[9fans] Threads: Sewing badges of honor onto a Kernel

[9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

A few of us had a discussion this evening on the state of Linux and
bloat in the Operating System development community. This isn't an
event isolated to this evening, however tonight had a bit of spice.

Andrey gleamed this bit of interesting text on the O'Reilly[1] website.
Let's have a look at a snippet, shall we?

Under the "Kernel Space and User Space" heading:

| ...  Another example of this happened in the Plan 9 operating system.
| They had this really cool system call to do a better process fork--a
| simple way for a program to split itself into two and continue processing
| along both forks. This new fork, which Plan 9 called R-Fork (and SGI later
| called S-Proc) essentially creates two separate process spaces that share
| an address space. This is helpful for threading especially.

| Linux does this too with its clone system call, but it was implemented
| properly. However, with the SGI and Plan9 routines they decided that
| programs with two branches can share the same address space but use
| separate stacks. Normally when you use the same address in both threads,
| you get the same memory location. But you have a stack segment that is
| specific, so if you use a stack-based memory address you actually get
| two different memory locations that can share a stack pointer without
| overriding the other stack.

| While this is a clever feat, the downside is that the overhead in maintaining
| the stacks makes this in practice really stupid to do. They found out too
| late that the performance went to hell. Since they had programs which used
| the interface they could not fix it. Instead they had to introduce an additional
| properly-written interface so that they could do what was wise with the stack
|  space. ...

This commentary from Linus Torvalds on rfork[2] proclaims that a single, unified
stack space for multiple threads is not only beneficial but imperative for proper
thread propagation. What we are wondering is ... why?

What scenario would make a unified stack segment beneficial in a multi-threaded
execution environment. Is this something that should be an optional parameter
on-the-fly in a call to rfork() ? Does clone() in Linux allow you to decide, or, does
it enforce the unified stack segment? Doesn't unification create collisions within
scopes of process namespaces through recursion and non-leaf code vectors?

Since we couldn't figure out the logic behind this Lunix tantra, we thought
trolling the public would be a slick idea. Who has an opinion on (or a defense
for ;-)) this statement?

Also, who is the "They" that "found out too late that the performance went to
hell".

Curious minds are interested.

Don (north_)

References:

[1] http://www.oreilly.com/catalog/opensources/books/linus.html
[2] fork(2)

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Scott Schwartz]

Beware that this mailing list doesn't permit postings
with more than a few cc-s.

  To: 9fans@cse.psu.edu
  cc: torvalds@osdl.org, sionide@beefed.org, fa1th@beefed.org

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> the url is wrong -- s/books/book/

Ah, yep. Can't paste over VNC. Thanks, Andrey

full URL is:
http://www.oreilly.com/catalog/opensources/book/linus.html

Don (north_)

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> also, you omitted the conclusion from the chapter, which is (imho) the best:

I didn't want to paste *too* much into the email ;-)

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[andrey mirtchovski]

the url is wrong -- s/books/book/

also, you omitted the conclusion from the chapter, which is (imho) the best:

| In fifteen years, I expect somebody else to come along and say, hey, I can
| do everything that Linux can do but I can be lean and mean about it because
| my system won't have twenty years of baggage holding it back. They'll say
| Linux was designed for the 386 and the new CPUs are doing the really
| interesting things differently. Let's drop this old Linux stuff. This is
| essentially what I did when creating Linux. And in the future, they'll be
| able to look at our code, and use our interfaces, and provide binary
| compatibility, and if all that happens I'll be happy.

andrey "just fanning the flames" mirtchovski

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

this is a peculiar distortion of a peculiar episode.  the original
issue linus
objected to, and rather rudely if i remember right, is that if the
stacks are
distinct, there can be confusing bugs if you pass pointers around that
accidentally happen to refer to stack addresses.  in practice, we never
do this because the thread library makes it all but impossible.  however
this issue really rankled linus,

on the other hand, having a separate piece of address space with
different contents is a powerful idea.  it's called something like
per-thread space, and it's really convenient for many implementation
details in threading.  but the feature is (or was; i really don't know
the
situation any more) hard to provide on linux and in fact the clone
interface used to be really nasty because the two processes returned
from the syscall with two processes sharing the same stack.  i
understand
something has been done to make the code after a clone on linux
easier to write, but when linus was dyspeptic about rfork, the linux
dance was really tricky.

so it's just a detail, a design choice with advantages and disadvantages
either way.  win some, lose some. yet somehow our code
is 'stupid' and linux is 'proper', so i guess the snarkiness endures.

-rob

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

in case that wasn't clear enough, i have no idea what linus is talking
about when he says we had 'overhead' that made it 'really stupid'
'in practice'.

-rob

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> so it's just a detail, a design choice with advantages and disadvantages
> either way.  win some, lose some. yet somehow our code
> is 'stupid' and linux is 'proper', so i guess the snarkiness endures.

What I am mainly confused about, is... We're talking about user
context threading. In user context threading we've got the abstractions
of a BSS, Data, Text, Stack, (etc in some cases) to maintain a sense
of program-persistant-object-classification. As a coder, we know
and can specify (for the most part) where these "objects" will end
up.

So the question to me is, why would you *want* a unified stack?
The only state in which this might be relevant is threads maintained
in ASM where data is passed back and forth via Stack and manipulated
via Registers. Otherwise, things just don't make much sense. In a
land of higher level languages, this semantic seems irrational.

This is mainly what I love about Plan 9's threading implementation, that
it unifies memory in the heap, and not at the stack. I agree that this
creates a powerful tool not only in efficiency, but also in reliability.

I see Linus' logic as more of a negation of both efficiency and reliability.

I'm not interested in starting a flame war, but I *am* interested in
understanding the logic, because I just can't see it clearly from my
point of view.

Don (north_)

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[ron minnich]

I did the original rfork for freebsd, and I wrote it so I split the stacks
too. It was lovely, since you just rforked and all was shared save the
stack -- no assembly required. When they did the committed version, after
fork, all was shared.  This sucked. It was quite a mess to fix up your
stack so you weren't walking on each other's stack -- required assembly
goo to make it go. I hated it and stopped using it.

So I like plan 9 rfork quite a bit. I think Linus/Linux are wrong.

ron

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:

> So the question to me is, why would you *want* a unified stack?

well, you don't. It makes programming a nightmare and you need assembly
goo to seperate the stacks after a fork, else you will be stepping on
each others' stack. As the freebsd users of their official version of
rfork() learned, that meant you hit segv walls very quickly. Ouch.

> I see Linus' logic as more of a negation of both efficiency and reliability.

I don't even understand it.

ron

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[George Michaelson]

what happens on MP systems?

-George

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

i agree with you. i think the model, although done for reasons of
expediency, has some advantages.  the split stack provides a
unique and powerful memory abstraction: same address, different
value.  (c.f. the up and mp registers in the kernel.)

> I see Linus' logic as more of a negation of both efficiency and
> reliability.

i don't understand his logic.

-rob

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Fri, 27 Feb 2004, George Michaelson wrote:

> what happens on MP systems?
>


not sure what you're driving at.

ron

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> > I see Linus' logic as more of a negation of both efficiency and
> > reliability.

> i don't understand his logic.

Well, I interpreted his logic that each process, itself, should be the
manager of how data is abstracted into each thread. Further, that
the kernel should not impose an interface on such userspace
processes. When I think in terms of future scalability and dynamic
code design, this *seems* to be a good thing.

However, over time, we've seen that the lack of a conformed
interface has prompted the design of such an interface: pthreads,
etc. So, in the end, the necessity for POSIX compliance (or what ever
compliance you prefer) is perceived as paramount.

Though, this could have been avoided and deemed unnecessary if
these kernels had only implemented an interface in the first place.

I believe the shortsightedness of not imposing a strict type created
too many possibilities. This creates a bigger problem when dealing
with libraries imported into a system with no defined interface, and
you end up with the desire for one.

So, in the end, Plan 9 looks like the better solution. The foresight was
there to see a necessity for conformity where all possible permutations
of a given set tended to yield outcomes meeting at the same axis:
thus an interface.

I duno, that's my two cents, anyway. But, that's dependant on me
getting Linus' logic correct, which is why he was CC'd.

Don (north_)

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[George Michaelson]

On Thu, 26 Feb 2004 22:59:36 -0700 (MST) ron minnich <rminnich@lanl.gov> wrote:

>On Fri, 27 Feb 2004, George Michaelson wrote:
>
>> what happens on MP systems?
>>
>
>
>not sure what you're driving at.

on an MP system, the forked process/thread has chances to be run on a distinct
CPU underneath. I would have thought the complexity of managing a shared stack
was magnified in that circumstance.

process migration -wise?

-George

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Scott Schwartz]

Al Viro once told me that if Plan 9 had shared libraries, rfork wouldn't
work as well.  Apparently they're concerned about page tables when there
are lots of different combinations of sharing going on.

[9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Thu, 26 Feb 2004 dbailey27@ameritech.net wrote:
>
> This commentary from Linus Torvalds on rfork[2] proclaims that a single, unified
> stack space for multiple threads is not only beneficial but imperative for proper
> thread propagation. What we are wondering is ... why?

(1) There are valid reasons to pass pointers between threads, and yes,
they can be pointers to thread stack areas.

(2) No existing common hardware can do "partial TLB flushes" efficiently.
The main performance advantage of threads is that you don't need to flush
the VM state on thread switches. If you have partial VM area sharing, you
lose that whole point.

(3) Implementation sucks. Irix and Plan-9 both get it wrong, and they
_pay_ for it. Heavily. The Linux code is just better.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> (1) There are valid reasons to pass pointers between threads, and yes,
> they can be pointers to thread stack areas.

Do you have code from existing Linux implementation that exemplifies
this scenario? What are your opinions on the benefits? If there is existing
code that exemplifies this scenario, how often does it propagate? Is it
often enough to denounce a standard interface in the kernel?

Also, what are your comments, Linus, on shared library execution altering
the scope of an interface from kernel context to user context? Even if there
isn't a unified stack, there is still an interface defined by such thread projects
as POSIX threads, etc. Doesn't that just create a standard interface in user-
land that could be sped up, if moved to kernel context?

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> Anything that accesses any process arguments would access the stack of the
> original thread.

I'm not sure I can agree here. Arguments are not meant for the scope of an
entire process, but for the initialization of a program's state. In a program that
parses a fully qualified file name path derived from command line arguments,
the initial state should only open a file descriptor and pass that descriptor to
a child thread while performing other duties. This simple example depicts
how a process' arguments should be restricted to initialization. After all, if
the process can't open the file, there is no need to thread.

It seems that allowing such a misuse of resource would be more of a
rationalization for allowing someone's bad programming, rather than a
strong argument against defined thread interface.

Can you elaborate on your synchronization example, please?

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
>
> Do you have code from existing Linux implementation that exemplifies
> this scenario?

Anything that accesses any process arguments would access the stack of the
original thread. That's the most obvious one. But synchronization with
completion events on the stack is also perfectly feasible (the kernel does
exactly that internally, for example - and user mode could do the same).

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Donald Brownlee]

Linus Torvalds wrote:


>
> (1) There are valid reasons to pass pointers between threads, and yes,
> they can be pointers to thread stack areas.
>

OK. If one of the threads is a "debugger" thread and another
is a "thread to be debugged," then it would be helpful if
the "debugger" thread can access the stack of the "thread to be debugged."

Why? So it can muck with the stack, say, to alter a return
address. Whatever!

But to inspect or modify the stack of another thread requires
some knowledge of that other thread's state. Like, is it
stopped?

D.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
>
> I'm not sure I can agree here. Arguments are not meant for the scope of an
> entire process, but for the initialization of a program's state.

Wrong. You have zero basis for your assertion, and it's simply bogus.

Would you say that the process environment[] array is also only applicable
to the first thread?

But it doesn't matter. Regardless, threads should see each others stacks.

> Can you elaborate on your synchronization example, please?

There are tons of examples in the linux kernel. Search for any use of
"struct completion" - most of them are allocatied on a threads stack
(which is obviously distinct between different threads), and other threads
and interrupts will use that data structure to wake up the thread.

But hey, I'm not interested in trying to convince you. Feel free to think
I'm wrong.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 01:48:09AM -0500, dbailey27@ameritech.net wrote:
>
> > Anything that accesses any process arguments would access the stack of the
> > original thread.
>
> I'm not sure I can agree here. Arguments are not meant for the scope of an
> entire process, but for the initialization of a program's state. In a program that
> [ ... ]

Not only.  Presumably, the stack is duplicated on rfork(2) which means
that reading the arguments is OK, writing them would be a problem.

But just because the arguments can be changed does not mean that it
can't be deprecated in the case of threads.

++L

PS: I do think this particular one is a quibble, though.  My choice
would be along the lines of: "having made choice "A", is there a
mechanism to implement choice "B" from within it?"  If there is,
then only efficiency criteria can be used to select "A" over "B".
If there isn't, then either the reverse applies in which case "B"
is the obvious contender, or the two are distinct solutions, each
requiring implementation.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> Wrong. You have zero basis for your assertion, and it's simply bogus.

Clean, logical programming isn't a sound basis? Or is it only not a sound
basis because you've decided so? Saying that I have zero basis for my
assertion is a bit bogus, in it self, don't you think? I CC'd you my
original email to 9fans because I was hoping to understand your logic
behind the rfork() comments, not to start a flame war.

> Would you say that the process environment[] array is also only applicable
> to the first thread?

Yes. That's why we have env(3) in Plan 9. So each thread can alter and
read the environment, globally, in a process.

> There are tons of examples in the linux kernel. Search for any use of
> "struct completion" - most of them are allocatied on a threads stack
> (which is obviously distinct between different threads), and other threads
> and interrupts will use that data structure to wake up the thread.

Okay, but is there a reason why it *needs* to be on the stack? What's the
rationale behind this usage.

> But hey, I'm not interested in trying to convince you. Feel free to think
> I'm wrong.

Well, I'm hoping to determine what really is the best thing for OS design,
as a whole. It isn't in my interest to determine who is right or wrong, only
what makes sense universally. If we are really all here to try and design
code that works the best it can, that's all that matters. Ego is irrelevant and
only gets in the way of those discoveries that can really help us materialize
the better code.

Even if there are no more comments on this thread, I still learned a lot
by asking the questions. If that stirrs up some contraversy, so be it. I'd
rather learn and ask questions than stay silent and experience nothing.

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

the argument about TLB flush times is interesting.

> But it doesn't matter. Regardless, threads should see each others
> stacks.

and on plan 9, they do.  they just can't see each other's *stack
segments*
within their own address space.

there seems to be confusion on this point.  plan 9's kernel splits the
stack segment after a fork, but in the normal state, the processes run
with the sp in shared memory.  the marvelous properties of the
same-address-different-contents split stack segment is used only
during the fiddly bits of process manipulation and to store per-process
data.

how does linux store per-process data in user space?

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>the argument about TLB flush times is interesting.

it might have been if Linux hadn't gone through an O(n) scheduler before it did that context switch.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[a]

// It isn't in my interest to determine who is right or wrong,
// only what makes sense universally.

let me help, then: nothing. nothing makes sense "universally".

Linux is Linux. Plan 9 is Plan 9. the two systems have very different
design goals. that's not to defend linux's fork design or any other
particular decision, but asking (by implication) why linux doesn't
just do something like env(3) misses those goals.

personally, i'm less interested in arguing point 1 from linus's post
(there might be valid reasons to do this sometimes), which seems at
least plausable, and much more interested in the other two. on the
second, i think clarification is in order: that's the main advantage
of threads over what, processes? and on that last one (the bit on
performance and code quality) i'm *really* curious what he's talking
about. do we have a particularly heavy fork? i'm confused. saying
"The Linux code is just better." just *begs* for support.

oh, and it's "Plan 9", or sometimes "plan9", but never "Plan-9".
thanks.
ア

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 02:06:57AM -0500, dbailey27@ameritech.net wrote:
>
> > Would you say that the process environment[] array is also only applicable
> > to the first thread?
>
> Yes. That's why we have env(3) in Plan 9. So each thread can alter and
> read the environment, globally, in a process.
>
Is Torvalds really saying that environment[] is held _in_ the stack?!
No wonder he was reluctant to copy it!  Specially when using "bash".
But I must be mistaken.

++L

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> > So the question to me is, why would you *want* a unified stack?
>
> well, you don't. It makes programming a nightmare and you need assembly
> goo to seperate the stacks after a fork, ...

yup

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> The rationale is that it's incredibly more sane, and it's the logical
> place to put something that (a) needs to be allocated thread-specific and
> (b) doesn't need any special allocator.

You've just proven my point. Thread specific. Being Thread specific, it
is data that is reserved to the scope of a single thread. Nothing more.
If you want more scope there are many more usages of memory that
are better utilized.

I'm not arguing against automatic variables. I'm arguing against using
automatic variables anywhere but the scope of stack in which they are
defined.

I don't know where you're getting religiousness, but, there's nothing
wrong with trying to do the best for all of your users. That's why
companies research their target audience, so they can improve a
product according to the most global needs of a given group.

