POPCNT = 0x80_0000u;
+enum uint CPUID_1_ECX_TSCDEADLINE = 0x100_0000u;
+enum uint CPUID_1_ECX_AES = 0x200_0000u;
+enum uint CPUID_1_ECX_XSAVE = 0x400_0000u;
+enum uint CPUID_1_ECX_OSXSAVE = 0x800_0000u;
+enum uint CPUID_1_ECX_AVX = 0x1000_0000u;
+enum uint CPUID_1_ECX_F16C = 0x2000_0000u;
+enum uint CPUID_1_ECX_RDRND = 0x4000_0000u;
+enum uint CPUID_1_ECX_HYPERVISOR = 0x8000_0000u;
+/** @} */
+
 void cli()
 {
     __asm("cli", "");
@@ -40,6 +155,11 @@ ulong read_cr0()
     return __asm!ulong("mov %cr0, %rax", "={rax}");
 }
 
+void write_cr0(ulong v)
+{
+    __asm("mov $0, %cr0", "r", v);
+}
+
 ulong read_cr2()
 {
     return __asm!ulong("mov %cr2, %rax", "={rax}");
@@ -60,6 +180,11 @@ ulong read_cr4()
     return __asm!ulong("mov %cr4, %rax", "={rax}");
 }
 
+void write_cr4(ulong v)
+{
+    __asm("mov $0, %cr4", "r", v);
+}
+
 ulong rdmsr(uint msr)
 {
     return __asm!ulong(`
@@ -73,6 +198,19 @@ void wrmsr(uint msr, ulong value)
     __asm(`wrmsr`, "{ecx},{edx},{eax}", msr, value >> 32u, value);
 }
 
+ulong xgetbv(uint xcr)
+{
+    return __asm!ulong(`
+            xgetbv
+            shl $$32, %rdx
+            or %rdx, %rax`, "={rax},{ecx},~{rdx}", xcr);
+}
+
+void xsetbv(uint xcr, ulong value)
+{
+    __asm(`xsetbv`, "{ecx},{edx},{eax}", xcr, value >> 32u, value);
+}
+
 ubyte in8(ushort ioaddr)
 {
     return __asm!ubyte("inb %dx, %al", "={al},{dx}", ioaddr);
@@ -102,3 +240,8 @@ void out32(ushort ioaddr, ulong v)
 {
     __asm("outl %eax, %dx", "{dx},{eax}", ioaddr, v);
 }
+
+void cpuid1(uint * ebx, uint * ecx, uint * edx)
+{
+    __asm("cpuid", "=*{ebx},=*{ecx},=*{edx},{eax}", ebx, ecx, edx, 1u);
+}

src/hulk/fb.d

@@ -13,8 +13,11 @@ import hulk.kfont;
  */
 struct fb
 {
-    /** Frame buffer base address. */
+    /** Device frame buffer base address. */
-    private __gshared static uint * m_buffer;
+    private __gshared static uint * m_device_buffer;
+
+    /** Frame buffer 1 base address. */
+    private __gshared static uint * m_buffer1;
 
     /** Frame buffer width. */
     private __gshared static uint m_width;
@@ -47,14 +50,16 @@ struct fb
     /**
      * Initialize a frame buffer.
      *
-     * @param buffer Frame buffer base address.
+     * @param device_buffer Device frame buffer base address.
+     * @param buffer1 Frame buffer 1 base address.
      * @param width Frame buffer width.
      * @param height Frame buffer height.
      * @param stride Frame buffer stride.
      */
-    static void initialize(uint * buffer, uint width, uint height, uint stride)
+    static void initialize(uint * device_buffer, uint * buffer1, uint width, uint height, uint stride)
     {
-        m_buffer = buffer;
+        m_device_buffer = device_buffer;
+        m_buffer1 = buffer1;
         m_width = width;
         m_height = height;
         m_stride = stride;
@@ -68,7 +73,8 @@ struct fb
      */
     static void clear(uint color = 0u)
     {
-        memset32(m_buffer, color, m_buffer_size);
+        memset32(m_buffer1, color, m_buffer_size);
+        memset32(m_device_buffer, color, m_buffer_size);
     }
 
     /**
@@ -85,7 +91,8 @@ struct fb
         for (size_t iy = 0u; iy < height; iy++)
         {
             size_t buffer_index = buffer_index(x, y);
-            memset32(&m_buffer[buffer_index], color, width);
+            memset32(&m_buffer1[buffer_index], color, width);
+            memset32(&m_device_buffer[buffer_index], color, width);
             y++;
         }
     }
@@ -101,7 +108,8 @@ struct fb
     static void hline(size_t x, size_t y, size_t width, uint color)
     {
         size_t buffer_index = buffer_index(x, y);
-        memset32(&m_buffer[buffer_index], color, width);
+        memset32(&m_buffer1[buffer_index], color, width);
+        memset32(&m_device_buffer[buffer_index], color, width);
     }
 
