M a c O S X CS-450-1: Operating Systems Fall 2005 Matt Grady – Mike O’Connor – Justin Rains

Overview: General Tenth version of the Mac OS developed by Apple Independent of all previous versions of Mac OS Noticeable surface changes –Brand new GUI “Aqua,” a radical departure from previous versions –Addition of the Dock, an application launcher Introduction of open-source Darwin –Mach forms the core of XNU kernel –Built on BSD Unix foundation

Overview: Tiger Current version of OS X, 10.4 Changes for the general user –Spotlight: powerful meta-data search tool –Dashboard: mini-application layer Additions “under the hood” –Startup daemon ‘launchd’ allows for quick boots –Kernel resource locking optimization –Core Image, Core Data, Core Video

CPU Scheduling Mach holds four different priority queues –Normal –System High –Kernel Mode Only –Real Time Non-kernel mode only threads’ priorities constantly changing Prevents bottlenecking by giving resource-intensive threads lower priorities

Processor Modes Supervisor Mode or User Mode –Supervisor calls made through Mach –Message Passing (no trap table) Interrupt Context not necessary –Uses a generic interrupt handling thread –Spends minimal time in supervisor mode –Minimal latency using this approach. I/O Kit –Part of Darwin that handles device drivers –Has all the support for multi-processor machines (SMP)

Process Management Multiprocessor Systems –Splits processes into independent threads –Takes advantage of critical sections (regions) –Multiprocessing Services API Features, code, and support for multiprocessor functionality (semaphores, critical regions, processor availability, etc) Allows programmers to set relative priority of tasks Process Tracking –Similar to Mach 3.0 –Mach Tasks –Mach Ports

File Management Provides support for numerous file systems –UFS, ISO 9660, MS-DOS, SMB, NFS –HFS and HFS+ (extended) Preferred file system 64 bit length files 32 bit allocation table (HFS was 16 – limits disk size drastically) Not case sensitive Virtual File System –On UFS, derived from BSD –Allows for booting from other file systems (need kernel extensions) –Vnode Data structure for directories and files Vnode Operations (VOP) allows access and provide operations

File Management Carbon –Default Volume type is HFS+ –Includes File Manager subsystem Provides functionality for manipulating directories and files Accessing Information about Files and Directories Allocating File Blocks Controlling Login and Directory AccessCopying and Moving Files Creating a File System Reference Creating and Deleting Forks Creating and Deleting Directories Creating File System Specifications Creating and Deleting Files Getting and Setting Volume Information

Memory Management Apple’s memory model is based on the one used in the Mach v3.0 kernel Uses virtual memory with the working set algorithm for determining page faults In addition, Translation Lookaside Buffers, page tables, and other techniques are implemented as hardware specific RAM is considered a cache for virtual memory – processes are initially loaded into virtual memory and pages are only brought into RAM as needed

Memory Management At the base level, memory is paged Above that level is the Memory Object, which may contain a page, a set of pages, a stack, or a file A Virtual Memory (VM) Object is a collection of memory objects A Memory Map is the total collection of VM objects for a process The Virtual Memory (VM) system manages the VM objects, while a subordinate, machine dependent, Process Mapping system (pmap) handles the actual page faults When two or more processes refer to the same page(s), a Shadow VM object is created, which is a pointer to the original VM object that actually contains the data Universal Page Lists (UPL) are data structures used to communicate with the VM system. A UPL is initialized with data from a VM object, the data is altered, and the UPL is forced back on the VM object

Memory Management There are two types of page faults: hard and soft A hard page fault is the usual, where a page is not RAM, but is needed A soft page fault is when the needed page is already in RAM, but does not belong to the calling process’s memory space. In this case a shadow VM object is created Shadow VM objects use “copy-on-write”, which means that the shadow VM object continues to point to the target VM object until a write is attempted, at which point the target pages within the VM object are copied The VM system maintains a “free list”, an “inactive list”, and an “active list” of RAM. The system attempts to maintain a certain number of pages in the free list by swapping pages in the inactive list out.

Deadlock OS X does not seem to have any standard strategies for dealing with deadlock. It is essentially left up to the application programmer to work out the bugs which may cause deadlock One strategy listed was to try to minimize the amount of a resource that could be contested at any one time. For example, break one large lock, which refers to a list of buffers, into several smaller locks for each buffer It is also recommended that critical sections of code be kept to a minimum Buffers may be eliminated, and thereby the possibility for their contention, by using InterProcess Communication (IPC), which is a kernel function that directly copies data from one process to another Apple mentions using a total order for read-write locks, but refers to this technique as “rather extreme”

Summary OS X v10.4 is Apple’s current OS for the Mac Designed for use by home users and creative professionals Built on open-source, very stable Darwin core