Machine-Independent Virtual Memory Management for Paged Uniprocessor
and Multiprocessor Architectures
--- Rashid et al., 1988


- Main idea: separate hardware support from software memory management
  without performance sacrifice;  have a simple, extensible kernel,
  concentrating on communication facilities;

- "Unix systems have traditionally restricted the facilities provided,
  basing implementation for new architectures on versions for older
  machines;"

- Mach provides:
	- large, sparse virtual address spaces
	- copy-on-write virtual copy operations
	- copy-on-write and read-write memory sharing between tasks
	- memory mapped files
	- user-provided backing store objects and pagers

- Mach makes few assumptions about memory management hardware

- Mach abstractions:

	- task:
		- environment in which threads may run;
		- basic unit of resource allocation;

	- thread:
		- basic unit of CPU utilization;
		- threads in a task access the same resources;

	- port:
		- communication channel (message queue);
		- similar to object reference in OOP;
		- used to perform object manipulation;
		- ports are protected by kernel-managed capabilities,
		  or port rights;

	- message:
		- basic communication unit;

	- memory object:
		- segment abstraction?
		- data collection that can be mapped into an address
		  space;

- operations on objects are performed by sending messages to ports.
  This abstraction allows objects to be arbitrarily placed in the
  network.  Also, by limiting all data operations to one mechanism, 
  Mach is able to provide systemwide protection to its users by 
  protecting the communications mechanism;

- Most of the inefficiency of message handling in traditional operating
  systems is due to either the copying of messages from one task to
  another (shared memory) or low network transfer speed.

- "The key to efficiency in Mach is the notion that virtual memory
  management can be integrated with a message-oriented communication
  facility."  To bypass low network transfer speeds, Mach uses 
  virtual-memory remapping to transfer the contents of large 
  messages. (copy-on-write is used to avoid or delay the actual
  copying of the data.)

- the Network Message Server (NMS) is a user-level capability-based
  networking daemon that forwards messages between hosts.  It also
  provides a primitive networkwide name service.  The NMS is protocol
  independent;


Memory Management
-----------------

- memory mapping basically maps a virtual address range to a memory
  object.  Address space is limited by hardware restrictions;

- the key to sparse address spaces is that page-table space is used for
  only currently allocated regions;

- Mach maintains a cache of memory-resident pages of all mapped objects.
  However, a page fault occurring when a thread accesses a nonresident 
  page is executed as a message to the object's port.  Therefore, memory
  can be paged by user-written memory managers;

- regions specified for virtual memory operations must be page
  aligned.  Note that the page size is a boot parameter in Mach, and
  is a power of two multiple of the hardware page size;

- read-write sharing is controled by the inheritance attribute, which
  can have the values "shared", "copy" or "none";

- like inheritance, protection is also specified on a per-page basis;

- "virtual memory related functions such as pagein and pageout, can be 
  performed directly by user-state tasks for memory objects they create."

- the implementation of the VM uses:

	- a resident page table;
	- address map;
	- memory object;
	- pmap.

- The resident page table keeps track of Mach pages residing in main 
  memory (a Mach page may contain several physical pages but their 
  number must be a power of 2);

- The memory object: a unit of backing storage managed by the kernel 
  or a user task; 

- The address map: a doubly linked list of map entries each which 
  describes a mapping from a range of virtual addresses to a region 
  of a memory object; all pages mapped by the same map entry have the 
  same protection and inheritance attributes (page is to be copied, 
  shared or ignored at fork() time) .

- The p-map: the memory-mapping data structure used by the specific 
  hardware (page tables for a VAX, inverted page table for the IBM RT, 
  more complex structures on other machines); it does not have to be kept 
  fully up-to-date except for the pages in the resident part of the kernel.

- memory objects are used to manage secondary storage, and generally 
  represent files, pipes, or other data that are mapped into virtual
  memory for reading and writing.  Memory objects may be backed by 
  user-level memory managers.

- "the machine dependent part of Mach... has no knowledge of machine 
  independent data structures and is not required to maintain full 
  knowledge of valid mappings from virtual to hardware pages."

- "all the page entries associated with a given object are linked together
  in a memory object list..."

- "fast lookup of a physical page associated with an object/offset is 
  performed using a bucket hash table..."

- "addresses within a task address space are mapped to byte offsets
  in memory objects by a data structure called an address map."  

- "when a copy-on-write copy is performed, the two address maps which
  contain copies point to the same memory object."  When one of the tasks
  writes the shared data, a shadow object is created (it basically holds
  modified pages which originally belonged to another object.)

- "the purpose of Mach's machine dependent code is the management of 
  physical address maps (called pmaps).  For a VAX, a pmap corresponds
  to a VAX page table".