     /**
@@ -117,7 +125,8 @@ struct fb
         size_t buffer_index = buffer_index(x, y);
         for (size_t iy = 0u; iy < height; iy++)
         {
-            m_buffer[buffer_index] = color;
+            m_buffer1[buffer_index] = color;
+            m_device_buffer[buffer_index] = color;
             buffer_index -= m_stride;
         }
     }
@@ -153,17 +162,19 @@ struct fb
         size_t bitmap_index;
         for (size_t iy = 0u; iy < height; iy++)
         {
+            size_t row_buffer_index = buffer_index(x, y);
             for (size_t ix = 0u; ix < width; ix++)
             {
-                size_t buffer_index = buffer_index(x + ix, y);
+                size_t buffer_index = row_buffer_index + ix;
                 ubyte alpha = alpha_bitmap[bitmap_index];
-                uint current_color = m_buffer[buffer_index];
+                uint current_color = m_buffer1[buffer_index];
                 uint in_color_scaled = scale_color(color, alpha);
                 uint old_color_scaled = scale_color(current_color, 255u - alpha);
                 uint new_color = old_color_scaled + in_color_scaled;
-                m_buffer[buffer_index] = new_color;
+                m_buffer1[buffer_index] = new_color;
                 bitmap_index++;
             }
+            memcpy32(&m_device_buffer[row_buffer_index], &m_buffer1[row_buffer_index], width);
             y--;
         }
     }
@@ -202,13 +213,15 @@ struct fb
         size_t bitmap_index;
         for (size_t iy = 0u; iy < height; iy++)
         {
+            size_t row_buffer_index = buffer_index(x, y);
             for (size_t ix = 0u; ix < width; ix++)
             {
-                size_t buffer_index = buffer_index(x + ix, y);
+                size_t buffer_index = row_buffer_index + ix;
                 ubyte alpha = alpha_bitmap[bitmap_index];
-                m_buffer[buffer_index] = (alpha << 16u) | (alpha << 8u) | alpha;
+                m_buffer1[buffer_index] = (alpha << 16u) | (alpha << 8u) | alpha;
                 bitmap_index++;
             }
+            memcpy32(&m_device_buffer[row_buffer_index], &m_buffer1[row_buffer_index], width);
             y--;
         }
     }
@@ -239,8 +252,13 @@ struct fb
      */
     static void copy_rows_up(size_t y, size_t height, size_t offset)
     {
-        memcpy32(&m_buffer[buffer_index(0u, y + height + offset - 1u)],
+        size_t dest_buffer_index = buffer_index(0u, y + height + offset - 1u);
-                 &m_buffer[buffer_index(0u, y + height - 1u)],
+        size_t src_buffer_index = buffer_index(0u, y + height - 1u);
+        memcpy32(&m_buffer1[dest_buffer_index],
+                 &m_buffer1[src_buffer_index],
+                 (m_stride * height));
+        memcpy32(&m_device_buffer[dest_buffer_index],
+                &m_buffer1[dest_buffer_index],
                 (m_stride * height));
     }
 

src/hulk/header.d

@@ -11,14 +11,14 @@ import hulk.bootinfo;
 struct HulkHeader
 {
     /**
-     * Total size of memory to be mapped.
+     * HULK BSS size.
+     *
+     * The OS loader must map memory to a zeroed range of this size.
      *
-     * The OS loader must ensure that virtual memory is mapped for this entire
-     * size.
      * This value is really an integer, but is stored as a pointer because it
      * is provided by the linker and LDC does not like us to cast it.
      */
-    void * total_size;
+    void * hulk_bss_size;
 
     /** Entry point. */
     void * entry;

src/hulk/hippo.d

@@ -1,20 +1,18 @@
 /**
  * HIPPO, the HOS In-place Physical Page Organizer.
+ *
+ * HIPPO maintains a list of free physical pages "in place", meaning that the
+ * available page itself is used as the linked list entry so separate memory
+ * is not needed to keep track of the available pages.
  */
 module hulk.hippo;
 
