A crash in computing is a condition where a program (either an application or part of the operating system) stops performing its expected function and also stops responding to other parts of the system. Often the offending program may simply appear to freeze. If this program is a critical part of the operating system kernel the entire computer may crash (a system crash).

Many crashes are the result of the execution of a single machine instruction, but the causes of this are manifold. Typical causes are when the program counter is set to an incorrect address or a buffer overflow overwrites a portion of program code due to an earlier bug. In either case, it is quite common for the processor to attempt to execute data or random memory values. Since all data values are possible but only some values are valid instructions, this often results in an illegal instruction exception. By chance such data or random values could be valid (though unplanned) instructions. If while successfully executing such instructions the processor happens onto a branch instruction that moves the program counter back a few bytes it would simply repeat those few instructions over and over, being stuck in an endless loop. One might say that the original bug that upset the program counter "caused" the crash, but the actual fault was an illegal instruction or endless loop some time later. The art of debugging such crashes is connecting the actual cause of the crash (easily determined) with the code that set off the chain of events. This is often very far from obvious—the original bug is usually perfectly valid code from the processor's perspective.

On earlier personal computers, it was actually possible to cause hardware damage through trying to write to hardware addresses outside of the system's main memory. Occasionally, execution of arbitrary data on a system will result in a breakup of screen display. This is widely considered a severe system crash.

Another cause of crashes is a race condition in communication between processes. One process may send a signal to a second process then stop execution until it receives a response. If the second process is busy the signal will be forced to wait until the process can get to it. However, if the second process was busy sending a signal to the first process then both processes would wait forever for the other to respond to signals and never see the other's signal. If the processes are uninterruptible they will hang and have to be shut down. If at least one of the processes is a critical kernel process the whole system may crash and have to be restarted.

Etymology

The term "crash" may originate in the phrase "head crash" which would occur when the read/write heads in a hard disk would physically come into contact with (i.e., crash into) the disk surface. The result of this would usually be that the computer would freeze up.