Question

this is Exceptions and Interrupts table(which I understand as IDT)
from the “Intel Architecture Software Developer Manual”

where is Timer interrupt which makes context switching possible??
(for multi-tasking)

if this is a stupid question, please fix my understanding.
thank you in advance

Answer 1

Well, yes, if we are talking about the traditional 8254 PIT timer, it is at IRQ0, which is vector 32. But that is not generally used as the timer in the Linux operating system on modern machines. [Note that the vector assignment of 32 is really quite arbitrary. It is set when programming the 8259 (PIC) or APIC – but it’s not a bad choice, since 32 is the first vector AFTER the reserved ones. It’s certainly better than mixing the hardware interrupts with exception vectors, as DOS would do – so there was no way to tell a General Protection fault (vector 13 in the table above) from a INTR 5 (also vector 13, as the INT0 was mapped to Vector 8, and 5 + 8 = 13). From memory, INTR5 wasn’t particularly well used – something like LPT2: (Second parallel port). But it’s still a good idea to not overlap them… Henc the “reserved” for the vectors 20 to 31.

The IRQ that actually controls the timing of the system is most likely a Local APIC timer, and it’s vector is not fixed in hardware in the same way as the original PC.

Also, with the advent of “message signalled interrupts”, it is entirely possible to have (much) more than 256 interrupt vectors.

I don’t agree with the wording “vector 0-19 are non-maskable interrupts”. Aside from NMI (vector 2), they are all EXCEPTIONS (aka TRAPS or FAULTS) – that is, an event driven by some error condition in the system – vector zero is the result of an integer divide by zero, vector 1 is a “single step” instruction interrupt [and a few other “debug” traps, such as “write to any address matching an enabled debug register”], vector 3 is the result of a “int3” instruction (opcode 0xcc), vector 4 is the result of executing “INTO”(that’s ‘o’ as in overflow, not 0 as in zero). When accessing a piece of memory not marked as present in the page-tables, vector 14 is used. They are indeed “non-maskable”, but they are, with a few exceptions, direct consequences of the instructon executing at the time – so they are synchronous to the program itself.

The exceptions are the “Double fault” exception and “machine check fault”.

Double fault is when the processor detects a fault during the handling of another exception – typically because the operating system has done something daft, like set the stack to somewhere invalid, and thus gets a page-fault, tries to use the stack to store the page-fault return address and that fails because the stack is not accessible. Double fault handlers, thus, tends to be set as “task switch interrupts”, and load a new stack to make sure the double fault can continue. If the double fault handler can’t run correctly, the processor will “triple-fault”. This usually means “reboot” on PC platforms. Double faults are normally not recoveriable – the handler will (try to) provide some information about what happened, and how it got into this state, but once that’s done, the system either reboots or waits for someone to come and push the reset button.

Machine check fault is where the processor detects an unrecoverable error – such as irrecoverable memory error or a cache parity error, etc. These are typically also non-recoverable, but not DIRECTLY coupled with the instruction being executed, but more on a combination of different events (memory read of an address where the memory content has gone bad, or similar).