-import hulk.header;
-import hulk.bootinfo;
-import hulk.klog;
-import hulk.linker_addresses;
-import hulk.pagetable;
-
 struct hippo
 {
     /**
      * Linked list node entry for a physical page.
      */
-    private struct PhysicalPage
+    private static struct PhysicalPage
     {
         PhysicalPage * next;
     }
@@ -27,96 +25,25 @@ struct hippo
     /**
      * Number of free physical pages.
      */
-    private static __gshared size_t n_free_pages;
+    private static __gshared size_t m_n_free_pages;
-
-    /**
-     * Physical address limit.
-     */
-    public static __gshared size_t physical_address_limit;
-
-    /**
-     * Initialize HIPPO.
-     *
-     * Ok, what we do here is iterate through all of the memory map regions in
-     * the bootinfo memory_map array, and create a linked list of all free
-     * physical pages that are available. Within the available memory regions,
-     * we have to watch out for the following items:
-     *   1) HULK binary (text + data)
-     *     - This includes the HULK header and bootinfo structures
-     *   2) HULK bss
-     *   3) HULK stack
-     *   4) Framebuffer
-     *   5) Page tables
-     *
-     * @param bootinfo HULK boot information structure.
-     */
-    public static void initialize(HulkHeader * header)
-    {
-        size_t usable_memory;
-        ulong[2][4] reserved = [
-            [header.bootinfo.hulk_phys, cast(ulong)header.total_size - LinkerAddresses.hulk_bss_size],
-            [header.bootinfo.bss_phys, LinkerAddresses.hulk_bss_size],
-            [header.bootinfo.stack_phys, header.stack_size],
-            [cast(ulong)header.bootinfo.fb.buffer, header.bootinfo.fb.height * header.bootinfo.fb.stride * 4u],
-        ];
-        for (size_t ri = 0u; ri < reserved.length; ri++)
-        {
-            reserved[ri][1] += reserved[ri][0];
-        }
-        for (size_t bii = 0u; bii < header.bootinfo.memory_map_count; bii++)
-        {
-            if ((header.bootinfo.memory_map[bii].type == BootInfo.MemoryRegion.Type.Bootloader) ||
-                (header.bootinfo.memory_map[bii].type == BootInfo.MemoryRegion.Type.Conventional))
-            {
-                ulong phys = header.bootinfo.memory_map[bii].base;
-                ulong phys_end = phys + header.bootinfo.memory_map[bii].size;
-                if (phys_end > physical_address_limit)
-                {
-                    physical_address_limit = phys_end;
-                }
-                usable_memory += header.bootinfo.memory_map[bii].size;
-                while (phys < phys_end)
-                {
-                    bool is_reserved = false;
-                    for (size_t ri = 0u; ri < reserved.length; ri++)
-                    {
-                        if ((reserved[ri][0] <= phys) && (phys < reserved[ri][1]))
-                        {
-                            is_reserved = true;
-                            break;
-                        }
-                    }
-                    if (!is_reserved)
-                    {
-                        free_page(phys);
-                    }
-                    phys += PAGE_SIZE;
-                }
-            }
-        }
-        size_t usable_kb = usable_memory >> 10u;
-        size_t usable_mb = usable_kb >> 10u;
-        size_t frac_mb = ((1000u * (usable_kb & 0x3FFu)) + 512u) >> 10u;
-        klog.writefln("Found %u.%03uMB of usable RAM", usable_mb, frac_mb);
-    }
 
     /**
      * Free a physical page.
      *
      * @param phys Physical page address.
      */
-    private static void free_page(ulong phys)
+    public static void free_page(T)(T phys)
     {
         PhysicalPage * pp = cast(PhysicalPage *)phys;
         pp.next = free_pages;
         free_pages = pp;
-        n_free_pages++;
+        m_n_free_pages++;
     }
 
     /**
      * Allocate a physical page.
      *
-     * @return Page address, or null if none available.
+     * @return Page address, or null if no pages are available.
      */
     public static void * allocate_page()
     {
@@ -125,7 +52,18 @@ struct hippo
         {
             pp = free_pages;
             free_pages = free_pages.next;
+            m_n_free_pages--;
         }
         return cast(void *)pp;
     }
+
+    /**
+     * Get the number of free pages.
+     *
+     * @return The number of free pages.
+     */
+    public @property size_t n_free_pages() const
+    {
+        return m_n_free_pages;
+    }
 }

src/hulk/hulk.d

@@ -20,12 +20,14 @@ import ldc.llvmasm;
 import hulk.pic;
 import hulk.acpi;
 import hulk.apic;
+import hulk.rtc;
+import hulk.pit;
 
-extern extern(C) __gshared ubyte _hulk_total_size;
+extern extern(C) __gshared ubyte _hulk_bss_size;
 
 @(ldc.attributes.section(".hulk_header"))
 private __gshared HulkHeader hulk_header = {
-    &_hulk_total_size, /* total_size */
+    &_hulk_bss_size, /* hulk_bss_size */
     &hulk_start, /* entry */
     16u * 1024u, /* stack_size */
     HULK_VIRTUAL_BASE_ADDRESS, /* virt_base */
@@ -33,27 +35,48 @@ private __gshared HulkHeader hulk_header = {
     HULK_VIRTUAL_FRAMEBUFFER_ADDRESS, /* virt_fb_buffer */
 };
 
+private void initialize_cpu()
+{
+    /* Enable SSE. */
+    /* Turn off CR0.EM and turn on CR0.MP. */
+    write_cr0((read_cr0() & ~CR0_EM) | CR0_MP);
+    /* Set CR4.OSFXSR and CR4.OSXMMEXCPT. */
+    write_cr4(read_cr4() | CR4_OSFXSR | CR4_OSXMMEXCPT);
+
+    /* Enable OSXSAVE. */
+    write_cr4(read_cr4() | CR4_OSXSAVE);
+
+    /* Turn on NXE (no execute enable) flag in the EFER MSR. */
+    wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
+}
+
 /**
  * HULK entry point.
  */
 void hulk_start()
 {
     cli();
+    initialize_cpu();
     gdt.initialize();
     idt.initialize();
-    fb.initialize(cast(uint *)HULK_VIRTUAL_FRAMEBUFFER_ADDRESS, hulk_header.bootinfo.fb.width, hulk_header.bootinfo.fb.height, hulk_header.bootinfo.fb.stride);
+    fb.initialize(cast(uint *)HULK_VIRTUAL_FRAMEBUFFER_ADDRESS,
+            cast(uint *)hulk_header.bootinfo.fb_buffer1_phys,
+            hulk_header.bootinfo.fb.width,
+            hulk_header.bootinfo.fb.height,
+            hulk_header.bootinfo.fb.stride);
     console.initialize();
     console.clear();
     klog.initialize();
 
     klog.writefln("Welcome to HULK, the HOS UltraLight Kernel!");
 
