Subject: Re: SMP re-entrancy in kernel drivers/"bottom half?"
To: Matt Fredette <firstname.lastname@example.org>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 12/19/2003 10:26:11
>If splbar() > splfoo(), this doesn't happen, right? Is it common practice,
>then, to simply have such an ordering?
Some hardware pretty much forces it (VAX comes to mind).
One reason I suggested totally-ordered SPLs, because thats well-known
to be deadlock-free.
I dont know how we currently support multiple CPUs in the kernel
handling clock interrupts whilst guaranteeing no deadlocks: the
manpage I have doesnt mention it. I'd guess splclock() through
splhigh() really are totally ordered?