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

> And at the same thing, the Linux VM (from an architecture standpoint) is
> simple.

Do you notice that you had to say 'from an architecture standpoint' ?
I don't think it's simple at all. I agree it seems to be more efficient
(although I don't have measures to support this), but it's
contorted code, at best.

Regarding what's broken or not, it mostly depends on bugs, and bugs
depend mostly on complexity.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> > what happens on MP systems?
> >
> not sure what you're driving at.

i do.  they have forgotten their lister and the old (now quick):

    s = splfoo();

    /* critical region */

    splx(s);

usually simplifies things, unlike the lunacy you have to code
on linix >= 2.2 -- i've coded it; it made me sick.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
>
> Okay, but is there a reason why it *needs* to be on the stack? What's the
> rationale behind this usage.

The rationale is that it's incredibly more sane, and it's the logical
place to put something that (a) needs to be allocated thread-specific and
(b) doesn't need any special allocator.

In short, it's an automatic variable.

You are arguing against automatic variables in C. You apparently do so for
some incredibly broken religious reason, and hey, I simply don't care.

> Well, I'm hoping to determine what really is the best thing for OS design,
> as a whole.

I can guarantee you that the broken behaviour of SGI sproc and plan-9
rfork i sa major pain in the ass for VM management.

I'm obviously very biased, but I claim that the Linux VM is the best VM
out there. Bar none. It's flexible, it's clearly very portable to every
single relevant hardware base out there (and quite a few that aren't
relevant), and it's efficient.

And at the same thing, the Linux VM (from an architecture standpoint) is
simple. There are lots of fundamentally hard problems in teh VM stuff, but
they tend to be things that _are_ fundamentally hard for other reasons (ie
page replacement algorithms in the presense of many different caches that
all fight for memory). But the actual virtual mapping side is very simple
indeed.

And the plan-9/irix thing isn't. It's an abomination.

And there are real _technical_ reasons why it's an abomination:

 - it means that you cannot share hardware page tables between threads
   unless you have special hardware (ie it is either fundamentally
   unportable, or it is fundamentally inefficient)
 - it means that you have to give different TLB contexts to threads,
   causing inefficient TLB usage. See above.
 - it means that you need to keep track of separate lists of "shared" and
   "private" VM mapping lists, and locking of your VM data structures is a
   complete nightmare.
 - it almost certainly means a lot of special cases, since on the magic
   hardware that does have segmented page tables and where you want to
   share the right segments, you now have magic hardware-dependent limits
   for which areas can be shared and which can be private.

But yes, I'm biased.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> personally, i'm less interested in arguing point 1 from linus's post
> (there might be valid reasons to do this sometimes), which seems at
> least plausable, and much more interested in the other two. on the
> second, i think clarification is in order: that's the main advantage
> of threads over what, processes? and on that last one (the bit on
> performance and code quality) i'm *really* curious what he's talking
> about. do we have a particularly heavy fork? i'm confused. saying
> "The Linux code is just better." just *begs* for support.

I wasn't interested in #2 because #2 didn't make much sense to me.
I wasn't interested in #3 because I had no desire to discuss things
resembling personal attacks.

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> (3) Implementation sucks. Irix and Plan-9 both get it wrong, and they
> _pay_ for it. Heavily. The Linux code is just better.

5M SLOC for the kernel?  look at list.h

better?  mon cul.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> Can you elaborate on your synchronization example, please?

FOTFLMAO :)

nice one, don.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Lucio De Re wrote:
>
> Is Torvalds really saying that environment[] is held _in_ the stack?!
> No wonder he was reluctant to copy it!  Specially when using "bash".
> But I must be mistaken.

Take a look if you don't believe me.

There is a _lot_ of state on the stack. That's how C works.

It's perfectly valid behaviour to do something like this:

	myfunction()
	{
		mytype myvariable;

		pthread_create(.. &myvariable);
		...
		pthread_join(..);
	}

where you give the thread you created its initial state on your own stack.
You obviously must keep it there until the thread is done with it (either
by waiting for the whole thread as in the above example, or by using some
other synchronization mechanism), but the point is that C programmers are
used to a fundamentally "flat" environment without segments etc.

And what a private stack is, is _nothing_ more than a segment.

And I have not _ever_ met a good C programmer that liked segments.

So in a C/UNIX-like environment, private stacks are wrong. You could
imagine _other_ environments where they might be valid, but even those
other environments would not invalidate my points about efficiency and
simplicity.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

> So in a C/UNIX-like environment, private stacks are wrong. You could
> imagine _other_ environments where they might be valid, but even those
> other environments would not invalidate my points about efficiency and
> simplicity.

as i said before, the stacks are not private.  you're right, that's a
bad thing.
that's why they're not private.

the segment called 'stack' is private, but that's a different thing.
i stress: stack != stack segment.  stack is where your sp is; stack
segment is a figment of the VM system.

i ask again: how does linux create per-thread storage?

the way the plan 9 thread library works is so different from linux's
that they're hard to compare.  program design in the two worlds is
radically different.  so your claim of 'better' is curious to me.  by
'better'
you seem to mean 'faster' and 'cleaner'.  faster at least can be
measured.
you speak with certainty.  have you seen performance comparisons?
i haven't, although it wouldn't surprise me to learn that there are
useful
programs for which linux outperforms plan 9, and vice versa of course.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> it might have been if Linux hadn't gone through an O(n) scheduler before
it did that context switch.

yes, the comments in the 2.4 scheduler are immediate:

    danger, danger, will robinson.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> But yes, I'm biased.

no, you're wrong.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> You don't like it. Fine. I don't care. You're myopic, and have an agenda
> to push, so you want to tell others that "you can't do that, it's against
> my agenda".

I'm not saying you can't, I'm just asking why, and you're not giving
me any sensible reason for it besides "because people do". That isn't
logical to me, and that's got nothing to do with any mysterious
agenda. I'm just interested in the theory.

> While I'm telling you that people _do_ do that, and that it makes sense,
> and if you didn't have blinders on, you'd see that.

Ok, so people do *do* that. That's fine, but that doesn't make it
correct, or sensible. I just wanted some semblance of logic, not
flames that equate to "I do it because I can". I can hop on one leg,
too, but that doesn't make hopping on one leg efficient.

Don (north_)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Thu, Feb 26, 2004 at 11:57:52PM -0800, Linus Torvalds wrote:
>
> On Fri, 27 Feb 2004, Lucio De Re wrote:
> >
> > Is Torvalds really saying that environment[] is held _in_ the stack?!
> > No wonder he was reluctant to copy it!  Specially when using "bash".
> > But I must be mistaken.
>
> Take a look if you don't believe me.
>
> There is a _lot_ of state on the stack. That's how C works.
>
Hm.  Yes, C allows it.  And your point is valid that an arbitrary
data segment may justifiably be copied.  But the stack is traditionally
the repository of special, let's call it "holy" information and
having a thread stomping all over it when another thread gets to
find out the hard way is not a thrilling prospect.

After all, there is nothing to prevent a coder from altering the
return address (traditionally a stack entry) in one thread and
totally wrecking the behaviour of another on return, right?

I still think the acid test lies with whether one version can model
the other.

++L

PS: and I really hope that the arguments and environment variables
are stored elsewhere and only the pointers appear on the stack :-)
I mean, even MS-DOS got that bit "sane", to use a very loaded word
previously used by Linus in this discussion.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> And I have not _ever_ met a good C programmer that liked segments.

err, the PDP-11?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Scott Schwartz]

I suspect Linus isn't subscribed to 9fans, so we'll have to cc him.

Rob writes:
| the argument about TLB flush times is interesting.
|
| > But it doesn't matter. Regardless, threads should see each others
| > stacks.
|
| and on plan 9, they do.  they just can't see each other's *stack
| segments*
| within their own address space.
|
| there seems to be confusion on this point.  plan 9's kernel splits the
| stack segment after a fork, but in the normal state, the processes run
| with the sp in shared memory.  the marvelous properties of the
| same-address-different-contents split stack segment is used only
| during the fiddly bits of process manipulation and to store per-process
| data.

That sharing is achieved by the thread library, not by the rfork
system call, though, right?

| how does linux store per-process data in user space?
|
| -rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
>
> > The rationale is that it's incredibly more sane, and it's the logical
> > place to put something that (a) needs to be allocated thread-specific and
> > (b) doesn't need any special allocator.
>
> You've just proven my point. Thread specific. Being Thread specific, it
> is data that is reserved to the scope of a single thread. Nothing more.
> If you want more scope there are many more usages of memory that
> are better utilized.

NO!

A "per-thread allocation" does NOT MEAN that other threads should not
access it. It measn that the ALLOCATION is thread-private, not that the
USE is thread-private.

per-thread allocations are quite common, and critical. If you have global
state, you need to protect them with locks, and you need to have nasty
global allocators.

One common per-thread allocation is the "I want to wait for an event". The
data is clearly for that one thread, and using a global allocator would be
WRONG. Not to mention inefficient.

But once the data has been allocated, other threads are what will actually
use the data to wake the original thread up. So while it wants a
per-thread allocator, it simply wouldn't _work_ if other threads couldn't
access the data.

That's what a "completion structure" is in the kernel. It's all the data
necessary to let a thread wait for something to complete. Another thread
will do "complete(xxx)", where "xxx" is that per-thread data.

You don't like it. Fine. I don't care. You're myopic, and have an agenda
to push, so you want to tell others that "you can't do that, it's against
my agenda".

While I'm telling you that people _do_ do that, and that it makes sense,
and if you didn't have blinders on, you'd see that.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 12:08:38AM -0800, Linus Torvalds wrote:
>
> On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
> >
> > > The rationale is that it's incredibly more sane, and it's the logical
> > > place to put something that (a) needs to be allocated thread-specific and
> > > (b) doesn't need any special allocator.
> >
> > You've just proven my point. Thread specific. Being Thread specific, it
> > is data that is reserved to the scope of a single thread. Nothing more.
> > If you want more scope there are many more usages of memory that
> > are better utilized.
>
> NO!
>
> A "per-thread allocation" does NOT MEAN that other threads should not
> access it. It measn that the ALLOCATION is thread-private, not that the
> USE is thread-private.
>
Wait a minute!  If the stack is not thread private, what is?  I
think this answers my question: stack duplication allows per-thread
private space, whereas stack sharing doesn't.  This is a one-way
path.  Unless you drop into register access, where it's the platform
that decides whether registers are duplicated or not.

But maybe they should also be shared, to be totally consistent.

Of course, I may be talking out of turn, but I really don't see
how threads can have private space if the stack isn't private.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

On Feb 27, 2004, at 12:06 AM, Scott Schwartz wrote:

> I suspect Linus isn't subscribed to 9fans, so we'll have to cc him.

sorry about that.

>
> Rob writes:
> | the argument about TLB flush times is interesting.
> |
> | > But it doesn't matter. Regardless, threads should see each others
> | > stacks.
> |
> | and on plan 9, they do.  they just can't see each other's *stack
> | segments*
> | within their own address space.
> |
> | there seems to be confusion on this point.  plan 9's kernel splits
> the
> | stack segment after a fork, but in the normal state, the processes
> run
> | with the sp in shared memory.  the marvelous properties of the
> | same-address-different-contents split stack segment is used only
> | during the fiddly bits of process manipulation and to store
> per-process
> | data.
>
> That sharing is achieved by the thread library, not by the rfork
> system call, though, right?

right. but where it's implemented is in some sense just a detail, as my
other
message, incuded below, implies.

another way of looking at it is that the kernel provide some nice
primitives upon which to build a thread library.  the plan 9 kernel
leaves out a lot of stuff that you're supposed to do in the kernel
for threads, but we were exploring options.

>
> | how does linux store per-process data in user space?
> |
> | -rob

here's my other message, with the cc: this time:


> So in a C/UNIX-like environment, private stacks are wrong. You could
> imagine _other_ environments where they might be valid, but even those
> other environments would not invalidate my points about efficiency and
> simplicity.

as i said before, the stacks are not private.  you're right, that's a
bad thing.
that's why they're not private.

the segment called 'stack' is private, but that's a different thing.
i stress: stack != stack segment.  stack is where your sp is; stack
segment is a figment of the VM system.

i ask again: how does linux create per-thread storage?

the way the plan 9 thread library works is so different from linux's
that they're hard to compare.  program design in the two worlds is
radically different.  so your claim of 'better' is curious to me.  by
'better'
you seem to mean 'faster' and 'cleaner'.  faster at least can be
measured.
you speak with certainty.  have you seen performance comparisons?
i haven't, although it wouldn't surprise me to learn that there are
useful
programs for which linux outperforms plan 9, and vice versa of course.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

On Feb 27, 2004, at 12:11 AM, Lucio De Re wrote:

> Of course, I may be talking out of turn, but I really don't see
> how threads can have private space if the stack isn't private.

well, perhaps the stack isn't the only place to do it, but it's
certainly an easy one, and one that makes the syscall interface
to fork easy to implement in a threaded environment: longjmp
to the private stack, fork, adjust, longjmp back.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Geoff Collyer]

I think we're talking past each other due to different terminologies.
Linus seems to use `thread' to mean `a process sharing address space
(other than normal text segment sharing)', whereas in Plan 9, that's
just a process; some share address space, some don't.  A Plan 9 thread
is entirely a user-mode creation of the Plan 9 thread library, which
doesn't implement POSIX threads.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 12:17:33AM -0800, Rob Pike wrote:
>
> On Feb 27, 2004, at 12:11 AM, Lucio De Re wrote:
>
> > Of course, I may be talking out of turn, but I really don't see
> > how threads can have private space if the stack isn't private.
>
> well, perhaps the stack isn't the only place to do it, but it's
> certainly an easy one, and one that makes the syscall interface
> to fork easy to implement in a threaded environment: longjmp
> to the private stack, fork, adjust, longjmp back.
>
But I can't think of even one possible alternative.  After all, the
stack is the only storage being duplicated (ignoring registers) so
where does one keep pointers to the private space?

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> We could have just made a trivial system call ("get the thread-local
> pointer" from the kernel stack), but obviously there are performance
> issues here.

just one?  surely another 32 or 64 ...

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Lucio De Re wrote:
>
> On Fri, Feb 27, 2004 at 12:17:33AM -0800, Rob Pike wrote:
> >
> > On Feb 27, 2004, at 12:11 AM, Lucio De Re wrote:
> >
> > > Of course, I may be talking out of turn, but I really don't see
> > > how threads can have private space if the stack isn't private.
> >
> > well, perhaps the stack isn't the only place to do it, but it's
> > certainly an easy one, and one that makes the syscall interface
> > to fork easy to implement in a threaded environment: longjmp
> > to the private stack, fork, adjust, longjmp back.
> >
> But I can't think of even one possible alternative.  After all, the
> stack is the only storage being duplicated (ignoring registers) so
> where does one keep pointers to the private space?

Think it through. You should _not_ duplicate the stack (because that
wreaks havoc with your TLB and normal usage), so what do you have left?

Once you've eliminated the impossible, what you have left, however
improbable, is the truth.

Registers.

Why are you ignoring registers? That's what you _should_ use.

For example, inside the Linux kernel, we tend to use the stack POINTER as
the thread-local state. When we allocate a new context of execution, we
allocate (depending on architecture) 8kB of memory, and it's aligned so
that if the architecture doesn't have any other registers free, we can get
at the "thread_info" structure by just doing bit masking on the stack
pointer.

That ends up being quite powerful - and it's cheap too, exactly because it
is a register, and thus fast to access. The stack itself is by no means
private - other threads can access the stack.

In fact, we used to put the whole "struct task_struct" (which is the thing
that defines a context of execution in Linux) that way, but it ends up
doing nasty things to caches when important global data structures are all
aligned on powers-of-two boundaries, so we ended up getting rid of that.

In user space, that doesn't tend to work too well, because the stack isn't
as well bounded as in the kernel, but most architectures either have lots
of registers (and then one is just used for the thread-local pointer) or
even an architected register that user space can read but not write.

One of the most problematic architectures is the x86, which doesn't have
lots of general-purpose registers (so using one of them to point to TLS
would be bad), and doesn't have any nice architected register either.
There we ended up using a segment register, however much I hate them.

We could have just made a trivial system call ("get the thread-local
pointer" from the kernel stack), but obviously there are performance
issues here.

In short: there is absolutely no reason to make the stack be private. The
only thing you need for thread-local-storage is literally just one
register, to indirect through. And it can be a fairly strange one at that,
ie it doesn't need to be able to hold a full 32-bit value.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 01:46:27AM -0800, Linus Torvalds wrote:
>
> Why are you ignoring registers? That's what you _should_ use.
>
Because they are not in the base language?  Because they impair
portability?

Because in my C code, I can't instantiate them as variables without
running the risk of my colleagues doing the same in a conflicting
manner?