-    hippo.initialize(&hulk_header);
     hurl.initialize(&hulk_header);
     pci.initialize();
     pic.initialize();
     acpi.initialize(hulk_header.bootinfo.acpi_xsdt_phys);
     apic.initialize();
+    rtc.initialize();
+    pit.initialize();
     sti();
 
     for (;;)

src/hulk/hulk.ld

@@ -2,6 +2,7 @@ SECTIONS
 {
     . = 0xFFFF800000000000;
     _hulk_mem_start = .;
+    _hulk_header_start = .;
 
     .hulk_header :
     {
@@ -9,6 +10,7 @@ SECTIONS
     }
 
     . = ALIGN(4K);
+    _hulk_header_end = .;
     _hulk_text_start = .;
 
     .text :
@@ -18,6 +20,7 @@ SECTIONS
 
     . = ALIGN(4K);
     _hulk_text_end = .;
+    _hulk_rodata_start = .;
 
     .rodata :
     {
@@ -26,12 +29,17 @@ SECTIONS
 
     . = ALIGN(4K);
 
+    _hulk_rodata_end = .;
+    _hulk_data_start = .;
+
     .data :
     {
         *(.data)
     }
 
     . = ALIGN(4K);
+    _hulk_data_end = .;
+    _hulk_binary_size = . - _hulk_mem_start;
     _hulk_bss_start = .;
 
     .bss :
@@ -41,7 +49,13 @@ SECTIONS
     }
 
     . = ALIGN(4K);
-    _hulk_bss_size = . - _hulk_bss_start;
+    _hulk_bss_end = .;
     _hulk_mem_end = .;
-    _hulk_total_size = _hulk_mem_end - _hulk_mem_start;
+
+    _hulk_header_size = _hulk_header_end - _hulk_header_start;
+    _hulk_text_size = _hulk_text_end - _hulk_text_start;
+    _hulk_rodata_size = _hulk_rodata_end - _hulk_rodata_start;
+    _hulk_data_size = _hulk_data_end - _hulk_data_start;
+    _hulk_bss_size = _hulk_bss_end - _hulk_bss_start;
+    _hulk_mem_size = _hulk_mem_end - _hulk_mem_start;
 }

src/hulk/hurl.d

@@ -10,6 +10,7 @@ import hulk.cpu;
 import hulk.hippo;
 import hulk.memory;
 import hulk.klog;
+import hulk.bootinfo;
 import hulk.header;
 import hulk.linker_addresses;
 
@@ -24,31 +25,56 @@ enum ulong HULK_VIRTUAL_FRAMEBUFFER_ADDRESS = 0xFFFF_A000_0000_0000u;
 
 struct hurl
 {
+    /**
+     * Pointer to the base page table.
+     */
     private static __gshared PageTable * m_pt_base;
 
     /**
-     * Initialize HURL.
+     * Build HULK page tables.
      *
-     * @param bootinfo HULK boot information structure.
+     * @param physical_address_limit
+     *   Limit of physical memory.
      */
-    public static void initialize(HulkHeader * header)
+    private static void build_page_tables(HulkHeader * header, size_t physical_address_limit)
     {
-        /* Turn on NXE (no execute enable) flag in the EFER MSR. */
-        wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
         m_pt_base = allocate_pt();
-        size_t hulk_bin_phys_size = cast(ulong)header.total_size - LinkerAddresses.hulk_bss_size;
         /* Identity map all physical RAM. */
         map_range(0u,
                 0u,
-                hippo.physical_address_limit,
+                physical_address_limit,
                 PT_WRITABLE | PT_NO_EXECUTE);
-        /* Map HULK binary region. */
+        ulong phys_address = header.bootinfo.hulk_phys;
-        map_range(HULK_VIRTUAL_BASE_ADDRESS,
+        ulong virt_address = HULK_VIRTUAL_BASE_ADDRESS;
-                header.bootinfo.hulk_phys,
+        /* Map HULK header region. */
-                hulk_bin_phys_size,
+        map_range(virt_address,
-                PT_WRITABLE);
+                phys_address,
+                LinkerAddresses.hulk_header_size,
+                PT_NO_EXECUTE);
+        phys_address += LinkerAddresses.hulk_header_size;
+        virt_address += LinkerAddresses.hulk_header_size;
+        /* Map HULK text region. */
+        map_range(virt_address,
+                phys_address,
+                LinkerAddresses.hulk_text_size,
+                0u);
+        phys_address += LinkerAddresses.hulk_text_size;
+        virt_address += LinkerAddresses.hulk_text_size;
+        /* Map HULK rodata region. */
+        map_range(virt_address,
+                phys_address,
+                LinkerAddresses.hulk_rodata_size,
+                PT_NO_EXECUTE);
+        phys_address += LinkerAddresses.hulk_rodata_size;
+        virt_address += LinkerAddresses.hulk_rodata_size;
+        /* Map HULK data region. */
+        map_range(virt_address,
+                phys_address,
+                LinkerAddresses.hulk_data_size,
+                PT_WRITABLE | PT_NO_EXECUTE);
+        virt_address += LinkerAddresses.hulk_data_size;
         /* Map HULK BSS region. */
-        map_range(HULK_VIRTUAL_BASE_ADDRESS + hulk_bin_phys_size,
+        map_range(virt_address,
                 header.bootinfo.bss_phys,
                 LinkerAddresses.hulk_bss_size,
                 PT_WRITABLE | PT_NO_EXECUTE);
@@ -57,15 +83,141 @@ struct hurl
                 header.bootinfo.stack_phys,
                 header.stack_size,
                 PT_WRITABLE | PT_NO_EXECUTE);
-        /* Map HULK framebuffer. */
+        /* Map device framebuffer. */
         map_range(HULK_VIRTUAL_FRAMEBUFFER_ADDRESS,
                 cast(ulong)header.bootinfo.fb.buffer,
                 header.bootinfo.fb.height * header.bootinfo.fb.stride * 4u,
                 PT_WRITABLE | PT_NO_EXECUTE);
+        /* Map framebuffer buffer1. */
+        map_range(cast(ulong)header.bootinfo.fb_buffer1_phys,
+                cast(ulong)header.bootinfo.fb_buffer1_phys,
+                header.bootinfo.fb.height * header.bootinfo.fb.stride * 4u,
+                PT_WRITABLE | PT_NO_EXECUTE);
         write_cr3(cast(ulong)m_pt_base);
     }
 
