Slashdot Mirror
Linus Pooh-Pooh's Real-Time Patch
An anonymous reader submits "Speaking with CNet via email, Linus Torvalds appears to be in no hurry to accept the latest real-time patches from embedded specialist MontaVista into the mainstream kernel, at least not "at this time." Nontheless, MontaVista's new open-source real-time Linux project could broadly expand commercial opportunities for the open source OS, especially in telecom initially, where real-time Linux will likely play on "both ends of the wire." For example, Linux is already making progress in smartphones."
Direct link to they story.
Wouldn't a real time patch violate software update patents
- i don't think so!
I learned something new!
Real Time Operating Systems. Now you know!
He might not be in a hurry, but I'd be surprised if he doesn't realize how this could help Linux. Maybe there are some stability problems with it, but then, I doubt that too. Does anyone have any experience with it? Maybe he's just waiting for the right time, not the earliest time.
Hurricane Ivan: A 17th century prison collapsed. All of the inmates escaped.
Linus is right not to accept this patch into the main kernel tree. What this would amount to is shoehorning Linux into a shoe it's too large to fill, and there is absolutely no reason burden all other Linux distros with this mess. Come on, MontaVista, don't try to cock things up for the rest of Linux just because you're too lazy to patch the kernel yourself.
BLING BLING. Meet the architecture that's changing everything.
Not exactly real time response here.
Linus' job is to say no. Here, he even gives rationale.
There are no trails. There are no trees out here.
I guess Linus does not realize the full potential of what real time functionality can do to Linux.
Yes, I'm sure he has no idea...
"A door is what a dog is perpetually on the wrong side of" - Ogden Nash
While useful for certain applications (if I press the "abort" button on a missile, I want it processed NOW and not 5 seconds after it explodes..), I don't see what a hard realtime operating system would do for desktop systems.. then again, maybe I'm completely missing the point?
I am the maverick of Slashdot
Gen. Melchett: Is this true, Blackadder? Did Captain Darling pooh-pooh you?
... by pooh-pooh!
Cpt. Blackadder: Well, perhaps a little.
Gen. Melchett: Well then, damn it all, how much more evidence do you need? The pooh-poohing alone is a court-martial offence!
Cpt. Blackadder: I can assure you, sir, that the pooh-poohing was purely circumstantial.
Gen. Melchett: Well, I hope so, Blackadder. You know, if there's one thing I've learned from being in the army, it's never ignore a pooh-pooh. I knew a major: got pooh-poohed; made the mistake of ignoring the pooh-pooh -- he pooh-poohed it. Fatal error, because it turned out all along that the soldier who pooh-poohed him had been pooh-poohing a lot of other officers, who pooh-poohed their pooh-poohs. In the end, we had to disband the regiment -- morale totally destroyed
Is it just me, or does this article sound like it's fueling steam for a fork of Linux development? If not adding steam for a fork, I have to say it's arrogant
I believe that it's appropriate to have a fork for realtime enhancements. I remember HP's philosophy in the 80's was to have a Real Time OS and a Business OS. They have competing goals. No need to blend them and end up with a compromise!
RTLinux takes executive hold of the system and runs the Linux kernel as a mere user process.
Perhaps Linus objects to this topsy-turvey approach and would prefer Linux to be re-written so it's actually completely preemptible, capable of handling interrupts with RT guarantees all by itself, etc?
What makes you think it wouldn't just be annother kernel config parameter to most people?
Dunno if this is even slightly relavent at all but this is from one of the other people doing Prempt things with the kernel.
finally, i went for correctness primarily, not latencies. I checked
out the MontaVista patches and they categorize roughly 30 spinlocks
as the ones that are necessary to be 'raw'. Unfortunately this is
inadequate, my patch excludes 90 such locks and it's still probably
not a 100% correct conversion. The core kernel needs changes in the
locking infrastructure to get rid of most of the these 90 non-mutex
locks.
Your hair look like poop, Bob! - Wanker.
You are missing the point insofar that Linux doesn't aim just for the desktop and/or server market, but also, as stated above, for the embedded market where hard real time often is simply necessary. Or why else do you think OSs like QNX still exist? :)
Whether this "aim for everything"-attitude is a good one for the Linux kernel as a whole remains unquestioned for now.
After digging around some (see the posts above), Linus seems to feel that the patches are too intrusive, which I can certainly understand. Hard Real time promises are not good in the general case, which is why most OSen don't bother. For the cases that require them, traditionally there are specialized OSen (QNX for example) that have the functionality needed. I'm not sure about this, but I believe that there are some specific hardware requirements for true RT. The scheduler is also radically different.
It would not suppise me at all if this lives a long and fruitful life outside of the standard kernel, as a stand-alone patch set. That's not even a bad place to live, especially since the requirements are rather esoteric.