Just out of curiosity, how do I get to the private space without
a lock?  It's been a long time since I studied these things and lots
of water has flown under bridges, so I could be missing something.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, boyd, rounin wrote:
>
> > We could have just made a trivial system call ("get the thread-local
> > pointer" from the kernel stack), but obviously there are performance
> > issues here.
>
> just one?  surely another 32 or 64 ...

Why?

You only need one thread-local pointer. You just put all your stuff offset
from that one. Realize that the kernel doesn't do anything at all with
that value - it's purely a random value that the user can set, and the
kernel doesn't even need to know that it is a pointer. You could use it as
an index into other pointers if you wanted to.

So anything but one doesn't make any sense. Where would you stop?

Am I missing something here? Linux certainly happily makes do with just
one TLS pointer.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>Just out of curiosity, how do I get to the private space without
>>a lock?  It's been a long time since I studied these things and lots
>>of water has flown under bridges, so I could be missing something.

static __inline Proc *getup(void) {
	Proc *p;
	__asm__(	"movl	%%esp, %%eax\n\t"
			: "=a" (p)
	);
	return *(Proc **)((unsigned long)p & ~(KSTACK - 1));
};

where the process that makes the new process's stack puts a pointer to
the private space (a Proc* in this case) at the base of that stack, and where KSTACK is
a power-of-2.  i don't think it can be accessed portably in Linux,
but you don't need a lock.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 10:00:32AM +0000, Charles Forsyth wrote:
>
> >>Just out of curiosity, how do I get to the private space without
> >>a lock?  It's been a long time since I studied these things and lots
> >>of water has flown under bridges, so I could be missing something.
>
> static __inline Proc *getup(void) {
> 	Proc *p;
> 	__asm__(	"movl	%%esp, %%eax\n\t"
> 			: "=a" (p)
> 	);
> 	return *(Proc **)((unsigned long)p & ~(KSTACK - 1));
> };
>
> where the process that makes the new process's stack puts a pointer to
> the private space (a Proc* in this case) at the base of that stack, and where KSTACK is
> a power-of-2.  i don't think it can be accessed portably in Linux,
> but you don't need a lock.

Yeow!  And where do I put the returned address, so that no other thread
can stomp on it?  While I attempt to use it?

++L

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Douglas A. Gwyn]

dbailey27@ameritech.net wrote:
> Under the "Kernel Space and User Space" heading:
[very confused description omitted]
> | While this is a clever feat, the downside is that the overhead in maintaining
> | the stacks makes this in practice really stupid to do. They found out too
> | late that the performance went to hell. Since they had programs which used
> | the interface they could not fix it. Instead they had to introduce an additional
> | properly-written interface so that they could do what was wise with the stack
> |  space. ...
I don't think any of that is correct.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Lucio De Re wrote:

> On Fri, Feb 27, 2004 at 01:46:27AM -0800, Linus Torvalds wrote:
> >
> > Why are you ignoring registers? That's what you _should_ use.
>
> Because they are not in the base language?  Because they impair
> portability?

You literally need _one_ operation: you need the operation of "give me the
TLS pointer" (well, your thread setup code obviously needs a way to set
the pointer when creating a thread too).

The rest _is_ in the language - although if you want nice syntax, you
sometimes want more explicit language support.

Have you ever looked at those system header files? They contain a lot of
magic cruft that is specific to your compiler and your particular
architecture.

> Because in my C code, I can't instantiate them as variables without
> running the risk of my colleagues doing the same in a conflicting
> manner?
>
> Just out of curiosity, how do I get to the private space without
> a lock?  It's been a long time since I studied these things and lots
> of water has flown under bridges, so I could be missing something.

In the kernel, the x86 implementation of the thread-local pointer is
literally:

	/* how to get the thread information struct from C */
	static inline struct thread_info *current_thread_info(void)
	{
	        struct thread_info *ti;
	        __asm__("andl %%esp,%0; ":"=r" (ti) : "0" (~(THREAD_SIZE - 1)));
	        return ti;
	}

That's it. It compiles to _two_ instructions, eg something like

	movl $-8192,%eax
	andl %esp,%eax

and it's all done. You literally _cannot_ do it faster or smaller. No
locking, no memory accesses, no TLB games, no NOTHING.

On some other architectures, you have

	register struct thread_info *__current_thread_info __asm__("$8");
	#define current_thread_info()  __current_thread_info

which just tells the compiler that the thread_info pointer is in hardware
register 8. And you're done. The compiler will just automatically use %r8
whenever you do a "current_thread_info()".

In user space, there are similar things going on. These days there is more
compiler support to make it easier to create these things, but it all
boils down to having a thread-local pointer somewhere.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 02:11:27AM -0800, Linus Torvalds wrote:
>
> On Fri, 27 Feb 2004, Lucio De Re wrote:
>
> > On Fri, Feb 27, 2004 at 01:46:27AM -0800, Linus Torvalds wrote:
> > >
> > > Why are you ignoring registers? That's what you _should_ use.
> >
> > Because they are not in the base language?  Because they impair
> > portability?
>
> You literally need _one_ operation: you need the operation of "give me the
> TLS pointer" (well, your thread setup code obviously needs a way to set
> the pointer when creating a thread too).
>
Yes, but...

> > Because in my C code, I can't instantiate them as variables without
> > running the risk of my colleagues doing the same in a conflicting
> > manner?
> >
Unless I'm missing the wood for the trees, this returns a pointer.
Where do I put _it_ so no other thread, choosing to do likewise,
decides to stomp all over it?

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>Yeow!  And where do I put the returned address, so that no other thread
>>can stomp on it?  While I attempt to use it?

the process is running by then on its own stack, so the result is either in
a register (EAX say) or in a temporary on the stack but either way, it's private.

how does it get to its own stack in the first place?  that's the bit
that's been causing some of the fuss here.  at least for a few years,
a variant of `clone' is available that takes a pointer to the new top-of-stack for the
new process, so the creating process allocates a stack somewhere somehow
and passes that to clone.  thus the new process is running on a private
stack once it starts.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Fri, Feb 27, 2004 at 10:14:32AM +0000, Charles Forsyth wrote:
>
> >>Yeow!  And where do I put the returned address, so that no other thread
> >>can stomp on it?  While I attempt to use it?
>
> the process is running by then on its own stack, so the result is either in
> a register (EAX say) or in a temporary on the stack but either way, it's private.
>
Sure, but I thought the idea was that all storage was shared, except
any that was privately allocated later.  Did I read wrong Torvalds'
comments about the stack being nothing special?  And then a new
private stack comes along?  How is it initialised?  Isn't it a
duplicate?

I think I'm out of my depth.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Lucio De Re wrote:
> >
> > You literally need _one_ operation: you need the operation of "give me the
> > TLS pointer" (well, your thread setup code obviously needs a way to set
> > the pointer when creating a thread too).
> >
> Yes, but...
>
> Unless I'm missing the wood for the trees, this returns a pointer.
> Where do I put _it_ so no other thread, choosing to do likewise,
> decides to stomp all over it?

Your register state? Or you save it to the (shared) VM, by using another
private pointer (ie the stack pointer).

It all boils down to one thing: the _only_ state that is always truly
private in your thread is your register state. But exactly because the
register state is private, you now have ways of turning the _rest_ of your
state (that isn't private) into your own private areas.

So while the VM is shared across all different threads, all threads
obviously have their own stack pointer registers, and they'd better be
pointing to different _parts_ of that shared VM, or your threads won't
work at all.

	"Any problem in computer science can be solved with another layer
	 of indirection."
			David Wheeler

In other words, your registers fundamentally _are_ your private space, and
everything else follows from that.

So your stack is "private"  in the sense that nobody else uses it unless
you explicitly give a pointer to it (modulo bugs, of course ;).

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[C H Forsyth]

[-- Attachment #1: Type: text/plain, Size: 568 bytes --]

sorry, when i said `private' i meant `for that process's exclusive use as a stack',
not that it was in a per-process address space.

it's per-process only in that by agreement the process to which
it is given is the only one that uses it as a stack (the others can
still access its contents if they like, but often they don't).

they are all still in the same address space.
they could be allocated (essentially) by malloc, except that
you need to ensure they are all aligned appropriately (eg,
at an address that is 0 mod KSTACK in that previous example).

[-- Attachment #2: Type: message/rfc822, Size: 2987 bytes --]

From: Lucio De Re <lucio@proxima.alt.za>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Fri, 27 Feb 2004 12:24:46 +0200
Message-ID: <20040227122445.H22848@cackle.proxima.alt.za>

On Fri, Feb 27, 2004 at 10:14:32AM +0000, Charles Forsyth wrote:
>
> >>Yeow!  And where do I put the returned address, so that no other thread
> >>can stomp on it?  While I attempt to use it?
>
> the process is running by then on its own stack, so the result is either in
> a register (EAX say) or in a temporary on the stack but either way, it's private.
>
Sure, but I thought the idea was that all storage was shared, except
any that was privately allocated later.  Did I read wrong Torvalds'
comments about the stack being nothing special?  And then a new
private stack comes along?  How is it initialised?  Isn't it a
duplicate?

I think I'm out of my depth.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Dave Lukes]

> > Do you have code from existing Linux implementation that exemplifies
> > this scenario?

... and I'm _still_ waiting for an answer.

> Anything that accesses any process arguments would access the stack of the
> original thread.

... which is copied, so what's the problem?
Do you want all the threads to be able to _update_ the process args?
That's harder, and also more dubious.

>  But synchronization with
> completion events on the stack is also perfectly feasible (the kernel does
> exactly that internally, for example - and user mode could do the same).

The kernel using it is an implementation detail.
User mode using it: show me the example!

Cheers,
	Dave.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

> I think we're talking past each other due to different terminologies.
> Linus seems to use `thread' to mean `a process sharing address space
> (other than normal text segment sharing)', whereas in Plan 9, that's
> just a process; some share address space, some don't.  A Plan 9 thread
> is entirely a user-mode creation of the Plan 9 thread library, which
> doesn't implement POSIX threads.

i was speaking using linux's terminology.

 > Having a pointer that sometimes works, and sometimes doesn't, based
on who
 > uses it - that's just crazy talk.

put in those terms, it sounds weird, but it's not.  consider the old u.
area in unix.
that was a piece of address space with the same virtual address in all
processes
but a different content.  the system used the fact that the addresses
aliased
that way.  plan 9's thread model does a similar thing by constructing a
special
storage class for data private to each process.  for instance, one can
have a
variable with the same address in each process, call it tp, that points
to the
thread-specific data, so you can write code like
	printf("my id is %d\n", tp->id);

> > i ask again: how does linux create per-thread storage?

 > The same way it creates any other storage: with mmap() and brk(). You
just
 > malloc the thing, and you pass in the new stack as an argument to the
 > thread creation mechanism (which linux calls "clone()", just to be
 > different).

that wasn't what i was asking.  i was referring to this special storage
class.
how does a thread identify who it is?  when we were porting inferno to
linux (back in 1996; things have likely changed) we resorted to using
unused user-visible MMU bits to store enough state to hold on with our
fingernails and claw back to private storage. another option we
considered was playing with magic address bits in the sp.

ah, i see in later mail that you answered this. there are now pointers
created in the user space (i think) to thread-local storage.  how is it
accessed, that is, how does the user process derive the pointer to it?
this state stuff did not exist when we did the inferno port.

oh, and now i see you answering that later.  it is 'cruft', as you say,
but
it will work; it's the magic address bits hack.  which kernel version
introduced this stuff?  i've heard people say that 2.6 is the first one
with the default thread model being 'efficient' and 'good', but i don't
know the specifics.  i've also heard that they can be retrofitted to
2.4.

i think a big part of my confusion is that my criticisms of linux
threads are
based on an older view of how they worked.  and a big part of the
commentary that started the discussion was incorrect about plan 9
history.  i hope we're on the same page now.

it's interesting you advocate using registers for the magic storage
class.
it's a great trick when you can do it - plan 9 uses it in the kernel on
machines
with lots of registers - but it's not so great on a machine with too few
registers, like the x86.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Dave Lukes wrote:
>
> > > Do you have code from existing Linux implementation that exemplifies
> > > this scenario?
>
> ... and I'm _still_ waiting for an answer.

Hmm.. I answered it the first time around. Look in the kernel for "struct
completion", and you will see it. It's common behaviour there to do (this
example is from the fork() case:

	struct completion vfork;

	if (clone_flags & CLONE_VFORK) {
		p->vfork_done = &vfork;
		init_completion(&vfork);
	}
	...
		wake_up_forked_process(p);      /* do this last */
	...
	if (clone_flags & CLONE_VFORK) {
		wait_for_completion(&vfork);
		...

and then the mm_release function (in the child) does:

	struct completion *vfork_done = tsk->vfork_done;

        /* Get rid of any cached register state */
        deactivate_mm(tsk, mm);

        /* notify parent sleeping on vfork() */
        if (vfork_done) {
                tsk->vfork_done = NULL;
                complete(vfork_done);
        }

ie the other thread reads and updates the "struct completion" that is on
another process's stack.

The above is neither insane nor bad in any other way. It's in my opinion
_the_ most readable way to do things, and it certainly is the most
efficient one too.

In user space, I don't work with threaded programs, but if I were to write
one, I'd do this all the time. With threaded programs - even more so than
with normal programs - you need to avoid global state, which means that
quite often the stack is the only easy place to put something.

For example, most thread interfaces only allow passing in one single data
structure to a thread (eg the pthread_create() "void * arg" argument. So
whenever you have an issue with passing down data to describe what you
want done to a helper thread, you commonly end up passing in a pointer to
a structure. And if you end up waiting for the result (imagine doing
threaded video encoding, for example - where the subthreads exist as
computational things to do encoding ove rmany CPU's), then it makes 100%
sense to do exactly the above.

> > Anything that accesses any process arguments would access the stack of the
> > original thread.
>
> ... which is copied, so what's the problem?

The problem is that communication is almost never one-way, except
apparently in the thread we're involved right now.

> Do you want all the threads to be able to _update_ the process args?
> That's harder, and also more dubious.

Hey, I don't know about you, but pretty much every _single_ argument
parser I've ever done tends to fill in some data structure with the
pointer to the argument in question.

So for example, when I get a filename argument, I don't do

	filename = strdup(argv[i]);

but I do

	filename = argv[i];

and then I just pass that filename around.

And nobody - and I mean NOBODY - should be in the position that they
should remember how exactly that "filename" pointer was created to know
what the semantics of trying to edit it are. It's a pointer.

> The kernel using it is an implementation detail.
> User mode using it: show me the example!

I'm not into user-mode - user mode is for whimps who can't handle the
truth (yeah yeah, I know, I'm crazy, but I just find it more interesting
to interact with the hardware).

The thing is, I don't see why you are even arguing. There are zero
advantages to a private stack, and there are tons of disadvantages. So
what's your beef?

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Dave Lukes]

> Hmm.. I answered it the first time around.

I said: the kernel using it is an implementation detail.

You said:
> Look in the kernel for "struct
> completion", and you will see it.

i.e. you've built the code to fit the thread implementation.
I stand by what I said.

I said: user mode using it: show me the example!

You said:
> In user space, I don't work with threaded programs, but if I were to
> write
> one, I'd do this all the time.

i.e. no example.

> The problem is that communication is almost never one-way, except
> apparently in the thread we're involved right now.

Ha ha very funny: let me just recover from my ruptured abdomen ...
Now let me point out that, in spite of your copious code inclusion,
you have not yet provided any real examples.

Also, communication in programs is often one-way
(that's why ANSI invented "const":-).

> Hey, I don't know about you, but pretty much every _single_ argument
> parser I've ever done tends to fill in some data structure with the
> pointer to the argument in question.

Oh, yes, of course, silly me:
Real Programmers always parse their arguments in a separate thread.

> I'm not into user-mode - user mode is for whimps who can't handle the
> truth (yeah yeah, I know, I'm crazy, but I just find it more interesting
> to interact with the hardware).

If it's PC hardware, that _is_ crazy:-).

> The thing is, I don't see why you are even arguing. There are zero
> advantages to a private stack, and there are tons of disadvantages. So
> what's your beef?

"There are zero disadvantages to incorporating an RSX-11M emulator
in the kernel, and there are tons of advantages.  What's your beef?"

Also, as a mantra:
the more shared data you have, the more problems you have.

One other detail: as far as I can see,
your examples all use shared memory as a cheap substitute
for message passing: why?

Cheers,
	Dave.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Rob Pike wrote:
>
>  > Having a pointer that sometimes works, and sometimes doesn't, based  on who
>  > uses it - that's just crazy talk.
>
> put in those terms, it sounds weird, but it's not.  consider the old u.
> area in unix. that was a piece of address space with the same virtual
> address in all processes but a different content.

I hate to put it to you, Rob, but that sucks. It sucks so hard that it's
not even funny.