-    public static void map(ulong virtual, ulong physical, ulong flags)
+    /**
+     * Reclaim the pages that were used for the bootloader page tables.
+     *
+     * @param pt
+     *   The page table to reclaim pages from.
+     * @param level
+     *   Page table level (internally used while recursing the page tables).
+     *
+     * @return Number of pages reclaimed.
+     */
+    private static size_t reclaim_bootloader_page_table_pages(PageTable * pt, size_t level = 0u)
+    {
+        map(pt, pt, PT_WRITABLE | PT_NO_EXECUTE);
+        size_t reclaimed_pages;
+        for (size_t i = 0u; i < PageTable.N_ENTRIES; i++)
+        {
+            PageTableEntry pte = (*pt)[i];
+            if (pte.present && (level < 2u) && !pte.huge)
+            {
+                /* For the first two levels of page tables, first
+                 * recurse and free pages from the lower level page
+                 * tables before reclaiming this entry. */
+                reclaimed_pages += reclaim_bootloader_page_table_pages(pte.follow(), level + 1u);
+            }
+        }
+        hippo.free_page(pt);
+        reclaimed_pages++;
+        return reclaimed_pages;
+    }
+
+    /**
+     * Initialize HURL.
+     *
+     * @param bootinfo HULK boot information structure.
+     */
+    public static void initialize(HulkHeader * header)
+    {
+        PageTable * bootloader_pt_base = cast(PageTable *)read_cr3();
+
+        /*
+         * Ok, what we do here is iterate through all of the memory map regions
+         * in the bootinfo memory_map array and mark each page that is not in
+         * use in a reserved range as available. Within the available memory
+         * regions, we have to watch out for the following items:
+         *   1) HULK binary (header + text + rodata + data)
+         *   2) Framebuffer
+         * In addition to these ranges, there are also pages for the following
+         * purposes that are not represented in the memory map from the
+         * bootloader:
+         *   3) HULK BSS
+         *   4) HULK stack
+         *   5) bootloader page table pages
+         *   6) framebuffer buffer1
+         */
+        size_t usable_memory;
+        size_t physical_address_limit;
+        const(size_t) fb_size = (header.bootinfo.fb.height * header.bootinfo.fb.stride * 4u + PAGE_SIZE - 1u) & ~(PAGE_SIZE - 1u);
+        ulong[2][2] reserved = [
+            [header.bootinfo.hulk_phys, LinkerAddresses.hulk_binary_size],
+            [cast(ulong)header.bootinfo.fb.buffer, fb_size],
+        ];
+        for (size_t ri = 0u; ri < reserved.length; ri++)
+        {
+            reserved[ri][1] += reserved[ri][0];
+        }
+        for (size_t bii = 0u; bii < header.bootinfo.memory_map_count; bii++)
+        {
+            if ((header.bootinfo.memory_map[bii].type == BootInfo.MemoryRegion.Type.Bootloader) ||
+                (header.bootinfo.memory_map[bii].type == BootInfo.MemoryRegion.Type.Conventional))
+            {
+                ulong phys = header.bootinfo.memory_map[bii].base;
+                ulong phys_end = phys + header.bootinfo.memory_map[bii].size;
+                if (phys_end > physical_address_limit)
+                {
+                    physical_address_limit = phys_end;
+                }
+                usable_memory += header.bootinfo.memory_map[bii].size;
+                while (phys < phys_end)
+                {
+                    bool is_reserved = false;
+                    for (size_t ri = 0u; ri < reserved.length; ri++)
+                    {
+                        if ((reserved[ri][0] <= phys) && (phys < reserved[ri][1]))
+                        {
+                            is_reserved = true;
+                            break;
+                        }
+                    }
+                    if (!is_reserved)
+                    {
+                        hippo.free_page(phys);
+                    }
+                    phys += PAGE_SIZE;
+                }
+            }
+        }
+
+        /*
+         * Now that we have available physical pages to allocate from, we can
+         * build new page tables to replace the bootloader page tables.
+         */
+        build_page_tables(header, physical_address_limit);
+
+        /*
+         * After we have switched to the newly constructed page tables, we can
+         * iterate through and free the bootloader page tables. They are most
+         * likely already mapped because we just identity mapped every page up
+         * to the physical_address_limit determined above. But just in case the
+         * bootloader used the last pages of RAM for the page table pages, we
+         * will make sure to map them as we traverse them.
+         */
+        size_t reclaimed_bootloader_page_table_memory = reclaim_bootloader_page_table_pages(bootloader_pt_base) << 10u;
+        usable_memory += reclaimed_bootloader_page_table_memory;
+
+        usable_memory += LinkerAddresses.hulk_bss_size;
+        usable_memory += header.stack_size;
+        usable_memory += fb_size;
+        klog.writefln("Usable memory: %uKB", usable_memory >> 10u);
+        klog.writefln("Kernel size: %uKB", (LinkerAddresses.hulk_binary_size + LinkerAddresses.hulk_bss_size + header.stack_size) >> 10u);
+    }
+
+    public static void map(T, U)(T virtual, U physical, ulong flags)
     {
         PageTable * pt = m_pt_base;
         for (size_t level = 0; level < 4u; level++)
@@ -87,7 +239,7 @@ struct hurl
             }
             else
             {
-                *ppte = PageTableEntry(physical, flags | PT_PRESENT);
+                *ppte = PageTableEntry(cast(ulong)physical, flags | PT_PRESENT);
             }
         }
     }

