Slashdot Mirror
Removing the Big Kernel Lock
Corrado writes "There is a big discussion going on over removing a bit of non-preemptable code from the Linux kernel. 'As some of the latency junkies on lkml already know, commit 8e3e076 in v2.6.26-rc2 removed the preemptable BKL feature and made the Big Kernel Lock a spinlock and thus turned it into non-preemptable code again. "This commit returned the BKL code to the 2.6.7 state of affairs in essence," began Ingo Molnar. He noted that this had a very negative effect on the real time kernel efforts, adding that Linux creator Linus Torvalds indicated the only acceptable way forward was to completely remove the BKL.'"
What's linux?
The future.
Seven puppies were harmed during the making of this post.
When the Linux kernel first supported multiprocessor systems, it was done with a single lock protecting access to all the kernel (the Big Kernel Lock); the kernel could still only do one thing at a time. Over time, most sections of the kernel have introduced their own fine-grained locking and moved out from under the BKL, allowing many parts of the kernel to be running at the same time on multiple processors. The BKL has shrunk over time, but it still exists over a chunk of the kernel. The kernel hackers recently tried to replace the hard lock with a preemptable lock, but that had some bad interactions with the scheduler (which determines what process/kernel thread runs when), so Linus switched back to the old-style BKL.
Now, a group is trying to see if it is possible to weed out all the remaining uses of the BKL and replace them with localized locking for specific sections of the kernel. This is tricky, as there are side-effects of the BKL that are not always obvious.
new semaphore code was introduced that simplified locking. Unfortunately in many kernel situations it's proven to affect performance at around something like 40% - which isn't just considerable its disastrous.
rather than merge the old locking code back in, and reintroduce the many different locking primitives they had, someone decided to simply reenable the BKL - the downside of which is they have to either fix the regression caused by the simpler semaphore code (not likely, it's very simple and clean - everyone's favourite pet/child) or remove instances of where the semaphore code is likely to be called (the BKL).
Matt
Ok so here's the deal:
Linux is a preemptive multi-tasking kernel. What this means is that a hardware interrupt like a keyboard click or the system timer will interrupt whatever is currently running on the CPU, and an interrupt handler in the kernel starts running code. In order to make sure that all the states of the kernel are consistent (ie: not corrupt), the different parts of the kernel are supposed to lock the data that they are using or modifying (ie, readlock or writelock) in case another code path gets run at the same time trying to modify the same data. It becomes even more important in a multi-cpu environment where locks have to be atomic (happen at the same time on all CPUs). So what you are supposed to do is only lock the resources you currently need (a file system drivers would only lock parts of the filesystem, not a character device). Because some programmers are lazy, or not sure what they are doing, they just use the big kernel lock which locks pretty much everything in the kernel. This is bad for multi-tasking and multi-processing because it means you can only have one codepath using the lock at a time.
Note: it's been a while since I've done kernel work, so I'm sure this is not 100% true, but hope it helps you understand.
Because these days the BKL is barely used in the kernel core, or so Linus says: the core kernel, VM and networking already don't really do BKL. And it's seldom the case that subsystems interact with other unrelated subsystems outside of the core areas. IOW, it's rare to hit BKL contention - and in those cases, you want the contention period to be as short as possible. And spinlocks are the faster locking primitive, so making the BKL a spinlock (which is not preemptable) makes the BKL contention periods faster. A mutex/spinlock brings you "preemptability" and hides a bit the fact that there's a global lock being used sometimes at the expense of performance, which may be a good thing for RT/lowlatency users, but apparently Linus prefers to choose the solution that is faster and doesn't hid the real problem.
Since the summary doesn't cut to the chase, and the article was starting to get a little boring and watered-down, I read Ingo's post and here's what I got from it: the BKL is released in the scheduler, so a lot of code is written that grabs the lock and assumes it will be released later, which is bad. Giving it the usual lock behavior of having explicit release will break lots of code. Ingo created a new branch that does this necessary breakage so that the broken code can be detected and fixed. He wants people to test this "highly experimental" branch and report error messages and/or fixes.
Assuming everything is stable and correct, the next step is to break the BKL into locks with finer granularity so that the BKL can go the way of the dodo.