Zapman
From the email threads and writeups on KernelTrap, it seems as though Linus (and his Lieutenants) have some issues with the invasiveness and maintainability of the patch, which are reasonable concerns from the maintainers.
Ingo Molnar - a RedHat employee/kernel hacker - has some patches that are similar in scope but different (and most likely preferable from a performance and maintainability viewpoint) in approach.
Read about them here and form your own opinion:
Linux: Real Time Kernel Prototype
Linux: Realtime Preemption
First we had a post on Hibernate with no explanation. Then a post on OQO and now we have one on 'Linux'. When will the madness end?!?!? ;)
Historically, Linus has never liked merging in great glops of code that touch the kernel in many places. It is disruptive to his maintenance of the kernel and it is disruptive to his lieutenants and their sub projects. The article even hinted at how Linus expects those with a major patch like this to handle things. Montavista needs to break this up into bite size chunks that can be slowly merged into the kernel and gives everybody time to get up to speed. Since it can have a major effect on how the kernel operates, it needs to at least be a compile-time option.
Linus has even told IBM "no" on occasion. Not hurrying things like this is far better for the quality of Linux than any feature a contributor may want in. Linus isn't flatly refusing Montavista. He most certainly isn't flatly refusing a major feature like hard real time. He is expecting Montavista to participate the way other developers are expected to participate. In particular, Montavista doesn't get to disrupt the work of hundreds of developers because their gargantuan patch was simply dumped in the main dev tree.
This isn't petty dictatorship. The kernel devs are a battle scarred lot who don't just chuck things in because it would be "cool".
What kind of smooth-it-over headline is this? Poo Poo? If this were Bill Gates instead of Linus, we'd say he's "blatantly ignoring", "throwing aside", "totally dismissing". But Poo Poo??
Tech, life, family, faith: Give me a visit
Tigger said to be hopping mad in protest; Piglet leading soft-spoken sit-in. Honey at 5:00.
-Rob
Marriage doesn't have to suck!
Has anyone ever taken a look at some of the stuff available in the 2.6 configuration options? Do a 'make menuconfig' and browse through the "General Setup" and "Processor Type and Features" submenus. A bunch of it is wholly useless to 99.9% of the installations out there.
But it's there as an option, if you want it, like most everything else. Have a tulip ethernet card? Include the driver. If you don't, leave it out. Old BIOS doesn't do ACPI? Leave it out. Don't need a keyboard driver because it's an appliance system? Leave it out.
Why the hell not just include the real-time business as options? Is the maintenance really that much more challenging?
Here's something that Montavista has contributed to the Linux kernel - PRAMFS. A quote (emphasis mine):
/var/log, or a user address book in a cell phone or PDA.
Many embedded systems have a block of non-volatile RAM seperate from normal system memory, i.e. of which the kernel maintains no memory page descriptors. For such systems it would be beneficial to mount a fast read/write filesystem over this "I/O memory", for storing frequently accessed data that must survive system reboots and power cycles. An example usage might be system logs under
[...]
2. If the backing-store RAM is comparable in access speed to system memory, there's really no point in caching the file I/O data in the page cache. Better to move file data directly between the user buffers and the backing store RAM, i.e. use direct I/O.
They've described that they want to use this stuff in a cell phone or PDA, yet have described an NVRAM technology that does not exist (as fast as system memory?). Methinks that they're working with Intel on some new fangled NVRAM, (hint, look for Ovonic). Samsung appears to be working with PRAM as well.
So this MontaVista file system is a PRAM-File System, maybe...
Life is the leading cause of death in America.
Having smart appliances and energy controls securely networked would be a boon to Linux. The real-time kernel would allow speedy monitoring of these type of devices.
At the risk of sounding redundant, how is MontaVista's implementation significantly better or different from pre-existing real-time Linux interfaces, such as FSMLabs' RT Linux or DIAPM's RTAI?
There's a hard, probebly irremediable fact about Real Time Operating Systems: the price for being able to [i]guarantee[/i] a specific response time is a [i]slower overall response time[/i]. "Realtime" isn't magic (though, as with all buzzwords, people tend to act like it is). It still must heed all the inherent limitations of the hardware.
Imagine that you run a pizza shop that MUST meet a certain delivery time guarantee or fail (go out of business--an RTOS MUST meet the guarantee to "be in business" at all). Before you were an RTOS, you could afford to promise pizzas in 15 minutes, with a 90%+ success rate, but if your head will roll if you fail at all, you won't advertise anything better than 30 -or even 60- minutes. I mean, what happens if a custom pizza gets ruined in the oven? You need time to make a new one.
