* Nick Piggin <npiggin@xxxxxxx> wrote: > On Fri, Nov 28, 2008 at 01:11:27PM +0100, Ingo Molnar wrote: > > > > * Nick Piggin <npiggin@xxxxxxx> wrote: > > > > > Hi, > > > > > > After yesterday noticing some code in mm/filemap.c accidentally perform > > > a __GFP_FS allocation when it should not have been, I thought it might > > > be a good idea to try to catch this kind of thing with lockdep. > > > > > > I coded up a little idea that seems to work. Unfortunately the system > > > has to actually be in __GFP_FS page reclaim, then take the lock, before > > > it will mark it. But at least that might still be some orders of > > > magnitude more common (and more debuggable) than an actual deadlock > > > condition, so we have some improvement I hope. > > > > > > I guess we could do the same thing with __GFP_IO and even GFP_NOIO > > > locks too, but I don't know how expensive it is to add these > > > annotations to lockdep. [...] > > > > Same cost as normal locking, i.e. as cheap and local as it gets. Lockdep > > is only expensive computationally when new rules are discovered and have > > to be validated - but that is rare. > > OK, good... I'll think about whether it makes sense to add those locks. > Actually, it probably makes sense to merge the __GFP_FS thing first, > with a design that will allow further types to be added easily. > > > > Nice feature - and we want more of this type of preventive dependency > > tracking - so feel free to add it whenever you run into an example like > > this. > > Well, lockdep has most of the support with iits "recusion possibility" > checking for interrupts. All the names in the lockdep code are geared > completely toward interrupts, but the concept is almost exactly the same > here (I can't think if there are any other important points in the kernel > where similar situation can arise, but it wouldn't surprise me if there > is). > > > > What merge route would you prefer? tip/core/locking would be the natural > > home of it (we already have a fair bit of lockdep stuff queued up there > > for v2.6.29) - it also touches a few FS bits. > > I'm happy for you or Peter to merge yet though there, sure. Just let me > get some more input and then I'll try fix it up and make it merge > worthy :) > > BTW. Do you have the might_lock annotations in there? I thought I'd see > them in 2.6.28, but they don't seem to be there. No problems with them? yes, they are still there, lined up for v2.6.29. They are working fine. Ingo -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html