A mutex/spinlock brings you "preemptability"
Duh, I meant mutex/semaphore. And Linux semaphores have become slower, meanwhile mutexes still are fast as old semaphores were, as #23446368 says. The options were to move from a semaphore to mutexes or spinlocks, but Linus chose spinlocks because the RT/low-latency crow will notice it and will try to remove the remaining BKL users.
here (for subscribers. I dare not post a free link here :)
you had me at #!
Matthew Wilcox replaced the per platform semaphore code with a generic implementation because it was likely to be less buggy, reduced code size and most places that are performance critical should be using mutexes now.
Unfortunately this caused a 40% regression in the AIM7 benchmark. The BKL was now a (slower) semaphore and the high lock contention on it was made worse by its ability to be preempted. As the ability to build a kernel without BKL preemption had been removed Linus decided that the BKL preemption would go. Ingo suggested semaphore gymnastics to try and recover performance but Linus didn't like this idea.
As the the BKL is no longer be preemptible it is now a big source of latency (since it could no longer be interrupted). People still want low latencies (that's why they made the BKL preemptible in the first place) so they took the only option left and started work to get rid of the BKL.
(Bah half a dozen other people have replied in the time it's taken me to edit and redit this. Oh well...)
"That part of the code" is the difficult part. The BKL assumption is present in thousands of place all around the kernel, and nobody really know where. You can have two pieces of code, that looks totally unrelated, that happen to work because in all the code path leading to them the BKL is taken. Removing the BKL and "code it all over again" will create this new race condition.
There would be thousands of such, and you'll probably never succeed in debugging it.
The approach suggested in the article is to replace the BKL by a true lock, then "pushing it down", which means understanding WHY that code want the BKL, and get smaller locks instead in subroutines.
For instance, one piece of code could take the BKL because it will change 3 data structure. You could then remove the BKL and use, in the 3 part of code that changes those 3 structure, and use a finer grained lock for each of those.
By iterating this way, you should always get a somewhat working kernel, and slowly kill the BKL.
The recent semaphore consolidation assumed that semaphores are not timing critical. Also it made semaphores fair. This interacted badly with the BKL (see [1]) which is a semaphore.
The consensus was to not revert the generic semaphore patch, but to fix it another way. Linus decided on a path that will make people focus on removing the BKL rather than a workaround in the generic semaphore code. Also, Linus doesn't think that the latency of the non-preemptable BKL is too bad [2].
[1] http://linux.derkeiler.com/Mailing-Lists/Kernel/2008-05/msg03526.html
[2] http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=8e3e076c5a78519a9f64cd384e8f18bc21882ce0
No.
You know, like, remove that part of the code and code it all over again, see what is broken, and continue this way?
It's not that simple. When it comes to locking, there is no "part of the code" that can be replaced. Locking governs interaction between two pieces of code, sometimes two pieces that are very different but have some small thing in common.
Besides, the kernel is too big to just start throwing parts of it out and redoing them from scratch. It's much better to make incremental improvements, because then the people working on them will actual learn how to solve the problem. The BKL is not just a coding problem, but also a people and project management problem.
And the men who hold high places must be the ones who start
To mold a new reality... closer to the heart
Your not wrong, and like you I am going to continue in an over simplified style so the non programs can understand. The part you are leaving out is why you want your locking to be granular.
The granularity is important because you want other threads(jobs) to beable to get something done. At some point there is this thing called a scheduler that assigns your thread to execute, because every job needs a CPU. You get to work until your time allotment is expired or you have to stop because something you need to continue is not availible, because its say "locked".
Think of this like working in a shop along side someone else. You have one set of tools, you need a little screw driver, and a big one to do your work. The other guy needs the little scew driver and a pair of pliers. You want him to put the screw driver down while he is using the pliers so that you can use it if you need to. If he instead puts it in his breast pocket you are going to have to wait to finish your job until he finishes his. Even though its your turn at the work bench(CPU) you can't do anything with it because you don't have what you need. So all you can do is yeild the rest of the time to the other task, and hope he finished up soon.
In the kernel world this really short circuits the work of the scheduler. It might want to give time to other threads and it will but they are going to just turn around and give that up because whichever thread is holding the BKL is likely the only one who can actually do any work. As an end user this means something like data gets read from your network card ok but your sound keeps skipping.
