Questions on linux's ISRs, contexts,... i386 perespective

Hi, I have many doubts regarding interrupts (not linux specific though), ISRs, concurrency, preemption, atomicity,.... Untill now i never took interest to clarify them. Now im determined to hunt them down. Some doubts might look trivial, but still are important for me, because i am not confident about my answers. I have listed some of them below. Can anyone shed some light on these please.

Are interrupt service routines (ISR) re-entrant? Can an ISR (of an interrupt #XYZ) be executing on two (or more) CPUs simultaneously? Are the above two questions same? (just to make sure, i understand the word re-entrant correctly)

Are ISRs preemptible? Who can preempt an ISR? Can a maskable interrupt's ISR be preempted by another *maskable* interrupt?

How many non-maskable interrupts are there in i386 arch? What are they? How can we block non-maskable interrupts? Are cli-sti sufficient? What are the priorities among non-maskable interrupts?

How many different contexts possible? (i can see only two: process context, interrupt context) How different contexts can stack up (one on top of another) due to preemption? How many levels (of such stacking up) are possible at worst case?

What are the *things* that one cannot do in interrupt context, but can in process context?

To make some bunch of statements atomic, do we need to *stop* all other CPUs? How is this possible in i386 arch? Can LOCK instruction be used to lock CPUs for executing bunch of instructions?

NOTE: these are not my buttignment questions.

thanks, -bvk.