src/hulk/idt.d

@@ -8,9 +8,18 @@ import ldc.llvmasm;
 import hulk.fb;
 import hulk.console;
 import hulk.klog;
+import hulk.apic;
+import hulk.rtc;
 
 struct idt
 {
+    /* The I/O APIC is configured to map IRQ 0 to interrupt 64 (0x40). */
+    public static enum ulong INT_RTC = 0x48u; /* IRQ 8 */
+
+    public static enum ulong INT_LAPIC_TIMER = 0x70u;
+    public static enum ulong INT_LAPIC_LINT0 = 0x71u;
+    public static enum ulong INT_LAPIC_LINT1 = 0x72u;
+
     struct idtr_t
     {
         ushort limit;
@@ -94,6 +103,19 @@ struct idt
 
 public extern(C) void isr(ulong vector, ulong arg)
 {
+    switch (vector)
+    {
+    case idt.INT_RTC:
+        rtc.isr();
+        break;
+
+    case idt.INT_LAPIC_TIMER:
+    case idt.INT_LAPIC_LINT0:
+    case idt.INT_LAPIC_LINT1:
+        apic.isr(vector);
+        break;
+
+    default:
         console.clear();
         fb.clear(0xFF8000u);
         klog.writefln("ISR %u, 0x%x", vector, arg);
@@ -103,3 +125,4 @@ public extern(C) void isr(ulong vector, ulong arg)
             __asm("hlt", "");
         }
     }
+}

src/hulk/klog.d

@@ -19,7 +19,7 @@ struct klog
     private enum size_t KLOG_SIZE = 1u << KLOG_SIZE_LOG;
 
     /** Kernel log buffer. */
-    private static __gshared align(4096) char[KLOG_SIZE] klog_buffer;
+    private static __gshared align(4096) ubyte[KLOG_SIZE] klog_buffer;
 
     /** Write index in the kernel log buffer. */
     private static __gshared size_t klog_index;

src/hulk/linker_addresses.d

@@ -3,39 +3,137 @@
  */
 module hulk.linker_addresses;
 
-private extern extern(C) __gshared ubyte _hulk_mem_start;
+private extern extern(C) __gshared ubyte _hulk_header_start;
+private extern extern(C) __gshared ubyte _hulk_header_end;
+private extern extern(C) __gshared ubyte _hulk_header_size;
+
+private extern extern(C) __gshared ubyte _hulk_text_start;
+private extern extern(C) __gshared ubyte _hulk_text_end;
+private extern extern(C) __gshared ubyte _hulk_text_size;
+
+private extern extern(C) __gshared ubyte _hulk_rodata_start;
+private extern extern(C) __gshared ubyte _hulk_rodata_end;
+private extern extern(C) __gshared ubyte _hulk_rodata_size;
+
+private extern extern(C) __gshared ubyte _hulk_data_start;
+private extern extern(C) __gshared ubyte _hulk_data_end;
+private extern extern(C) __gshared ubyte _hulk_data_size;
+
 private extern extern(C) __gshared ubyte _hulk_bss_start;
+private extern extern(C) __gshared ubyte _hulk_bss_end;
 private extern extern(C) __gshared ubyte _hulk_bss_size;
+
+/* hulk_binary includes header, text, rodata, and data */
+private extern extern(C) __gshared ubyte _hulk_binary_size;
+
+/* hulk_mem includes header, text, rodata, data, and bss */
+private extern extern(C) __gshared ubyte _hulk_mem_start;
 private extern extern(C) __gshared ubyte _hulk_mem_end;
-private extern extern(C) __gshared ubyte _hulk_total_size;
+private extern extern(C) __gshared ubyte _hulk_mem_size;
 
 /**
  * This struct provides access to linker-defined symbols.
  */
 public struct LinkerAddresses
 {
-    public static @property ulong hulk_mem_start()
+    public static @property ulong hulk_header_start()
     {
-        return cast(ulong)&_hulk_mem_start;
+        return cast(ulong)&_hulk_header_start;
     }
 
