HLT stops the processor. It fetches and executes nothing until one of three things happens: a maskable interrupt while IF = 1, a non-maskable interrupt, or a reset. On an interrupt the CPU runs the handler, and the handler's IRET returns to the instruction after HLT. From the program's point of view HLT is "wait for the next interrupt".

On a PC the timer interrupts about 18.2 times a second and every key press interrupts too, so a HLT with interrupts on lasts at most about 55 ms. Since any interrupt wakes it, put it in a loop that checks for the thing you are waiting for. The example waits about one second by watching the BIOS tick counter at 0040:006Ch.

STI                   ; HLT needs interrupts on
PUSH    DS
MOV     AX, 40h
MOV     DS, AX        ; the BIOS data area
MOV     BX, DS:[6Ch]  ; timer ticks so far
TICK1:
HLT                   ; sleep until an interrupt
MOV     AX, DS:[6Ch]
SUB     AX, BX
CMP     AX, 18
JB      TICK1         ; about one second
POP     DS
MOV     DX, OFFSET MSG1
MOV     AH, 09h       ; DOS: print the string at DS:DX
INT     21h
 
MSG1    DB      'One second later', 13, 10, '$'

Pitfalls

Never HLT after CLI: with IF = 0 only an NMI or a reset can wake the CPU, and the machine looks frozen. That pairing is used deliberately only to stop a system for good. The reverse order, STI then HLT, is safe thanks to the one-instruction delay after STI.

Compared with a busy loop, HLT lets the processor idle, and later CPUs use that time to save power. Later CPUs: in protected mode HLT is privileged (ring 0 only).