Playing tricks with the VM is a cool hack, and we've considered it
multiple times, but in the end sanity has always prevailed. Playing VM
tricks is _expensive_. It looks cheap because the access code "just
works", but the expense is in

 - wasting TLB entries (all sane CPU's have big "fixed" areas that are
   better on the TLB for system mode - large pages, BAT registers, or just
   architected 1:1 mappings). The kernel should use them, because the
   fewer TLB entries the kernel uses, the more there are for user space to
   waste. But that implies that the kernel shouldn't play any VM tricks
   with its own data - the VM is for user space (yeah, the kernel ends up
   wanting to use the VM hardware for some things, but it's discouraged)

 - CPU's with hardware TLB lookup have to have separate page tables for
   different CPU's. Which means not only that you're wasting memory on
   having <n> copies of the thing, it also means that now you have to have
   code to maintain coherency between those separate page tables, and have
   to have locking.

   Having just one copy is just better. Two copies of the same thing is
   bad.

 - TLB invalidates are just a lot more expensive than changing a register
   around. It's bad even on "good" hardware, it's totally unacceptable on
   anything with a virtual cache, for example.

So the kernel internally has always had the stack pointer register as the
main "thread pointer". It did that even before SMP, just because it was
simpler and faster. With SMP, doing anything else becomes unacceptable.

>  the system used the fact that the addresses aliased that way.  plan 9's
> thread model does a similar thing by constructing a special storage
> class for data private to each process.  for instance, one can have a
> variable with the same address in each process, call it tp, that points
> to the thread-specific data, so you can write code like
>
> 	printf("my id is %d\n", tp->id);

Yes. And you pay the price. For no good reason, I may add, since you
traditionally have been able to do the same by just having to add some
explicit thread tracking (it wouldn't be "tp->id", it would be
"mythread()->tp->id") or by adding compiler support to make the syntax be
easier.

These days, if you want to avoid the syntax of carrying that per-thread
pointer around, we have compiler support, ie you can do

	__thread struct mystruct *tp = NULL;

and now "tp" is a per-thread variable. Behind the schenes the compiler
will change it to be an offset off the TLS pointer, pretty much exactly
the same way it does position-independent code.

>  > The same way it creates any other storage: with mmap() and brk(). You just
>  > malloc the thing, and you pass in the new stack as an argument to the
>  > thread creation mechanism (which linux calls "clone()", just to be
>  > different).
>
> that wasn't what i was asking.  i was referring to this special storage
> class. how does a thread identify who it is?

A long time ago, it was literally a "gettid()" system call. If you wanted
the thread-local space, you followed that by a index lookup.

It's not insanely expensive if you avoid re-generating the thread-local
pointer all the time, and pass it down as a regular argument, but it is
obviously syntactically not pretty.

These days - mostly thanks to compiler and library advances, not so much
any real kernel changes - the thread infrastructure sets up its local
pointers in registers, so that you can use the above "__thread"
specifier in the compiler, and when you access

	tp->id

the compiler will actually generate

	movl    %gs:tp@NTPOFF, %eax
	movl    (%eax), %eax

for you (on other platforms that have compiler support of thread-local
storage it usually would end up being a indirect access through a regular
register).

The linker fixes these things up, the same way it does things like GOT
tables etc.

> ah, i see in later mail that you answered this. there are now pointers
> created in the user space (i think) to thread-local storage.  how is it
> accessed, that is, how does the user process derive the pointer to it?
> this state stuff did not exist when we did the inferno port.

See above. If you control your environment (ie you don't have to worry
about having arbitrary TLS space), you can do better with the stack
register trick the kernel uses, but indirection will handle the general
case.

> it will work; it's the magic address bits hack.  which kernel version
> introduced this stuff?  i've heard people say that 2.6 is the first one
> with the default thread model being 'efficient' and 'good', but i don't
> know the specifics.  i've also heard that they can be retrofitted to
> 2.4.

The new threading model in 2.6.x is really more about signal handling than
anything else. The _real_ problem with the original clone() implementation
had nothing to do with the VM, and had everything to do with insane POSIX
shared signal semantics. It's really hard to get the POSIX thread signal
semantics rigt, since the whole pthreads thing really was designed for
having all threads run within one master process, and Linux never had the
notion of "process vs threads".

The signal case that is hard to get right in POSIX is the fact that signal
masks are thread-local, yet their effect is "process global" (ie when you
change the signal mask of your thread, that means that you suddenly now
potentially start accepting pending signals that were shared process
global). I still don't like that POSIX model, and I didn't see any sane
way to do it efficiently with truly independent threads that don't have
the notion of a "process" that encompasses them.

What 2.6.x (and the 2.4.x back-port) does is to just accept the fact that
there is a "thread group leader" (that's what the CLONE_THREAD flag does:
if it is set, you share the thread group leader, if it is clear you create
a new thread group), and that pending signal state really is shared in the
thread group.

The VM side has always been the same: if you share the VM, you share
everything. There literally isn't any thread-local storage from a VM
standpoint, there are only thread-local registers that point to different
areas of memory.

> it's interesting you advocate using registers for the magic storage
> class. it's a great trick when you can do it - plan 9 uses it in the
> kernel on machines with lots of registers - but it's not so great on a
> machine with too few registers, like the x86.

Well, even in the absense of a register, you can always just have a system
call to ask what the pointer should be. That really does work very well,
as long as your programming model is about explicit thread pointers (which
pthreads is) so that you don't have to do it all the time.

And the x86 really is the worst possible case, in this situation, because
it is so register-starved anyway. But happily, it has some (very ugly)
legacy registers that have to be user-visible, and have to be saved and
restored anyway, and that nobody sane really wants to use, so the thread
model can use them.

Making the threaded stuff explicit helps avoid confusion. Now, if somebody
takes an address of a per-thread variable, it is clear that that address
is the address of the variable IN THAT THREAD. You can pass it along, but
when you pass it along to another thread, it doesn't change value - it
still points to the exact same thread-local variable in the _original_
thread.

(Obviously you can pass around offsets to the thread-local space if you
want to, although I can't really see why you'd do it).

And I hope it's clear by now that because the thing is entirely in
registers, that "thread model" is pretty much all in user space. It needs
no kernel support, although some of the interfaces are obviously done
certain ways to make it easier to do (ie the kernel does know about a TLS
pointer at thread setup, even if the kernel doesn't actually _use_ it for
anything, it just sets up the register state as indicated by the parent of
the thread).

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

> Hint: you can't message-pass a hash table that describes 200 megabytes
> worth of filesystem names.
>
> Welcome to the real world, Neo.

You can. You can pass a pointer to it.
If you're in the mood for using locks, you can put a proc
in charge of your hash table, and avoid locks. If you're not,
you can use non-preemptive threads and avoid locks too.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Dave Lukes wrote:
>
> I said: user mode using it: show me the example!

Hey. Ask why Plan-9 doesn't use the shared stack either. I bet it's
because tons of programs broke.

 [ Hint: Dave just explained the Plan9 thread model to me: the "private
   stack" ends up not being used as a stack at all. The real stacks end up
   being malloc'ed for each thread (in _shared_ space), and the "private
   stack" area only ends up being a TLS segment. ]

I wouldn't know about whatever apps a private stack would break, because I
was never crazy enough to think that a private stack was a good diea.

> i.e. no example.

Hey, _you're_ the one with the crazy idea. As such, the burden of proof is
on you to show that it isn't crazy.

> One other detail: as far as I can see,
> your examples all use shared memory as a cheap substitute
> for message passing: why?

Because message passing is idiotic, when the real hardware just passes
pointers around?

Because you can't put complex data structures in a message without silly
encodings that make performance plummet like a stone?

Methinks you have read a few too many papers about microkernels, without
actually seeing the real world.

Hint: you can't message-pass a hash table that describes 200 megabytes
worth of filesystem names.

Welcome to the real world, Neo.

Stop playing around with those examples your professors showed you. They
had no relevance.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[C H Forsyth]

>>I said: user mode using it: show me the example!

my impression, and it's only that, is that relatively few user mode
applications use clone-for-threads at all on Linux, as yet.  thus,
it's a little unfair, or at least pointless, to ask Torvalds for
specific Linux user mode examples since there aren't many generally.
(i am curious to know what applications there are that do use it,
though.)

examples seem few partly because Unix programs aren't typically
structured as cooperating threads (however they cooperate).  that's
partly because the most obviously portable way to do it, pthreads, has
several awful or incomplete implementations.  you don't know what
you're going to get (or rather, you know only too well).  (i don't
refer here to the pthreads specification, just its implementation in
practice, because that's what is relevant here.) for instance, some in
the past tried to do it all with coroutines and an internal scheduler
that (say) intercepted operations on file descriptors and multiplexed
coroutines using select/poll.

>>One other detail: as far as I can see,
>>your examples all use shared memory as a cheap substitute
>>for message passing: why?

in practice, you often end up sharing something even for message passing
(for instance a Channel or a mailbox if those are your models), unless you
meant reading and writing pipe file descriptors, in which case the sharing
(if it's a pipe) is tucked away inside the kernel but it's still there.

>>how does a thread identify who it is?  when we were porting inferno to
>>linux (back in 1996; things have likely changed) we resorted to using
>>unused user-visible MMU bits to store enough state to hold on with our
>>fingernails and claw back to private storage. another option we
>>considered was playing with magic address bits in the sp.

inferno on linux currently conditions the stack and sp as previously discussed,
not least because the place used by the original scheme wasn't any longer visible.
that does require machine-specific code to access the sp, although that's
in a linux-386-dependent include file.  fortunately clone accepts the stack
pointer for the new process (although that's machine-specific as to whether
it's top or bottom), so there's no need for assembly-language to bounce
to the new stack (as there was in FreeBSD for a few years, until they extended
the rfork interface).

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Dave Lukes]

> Hey, _you're_ the one with the crazy idea. As such, the burden of proof is
> on you to show that it isn't crazy.

Firstly,
an old saying: "Everyone is insane but yourself".

Secondly,
I'm not the one who wants to put 200Mb on the stack (see below):
who's crazy, again?

> Because message passing is idiotic, when the real hardware just passes
> pointers around?

Ohh, yea, and pointers scale really well to NUMAs, right?

> Methinks you have read a few too many papers about microkernels, without
> actually seeing the real world.

Methinks you don't know much about me.

> Hint: you can't message-pass a hash table that describes 200 megabytes
> worth of filesystem names.

No shit, Sherlock!  Thanks for the enlightenment!
Next question: are you _seriously_ suggesting having 200Mb on a stack?

> Welcome to the real world

I _am_ in the real world.
Check out who we are: we don't sit in labs playing with our pointers:
we build real systems that we sell to real customers for real money.

> , Neo.

Hmmm ...
Someone who fought against the bland uniformity of a regimented world:
I'll tolerate that epithet.

> Stop playing around with those examples your professors showed you.

This attitude is beginning to _seriously_ piss me off:
you know _nothing_ about my knowledge, abilities or opinions.
To be specific: I have _never_ attended a CompSci lecture in my life.

>  They
> had no relevance.

... and what _does_ have relevance?

Cheers,
	Dave.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Dave Lukes wrote:
>
> I'm not the one who wants to put 200Mb on the stack (see below):
> who's crazy, again?

Neither am I.

I'm saying that you CANNOT "message pass" a hash table. You pass its
address, and that has NOTHING to do with message passing. You use a
function call that passes the address to a data structure around.

> > Because message passing is idiotic, when the real hardware just passes
> > pointers around?
>
> Ohh, yea, and pointers scale really well to NUMAs, right?

Hey, when did you last write an operating system that worked on NUMA
machines?

Trust me. Passing pointers around scales a _hell_ of a lot better than
copying data around as messages.

> > Hint: you can't message-pass a hash table that describes 200 megabytes
> > worth of filesystem names.
>
> No shit, Sherlock!  Thanks for the enlightenment!
> Next question: are you _seriously_ suggesting having 200Mb on a stack?

No.

I'm suggesting passing the pointer around, and not messing with messages
at all. You're the one who complained about me using pointers:

  "your examples all use shared memory as a cheap substitute
   for message passing: why?"

and I'm telling you that pointers are NOT a "cheap substitute for message
passing". Pointers are fundamentally MORE POWERFUL than message passing
is, and anybody who calls them a "cheap substitute" is a moron.

That was my point: a pointer can point to hundreds of megabytes of complex
data structures with lots of interdependencies and interesting locking
rules. THAT is the real world. A message it is NOT.

A message is a way to pass data by value. It has its place too, of course,
especially in networks, but comparing it to a pointer is just misguided.

Th eonly people who confuse pointers and messages are the microkernel
people who noticed that real messages are too expensive to use, so they
started calling pointers "messages", and play other semantic games.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[matt]

[-- Attachment #1: Type: text/plain, Size: 81 bytes --]

I can't add much to the technical discussion but I did make a Linus "face"

m

[-- Attachment #2: Type: application/octet-stream, Size: 2031 bytes --]

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[andrey mirtchovski]

> I can't add much to the technical discussion but I did make a Linus "face"
>
> m

hehe, you should've used this one:

	http://pages.infinit.net/rave/FOTO/linus.gif

source:

	http://www.robotwisdom.com/linux/timeline.html

the source is worth spending some time with, especially in the pre-'91
sections...

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Donald Brownlee]

Is "completion" in Linux like VMS' fork routines?

Linus Torvalds wrote:
>
> On Fri, 27 Feb 2004 dbailey27@ameritech.net wrote:
>
>>>The rationale is that it's incredibly more sane, and it's the logical
>>>place to put something that (a) needs to be allocated thread-specific and
>>>(b) doesn't need any special allocator.
>>
>>You've just proven my point. Thread specific. Being Thread specific, it
>>is data that is reserved to the scope of a single thread. Nothing more.
>>If you want more scope there are many more usages of memory that
>>are better utilized.
>
>
> NO!
>
> A "per-thread allocation" does NOT MEAN that other threads should not
> access it. It measn that the ALLOCATION is thread-private, not that the
> USE is thread-private.
>
> per-thread allocations are quite common, and critical. If you have global
> state, you need to protect them with locks, and you need to have nasty
> global allocators.
>
> One common per-thread allocation is the "I want to wait for an event". The
> data is clearly for that one thread, and using a global allocator would be
> WRONG. Not to mention inefficient.
>
> But once the data has been allocated, other threads are what will actually
> use the data to wake the original thread up. So while it wants a
> per-thread allocator, it simply wouldn't _work_ if other threads couldn't
> access the data.
>
> That's what a "completion structure" is in the kernel. It's all the data
> necessary to let a thread wait for something to complete. Another thread
> will do "complete(xxx)", where "xxx" is that per-thread data.
>
> You don't like it. Fine. I don't care. You're myopic, and have an agenda
> to push, so you want to tell others that "you can't do that, it's against
> my agenda".
>
> While I'm telling you that people _do_ do that, and that it makes sense,
> and if you didn't have blinders on, you'd see that.
>
> 		Linus
>

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> Because message passing is idiotic, when the real hardware just passes
> pointers around?

pass pointers around?  err, dump core or take a kernel mode protection
fault.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> I'm saying that you CANNOT "message pass" a hash table.

you can.  i've done it.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Martin C.Atkins]

If I understand right, I could summarise (one of) Linus's arguments,
as follows:

	The cost of sharing some of VM, but not all (in terms of peformance, due
	to less-visible things like TLB flushes, etc), out-weighs having to
	write a little bit of assembler wrapping the clone system call (which
	only has to be got right once, however horrible it might be).

I haven't seen any argument rebutting this in any way. If it is true,
then surely a performance hit on all (forked?) processes, is more
important than having to shim a system call? If it is not true, then it
would be nice to know!

Tell me where I'm wrong, please?

Martin

--
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

i had trouble believing that the screeds of banter were from linus.
maybe he has a ghost writer.  when he did the "i don't write
user programs with threads" and thru in some kernel examples,
with obligatory assembly language results - well i thought
what are these guys discussing? gee, i just write in limbo.
threads are fun, linus, when you don't have to do all that crap
to manage them.  i just use "spawn", the only language support
for threads.  the rest is easy - i'm happy.

correct or challenge on any presummption.

brucee
----- Original Message -----
From: "Rob Pike" <rob@mightycheese.com>
To: <9fans@cse.psu.edu>
Sent: Friday, February 27, 2004 4:28 PM
Subject: Re: [9fans] Threads: Sewing badges of honor onto a Kernel


> in case that wasn't clear enough, i have no idea what linus is talking
> about when he says we had 'overhead' that made it 'really stupid'
> 'in practice'.
>
> -rob

RE: [9fans] Threads: Sewing badges of honor onto a Kernel

[Nigel Roles]

9fans-admin@cse.psu.edu wrote:
> If I understand right, I could summarise (one of) Linus's arguments,
> as follows:
>
> 	The cost of sharing some of VM, but not all (in terms of peformance,
> 	due to less-visible things like TLB flushes, etc), out-weighs having
> 	to write a little bit of assembler wrapping the clone system call
> 	(which only has to be got right once, however horrible it might be).
>
> I haven't seen any argument rebutting this in any way. If it is true,
> then surely a performance hit on all (forked?) processes, is more
> important than having to shim a system call? If it is not true, then
> it would be nice to know!
>
> Tell me where I'm wrong, please?
>

The performance argument may well still be regarded by Linus as
stronger, but there are other differences. One is that the stack
used by the clone, being allocated on the heap, is fixed in size,
and unprotected from overflow.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

it was much more fun porting inferno to linux a few years ago.
no privilege of private data after a clone so the only way we
could find "up" was to grab the TSS, which linux actually used,
and hash it to get a up.  good stuff.

brucee
----- Original Message -----
From: "Charles Forsyth" <forsyth@terzarima.net>
To: <9fans@cse.psu.edu>
Sent: Friday, February 27, 2004 10:00 AM
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel


> >>Just out of curiosity, how do I get to the private space without
> >>a lock?  It's been a long time since I studied these things and lots
> >>of water has flown under bridges, so I could be missing something.
>
> static __inline Proc *getup(void) {
> Proc *p;
> __asm__( "movl %%esp, %%eax\n\t"
> : "=a" (p)
> );
> return *(Proc **)((unsigned long)p & ~(KSTACK - 1));
> };
etc ...