You'll also need more hardware for the same tasks (more delivery cars), restrict services (smaller delivery area, fewer options), and institute effort-intensive safeguards to assure that no pizza order slips through the cracks. As I said: RTOS isn't magic; adding NEW performance demands won''t magically enable you to do more with less. Quite the contrary, it usually means doing less with more -- but presumably doing it better (assuming that the new requirement *is* better for your specific needs).
Would you embrace a hardware technology that slowed down your computers, and offered little or no benefit for most (or all) of the tasks you do? There are plenty of examples in he market, and we rightfully shun them as "unnecessary for us". That's the choice Linus faces: most users won't experience any benefit, so why include it in the kernel and make everyone pay the (performance and complexity) price?
I applaud the availability of a Real-Time patch or variant (I've wanted one for a long time, and I've used Wind River for those applications), but for most people or even 99% of my applications, it's pure downside, even if reworking the kernel to allow its inclusion only decreases performance or complicates programming by 1%.
Sure, in time --maybe a couple of years-- it may be streamlined until the RTOS burden is miniscule. Until then, Let the Real Time people deal with the issues and limitations inherent in their task. 99.99% of us don't need the unnecessary baggage in our OS. It'd be like mandating infan/child car seats in all cars, whether they carry kids or not.
What can an RTOS do that Linux can't that wouldn't be better handled in hardware? If responding to a given event is really THAT time critical, then perhaps you shouldn't be handing the event all the way up to your systems software for a response... In my experience, most problems that people claimed demanded "real time" to solve could be more easily solved by adding more buffers.
"Freedom means freedom for everybody" -- Dick Cheney
Yes. All the other obscure things which only 0.1% of everybody uses, they are small isolated pieces of code (like some random driver). What we're talking about here is adding lots of highly non-trivial code to the core of linux (you know the kernel/ subdirectory of the kernel source) which only 0.01% of people will actually need/use. So, yes.
I also think it would be quite arrogant of the RT people to expect this to be added without serious thought (and possible reworking). (NOTE: I'm not saying they do/did expect it to, just that it would be arrogant to do so)
HAND.
The existence of the code in the main kernel, regardless of whether it is enabled at compile time, affects everybody. It's not like you're going to have a single place with
...lots of code
#ifdef REAL_TIME_KERNEL
#endif
It's going to change a lot of things in a lot of places. Ideally, it will make infrastructure changes which benefit everyone, by abstracting certain elements of the code, which then makes its own specific features fit in nicely. Unfortunately, this is very difficult to do with real-time scheduling.
Here, have a cookie.
My other first post is car post.
Linus merely said "not at this time," and gave his rationale. To me, this hardly qualifies as "pooh-poohing." Therefore, I'd say the article headline is misleading, and designed merely to stir up emotions rather than foster rational dialog.
If you want hard real time and protected mode, you need an architecture like that of QNX, where almost everything runs in user space. File systems, drivers, and networking are all user programs, intercommunicating by message passing. The kernel only handles CPU dispatching, memory management, and message passsing.
In an architecture like that, everything in user space is preemptable, without any extra work in the system services. There are no long latencies in the QNX kernel; they were all taken care of years ago.
As Linus points out, though, few consumer embedded devices really need hard real time. Most media-related stuff can paper over delays with buffering. A classic comment is, "You run your web server on Linux. You run your nuclear reactor on QNX".
Automotive systems, though, really need it. QNX is big in that market.
Here is the LKML thread discussing this (including explanation of why it isnt accepted into mainline).
Speaking as someone who actually downloaded and tested these patches, I would not worry too much. This stuff is all very rough around the edges, though it has amazing potential.
If the patches were mature and worked well, and Linus rejected them, it would be news. For now he is just saying "Show me the money". Nothing new, the burden of proof is on people who introduce new features like this to prove them stable, and it just hasn't happened, yet.
while (CheckOKtoProceed());
You see, the program "spins" until CheckOKtoProceed() returns true. The alternative is a call to a Yield(), Wait() or Sleep() function that 1) blocks execution until some condition is satisfied, and 2) tries to yield control to some other pending process while that is happening.
The trouble with a spin lock is that it hogs the processor. The trouble with the other kind of lock is that it allows another process to proceed, but it may not be safe to allow that on account of a data structure not being in a coherent state of update. A mutex is a kind of lock that by agreement of its use allows only one such process to proceed. A non-mutex lock doesn't offer such protection.
The argument is that the proposed modification make the kernel much more preemptable and do less spin locking that can kill response, but each element of the proposed modification would need to be checked and tested very carefully that after the change there aren't issues regarding the protection of data structures from multiple processes that could change it along with all kinds of mind-bending subtle bugs that can arise.
People throw around bloat with great abandon, and usually without any real rationale behind the term.
My OS is full-featured. Yours has feature creep. His is bloated.
Next time you design a reactor, you go ahead and do it your way.
If you were blocking sigs, you wouldn't have to read this.