+    public static @property ulong hulk_header_end()
+    {
+        return cast(ulong)&_hulk_header_end;
+    }
+
+    public static @property ulong hulk_header_size()
+    {
+        return cast(ulong)&_hulk_header_size;
+    }
+
+
+    public static @property ulong hulk_text_start()
+    {
+        return cast(ulong)&_hulk_text_start;
+    }
+
+    public static @property ulong hulk_text_end()
+    {
+        return cast(ulong)&_hulk_text_end;
+    }
+
+    public static @property ulong hulk_text_size()
+    {
+        return cast(ulong)&_hulk_text_size;
+    }
+
+
+    public static @property ulong hulk_rodata_start()
+    {
+        return cast(ulong)&_hulk_rodata_start;
+    }
+
+    public static @property ulong hulk_rodata_end()
+    {
+        return cast(ulong)&_hulk_rodata_end;
+    }
+
+    public static @property ulong hulk_rodata_size()
+    {
+        return cast(ulong)&_hulk_rodata_size;
+    }
+
+
+    public static @property ulong hulk_data_start()
+    {
+        return cast(ulong)&_hulk_data_start;
+    }
+
+    public static @property ulong hulk_data_end()
+    {
+        return cast(ulong)&_hulk_data_end;
+    }
+
+    public static @property ulong hulk_data_size()
+    {
+        return cast(ulong)&_hulk_data_size;
+    }
+
+
     public static @property ulong hulk_bss_start()
     {
         return cast(ulong)&_hulk_bss_start;
     }
 
+    public static @property ulong hulk_bss_end()
+    {
+        return cast(ulong)&_hulk_bss_end;
+    }
+
     public static @property ulong hulk_bss_size()
     {
         return cast(ulong)&_hulk_bss_size;
     }
 
+
+    public static @property ulong hulk_binary_size()
+    {
+        return cast(ulong)&_hulk_binary_size;
+    }
+
+
+    public static @property ulong hulk_mem_start()
+    {
+        return cast(ulong)&_hulk_mem_start;
+    }
+
     public static @property ulong hulk_mem_end()
     {
         return cast(ulong)&_hulk_mem_end;
     }
 
-    public static @property ulong hulk_total_size()
+    public static @property ulong hulk_mem_size()
     {
-        return cast(ulong)&_hulk_total_size;
+        return cast(ulong)&_hulk_mem_size;
     }
 }

src/hulk/mtrr.d

@@ -0,0 +1,70 @@
+/**
+ * MTRR (Memory Type Range Register) support for HULK.
+ */
+module hulk.mtrr;
+
+import hulk.cpu;
+import hulk.klog;
+
+enum ulong MTRR_TYPE_UC = 0u; /* Uncacheable */
+enum ulong MTRR_TYPE_WC = 1u; /* Write-Combining */
+enum ulong MTRR_TYPE_WT = 4u; /* Writethrough */
+enum ulong MTRR_TYPE_WP = 5u; /* Write-Protect */
+enum ulong MTRR_TYPE_WB = 6u; /* Writeback */
+
+enum uint MSR_MTRRCAP = 0xFEu;
+enum uint MSR_MTRRDEFTYPE = 0x2FFu;
+enum uint MSR_MTRRPHYSBASE0 = 0x200u;
+enum uint MSR_MTRRPHYSMASK0 = 0x201u;
+
+enum ulong MTRRCAP_WC = 0x400u;
+enum ulong MTRRCAP_FIX = 0x100u;
+enum ulong MTRRCAP_VCNT_MASK = 0xFFu;
+
+enum ulong MTRRDEFTYPE_E = 0x800u;
+enum ulong MTRRDEFTYPE_FE = 0x400u;
+enum ulong MTRRDEFTYPE_TYPE_MASK = 0xFFu;
+
+enum ulong MTRRPHYSBASE_PHYSBASE_MASK = 0x000F_FFFF_FFFF_F000u;
+enum ulong MTRRPHYSBASE_TYPE_MASK = 0xFFu;
+
+enum ulong MTRRPHYSMASK_PHYSMASK_MASK = 0x000F_FFFF_FFFF_F000u;
+enum ulong MTRRPHYSMASK_VALID = 0x800u;
+
+struct mtrr
+{
+    public static void initialize()
+    {
+        uint ebx;
+        uint ecx;
+        uint edx;
+        cpuid1(&ebx, &ecx, &edx);
+        if ((edx & CPUID_1_EDX_MTRR) == 0u)
+        {
+            klog.writefln("CPU does not support MTRR");
+            return;
+        }
+        const(ulong) mtrrcap = rdmsr(MSR_MTRRCAP);
+        const(ulong) mtrrdeftype = rdmsr(MSR_MTRRDEFTYPE);
+    }
+
+    private static ulong get_phys_base(uint n)
+    {
+        return rdmsr(MSR_MTRRPHYSBASE0 + 2u * n);
+    }
+
+    private static void set_phys_base(uint n, ulong physbase)
+    {
+        wrmsr(MSR_MTRRPHYSBASE0 + 2u * n, physbase);
+    }
+
+    private static ulong get_phys_mask(uint n)
+    {
+        return rdmsr(MSR_MTRRPHYSMASK0 + 2u * n);
+    }
+
+    private static void set_phys_mask(uint n, ulong physmask)
+    {
+        wrmsr(MSR_MTRRPHYSMASK0 + 2u * n, physmask);
+    }
+}

src/hulk/pagetable.d

@@ -136,8 +136,11 @@ static assert(PageTableEntry.sizeof == 8u);
  */
 struct PageTable
 {
+    /** Number of page table entries in a page table. */
+    enum size_t N_ENTRIES = 512u;
+
     /** Page table entries. */
-    private PageTableEntry[512] entries;
+    private PageTableEntry[N_ENTRIES] entries;
 