The tricky part with more granular locks though is avoiding circular conditions, these can crash the system. Imagine: Job One needs resource A and B and has A locked, its waiting for B. Job Two needs B and C and has B locked and is waiting on C. Job Three needs A and C and has C locked and is waiting on A. Unless the system can detect this condition which is hard to do in many cases none of these threads will ever be able to run. The kernel contributors likely have some work ahead to eliminate the BKL and not cause these types of problems.
Repeal the 17th Amendment TODAY! Also Please Read http://www.gnu.org/philosophy/right-to-read.html
Wow. It sounds like it's about time someone on the kernel team reads Working Effectively With Legacy Code by Michael Feathers.
I'm a software developer myself on a very large project myself, and this book has absolutely revolutionized what I do. Having things break silently in the kernel is a sure sign that dependency problems in the code exist, and most of this book is about ways to break dependencies effectively and get code under test. And that's the other thing... if they aren't writing tests for everything they do, then even the code they write today is legacy code. Code without tests can't be easilly checked for correctness when a change is made, can fail silently easilly, and can't be understood as easilly.
That's what this book is about, and if things in the kernel have deteriorated to such a state then they need to swallow their pride and take a look at resources designed to cope with this. I know they are all uber-coders in many respects, but everyone has something they can improve on, and from the description they give of their own code, this is their area for improving.
Beware of bugs in the above code; I have only proved it correct, not tried it.
It's hard to test whether you've broken a driver when you don't have the hardware to test with. Perhaps the future will be Qemu emulation of all the different hardware in your system : )
This is not to say that there need to be tests for things that can be caught at compile time or run time regardless of hardware but there is only so far you can take it.
It's not like the kernel doesn't have any testing done on it though. There's the Linux Test Project which seems to test new kernel's nightly. If you ever look in the kernel hacking menu of the kernel configuration you will see tests ranging from Ingo Molnar's lock dependency tester (which checks to see locks are taken in the right order at run time), memory poisoning, spurious IRQ at un/registration time, rcu torture testing, softlockup testing, stack overflow checking, marking parts of the kernel readonly, changing page attributes every 30 seconds... Couple that with people like Coverity reporting static analysis checks on the code. Tools like sparse have been developed to try and so some of the static checks on kernel developer machines while they are building the code.
But this is not enough. Bugs STILL get through and there are still no go areas of code. If you've got the skills to write tests for the Linux kernel PLEASE do! Even having more people testing and reporting issues with the latest releases of the kernel would also help. It's only going to get more buggy without help...
Linux is something like nearly half the servers in existence and most of the top supercomputers. Desktop is a slower road of course, but it is still chugging along slowly but surely. Look at apple, originally a big percentage of desktops, then dropped to almost nothing, now inching its way back up because it got good. Stuff changes. The linux desktop market is big enough for there to be a lot of credible choices just within "linux" itself, there are half a dozen or so really good desktops and dozens of pretty good desktop linuxes out there now. And word gets around. It will be like FF, 0% to now upwards of one quarter to one half depending on where you look around the planet. There's some magic number that is hard to pinpoint but once anything reaches a certain level of use/adoption it really takes off then, usually near as I can see around 10%, then it makes huge jumps. Bad car analogy time, toyota prius is now more than one million cars sold from zero cars ten years ago, and the first with a mass market hybrid system that they really tried to make and sell in decent numbers (compared to honda for example who only fooled around with their insight). Now look, all the major manufacturers either have their own hybrids or will have them shortly. Ten years, that's all it takes once some threshold hits and it looks "real" to joe consumer to go from exotic to normal. I think this year the asus eeePC made linux "real" to a lot of people, so I am expecting ubiquitous linux as a choice to be along shortly with most computer makers as an option. And that is leaving out all the gadgets people use day to day running some smallish embedded linux, gps systems, cellphones, etc.
Comment removed based on user account deletion
There are a few added benefits from a fully real-time OS that most people gloss over.
