Subversion Repositories HelenOS-doc

    <title>AMD64/Intel EM64T</title>

    <para>The AMD64 architecture is a 64-bit extension of the older IA-32

    architecture. Only 64-bit applications are supported. Creating this port

    was relatively easy, because it shares a lot of common code with IA-32

    platform. However, the 64-bit extension has some specifics, which made the

    porting interesting.</para>

    <section>

      <title>Virtual Memory</title>

      <para>The AMD64 architecture uses standard processor defined 4-level

      page mapping of 4KB pages. The NX(no-execute) flag on individual pages

      is fully supported.</para>

    </section>

    <section>

      <title>TLB-only Paging</title>

      <para>All memory on the AMD64 architecture is memory mapped, if the

      kernel needs to access physical memory, a mapping must be created.

      During boot process the boot loader creates mapping for the first 20MB

      of physical memory. To correctly initialize the page mapping system, an

      identity mapping of whole physical memory must be created. However, to

      create the mapping it is unavoidable to allocate new - possibly unmapped

      - frames from frame allocator. The ia32 solves it by mapping first 2GB

      memory during boot process. The same solution on 64-bit platform becomes

      unfeasible because of the size of the possible address space.</para>

      <para>As soon as the exception routines are initialized, a special page

      fault exception handler is installed which provides a complete view of

      physical memory until the real page mapping system is initialized. It

      dynamically changes the page tables to always contain exactly the

      faulting address. The page then becomes cached in the TLB and on the

      next page fault the same tables can be utilized to handle another

      mapping.</para>

    </section>

    <section>

      <title>Mapping of Physical Memory</title>

      <para>The AMD64 ABI document describes several modes of program layout.

      The operating system kernel should be compiled in a

      <emphasis>kernel</emphasis> mode - the kernel is located in the negative

      2 gigabytes (0xffffffff80000000-0xfffffffffffffffff) and can access data

      anywhere in the 64-bit space. This wouldn't allow kernel to see directly

      more than 2GB of physical memory. HelenOS duplicates the virtual mapping

      of the physical memory starting at 0xffff800000000000 and accesses all

      external references using this address range.</para>

    </section>

    <section>

      <title>Thread Local Storage</title>

      <para>The code accessing thread local storage uses a segment register FS

      as a base. The thread local storage is stored in the hidden 64-bit part

      of the FS register which must be written using priviledged machine

      specific instructions. Special syscall to change this register is

      provided to user applications. The TLS address for this platform is

      expected to point just after the end of the thread local data.</para>

    </section>

    <section>

      <title>Fast SYSCALL/SYSRET Support</title>

      <para>The entry point for system calls was traditionally a speed problem

      on IA32 architecture. AMD64 supports a SYSCALL/SYSRET instructions. Upon

      encountering SYSCALL instruction, the processor changes privilege mode

      and transfers control to an address stored in machine specific register.

      Unlike other similar instructions it does not change stack to a known

      kernel stack, which must be done by the syscall entry routine. A hidden

      part of a GS register is provided to support the entry routine with data

      needed for switching to kernel stack.</para>

    </section>

    <section>

      <title>Debugging Support</title>

      <para>To provide developers tools for finding bugs, hardware breakpoints

      and watchpoints are supported. The kernel also supports self-debugging -

      it sets watchpoints on certain data and upon every modification

      automatically checks whether a correct value was written. It is

      worthwhile to mention, that since this feature was implemented, the

      watchpoint was never fired.</para>

    </section>

  </section>

  <section>

    <title>Intel IA32</title>

    <para>The IA32 architecture uses 4K pages and processor supported 2-level

    page tables. Along with AMD64 It is one of the 2 architectures that fully

    supports SMP configurations. IA32 is mostly similar to AMD64, it even

    shares a lot of code. The debugging support is the same as with AMD64. The

    thread local storage uses GS register.</para>

  </section>

  <section>

    <title>MIPS32</title>

    <para>Both little and big endian kernels are supported. In order to test

    different page size it was set to 16K. The MIPS architecture is TLB-only,

    the kernel simulates 2-level page tables. On processors that support it,

    lazy FPU context switching is implemented.</para>

    <section>

      <title>Thread Local Storage</title>

      <para>The thread local storage support in compilers is a relatively

      recent phenomena. The standardization of such support for MIPS platform

      is very new and even the newest versions of GCC cannot generate 100%

      correct code. Because of some weird MIPS processor variants, it was

      decided, that the TLS pointer will be gathered not from some of the free

      registers, but a special instruction was devised and the kernel is

      supposed to emulate it. HelenOS expects that the TLS pointer is in the

      K1 register. Upon encountering the reserved instruction exception and

      checking that the application is requesting a TLS pointer, it returns

      the contents of the K1 register. The K1 register is expected to point

      0x7000 bytes after the beginning of the thread local data.</para>

    </section>

  </section>

  <section>

    <title>Power PC</title>

    <para></para>

  </section>

  <section>