On 11 February 2013 12:22, Vineet Gupta <Vineet.Gupta1@xxxxxxxxxxxx> wrote: > On Monday 11 February 2013 04:23 PM, Jonas Bonn wrote: >> On 11 February 2013 11:28, James Hogan <james.hogan@xxxxxxxxxx> wrote: >>> On 11/02/13 10:13, Vineet Gupta wrote: >>>> On Monday 11 February 2013 03:06 PM, Jonas Bonn wrote: >>>>> On 11 February 2013 08:26, Vineet Gupta <Vineet.Gupta1@xxxxxxxxxxxx> wrote: >>>>> >>>>>> The only downside of this patch is that userspace signal stack grows in size, >>>>>> since signal frame only cares about scratch regs (pt_regs), but has to accommodate >>>>>> unused placeholder for callee regs too by virtue of using user_regs_struct. >>>>> Is this really true? Don't setcontext and friends require that _all_ >>>>> the registers be part of sigcontext? >>>> >>>> But for an ABI - callee saved regs will anyhow be saved/restored even in >>>> setcontext case ! So collecting it for that purpose seems useless, or am I missing >>>> something here. >>> >>> I think Jonas' point was that signals are asynchronous, i.e. you could >>> get interrupted by a signal at virtually any time during the program's >>> execution. >> >> No, I agree that the callee-saved regs don't need to be saved across a >> signal handler invocation. It's really just the setcontext case that >> wants to be able to swap out the callee-saved regs. > > I don't think that's needed either - and if thats mandated somewhere, it would > seem a unnecessary mis-optimization IMHO. > > See, even a setcontext enabled control flow needs to be ABI compliant so that it > plays nicely with other normal flows of execution. Thus e.g. it can't fudge a > callee reg - it needs to save orig callee reg(s) and restore them in the end. And > if we agree to those semantics - I don't see any value in swapping the callee reg > context around usage of setcontext as it would be a wasted effort. Yeah, that makes sense. I can see where you're coming from... and the fact that you switch the stack, as James pointed out, means that you end up restoring whatever callee-saved registers you need in the new control flow on the way out of your setcontext wrapper. BUT... a successful call to setcontext() does not return and whatever code you end up jumping to as a result of the call has its own expectations about the state of the registers. Somebody has to set up the registers to meet these expectations and, as far as I can see, this means: i) sigreturn fixes up the internal pt_regs with the new userspace state ii) the syscall return path restores _all_ the regs, as though there had been a context switch What am I missing? I'm totally open to the idea that I'm the one who's confused here... ...and perhaps this is all moot since it seems that getcontext/setcontext are obsolete anyway(???). /Jonas -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