:)
For example, you will *know* your PC will never become utterly unusable to the point it's unsafe with your data. ie: while it's handling many IO operations (say you're being ddosed whilst transcoding a dvd and flossing with a sata cable) unless you run completely out of system memory. Nothing should run away wasn't design to. This stems from the predictability of code execution times that pre-empting offers.
The predictably allows devices to make guarantees, for example, if your mouse is aware it's going to get a time slice, at worst, every 100ms, at least it'll be doing something every 100ms and you gain a visually responsive mouse (aye, it's not as great as it could be). The non-preempt side of life has your CPU tied up doing work that was sits inside a BKL - ie: dealing with a character device or ioctl - your mouse could be waiting 500ms or 1000ms before updating it's position: giving you the impression your PC is dying.
Code that is stuck inside the BKL isn't pre-emptable (you *must* wait for it to finish.) - there's a lot of it that does a lot of regular stuff. Often this will hold up other cores if you've got a cooperative multi-threaded program. The net effect is a slow PC.
RT systems have a different use: they want a guarentee that something is unable to ever delay the system, in particular interrupts. The BKL allows code to take a time slice and run away with it, because you thought it was very important and wrapped it in [un]lock_kernel. This then delays IRQs (IRQs cant run until the lock has finish - at least, I'd like to believe second level IRQs can't run, I'm unsure of the specifics) which will delay data coming in and out of your PC - hard file, disk and display buffers suffer: they fill and you start to loose data because there's nothing dealing with it.
Preempt kernels are good
(viva the Desktop)
Matt
Dragonfly is just the latest branch of development of a codebase with decades of history!
On the other point (which I wasn't disputing or answer)--about the giant lock, I don't know much about the linux internals, but it sounds like the bsd GIANT is the same as the linux BKL. FreeBSD is in something like its 6th year of removing giant!
It's worth pointing out here that the kind of races (bugs) introduced by faulty locking in general suffer from a very important problem: YOU CANNOT TEST FOR THEM.
Race conditions are mostly eliminated by design, not by testing. Testing will find the most egregious ones but the rest cause bizarre and hard-to-trace symptoms that usually end up with someone fixing them by reasoning about them. "Hmm" you think to yourself "that sounds like a race problem. Wonder where it might be?" and thinking about it, looking at the code, inventing scenarios that might trigger a race; that's how you find them.
The proposed outcome is for there to be increased opportunities to switch between programs/kernel or to run multiple things at the same time.
For those who enable the option this should reduce the chance of a hardware's buffer not being filled in time (so audio is less likely to skip in demanding environments). If you are an audio recording person or need VERY (less than hundredths of seconds) fast responses all the time your experience should improve. If you run VERY big workloads that have lots of pieces that can happen simultaneously on computers with 2-1024 CPUs, your experience (increase in work finished per second say) should improve. Typical desktop performance may improve a little if you have multiple CPUs/cores but one would guess not enough to be noticeable without careful measurement.
The trade off is increased risk of system hangs / data corruption due to the programming being trickier although instances of this happening should fall over time with popular hardware.
Yup and Mica designed in the late 80's as a Microkernel replacement for VMS.
http://en.wikipedia.org/wiki/DEC_PRISM
It was canned by DEC and Cutler and most of the engineers were hired by Microsoft where they wrote Windows NT.
Interesting Mica was supposesed to have multiple personalities - posix and VMS. Similarly Windows NT had Win32, Win16/Dos, Posix and originally OS/2.
The NT model is to packetize IO as IRPs and pipeline requests across multiple CPUs. E.g. on Windows NT under very heavy load a disk driver can be starting a request on one CPU, in an ISR on another and in an APC on yet another. That makes the rules for writing drivers quite complex of course, but that's a price worth paying. That's an extreme case though, normally DPCs have an affinity for the CPU that generated the interrupt because that CPU is likely to have the driver code in its cache. But the fact it's possible illustrates how far Mica/NT is from the tradional Unix model of treating device drivers as synchronous subroutines that only one kernel thread can be in.
And it seems pretty obvious to me that Mica/NT works like this because they were designed from the start to run on a lot of small Risc processors wired up as an SMP machine. Unix wasn't - back when Unix was designed the dominant paradigm was one big CISC CPU. In that world, drivers as a synchronous library is actually better since you don't have the overhead of building packets.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;