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 438 bytes --]

i didn't say this at the time, but now that you mention inferno, i will.
if someone is seriously worried about TLB flushing on context switches,
likes setting up the MMU pretty much once-for-all,
thinks all threads should share all the address space, and absolutely
disdains user mode, he should be writing in limbo for
native inferno, where all of that is the usual state,
and indeed there is no other.  we're all in it together.

[-- Attachment #2: Type: message/rfc822, Size: 3045 bytes --]

From: "Bruce Ellis" <brucee@chunder.com>
To: <9fans@cse.psu.edu>
Subject: Re: [9fans] Threads: Sewing badges of honor onto a Kernel
Date: Sat, 28 Feb 2004 20:38:35 +1100
Message-ID: <006201c3fdde$a6339030$8201a8c0@cc77109e>

i had trouble believing that the screeds of banter were from linus.
maybe he has a ghost writer.  when he did the "i don't write
user programs with threads" and thru in some kernel examples,
with obligatory assembly language results - well i thought
what are these guys discussing? gee, i just write in limbo.
threads are fun, linus, when you don't have to do all that crap
to manage them.  i just use "spawn", the only language support
for threads.  the rest is easy - i'm happy.

correct or challenge on any presummption.

brucee
----- Original Message -----
From: "Rob Pike" <rob@mightycheese.com>
To: <9fans@cse.psu.edu>
Sent: Friday, February 27, 2004 4:28 PM
Subject: Re: [9fans] Threads: Sewing badges of honor onto a Kernel


> in case that wasn't clear enough, i have no idea what linus is talking
> about when he says we had 'overhead' that made it 'really stupid'
> 'in practice'.
>
> -rob

[9fans] limbo?

[David Tolpin]

Novice question: do I need Inferno to program in limbo?
That is, I know I can if I have (I did on FreeBSD-hosted one).
But there is no limbo without Inferno under Plan9 too, right?

David

Re: [9fans] limbo?

[Geoff Collyer]

I believe that you still need Inferno.  There's been talk of
freestanding limbo but I don't believe it exists yet.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

yeah, it made no sense.  both the froggie and the ps2 ports
have very static page tables.  and the tlb is always right
after a first access.  private stacks?  every thread.

brucee
----- Original Message -----
From: "Charles Forsyth" <forsyth@terzarima.net>
To: <9fans@cse.psu.edu>
Sent: Saturday, February 28, 2004 10:10 AM
Subject: Re: [9fans] Threads: Sewing badges of honor onto a Kernel


> i didn't say this at the time, but now that you mention inferno, i will.
> if someone is seriously worried about TLB flushing on context switches,
> likes setting up the MMU pretty much once-for-all,
> thinks all threads should share all the address space, and absolutely
> disdains user mode, he should be writing in limbo for
> native inferno, where all of that is the usual state,
> and indeed there is no other.  we're all in it together.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[viro]

On Sat, Feb 28, 2004 at 09:44:24AM -0000, Nigel Roles wrote:

> The performance argument may well still be regarded by Linus as
> stronger, but there are other differences. One is that the stack
> used by the clone, being allocated on the heap, is fixed in size,
> and unprotected from overflow.

clone() uses whatever you pass to it; man 2 mmap for further inspiration...

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> correct or challenge on any presummption.

any number between 1 and 9 is a pint!!

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> The performance argument may well still be regarded by Linus as
> stronger, but there are other differences.

get real, we're no longer using 1 MIP VAXes, where TLB flushes
etc were a real problem.  128 users on an 11/780, anyone?

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> To paraphrase Captain Mainwaring, "I wondered if you'd spot that".

just "don't panic, don't panic"!!

RE: [9fans] Threads: Sewing badges of honor onto a Kernel

[Nigel Roles]

9fans-admin@cse.psu.edu wrote:
> On Sat, Feb 28, 2004 at 09:44:24AM -0000, Nigel Roles wrote:
>
>> The performance argument may well still be regarded by Linus as
>> stronger, but there are other differences. One is that the stack
>> used by the clone, being allocated on the heap, is fixed in size,
>> and unprotected from overflow.
>
> clone() uses whatever you pass to it; man 2 mmap for further
> inspiration...

To paraphrase Captain Mainwaring, "I wondered if you'd spot that".
Whilst writing the email, it occurred to me that you could
probably pull stunts with mmap/mremap and catching signals,
and get what is wanted.

We are now composing several system calls in some moderately
clever ways to get a behaviour which, whilst not equivalent to
rfork() is as flexible. It's not exactly obvious though is it?
I think I would expect to see a helper function.

Nice to see you on the list again Al.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[David Presotto]

The overhead on fork/exec of having to copy the stack descriptors
into the forked process is pretty minimal compared to the exec.

If you are just forking without execing then getting a new
stack is what you want.

If you are forking and sharing memory twixt the two forked processes,
we do indeed take a hit every time the processes context switch
especially on the x86 where we lose the previous tlb context as
soon as we putcr3().  If the two processes shared the TLB state
(pid on mips, page table on x86) we'ld be able to avoid that.
Not having any private segments would allow us to do it.

If you are serious about caring, throw another bit into rfork
that says dump the stack segment.  You'll also have to find
someplace in all architectures to hide the pointer to the
thread private memory.  Add to the kernel support for sharing
the TLB state.  Then measure the two and tell us how much you
saved in various programs.  If its non-neglibigle, you'll
have a reason argument instead of this endless whining back
and forth.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[viro]

On Sat, Feb 28, 2004 at 01:40:48PM -0000, Nigel Roles wrote:
> Whilst writing the email, it occurred to me that you could
> probably pull stunts with mmap/mremap and catching signals,
> and get what is wanted.

mmap(2)
       MAP_GROWSDOWN
              Used for stacks. Indicates to the kernel VM system that the map-
              ping should extend downwards in memory.

See mm/mmap.c:expand_stack() for details.

So no signals involved and that's precisely what is used for stack anyway -
anonymous VM_GROWSDOWN mapping.

Note: from my reading of the calc_vm_flag_bits() there is a problem on
parisc, what with the stack growing up - no way to get VM_GROWSUP from
mmap() and that's what would be needed to act as stacks there.  For that
matter, parisc do_page_fault() looks fishy in that area, but I'm not
familiar enough with that platform.

RE: [9fans] Threads: Sewing badges of honor onto a Kernel

[Nigel Roles]

9fans-admin@cse.psu.edu wrote:
> Subject: Re: [9fans] Threads: Sewing badges of honor onto a Kernel
>
>
> On Sat, Feb 28, 2004 at 01:40:48PM -0000, Nigel Roles wrote:
>> Whilst writing the email, it occurred to me that you could
>> probably pull stunts with mmap/mremap and catching signals,
>> and get what is wanted.
>
> mmap(2)
>        MAP_GROWSDOWN
>               Used for stacks. Indicates to the kernel VM system that
>               the map- ping should extend downwards in memory.
>
> See mm/mmap.c:expand_stack() for details.
>
> So no signals involved and that's precisely what is used for stack
> anyway - anonymous VM_GROWSDOWN mapping.
>
> Note: from my reading of the calc_vm_flag_bits() there is a problem on
> parisc, what with the stack growing up - no way to get VM_GROWSUP from
> mmap() and that's what would be needed to act as stacks there.  For
> that matter, parisc do_page_fault() looks fishy in that area, but I'm
> not familiar enough with that platform.

Fair point; I had missed the utility of MAP_ANON.

So there it is, Martin. Near enough the same semantics.
It's not pretty, and not obvious, but equivalent. With a helper
function it could be transparent to the user level (i.e. as
easy to deploy as rfork()), and apparently more MMU efficient.
I say apparently because I don't feel qualified to judge.

The question is, does this make Linux 'right' and Plan 9 'wrong',
which was quite a bit of what touched off the thread?

Neither, I would have thought.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[viro]

On Sat, Feb 28, 2004 at 06:16:50PM -0000, Nigel Roles wrote:

> The question is, does this make Linux 'right' and Plan 9 'wrong',
> which was quite a bit of what touched off the thread?
>
> Neither, I would have thought.

<shrug>

There is only one way to figure out and it had been described upthread.
For Linux such experiments had been done and results were very clear -
price of TLB flushes was considerable and that's aside of the complexity
of supporting a lot of mappings with partial VM sharing.  For Plan 9 the
answer might be different, might be the same, but there's no way to
pull that answer out of thin air.  FWIW, I suspect that full-shared
VM would be a win, but it can only be decided by trying and comparing.
End of story, unless somebody has time for that work and is willing to
do it.

Re: [9fans] limbo?

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 500 bytes --]

several of us are interested in blurring the boundaries enough
that quasi-real quasi-applications can be done in Plan 9 in Limbo
under hosted Inferno without the feel of a complete-box-in-a-box.
in fact, it can and has been done but some of the bits aren't implemented
so as to be distributed (which could be taken either way).

i'm surprised, though, that you'd want to put your TLBs so much at risk.
poor little things.  show them evil limbo and i'm sure they'll collapse
from heapatitis.

[-- Attachment #2: Type: message/rfc822, Size: 2152 bytes --]

From: David Tolpin <dvd@davidashen.net>
To: 9fans@cse.psu.edu
Subject: [9fans] limbo?
Date: Sat, 28 Feb 2004 14:11:32 +0400 (AMT)
Message-ID: <200402281011.i1SABWNP040924@adat.davidashen.net>

Novice question: do I need Inferno to program in limbo?
That is, I know I can if I have (I did on FreeBSD-hosted one).
But there is no limbo without Inferno under Plan9 too, right?

David

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>For Linux such experiments had been done and results were very clear -
>>price of TLB flushes was considerable and that's aside of the complexity
>>of supporting a lot of mappings with partial VM sharing.  For Plan 9 the

were those real applications or a synthetic test?  i'm curious what it actually did.
sorry, wrong question.  can you please point me to the paper and/or file that might answer them?
i wonder about the claimed `complexity'.  then again, i've seen signal.h
not to mention sys/types.h

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Rob Pike]

>>> For Linux such experiments had been done and results were very clear
>>> -
>>> price of TLB flushes was considerable and that's aside of the
>>> complexity
>>> of supporting a lot of mappings with partial VM sharing.  For Plan 9
>>> the
>
> were those real applications or a synthetic test?  i'm curious what it
> actually did.
> sorry, wrong question.  can you please point me to the paper and/or
> file that might answer them?
> i wonder about the claimed `complexity'.  then again, i've seen
> signal.h
> not to mention sys/types.h

mike burrows and i talked about this yesterday and concluded that there
is certainly a cost for plan 9's model, since you must, depending on the
MMU, do at least one of:
	- share the TLB among >1 task ids
	- fault in some MMU entries from memory after a context switch,
	  since the root of the MMU tree must be different for each process.
however, there are two mitigating factors.  first, the cost of sharing
the
TLB will be very program dependent.  sometimes it will hurt, but often
it will not.  for typical plan 9 programs with most of the non-main
procs
just doing an i/o loop, the cost will be minimal.  for a program that's
a multithreaded memset or its equivalent, it's probably a measurable
factor but not deal-breaking except in unrealistic pathological
situations.  in any case, this issue
matters more on MIPS-like MMUs than on the x86, i believe.

second, the total
overhead to fault in the TLB path to the pages you're using after a
context switch is something like a couple of microseconds in current
processors, perhaps as high as 10 microseconds for a very bad and
unusual case.  if those few microseconds are a performance issue for
your
program, you should be pretty happy.

if performance is your dominant concern, you could turn these points
into
a condemnation of plan 9's model.  i've never been in that position,
though.  for me, performance is but one factor in the figure of merit
of a system and is often offset by other factors.

-rob

Re: [9fans] limbo?

[Russ Cox]

> Novice question: do I need Inferno to program in limbo?
> That is, I know I can if I have (I did on FreeBSD-hosted one).
> But there is no limbo without Inferno under Plan9 too, right?

Geoff is right -- you need Inferno for now.  However, if you want to
do it without Inferno, I have that lying around.  I got the basic
running of modules (including loading "native-code" modules written
in C) working and then stopped.  This required some changes
to the Plan 9 linker in order to create the native-code module
in the first place.  I think these changes are on sources, just
undocumented.  If you wanted to run this on a non-Plan 9 system,
it shouldn't be hard to use the Unix dl interface instead.

The idea was to provide native-code module interfaces to the common
C libraries and have some fun writing Limbo code that interacted
with the OS directly instead of via the Inferno buffer.

The main thing missing is fleshing out the sys module and other
native-code modules.

	http://pdos.lcs.mit.edu/~rsc/dis.tar.gz

The LICENSE in that archive is the VN Liberal Source Licence, since
that's the current Inferno distribution license, but the code base
I used was actually a little earlier (somewhere between 3rd and 4th
edition Inferno).  Almost none of the code is mine -- I just repackaged
the dis interpreter.

Russ

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> i had trouble believing that the screeds of banter were from linus.
> maybe he has a ghost writer.

this is the bigest kernel (sic) in the southern hemisphere.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

i think it was rob, but i'm not sure, who said:

    linear is programming is hard enough, but
    concurrent programming is beyond the
    'average' programmer.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> I believe the shortsightedness of not imposing a strict type created
> too many possibilities. This creates a bigger problem when dealing
> with libraries imported into a system with no defined interface, and
> you end up with the desire for one.

limbo does this right.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> > Wrong. You have zero basis for your assertion, and it's simply bogus.

the 'everything is a file' (invented 30? years ago) is a plan.

unlike that hideous lunix /proc [yeah, we had that 1984].

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Kenji Okamoto]

By the way, it took half a day for me to read plenty of mails during
these three days while I was knocked down by flu...  Yeah, it was
interesting to read those.

Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>i think it was rob, but i'm not sure, who said:
>    linear is programming is hard enough, but
>    concurrent programming is beyond the
>    'average' programmer.

actually, i'd say that programming generally is hard enough, and not
just for the average programmer,
and concurrent programming is just a special case within that.

more seriously, some things are more clearly expressed and more easily implemented
using concurrent processes or perhaps coroutines, so concurrency ought to be taught,
and learned, and well.

in contrast to the quote above, in an ancient usenet article,
in the context of concurrent programming, i am reasonably certain that
rob made the observation that as a discipline, we can learn.  he
used the example of fork(), which was considered `difficult' when it
first appeared, compared to existing notions such as `jobs', but after
being studied and taught in advanced programming for a time, it become
familiar enough that it was suitable to be taught/used much earlier.
that seems to me to be a better quote to use.

i'd say from my experience that unless people insist on pronouncing themselves `full' and
incapable of accepting any new ideas, which certainly does happen, they can typically
learn.    (otherwise, they end up saying things such as ``in every other X i've
ever seen i could always do Y in this particular way''.)
i think in many ways concurrent programming is more general
than (say) object-oriented programming, and certainly the language
constructions can be much simpler than some object-oriented ones.

in other words: if you're capable of understanding `finalised virtual hyperstationary factory class',
remembering the Java class hierarchy, and all the details of the Java Media Framework,
you are (a) a better man than i am (b) capable of filling your mind with large
chunks of complexity, so concurrent programming should be simple by comparison.
go for it.

ps. i made up the hyperstationary, but then again, it's probably a design pattern.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

>> I'm saying that you CANNOT "message pass" a hash table.
> 
> you can.  i've done it.

I told him, but he them said that calling a function was also
message passing, which is not. Passing a fn pointer through
a pointer, which IMHO he missed, of course is. IMHO, he should
read the source of acme. I did learn a lot by doing so (after decades
of doing concurrent programs).

