lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
Open Source and information security mailing list archives
| ||
|
Date: Mon, 18 Jan 2010 12:23:09 +0900 (JST) From: KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com> To: "Andrew C. Morrow" <andrew.c.morrow@...il.com> Cc: kosaki.motohiro@...fujitsu.com, Gleb Natapov <gleb@...hat.com>, linux-mm@...ck.org, linux-kernel@...r.kernel.org, linux-api@...r.kernel.org, akpm@...ux-foundation.org Subject: Re: [PATCH v5] add MAP_UNLOCKED mmap flag > On Thu, Jan 14, 2010 at 3:17 AM, KOSAKI Motohiro > <kosaki.motohiro@...fujitsu.com> wrote: > >> > Hmm.. > >> > Your answer didn't match I wanted. > >> Then I don't get what you want. > > > > I want to know the benefit of the patch for patch reviewing. > > > > The benefit of the patch is that it makes it possible for an > application which has previously called mlockall(MCL_FUTURE) to > selectively exempt new memory mappings from memory locking, on a > per-mmap-call basis. As was pointed out earlier, there is currently no > thread-safe way for an application to do this. The earlier proposed > workaround of toggling MCL_FUTURE around calls to mmap is racy in a > multi-threaded context. Other threads may manipulate the address space > during the window where MCL_FUTURE is off, subverting the programmers > intended memory locking semantics. > > The ability to exempt specific memory mappings from memory locking is > necessary when the region to be mapped is larger than physical memory. > In such cases a call to mmap the region cannot succeed, unless > MAP_UNLOCKED is available. > > > > > >> > few additional questions. > >> > > >> > - Why don't you change your application? It seems natural way than kernel change. > >> There is no way to change my application and achieve what I've described > >> in a multithreaded app. > > > > Then, we don't recommend to use mlockall(). I don't hope to hear your conclusion, > > it is not objectivization. I hope to hear why you reached such conclusion. > > > > I agree that mlockall is a big hammer and should be avoided in most > cases, but there are situations where it is exactly what is needed. In > Gleb's instance, it sounds like he is doing some finicky performance > measurement and major page faults skew his results. In my case, I have > a realtime process where the measured latency impact of major page > faults is unacceptable. In both of these cases, mlockall is a > reasonable approach to eliminating major faults. > > However, Gleb and I have independently found ourselves unable to use > mlockall because we also need to create a very large memory mapping > (for which we don't care about major faults). The proposed > MAP_UNLOCKED flag would allow us to override MCL_FUTURE for that one > mapping. > > > > >> > - Why do you want your virtual machine have mlockall? AFAIK, current majority > >> > virtual machine doesn't. > >> It is absolutely irrelevant for that patch, but just because you ask I > >> want to measure the cost of swapping out of a guest memory. > > > > No. if you stop to use mlockall, the issue is vanished. > > > > And other issues arise. Gleb described a situation where the use of > mlockall is justified, identified an issue which prevents its use, and > provided a patch which resolves that issue. Why are you focusing on > the validity of using mlockall? > > > > >> > - If this feature added, average distro user can get any benefit? > >> > > >> ?! Is this some kind of new measure? There are plenty of much more > >> invasive features that don't bring benefits to an average distro user. > >> This feature can bring benefit to embedded/RT developers. > > > > I mean who get benifit? > > > > > >> > I mean, many application developrs want to add their specific feature > >> > into kernel. but if we allow it unlimitedly, major syscall become > >> > the trushbox of pretty toy feature soon. > >> > > >> And if application developer wants to extend kernel in a way that it > >> will be possible to do something that was not possible before why is > >> this a bad thing? I would agree with you if for my problem was userspace > >> solution, but there is none. The mmap interface is asymmetric in regards > >> to mlock currently. There is MAP_LOCKED, but no MAP_UNLOCKED. Why > >> MAP_LOCKED is useful then? > > > > Why? Because this is formal LKML reviewing process. I'm reviewing your > > patch for YOU. > > > > If there is no objective reason, I don't want to continue reviewing. > > > > There is an objective reason: the current interaction between > mlockall(MCL_FUTURE) and mmap has a deficiency. In 'normal' mode, > without MCL_FUTURE in force, the default is that new memory mappings > are not locked, but mmap provides MAP_LOCKED specifically to override > that default. However, with MCL_FUTURE toggled to on, there is no > analogous way to tell mmap to override the default. The proposed > MAP_UNLOCKED flag would resolve this deficiency. Very thank you, Andrew! Your explanation help me lots rather than original patch description. OK, At least MAP_UNLOCED have two users (you and gleb) and your explanation seems makes sense. So, if gleb resend this patch with rewrited description, I might take my reviewed-by tag to it, probagly. Thanks. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@...r.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists