My initial guess is wrong. I would have thought it would simply return from loop and the core library would just call loop() again. However, I see the following code was created. Noticing that __stop_program is a hard loop...
An extract of Blink.ino's listing, with exit(0) added:
// the loop routine runs over and over again forever:
void loop() {
digitalWrite(led, HIGH); // turn the LED on (HIGH is the voltage level)
delay(1000); // wait for a second
digitalWrite(led, LOW); // turn the LED off by making the voltage LOW
delay(1000); // wait for a second
exit(0);
}
The disassembly of the above:
00000100 <loop>:
100: 80 91 00 01 lds r24, 0x0100
104: 61 e0 ldi r22, 0x01 ; 1
106: 0e 94 ca 01 call 0x394 ; 0x394 <digitalWrite>
10a: 68 ee ldi r22, 0xE8 ; 232
10c: 73 e0 ldi r23, 0x03 ; 3
10e: 80 e0 ldi r24, 0x00 ; 0
110: 90 e0 ldi r25, 0x00 ; 0
112: 0e 94 23 01 call 0x246 ; 0x246 <delay>
116: 80 91 00 01 lds r24, 0x0100
11a: 60 e0 ldi r22, 0x00 ; 0
11c: 0e 94 ca 01 call 0x394 ; 0x394 <digitalWrite>
120: 68 ee ldi r22, 0xE8 ; 232
122: 73 e0 ldi r23, 0x03 ; 3
124: 80 e0 ldi r24, 0x00 ; 0
126: 90 e0 ldi r25, 0x00 ; 0
128: 0e 94 23 01 call 0x246 ; 0x246 <delay>
12c: 80 e0 ldi r24, 0x00 ; 0
12e: 90 e0 ldi r25, 0x00 ; 0
130: 0e 94 2e 02 call 0x45c ; 0x45c <_exit>
...
0000045c <_exit>:
45c: f8 94 cli
0000045e <__stop_program>:
45e: ff cf rjmp .-2 ; 0x45e <__stop_program>
Note that if _exit had not called cli, interrupts would be able to do stuff. But that is not the case.