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[dbailey27]

[-- Attachment #1: Type: text/plain, Size: 6 bytes --]

Nice

[-- Attachment #2: Type: message/rfc822, Size: 4189 bytes --]

From: Charles Forsyth <forsyth@terzarima.net>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Mon, 1 Mar 2004 08:19:16 0000
Message-ID: <757a63de546942340d005e0e4fcc471e@terzarima.net>

>>i think it was rob, but i'm not sure, who said:
>    linear is programming is hard enough, but
>    concurrent programming is beyond the
>    'average' programmer.

actually, i'd say that programming generally is hard enough, and not
just for the average programmer,
and concurrent programming is just a special case within that.

more seriously, some things are more clearly expressed and more easily implemented
using concurrent processes or perhaps coroutines, so concurrency ought to be taught,
and learned, and well.

in contrast to the quote above, in an ancient usenet article,
in the context of concurrent programming, i am reasonably certain that
rob made the observation that as a discipline, we can learn.  he
used the example of fork(), which was considered `difficult' when it
first appeared, compared to existing notions such as `jobs', but after
being studied and taught in advanced programming for a time, it become
familiar enough that it was suitable to be taught/used much earlier.
that seems to me to be a better quote to use.

i'd say from my experience that unless people insist on pronouncing themselves `full' and
incapable of accepting any new ideas, which certainly does happen, they can typically
learn.    (otherwise, they end up saying things such as ``in every other X i've
ever seen i could always do Y in this particular way''.)
i think in many ways concurrent programming is more general
than (say) object-oriented programming, and certainly the language
constructions can be much simpler than some object-oriented ones.

in other words: if you're capable of understanding `finalised virtual hyperstationary factory class',
remembering the Java class hierarchy, and all the details of the Java Media Framework,
you are (a) a better man than i am (b) capable of filling your mind with large
chunks of complexity, so concurrent programming should be simple by comparison.
go for it.

ps. i made up the hyperstationary, but then again, it's probably a design pattern.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

> I told him, but he them said that calling a function was also
> message passing, which is not. Passing a fn pointer through
> a pointer, which IMHO he missed, of course is. IMHO, he should
> read the source of acme. I did learn a lot by doing so (after decades
> of doing concurrent programs).

I meant this, sorry, not enough coffee yet...

I told him, but he then said that calling a function was also
message passing, which is not. Passing a fn pointer through
a channel, which IMHO he missed, of course is. IMHO, he should
read the source of acme. I did learn a lot by doing so (after decades
of doing concurrent programs).

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Lucio De Re]

On Mon, Mar 01, 2004 at 09:48:47AM +0100, Fco.J.Ballesteros wrote:
> 
> I told him, but he then said that calling a function was also
> message passing, which is not. Passing a fn pointer through
> a channel, which IMHO he missed, of course is. IMHO, he should
> read the source of acme. I did learn a lot by doing so (after decades
> of doing concurrent programs).

Surely message passing differs from pointer passing by the mere
addition of synchronisation?!  In which case I fail to see how
Torvalds could consider himself a kernel designer and not recognise
the chasm between them.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

> Surely message passing differs from pointer passing by the mere
> addition of synchronisation?!

Maybe you don't even need that:

	msg = mallocmsg(...);
	sendp(chan, msg);

Is it a pointer? Sure. Is it a message? Sure. Do we need locks
if we transfer ownership (which is an abstract concept)? Not sure.

IMHO, Linus really needs to read /sys/src/cmd/acme/xfid.c

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Kenji Okamoto]

> IMHO, Linus really needs to read /sys/src/cmd/acme/xfid.c

One of the most difficult source to me.☺

Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Kenji Okamoto]

>> IMHO, Linus really needs to read /sys/src/cmd/acme/xfid.c
> 
> One of the most difficult source to me.☺

I meant clear and difficult.

Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> in other words: if you're capable of understanding `finalised virtual hyperstationary factory class',
> remembering the Java class hierarchy, and all the details of the Java Media Framework,

In fact, there is no such thing as 'virtual' in Java, it is from other
languages. Finalised cannot be a feature of a class in a hierarchy,
since finalized is a state of an object after it was deleted and it
is only needed to define the semantics of the language, not to
program in it; remembering all the details
of Java Media Framework is no more necessary than remembering all
the details of PDF 1.5 specification -- unless you earn money by
winning contests on writing multimedia java applets or
checking PDF generators for conformance.

Factory in Java is a simple concept, one more level of indirection.
Think of it as something similar to 'bind' in Plan9.

In general, Java is very close to Plan9 in many things. It is built
to provide a cleaner and smaller alternative to C++; it offers very
few concepts to learn -- and the extensive libraries are not meant
to learn -- there are javadocs for them. The core language and system
design is small and logical. It bears many ideas from Oberon (the system
acme took its interface and interaction design), including
the language design and the idea of embedded applets.

Parallel programming in Java is easy and natural, as easy and natural
as in other language or system or even easier, dare I say it.  You
just use threads, and use basic language features, such as variable
scopes, to have shared and private thread resources. No additional
flags and bits to remember; everything is easy and straightforward.

I think that the major objection against Java is its closedness,
but Plan9 has this problem too -- and it is being solved for both
designs.

David Tolpin

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Geoff Collyer]

I think the original quote Boyd was looking for is this, from the 8½
paper:

	As discussed in a previous paper [Pike89] I prefer to free
	applications from event-based programming.  Unfortunately, though, I
	see no easy way to achieve this in single-threaded C programs, and am
	unwilling to require all programmers to master concurrent programming.

Fair enough.  Such a sentiment may seem old-fashioned nowadays when
everybody seems to want to write multi-threaded, hyper-threaded,
multi-tasking programs (it's faster, you know); just look at the links
browser.  One of the services most operating systems provide is to
convert asynchrony (notably activity on other processors and
interrupts) into something more manageable.  I find that enthusiasm
for writing multi-threaded, multi-process programs fades after finding
and fixing a few kernel race conditions on multi-processor systems.

Finally, from a fortune file:

Von Neumann's greatest gift was to get us OUT of parallel processors. -E. Miya

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

>>languages. Finalised cannot be a feature of a class in a hierarchy,
>>since finalized is a state of an object after it was deleted and it
>>is only needed to define the semantics of the language, not to

``[the programmer has] the ability to declare classes or methods as "final". [they cannot
be overriden in a subclass].  The language guarantees that the actual method invoked on
the object is the {finalised method} [my {}] which was written for that class...''

it was that sense i meant, not object destruction.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> >>since finalized is a state of an object after it was deleted and it
> >>is only needed to define the semantics of the language, not to
>
> ``[the programmer has] the ability to declare classes or methods as "final". [they cannot
> be overriden in a subclass].  The language guarantees that the actual method invoked on
> the object is the {finalised method} [my {}] which was written for that class...''
>
> it was that sense i meant, not object destruction.

Isn't it like bind without -c?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bengt Kleberg]

Linus Torvalds wrote:
...deleted
> 
> Methinks you have read a few too many papers about microkernels, without 
> actually seeing the real world.

do not forget that applications running on microkernels are faster than 
on monolithic kernels. see http://www.pdos.lcs.mit.edu/exo


bengt

This communication is confidential and intended solely for the addressee(s). Any unauthorized review, use, disclosure or distribution is prohibited. If you believe this message has been sent to you in error, please notify the sender by replying to this transmission and delete the message without disclosing it. Thank you.

E-mail including attachments is susceptible to data corruption, interruption, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any such corruption, interception, amendment, tampering or viruses or any consequences thereof.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Douglas A. Gwyn]

Nigel Roles wrote:
> The performance argument may well still be regarded by Linus as
> stronger, but there are other differences. One is that the stack
> used by the clone, being allocated on the heap, is fixed in size,
> and unprotected from overflow.

That would be a serious flaw on a system with a small address space.
It's problematic anyway in its inefficient use of PTEs for the
process, since far more table entries are needed (extra stack
for each thread).

However, overflow detection can in principle be achieved by
mapping all pages of each thread's private stack region except
for the last page, which allows the MMU to flag any overflow.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Douglas A. Gwyn]

boyd, rounin wrote:
> get real, we're no longer using 1 MIP VAXes, where TLB flushes
> etc were a real problem.  128 users on an 11/780, anyone?

There are a large number of platforms where there are
important constraints imposed by the MMU architecture.
Consider what it means when an OS requires such resources
that it cannot feasibly be implemented on a VAX, yet does
almost nothing more than could have readily been done
(with a different design) on a VAX.  It's nearly as bad
as a 1GHz machine getting so bogged down with GUI code
and poor interface design that some common tasks take
longer to accomplish than on a 1MHz PDP-11 running Unix.
(Whose "Law" was it that said that file storage always
expands to fill whatever disk resources are available?
The same thing seems to apply to CPU cycles.)

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

my point was just that programmers, presumably `average' ones as well,
often need to come to grips with a fair number of non-trivial things,
and they are often rather specific (as with the JMF).  it seems to me that
if programmers can deal with things such as those (not just those particular things),
there's a reasonable chance that at least a good percentage can indeed deal with
something that is more fundamental, and more general.
of course part of what is discovered, taught and learnt,
is what discipline to follow to avoid trouble (as in structured programming),
and even when concurrency is best avoided.   of course talents will vary, but
even so.

>>remembering all the details
>>of Java Media Framework is no more necessary than remembering all

i chose it as an example of a moderately complex interface only
because i was on the JMF email list years ago and still
retain the following e-mail message from that list, as an example of
how one can produce fairly intricate interfaces (that did get in the way).
my favourite part is the last paragraph, which reminds me of the sketch
in the Life of Brian: ``the people's liberation army of judea'', 
``the judean people's liberation army'', ... .  splitters!


	Subject: DataSource and Player coupling
		...

	I'm writing a java.media.content.reliable.DataSource implementation for
	use with Intels Java Media implementation. I'm not sure how to get the
	framework to use my DataSource without resorting to writing my own
	Player implementation (or instantiating an undocumented internal Player
	implementation)

	I could use a unique protocol (e.g. softcom:) and implement a
	java.media.protocol.softcom.PlayerFinder which somehow determines the
	MIME type and then calls
	java.media.content.reliable.<mime-type>.PlayerFinder.createPlayer()
	passing an instance of my DataSource implementation. Or I could create
	an instance of my DataSource and create an instance of the correct
	content PlayerFinder directly, e.g.:

	 DataSource ds = new MyDataSource(url);
	 java.media.content.reliable.PlayerFactory pf =
	                new
	java.media.content.reliable.video.mpeg.PlayerFinder();
	 m_jmPlayer = pf.createPlayer(ds);

	I would still need to determine the MIME type of the source URL in order
	to locate the PlayerFinder.

		...

	DataSource and Player implementations seem tightly bound together in the
	framework.  I would think they should be independent so I can mix and
	match DataSources with Player implementations.

	Also, it seems strange that the DataSource is in a content specific
	package, e.g. java.media.content.reliable.<mime-type>.DataSource. The
	DataSource doesn't need to know anything about the content - it just
	reads raw bytes - why does it care if they are MPEG, AVI, WAV etc.?  It
	would seem more sensible to tie the DataSource to the protocol, e.g.
	java.media.protocol.<protocol>.DataSource.

	In fact, why are the reliable and streaming packages tied to content
	type (java.media.content.reliable and java.media.content.streaming)?
	It's the protocol that is reliable or streaming, not the content (is
	MPEG reliable or streaming? it depends on the protocol). Shouldn't these
	packages be named java.media.protocol.reliable and
	java.media.protocol.streaming?

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Geoff Collyer]

from a fortune file:

	The steady state of disks is full.  --Ken Thompson

I believe it was Mike O'Dell who asked (approximately) `Why is it that
despite processors getting faster, there never seems to be any
additional cycles available to me?'.  His answer was window systems
and networks, among others.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[dbailey27]

> Hmm...  Probably Ingo Molnar would have all details.  IIRC, that was on real
> applications, but back then I was dealing with filesystem side of things and
> not much else, so that's second-hand information.  I can ask him for details
> if you want them; 

Yes, please.

Don

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[viro]

On Sat, Feb 28, 2004 at 07:44:31PM +0000, Charles Forsyth wrote:
> >>For Linux such experiments had been done and results were very clear -
> >>price of TLB flushes was considerable and that's aside of the complexity
> >>of supporting a lot of mappings with partial VM sharing.  For Plan 9 the
> 
> were those real applications or a synthetic test?  i'm curious what it actually did.
> sorry, wrong question.  can you please point me to the paper and/or file that might answer them?

Hmm...  Probably Ingo Molnar would have all details.  IIRC, that was on real
applications, but back then I was dealing with filesystem side of things and
not much else, so that's second-hand information.  I can ask him for details
if you want them; I certainly agree that e.g. presense of shared libraries
changes the picture, so those results do not apply directly to Plan 9 and if
somebody really cares they should try and compare for normal Plan 9 workloads.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> i chose it as an example of a moderately complex interface only
> because i was on the JMF email list years ago and still

Yes, it was a good example. 

Do you think limbo is better for learning/teaching parallel programming
ideas than C+rfork?

David Tolpin

[9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd]

Geoff Collyer writes:
> I find that enthusiasm
> for writing multi-threaded, multi-process programs fades after finding
> and fixing a few kernel race conditions on multi-processor systems.

yes, <sigh> ... 

[9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd]

Charles Forsyth writes:
> ps. i made up the hyperstationary, but then again, it's probably a design pattern.

ROTFLMAO 

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 11 bytes --]

ROTFLMAO ??

[-- Attachment #2: Type: message/rfc822, Size: 1968 bytes --]

From: boyd@insultant.net
To: 9fans@cse.psu.edu
Subject: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Mon, 01 Mar 2004 07:18:58 -0500
Message-ID: <04Mar1.071858-0500_est.387693-24448+36064@ams.ftl.affinity.com>

Charles Forsyth writes:
> ps. i made up the hyperstationary, but then again, it's probably a design pattern.

ROTFLMAO 

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[lucio]

> ROTFLMAO ??

Roll on the floor laughing my arse off!

++L

[9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd]

9fans-admin@cse.psu.edu writes: 

> ROTFLMAO ??

Rolling On The Floor Laughing My Arse Off 

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Russ Cox]

> do not forget that applications running on microkernels are faster than 
> on monolithic kernels. see http://www.pdos.lcs.mit.edu/exo

no.  applications running on a kernel tailored to the benchmark
are faster than applications running on a generic kernel.  
exokernel != microkernel.

- rsc@pdos.lcs.mit.edu

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Presotto]

Its like anything else, once you get used to it, you make less
mistakes.

[9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd]

Russ Cox writes:
> exokernel != microkernel.

Correct! -- Justin Hawkins, The Darkness. 

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Mon, 1 Mar 2004, Lucio De Re wrote:

> Surely message passing differs from pointer passing by the mere addition
> of synchronisation?!  In which case I fail to see how Torvalds could
> consider himself a kernel designer and not recognise the chasm between
> them.

In most people's minds message passing passing usually means 'move the
data from this hole here to that hole there'. Of course in any optimized
situation (shared memory) this can be done with pointer hacks. And even in
the optimized situations, message passing libraries as measured are at
least one or two orders of magnitude slower than simple 'hand the pointer
around', esp. on shared-memory and CC-NUMA systems -- I can give you
references, if you wish, or you can google Jim Taft's work at NASA AMES.  
There's a hardware reason for the performance difference. 

Do people care about that difference? Yes! If you just dropped $10-20M on
a computer, do you want to get $500K worth of performance out of it, or
would you rather get more like $10-20M out of it? Your call. Better yet, 
assume your job rests on the decision, then make the decision :-)

Point to Linus on this one, I think. Sorry.

On another note, the Linux guys are not kernel designers in many senses of
that word. From my point of view Linux is an extremely competent
implementation of some really good and some really bad OS ideas -- ranging
from V6 Unix syscall interface to later ideas such as VFS.  And, like it
or not, it works very very well for many people, including the several
thousand systems we have here that run it for high performance computing.
Credit where credit is due.

And, that said, I'm still gradually moving my existence over to Plan 9. 

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Mon, 1 Mar 2004, Bengt Kleberg wrote:

> > 
> > Methinks you have read a few too many papers about microkernels, without 
> > actually seeing the real world.
> 
> do not forget that applications running on microkernels are faster than 
> on monolithic kernels. see http://www.pdos.lcs.mit.edu/exo

so I assume you are running this on your machines?

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[lucio]

> In most people's minds message passing passing usually means 'move the
> data from this hole here to that hole there'. Of course in any optimized
> situation (shared memory) this can be done with pointer hacks. And even in
> the optimized situations, message passing libraries as measured are at
> least one or two orders of magnitude slower than simple 'hand the pointer
> around', esp. on shared-memory and CC-NUMA systems -- I can give you
> references, if you wish, or you can google Jim Taft's work at NASA AMES.  
> There's a hardware reason for the performance difference. 
> 
I don't understand what is being measured in this case.  Surely one
doesn't go to Linus Torvalds to get specifications for the design of
on operating system for a specialised piece of equipment?  Ever since
before Unix, the emphasis has been on portability, OS design has
sacrificed performance to the objective of being able to continue
using applications on the newer, faster hardware.  At least, that is
how it seems to me.