     /**
      * Access the PageTableEntry for the given address and page table level.

src/hulk/pit.d

@@ -0,0 +1,42 @@
+/**
+ * PIT (Programmable Interval Timer) functionality.
+ */
+module hulk.pit;
+
+import hulk.cpu;
+import hulk.klog;
+
+struct pit
+{
+    /** PIT input frequency in 1/1000 Hz. */
+    private enum ulong PIT_FREQUENCY_1000HZ = 1_193_181_667u;
+    private enum ulong TARGET_INTERRUPT_HZ = 100u;
+    private enum ulong TARGET_INTERRUPT_1000HZ = TARGET_INTERRUPT_HZ * 1000u;
+    private enum ulong PIT_DIVIDER = (PIT_FREQUENCY_1000HZ + TARGET_INTERRUPT_1000HZ / 2u) / TARGET_INTERRUPT_1000HZ;
+
+    /** PIT channel 0 I/O port. */
+    private enum ubyte PORT_CH_0 = 0x40u;
+    /** PIT channel 1 I/O port. */
+    private enum ubyte PORT_CH_1 = 0x41u;
+    /** PIT channel 2 I/O port. */
+    private enum ubyte PORT_CH_2 = 0x42u;
+    /** PIT mode/command register I/O port. */
+    private enum ubyte PORT_MC = 0x43u;
+
+    private enum ubyte MC_CH_0 = 0x00u;
+    private enum ubyte MC_LO_HI = 0x30u;
+    private enum ubyte MC_RATE_GENERATOR = 0x04u;
+    private enum ubyte MC_BINARY = 0x00u;
+
+    public static void initialize()
+    {
+        out8(PORT_MC, MC_CH_0 | MC_LO_HI | MC_RATE_GENERATOR | MC_BINARY);
+        out8(PORT_CH_0, PIT_DIVIDER & 0xFFu);
+        out8(PORT_CH_0, PIT_DIVIDER >> 8u);
+    }
+
+    public static void isr()
+    {
+        klog.writefln("PIT ISR");
+    }
+}

src/hulk/rtc.d

@@ -0,0 +1,58 @@
+/**
+ * RTC (Real-Time Clock) functionality.
+ */
+module hulk.rtc;
+
+import hulk.cpu;
+import hulk.klog;
+import hulk.apic;
+
+struct rtc
+{
+    private enum ubyte PORT_SELECT = 0x70u;
+    private enum ubyte PORT_DATA = 0x71u;
+
+    private enum ubyte DISABLE_NMI = 0x80u;
+
+    private enum ubyte SR_A = 0xAu;
+    private enum ubyte SR_B = 0xBu;
+    private enum ubyte SR_C = 0xCu;
+
+    private enum ubyte SR_B_ENABLE_IRQ = 0x40u;
+
+    public static void initialize()
+    {
+        /* Enable IRQ 8 to receive RTC interrupts. */
+        out8(PORT_SELECT, DISABLE_NMI | SR_B);
+        ubyte sr_b = in8(PORT_DATA);
+        out8(PORT_SELECT, DISABLE_NMI | SR_B);
+        out8(PORT_DATA, sr_b | SR_B_ENABLE_IRQ);
+
+        out8(PORT_SELECT, DISABLE_NMI | SR_A);
+        klog.writefln("SR_A = 0x%x", in8(PORT_DATA));
+
+        /* Send EOI to enable more RTC interrupts and re-enable NMIs. */
+        eoi();
+    }
+
+    private static void eoi()
+    {
+        /* Read from status register C to clear the interrupt. */
+        out8(PORT_SELECT, SR_C);
+        in8(PORT_DATA);
+    }
+
+    public static void isr()
+    {
+        static __gshared ulong count;
+        static __gshared ulong seconds;
+        count++;
+        if ((count % 1024) == 0u)
+        {
+            seconds++;
+            klog.writefln("Seconds: %u", seconds);
+        }
+        eoi();
+        apic.eoi();
+    }
+}

src/hulk/writef.d

@@ -88,9 +88,9 @@ size_t writef(string s, va_list args, ch_out_fn ch_out)
  */
 private size_t write_hex(ulong v, bool upper, char pad, size_t width, ch_out_fn ch_out)
 {
-    static __gshared string hex_chars_lower = "0123456789abcdef";
+    static __gshared const(char)[16] hex_chars_lower = "0123456789abcdef";
-    static __gshared string hex_chars_upper = "0123456789ABCDEF";
+    static __gshared const(char)[16] hex_chars_upper = "0123456789ABCDEF";
-    static __gshared char[16] buf;
+    char[16] buf;
     size_t i;
     do
     {
@@ -126,7 +126,7 @@ private size_t write_hex(ulong v, bool upper, char pad, size_t width, ch_out_fn
  */
 private size_t write_udec(ulong v, char pad, size_t width, ch_out_fn ch_out)
 {
-    static __gshared char[20] buf;
+    char[20] buf;
     size_t i;
     do
     {