[ltt-dev] [RFC git tree] Userspace RCU (urcu) for Linux (repost)

[Paul E. McKenney]

On Sat, Feb 14, 2009 at 12:46:02AM -0500, Mike Frysinger wrote:
> On Sat, Feb 14, 2009 at 00:20, Paul E. McKenney wrote:
> > On Sat, Feb 14, 2009 at 12:07:46AM -0500, Mike Frysinger wrote:
> >> On Fri, Feb 13, 2009 at 14:36, Paul E. McKenney wrote:
> >> > On Fri, Feb 13, 2009 at 01:54:11PM -0500, Mathieu Desnoyers wrote:
> >> >> * Linus Torvalds (torvalds at linux-foundation.org) wrote:
> >> >> > Btw, for user space, if you want to do this all right for something like
> >> >> > BF. I think the only _correct_ thing to do (in the sense that the end
> >> >> > result will actually be debuggable) is to essentially give full SMP
> >> >> > coherency in user space.
> >> >> >
> >> >> > It's doable, but rather complicated, and I'm not 100% sure it really ends
> >> >> > up making sense. The way to do it is to just simply say:
> >> >> >
> >> >> >  - never map the same page writably on two different cores, and always
> >> >> >    flush the cache (on the receiving side) when you switch a page from one
> >> >> >    core to another.
> >> >> >
> >> >> > Now, the kernel can't really do that reasonably, but user space possibly could.
> >> >> >
> >> >> > Now, I realize that blackfin doesn't actually even have a MMU or a TLB, so
> >> >> > by "mapping the same page" in that case we end up really meaning "having a
> >> >> > shared mapping or thread". I think that _should_ be doable. The most
> >> >> > trivial approach might be to simply limit all processes with shared
> >> >> > mappings or CLONE_VM to core 0, and letting core 1 run everything else
> >> >> > (but you could do it differently: mapping something with MAP_SHARED would
> >> >> > force you to core 0, but threads would just force the thread group to
> >> >> > stay on _one_ core, rather than necessarily a fixed one).
> >> >> >
> >> >> > Yeah, because of the lack of real memory protection, the kernel can't
> >> >> > _know_ that processes don't behave badly and access things that they
> >> >> > didn't explicitly map, but I'm hoping that that is rare.
> >> >> >
> >> >> > And yes, if you really want to use threads as a way to do something
> >> >> > across cores, you'd be screwed - the kenrel would only schedule the
> >> >> > threads on one CPU. But considering the undefined nature of threading on
> >> >> > such a cpu, wouldn't that still be preferable? Wouldn't it be nice to have
> >> >> > the knowledge that user space _looks_ cache-coherent by virtue of the
> >> >> > kernel just limiting cores appropriately?
> >> >> >
> >> >> > And then user space would simply not need to worry as much. Code written
> >> >> > for another architecture will "just work" on BF SMP too. With the normal
> >> >> > uclinux limitations, of course.
> >> >>
> >> >> I don't know enough about BF to tell for sure, but the other way around
> >> >> I see that would still permit running threads with shared memory space
> >> >> on different CPUs is to call a cache flush each time a userspace lock is
> >> >> taken/released (at the synchronization points where the "magic
> >> >> test-and-set instruction" is used) _from_ userspace.
> >> >>
> >> >> If some more elaborate userspace MT code uses something else than those
> >> >> basic locks provided by core libraries to synchronize data exchange,
> >> >> then it would be on its own and have to ensure cache flushing itself.
> >> >
> >> > How about just doing a sched_setaffinity() in the BF case?  Sounds
> >> > like an easy way to implement Linus's suggestion of restricting the
> >> > multithreaded processes to a single core.  I have a hard time losing
> >> > sleep over the lack of parallelism in the case where the SMP support is
> >> > at best rudimentary...
> >>
> >> the quick way is to tell people to run their program through `taskset`
> >> (which is what we're doing now).
> >
> > Not sure what environment Mathieu is looking to run his program from,
> > but he would need to run it on multiple architectures.
> 
> right, that is exactly the kind of thing we strive to avoid on our
> (the Blackfin) side of things
> 
> >> the next step up (or down depending on how you look at it) would be to
> >> hook the clone function to do this automatically.  i havent gotten
> >> around to testing this yet which is why there isnt anything in there
> >> yet though.
> >>
> >> asmlinkage int bfin_clone(struct pt_regs....
> >>        unsigned long clone_flags;
> >>        unsigned long newsp;
> >>
> >> +#ifdef CONFIG_SMP
> >> +       if (current->rt.nr_cpus_allowed == NR_CPUS) {
> >> +               current->cpus_allowed = cpumask_of_cpu(smp_processor_id());
> >> +               current->rt.nr_cpus_allowed = 1;
> >> +       }
> >> +#endif
> >> +
> >>        /* syscall2 puts clone_flags in r0 and usp in r1 */
> >>        clone_flags = regs->r0;
> >>        newsp = regs->r1;
> >
> > Wouldn't you also have to make sched_setaffinity() cut back to only one
> > CPU if more are specified?
> 
> mmm, yes and no.  if we wanted to keep the transparency thing going,
> then adding a check to the affinity functions to make sure threaded
> apps dont span cpus would be needed.  but i would think we'd want to
> have it return an error rather (EINVAL prob) than attempting to make
> any automatic selections.  the only real blocker here would be
> figuring out how to detect the application in question is threaded
> with 100% accuracy.  the Blackfin port does not yet have TLS support
> which means we're using linuxthreads rather than NPTL ...
> 
> hooking clone gives us the biggest bang for the buck: majority of
> stuff today are threaded applications that dont look at affinity.
> 
> > If Blackfin handles hotplug CPU, that may
> > need attention as well, since tasks affinitied to the CPU being removed
> > can end up with their affinity set to all CPUs.  And there are probably
> > other issues.
> 
> no, we dont support hotplugging of CPUs.  there is no hardware support
> for it, so i think the only thing you'd gain is perhaps power savings
> ?  not sure it would even work in our case though as the hardware does
> not support restarting or shutdown of one core ... they both have to
> restart/shutdown.  putting one core into a constant idle loop would
> save power, but that can already be accomplished by reducing the apps
> that go onto a specific core.

OK, that removes that issue, at least aside from any people who will
take the software approach to CPU hotplug (leaving the unplugged CPU
spinning with irqs disabled or some such).

Other potential issues include unrelated processes that share memory via
shmget() or mmap() -- presumably groups of such processes would need to
be bound to a single CPU?

							Thanx, Paul