> Do people care about that difference? Yes! If you just dropped $10-20M on
> a computer, do you want to get $500K worth of performance out of it, or
> would you rather get more like $10-20M out of it? Your call. Better yet, 
> assume your job rests on the decision, then make the decision :-)
> 
Sure.  But then did the price include optimised software, or do you
expect to run some off-the-street operating software on your
performance host?  By the time you spend money in that order of
magnitude, you can afford to employ the few software architects that
still understand low-level programming and get them to squeeze every
bit of performance out of your horse.  Or would you ask Torvalds to
drive the next generation of Grand Prix Ferrari to win the racing
season?

> Point to Linus on this one, I think. Sorry.
> 
No ways, it's a mismatch.  Torvalds just organically grew an OS from
Minix, for the Intel architecture.  The Great Barrier Reef likewise
grew organically, but is unlikely to be recommended as the dam wall
for the Yellow River.

> On another note, the Linux guys are not kernel designers in many senses of
> that word. From my point of view Linux is an extremely competent
> implementation of some really good and some really bad OS ideas -- ranging
> from V6 Unix syscall interface to later ideas such as VFS.  And, like it
> or not, it works very very well for many people, including the several
> thousand systems we have here that run it for high performance computing.
> Credit where credit is due.
> 
One can't hold against Torvalds the decision to stick to the Unix
paradigm, no one else had much of a better choice at the time.  But to
defend it in the face of obvious obsolescence is unforgivable.  And to
suggest that performance is a significant factor in the design of a
generic operating system is laughable.

> And, that said, I'm still gradually moving my existence over to Plan 9. 
> 
Which makes you a lot more practical than Torvalds.  Presumably
microoptimisations are not saving you millions of dollars.  Nor me :-)

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[rog]

> Do you think limbo is better for learning/teaching parallel programming
> ideas than C+rfork?

having been through a few years of doing C programming workshops,
i'd love to use Limbo as a teaching language.

parallel programming is not the only reason (although it's a good one);
just the overall cleanliness, the general modularity (no sprawling
class hierarchy to deal with), and the straightforward syntax.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[viro]

On Mon, Mar 01, 2004 at 06:23:33PM +0200, lucio@proxima.alt.za wrote:
> No ways, it's a mismatch.  Torvalds just organically grew an OS from
> Minix, for the Intel architecture.  The Great Barrier Reef likewise
> grew organically, but is unlikely to be recommended as the dam wall
> for the Yellow River.
 
Oh, for crying out loud...  Have you ever read either kernel?  Yes or no?
Or Plan 9 one, for that matter...

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Taj Khattra]

> Factory in Java is a simple concept, one more level of indirection.

almost every concept in computing is about one or more levels of
indirection.

> It is built to provide a cleaner and smaller alternative to C++;

yes, but that's not saying much - almost every language in existence
provides a cleaner and smaller alternative to c++.

> it offers very few concepts to learn 

i guess that's why the language spec alone is a svelte 500 pages.

> -- and the extensive libraries are not meant
> to learn -- there are javadocs for them. The core language and system

not meant to learn?  do javadocs have an ESP mechanism built into them?

> Parallel programming in Java is easy and natural, as easy and natural
> as in other language or system or even easier, dare I say it.  You
> just use threads, and use basic language features, such as variable
> scopes, to have shared and private thread resources. No additional
> flags and bits to remember; everything is easy and straightforward.

no it isn't.  it's nowhere near as easy and natural as the limbo, alef,
erlang etc. model for concurrent programming.  otoh, once you get used
to it, everything looks easy and straightforward.

> I think that the major objection against Java is its closedness,

no, that's the major objection only for a few groups.

Re: [9fans] Threads: Sewing badges of honor onto a Kernel

[Joel Salomon]

Douglas A. Gwyn said:
> (Whose "Law" was it that said that file storage always
> expands to fill whatever disk resources are available?
> The same thing seems to apply to CPU cycles.)

I've seen it quoted: "Intel giveth, Microsoft taketh away."

Is interesting to look at various open-source projects and see which ones
get larger & slower with newer versions (gcc, KDE, linux gets larger, but
if Linus is to be believed, they *are* getting faster) and those that
*decrease* in size -- any examples, anyone? Gnome, I *think*, but that's
just because they started of with infinite bloat, nowhere to go but
smaller/faster ;-)

--Joel

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> > Factory in Java is a simple concept, one more level of indirection.
>
> almost every concept in computing is about one or more levels of
> indirection.

Most concepts are more levels of indirections than one.
>
> > It is built to provide a cleaner and smaller alternative to C++;
>
> yes, but that's not saying much - almost every language in existence provides
> a cleaner and smaller alternative to c++.

Not every language is built to provide a smaller and cleaner alternative to a
successful one, and is successful. I would be glad to see limbo or Alef widely
used. They are not.

>
> > it offers very few concepts to learn 
>
> i guess that's why the language spec alone is a svelte 500 pages.

No. It is because the spec is written by people who come from culture of
detailed specifications; and it is because a part of the specification is
dedicated to binary compatibility, which few other languages provide.  Still
yet, the specification for Java, 504 pages, is SMALLER than specification for
ANSI/ISO C, which 554 pages.

> > -- and the extensive libraries are not meant to learn -- there are javadocs
> > for them. The core language and system
>
> not meant to learn?  do javadocs have an ESP mechanism built into them?

I do not know what is ESP. 

>
> > Parallel programming in Java is easy and natural, as easy and natural
> > as in other language or system or even easier, dare I say it.  You
> > just use threads, and use basic language features, such as variable
> > scopes, to have shared and private thread resources. No additional
> > flags and bits to remember; everything is easy and straightforward.
>
> no it isn't.  it's nowhere near as easy and natural as the limbo, alef,
> erlang etc. model for concurrent programming.  otoh, once you get used
> to it, everything looks easy and straightforward.

I wrote many thousands lines of code in C, Java, Perl, Scheme and Oberon. 
I used to do parallel programming in Modula-2. Parallel programming in model
is natural and straightforward, and it is an achievement of language designers
to come up with the model.

It is strange how often other systems are criticized for various features
on this list. Is it a part of plan9 culture? Is there a garbage-collecting
language with concepts for parallel programming born in Plan9 environment
which can be used with Plan9 to try out? 

David

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Joel Salomon]

David Tolpin said:

> I do not know what is ESP.

Extra-Sensory Preception i.e. psychic powers.

> Is there a garbage-collecting
> language with concepts for parallel programming born in Plan9 environment
> which can be used with Plan9 to try out?
>

Limbo/Inferno has been mentioned in this thread. Alef is dead, and cannot
be brought back (licencing issues with SGI iirc) -- was it GC, though?
Then there's Newsqueak, but I think that's just a proof-of-concept toy,
and not very useable.

But limbo is alive and available, and can be tried out under Windows,
plan9, linux, mac (I think).

--Joel

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> > I do not know what is ESP.
>
> Extra-Sensory Preception i.e. psychic powers.

No javadoc does not have ESP. But one can always browse the
documentation and see what interfaces are available. The javadoc
generated documentation is hyperlinked in its basic format,
which is very useful too.

> But limbo is alive and available, and can be tried out under Windows,
> plan9, linux, mac (I think).

Yes, I have Inferno installed under FreeBSD. The problem with limbo
for me is that it is much more 'a box in a box' than Java. This is
not a language available in Plan 9. It is a language of Inferno, a 
different system that happens to have common roots with Plan 9. And,
under FreeBSD, I cannot type in anything but English (that is,
I need Russian, Armenian and Hebrew keyboard input on daily basis)
without patching the kernel, I think.

I am not saying I cannot live without limbo/Java/whatever. I am
very happy with C, I've just compiled SCM under Plan9 and am going
to bring system interfaces in, and there is hugs ported.

But I would love to see something like limbo among basic 
tools of Plan9. Or like Java. Or, even better, both. And compare
how good they go in the same environment.

David

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Derek Fawcus]

On Mon, Mar 01, 2004 at 11:43:01PM +0400, David Tolpin wrote:
> Yes, I have Inferno installed under FreeBSD. The problem with limbo
> for me is that it is much more 'a box in a box' than Java.

But wasn't that supposed to be the aim of Java?  An identical virtual
machine running on all systems,  such that code could be written once?

At least in concept,  then inferno/limbo would seem to be "java done right".

DF

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> On Mon, Mar 01, 2004 at 11:43:01PM +0400, David Tolpin wrote:
> > Yes, I have Inferno installed under FreeBSD. The problem with limbo
> > for me is that it is much more 'a box in a box' than Java.
>
> But wasn't that supposed to be the aim of Java?  An identical virtual
> machine running on all systems,  such that code could be written once?
>
> At least in concept,  then inferno/limbo would seem to be "java done right".

Java is done right in concept. In reality, inferno is no more right
than Java. Unfortunately. I can do system programming and command-line
apps with Java. 

How am I supposed with Inferno on FreeBSD to do that?

David

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Charles Forsyth]

[-- Attachment #1: Type: text/plain, Size: 135 bytes --]

you're telling me.

nevertheless, there is more that could and no doubt will be done
to make it less obtrusive and more intrusive.

[-- Attachment #2: Type: message/rfc822, Size: 2735 bytes --]

From: Derek Fawcus <dfawcus@cisco.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Mon, 1 Mar 2004 21:07:19 +0000
Message-ID: <20040301210719.B10993@edinburgh.cisco.com>

On Mon, Mar 01, 2004 at 11:43:01PM +0400, David Tolpin wrote:
> Yes, I have Inferno installed under FreeBSD. The problem with limbo
> for me is that it is much more 'a box in a box' than Java.

But wasn't that supposed to be the aim of Java?  An identical virtual
machine running on all systems,  such that code could be written once?

At least in concept,  then inferno/limbo would seem to be "java done right".

DF

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[rog]

> But wasn't that supposed to be the aim of Java?  An identical virtual
> machine running on all systems,  such that code could be written once?
>
> At least in concept,  then inferno/limbo would seem to be "java done right".

i think the real problem people have with it is the visual one: there
really is a box (window) on screen that holds all your inferno
windows.

the new inferno window manager allows one to break out of this box, in
principle (i've done such a thing for rio, except later changes broke
it; charles suggested a fix which i should implement sometime).

under other systems, it would need a change in the draw device to
allow multiple windows (it's not entirely clear what the base
abstraction should be), but that shouldn't be too hard, it's just
setting the time aside to do it!

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[rob pike, esq.]

> Passing a fn pointer through
> a pointer, which IMHO he missed, of course is. IMHO, he should
> read the source of acme. I did learn a lot by doing so (after decades
> of doing concurrent programs).

philw was sitting at the next terminal when i wrote that code, which
was originally in alef.  i showed it to him and he was surprised that it
even worked.  no faith in his own language.  but the language worked
and so did the program.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

From: "David Tolpin" <dvd@davidashen.net>
> Java is done right in concept. In reality, inferno is no more right
> than Java.

bullshit

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Joel Salomon]

rog@vitanuova.com said:
> the new inferno window manager allows one to break out of this box, in
> principle (i've done such a thing for rio, except later changes broke
> it; charles suggested a fix which i should implement sometime).
>
Two questions about integrating inferno/limbo into plan9:
1) is there any way to make dis programs "runnable" like shell scripts are?
eg
  %dis/charon
rather than
  %emu /bin/dis/charon

2) could the inferno system be made to serve its /prog to the machine's
/proc? this would blur the borders between the host plan9 and the inferno
subsystem, which might be a good thing or a terrible idea.

--Joel

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Taj Khattra]

> in contrast to the quote above, in an ancient usenet article,
> in the context of concurrent programming, i am reasonably certain that
> rob made the observation that as a discipline, we can learn.  he

http://groups.google.ca/groups?q=ie=UTF-8&selm=13fi7tINNk1i%40darkstar.UCSC.EDU&rnum=1

> that seems to me to be a better quote to use.

i like this line from todd proebsting: ``Concurrency models many
applications better than objects, yet the world is mired in OO
religion.''  of course, joe armstrong has been preaching this for
many years too.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Roman Shaposhnick]

What do you guys think of OpenMP extension for C/C++ (http://www.openmp.org) ?
Of course, it's nothing radical, but still being able to write:

#pragma omp for
for (i=0; i<1000; i++) {
     do_something;
}

and not having to worry about nuts'n'bolts of the implementation is pretty
neat. Plus there's a bonus of worry-free synchronization (at least in it's
simplest form of a barrier). 

I do realize, that OpenMP is mostly suited for a domain with short-lived 
fire-and-forget sort of threads, but still, the fact that it's integrated
with the language should be appealing enough.

Thanks,
Roman.

On Mon, Mar 01, 2004 at 08:13:09PM -0800, Taj Khattra wrote:
> > in contrast to the quote above, in an ancient usenet article,
> > in the context of concurrent programming, i am reasonably certain that
> > rob made the observation that as a discipline, we can learn.  he
> 
> http://groups.google.ca/groups?q=ie=UTF-8&selm=13fi7tINNk1i%40darkstar.UCSC.EDU&rnum=1
> 
> > that seems to me to be a better quote to use.
> 
> i like this line from todd proebsting: ``Concurrency models many
> applications better than objects, yet the world is mired in OO
> religion.''  of course, joe armstrong has been preaching this for
> many years too.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Mon, 1 Mar 2004, Roman Shaposhnick wrote:

> What do you guys think of OpenMP extension for C/C++
> (http://www.openmp.org) ?

And to think I just ate. 

Too bad. Too bad!

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Roman Shaposhnick]

On Mon, Mar 01, 2004 at 09:47:01PM -0700, ron minnich wrote:
> On Mon, 1 Mar 2004, Roman Shaposhnick wrote:
> 
> > What do you guys think of OpenMP extension for C/C++
> > (http://www.openmp.org) ?
> 
> And to think I just ate. 

  Hungry or not, what *do* you think ?

Thanks,
Roman.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Mon, 1 Mar 2004, Roman Shaposhnick wrote:

> > And to think I just ate. 
> 
>   Hungry or not, what *do* you think ?

openmp is a bad idea.

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> From: "David Tolpin" <dvd@davidashen.net>
> > Java is done right in concept. In reality, inferno is no more right
> > than Java.
>
> bullshit

In particular, what part of limbo in reality is better than Java
in such a way that reminded you bull's excrements?

David Tolpin

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[9nut]

> which reminds me of the sketch
> in the Life of Brian: ``the people's liberation army of judea'', 
> ``the judean people's liberation army'', ... .  splitters!

"Judean People's Front" and "People's Front of Judea."  Brilliantly funny!

http://www.mwscomp.com/movies/brian/brian-07.htm

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[rob pike, esq.]

object-oriented design is the roman numerals of computing.

-rob

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Douglas A. Gwyn]

lucio@proxima.alt.za wrote:
> And to
> suggest that performance is a significant factor in the design of a
> generic operating system is laughable.

Not entirely:  If an OS design *precludes* efficient
operation on any architecture, surely that is bad design.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bengt Kleberg]

Russ Cox wrote:
>>do not forget that applications running on microkernels are faster than 
>>on monolithic kernels. see http://www.pdos.lcs.mit.edu/exo
> 
> 
> no.  applications running on a kernel tailored to the benchmark
> are faster than applications running on a generic kernel.  
> exokernel != microkernel.

i thought that the idea with an micro kernel is that it makes it 
_possible_ to tailor the application interface towards many different 
applications. a monolithic kernel has only one interface.

how does one distinguish between a micro kernel (like l4) and the exokernel?


bengt

This communication is confidential and intended solely for the addressee(s). Any unauthorized review, use, disclosure or distribution is prohibited. If you believe this message has been sent to you in error, please notify the sender by replying to this transmission and delete the message without disclosing it. Thank you.

E-mail including attachments is susceptible to data corruption, interruption, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any such corruption, interception, amendment, tampering or viruses or any consequences thereof.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bengt Kleberg]

ron minnich wrote:
> On Mon, 1 Mar 2004, Bengt Kleberg wrote:
> 
> 
>>>Methinks you have read a few too many papers about microkernels, without 
>>>actually seeing the real world.
>>
>>do not forget that applications running on microkernels are faster than 
>>on monolithic kernels. see http://www.pdos.lcs.mit.edu/exo
> 
> 
> so I assume you are running this on your machines?

afaik no exokernel version available meets the neccessary criteria.
one of which is that it is supported by ''HP user support for Ericsson''

otherwise i am sure i would.


bengt

This communication is confidential and intended solely for the addressee(s). Any unauthorized review, use, disclosure or distribution is prohibited. If you believe this message has been sent to you in error, please notify the sender by replying to this transmission and delete the message without disclosing it. Thank you.

E-mail including attachments is susceptible to data corruption, interruption, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any such corruption, interception, amendment, tampering or viruses or any consequences thereof.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

> how does one distinguish between a micro kernel (like l4) and the exokernel?

The  µkernel has abstractions.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[lucio]

> lucio@proxima.alt.za wrote:
>> And to
>> suggest that performance is a significant factor in the design of a
>> generic operating system is laughable.
> 
> Not entirely:  If an OS design *precludes* efficient
> operation on any architecture, surely that is bad design.

OK, so I have to be more careful in the way I phrase things.  Let's say that performance isn't going to be the most significant criterion and that it would take intent to "preclude" efficient operation on any architecture, specially when one might be speculating about the architecture in the first place.

Plan 9 and NetBSD are sufficient examples that portable design is unlikely to be intentionally inefficient.

++L

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Dick Davies]

David Tolpin wrote:

> Java is done right in concept.

wtf?

[ Let me start by saying I don't speak for any of these guys,
I can't even get their OS to run a GUI.]


Java is better than C++, but as others have mentioned that's like saying
toothache is better than herpes.

It's just C without pointer arithmetic.  Leaving aside actual 
implementation bugs, it :

* runs on less platforms than any other language I've ever used,
* wimped out of having numeric types implemented as objects, then
bolted on 'containers' for them in 1.1
*has changed its GUI twice and *still* doesn't do it right,
* it's thread model is essential to its operation but differs widely 
between VMs even on the same platform
* the same it true for its garbage collector.
* The APIs are huge and inconsistent.
* it *finally* - in 1.4 - has a regex API (which is unusably complex).
* it's native code bindings are absolutely the worst I've ever seen,
even Perl does a better job

And don't forget m5 current personal favourite -
  its pointlessly restrictive type system.

Type is stored in the *pointer*, not the object.
So I have to hold the VMs hand each time I pass an argument.
I'm not allowed to reuse a variable to
point to an List and a Vector, even if I'm going to call a method
with the same name on both.

And then to really piss in my chips, 90% of the Collections API
which lets me store these objects and actually *do* something with them
only lets me pull them out as Object.

I HAVE TO CAST THEM. So when I get an object 'dynamically' - via RMI or
out of a Vector or whatever, I have to know what it is in advance.
The poor frigging object doesn't know (no, don't get me started on the 
abomination that is the reflection API, people are already starting to 
stare).

And this is a dynamic language. KERR-RIST.

None of this matters of course, because "it's better than C++".
And IT managers think the Sun shines out of its ass.

Java is the f*cking COBOL of the 90s and
future generations of geeks are going to fly back from Mars to piss on 
our graves for inflicting it on them.

> Unfortunately. I can do system programming and command-line
> apps with Java. 

Sure. If you don't mind a fifteen second delay while the VM drags itself 
up from the bottom of the sea each time you run a command line app.

Or should that '.' be a ',' ?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

i'll answer wrecklessly, 'cause i haven't read the zillions of
e-mails after this one.  java sux.  it's syntax and semantics
are abominable because it was not designed ... it is bundle
of failed sun projects that sold a lot of books.

limbo?  i could write a 20 page ref manual, which i think was
the requirement for modula-2, and let someone else spazz
it out to a 500 page doc.

good evening,

brucee

> In particular, what part of limbo in reality is better than Java
> in such a way that reminded you bull's excrements?
> 
> David Tolpin

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Tue, 2 Mar 2004, Bengt Kleberg wrote:

> i thought that the idea with an micro kernel is that it makes it 
> _possible_ to tailor the application interface towards many different 
> applications. a monolithic kernel has only one interface.

no, I recall Rashid saying this about Mach 3.0: microkernels are not
small, they just don't do much. This in response to the complaints that
this low-capability "micro kernel" was about 3 Mbytes in size. I wish I
had saved the exact email :-)

you get way more ability to tailor etc. with Plan 9 than I was every able 
to figure out with Mach.

ron
p.s. why do all your short emails have a 20-line disclaimer?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> And to think I just ate. 

fortunately, i was sleeping.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> > Java is done right in concept.
> 
> wtf?

the story goes that Sun had a group developing that junk
with no plan and then the web came along and they
bolted it on.  ref:  the project group themselves.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> good evening,

yeah it was.  brucee played worms and i coded a pop3
client over a few beers.  after being beaten into doing
it and then having done it, i was sold.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Andrew Simmons]

>object-oriented design is the roman numerals of computing.

Would you care to expand on that?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

>object-oriented design is the roman numerals of computing.
> 
> Would you care to expand on that?

should be made a fortune.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Kenji Okamoto]

> And, that said, I'm still gradually moving my existence over to Plan 9. 

Because some people loves art over everyday life.

Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Martin C.Atkins]

On Mon, 1 Mar 2004 23:15:58 +0400 (AMT) David Tolpin <dvd@davidashen.net> wrote:
>...
> No. It is because the spec is written by people who come from culture of
> detailed specifications; and it is because a part of the specification is
> dedicated to binary compatibility, which few other languages provide.  Still
> yet, the specification for Java, 504 pages, is SMALLER than specification for
> ANSI/ISO C, which 554 pages.

No. The Revised^5 report on Scheme is only 50 pages - and that
includes a full denotation semantics for the language - almost by
definition, nothing can be more "detailed" than that!

OK, so maybe Scheme is a "smaller" language - whatever that means -
but not 10 times smaller (and perhaps being smaller is a good thing,
not a bad thing!).

Martin

-- 
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Kenji Okamoto]

>>object-oriented design is the roman numerals of computing.
> 
> Would you care to expand on that?

Just from my guess, it looks fancier to someone, however, it solves no
essencial problem. ☺

Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> Just from my guess, it looks fancier to someone, however, it solves no
> essencial problem. ☺

roman numerals were all very fine for writing down numbers, but were
a nightmare to do mathematics with. this was due to their awful
representation.

brucee did once add roman numbers to csh.  you could:

    set roman

and it would spit roman numerals where it thought it needed
to spit numbers.  iirc, history was one example.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Anastasopoulos S]

On Tue, 2 Mar 2004, boyd, rounin wrote:

> >object-oriented design is the roman numerals of computing.
> > 
> > Would you care to expand on that?
> 
> should be made a fortune.
> 
> 

There is an post post from Rob Pike on OO.

Search groups.google.com for
	Rob Pike The emperor's new O-O O.S

Spyros

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[9nut]

>>object-oriented design is the roman numerals of computing.

Definitely one for the fortune file.

> Would you care to expand on that?

You have something against tersely eloquent, and expressive
statements? ☺

How I interpret it, and what my experience tells me is that OO doesn't
scale up (or down) and is the wrong model for a lot of things.  Java's
issues are not just with the language.  It is also the fact that it
insists on (ordained to) representing EVERYTHING in an object
abstraction.  Java suffers from the modern illusion that if a little
of something is good, more of it must be better.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bengt Kleberg]

ron minnich wrote:
> On Tue, 2 Mar 2004, Bengt Kleberg wrote:
> 
> 
>>i thought that the idea with an micro kernel is that it makes it 
>>_possible_ to tailor the application interface towards many different 
>>applications. a monolithic kernel has only one interface.
> 
> 
> no, I recall Rashid saying this about Mach 3.0: microkernels are not
> small, they just don't do much. This in response to the complaints that
> this low-capability "micro kernel" was about 3 Mbytes in size. I wish I
> had saved the exact email :-)

so different micro kernels have different ideas behind them. i like 
speed, and therefore prefer the ones that have faster applications as 
their idea.


> you get way more ability to tailor etc. with Plan 9 than I was every able 
> to figure out with Mach.

please consider using a modern micro kernel as example of a micro kernel.
you would not want me to use multics as the one true monolithic kernel, 
would you :-)


> ron
> p.s. why do all your short emails have a 20-line disclaimer?

the disclaimer is added by the mail server for outgoing email.


bengt

This communication is confidential and intended solely for the addressee(s). Any unauthorized review, use, disclosure or distribution is prohibited. If you believe this message has been sent to you in error, please notify the sender by replying to this transmission and delete the message without disclosing it. Thank you.

E-mail including attachments is susceptible to data corruption, interruption, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any such corruption, interception, amendment, tampering or viruses or any consequences thereof.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

wise man

----- Original Message ----- 
From: "Kenji Okamoto" <okamoto@granite.cias.osakafu-u.ac.jp>
To: <9fans@cse.psu.edu>
Sent: Wednesday, March 03, 2004 4:11 PM
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel


> >>object-oriented design is the roman numerals of computing.
> >
> > Would you care to expand on that?
>
> Just from my guess, it looks fancier to someone, however, it solves no
> essencial problem. ☺
>
> Kenji

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

actually ir was an april fools thing.

if you set english ... then you would get stylish pids

Four hundred and fourteen.

tho i think some of the 'vi' messages were funnier.

must have been the nine schooners.

brucee
----- Original Message ----- 
From: "boyd, rounin" <boyd@insultant.net>
To: <9fans@cse.psu.edu>
Sent: Wednesday, March 03, 2004 4:26 PM
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel


> > Just from my guess, it looks fancier to someone, however, it solves no
> > essencial problem. ☺
>
> roman numerals were all very fine for writing down numbers, but were
> a nightmare to do mathematics with. this was due to their awful
> representation.
>
> brucee did once add roman numbers to csh.  you could:
>
>     set roman
>
> and it would spit roman numerals where it thought it needed
> to spit numbers.  iirc, history was one example.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> must have been the nine schooners.

yeah, probably those IX beverages.

    careful man, there's a beverage here -- the dude

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Wed, 3 Mar 2004, Bengt Kleberg wrote:

> please consider using a modern micro kernel as example of a micro
> kernel. you would not want me to use multics as the one true monolithic
> kernel, would you :-)

example? I have not seen that much fundamental change in the idea. I'll 
pick a modern one if you want.

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 4 bytes --]

L4

[-- Attachment #2: Type: message/rfc822, Size: 2621 bytes --]

From: ron minnich <rminnich@lanl.gov>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Wed, 3 Mar 2004 05:59:30 -0700 (MST)
Message-ID: <Pine.LNX.4.44.0403030558220.24778-100000@maxroach.lanl.gov>

On Wed, 3 Mar 2004, Bengt Kleberg wrote:

> please consider using a modern micro kernel as example of a micro
> kernel. you would not want me to use multics as the one true monolithic
> kernel, would you :-)

example? I have not seen that much fundamental change in the idea. I'll 
pick a modern one if you want.

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Wed, 3 Mar 2004, Fco.J.Ballesteros wrote:

> L4

thanks.

Looks interesting, although I note that what people run on top of it in 
one case is ... Linux.

But the single address space os thing looks interesting. Nice to see one 
of them again (Data General AOS/VS did this in the early 80s).

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[rog]

this paper goes well with the benchmarking paper referred to earlier,
i think:

http://i30www.ira.uka.de/research/documents/l4ka/irreproducible-benchmarks.pdf

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[a]

L7?

"You made my..."
ア

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Martin C.Atkins]

On Tue, 02 Mar 2004 12:19:24 +0000 Dick Davies <rasputnik@hellooperator.net> wrote:
>...
> And then to really piss in my chips, 90% of the Collections API
> which lets me store these objects and actually *do* something with them
> only lets me pull them out as Object.
> 
> I HAVE TO CAST THEM. So when I get an object 'dynamically' - via RMI or
>...

C# seems to have duplicated the same mistake, although at least *they*
seem to realise it was a mistake, and intend to add 'generics'. It's a
pity that means they have to migrate all the existing code...

(Java generic types are not, AFAIK, official, in any sense - will they ever be?)

BTW: does anyone know if C# duplicated all Java's mistakes in multi-treading,
as summarised in Roj's excerpts of Holub's book?

Martin

-- 
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bruce Ellis]

not all of them, but they threw in a few more.

----- Original Message ----- 
From: "Martin C.Atkins" <martin@parvat.com>
To: <9fans@cse.psu.edu>
Sent: Thursday, March 04, 2004 2:52 PM
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel


...

> BTW: does anyone know if C# duplicated all Java's mistakes in multi-treading,
> as summarised in Roj's excerpts of Holub's book?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Yi Li]

"ron minnich" <rminnich@lanl.gov> wrote in message
news:Pine.LNX.4.44.0403030620180.24778-100000@maxroach.lanl.gov...
> On Wed, 3 Mar 2004, Fco.J.Ballesteros wrote:
>
> > L4
>
> thanks.
>
> Looks interesting, although I note that what people run on top of it in
> one case is ... Linux.
>

And I suspect that the design philosophy of L4 has a certain
degree of impact on Linux. No proof, just a guess.

E

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 290 bytes --]

I don't think so. The claim of L4 is that µkernels are non-portable
(which makes sense) and that they must be coded in a non-portable way,
The orignial one was assembly, and then fiasco is a reimplementation of L4
in C (mostly).

Linux is UNIX, mostly, if we forget about its bloat.

[-- Attachment #2: Type: message/rfc822, Size: 2336 bytes --]

From: Yi Li <vangoh12@singnet.com.sg>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel
Date: Thu, 4 Mar 2004 10:00:09 GMT
Message-ID: <c266s1$2g9$1@mawar.singnet.com.sg>


"ron minnich" <rminnich@lanl.gov> wrote in message
news:Pine.LNX.4.44.0403030620180.24778-100000@maxroach.lanl.gov...
> On Wed, 3 Mar 2004, Fco.J.Ballesteros wrote:
>
> > L4
>
> thanks.
>
> Looks interesting, although I note that what people run on top of it in
> one case is ... Linux.
>

And I suspect that the design philosophy of L4 has a certain
degree of impact on Linux. No proof, just a guess.

E

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Yi Li]

> And I suspect that the design philosophy of L4 has a certain
> degree of impact on Linux. No proof, just a guess.

On Thu, 04 Mar 2004 11:23:44 +0000, Fco.J.Ballesteros wrote:

>> I don't think so. The claim of L4 is that µkernels are non-portable
>> (which makes sense) and that they must be coded in a non-portable way,
>> The orignial one was assembly, and then fiasco is a reimplementation of L4
>> in C (mostly).
>> 
>> Linux is UNIX, mostly, if we forget about its bloat.

I understand that. Having impact does not necessarily mean to agree
or even to be similar. What I meant was that from Linus's posting
here I got the impression that he might have read those L4 papers.

Best,

E

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Andrew Simmons]

>You have something against tersely eloquent, and expressive
>statements? 

Yes. I'm too stupid to understand them. I couldn't even understand what that
blind guy in "Kung Fu" was talking about.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Bengt Kleberg]

ron minnich wrote:
> On Tue, 2 Mar 2004, Martin C.Atkins wrote:
> 
...deleted
>>back, but don't know what happened to his work. (Sure enough, it
>>does seem to do something in 2.4.24 - is this a "best kept secret"?
>>If it works, why aren't we all using it?)
> 
> 
> It works. I don't know why we're not all using it.

the same reason that so few people switched to plan9 when so many 
switched to linux.
resistance to change.


bengt

This communication is confidential and intended solely for the addressee(s). Any unauthorized review, use, disclosure or distribution is prohibited. If you believe this message has been sent to you in error, please notify the sender by replying to this transmission and delete the message without disclosing it. Thank you.

E-mail including attachments is susceptible to data corruption, interruption, unauthorized amendment, tampering and viruses, and we only send and receive e-mails on the basis that we are not liable for any such corruption, interception, amendment, tampering or viruses or any consequences thereof.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[David Tolpin]

> > how can I make my text windows in rio show indents in increments of 8
> > for tabs, and not of 4?
> > 
> 	% tabstops=4

In acme and samterm, it is tabstop, not tabstops.

> is all sam needs to adjust to 4 spaces to the tab.  



> Acme does it by
> default, but it also obeys $tabstops.  Learn acme in preference to
> sam, it will astonish you.  

Yes, I know how to use acme for editing, and had learned it 
before acme had been written. In 1991, I think.

> I keep using sam by habit, and it seems to
> me to be a bad habit.

I don't think sam is a bad habit. It is a different editor, less
integrated into Plan9/rio, but I use it not just in Plan 9.

The question, however, was not how to do 4 spaces per tab in sam.
It was the opposite. I was asking how to make window show 8 spaces
per tab.

I've looked into the sources. It turns out that I cannot do it for
one window, I have to set tabstop and run rio. It can be obvious for
you, but it was not for me, untill I looked at what /bin/window is.

The code in acme.c appears to be copy'n'pasted from the same source
as code in rio.c, with independent default value for maxtab. 
Why it is needed? Is acme supposed to run without rio?

David Tolpin

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[ron minnich]

On Tue, 2 Mar 2004, Martin C.Atkins wrote:

> Does this mean "namespace" in the Plan 9 sense? Is it implemented
> yet, or was this file written in the "future optimistic" tense? :-)

in many ways. We exploit this new stuff for v9fs. It has allowed us to 
have private name space mounts on the 1024 node cluster. It's pretty nice.

> I heard that Al Viro was doing something in this area some time
> back, but don't know what happened to his work. (Sure enough, it
> does seem to do something in 2.4.24 - is this a "best kept secret"?
> If it works, why aren't we all using it?)

It works. I don't know why we're not all using it.

ron

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Martin C.Atkins]

On Fri, 27 Feb 2004 01:30:57 -0800 (PST) Linus Torvalds <torvalds@osdl.org> wrote:
>...
> The "clone()" system call in Linux just creates a new such "context of 
>...
> The bits you can control the context copy with are:
>...
>   CLONE_NEWNS     	/* New namespace group? */
>...

Does this mean "namespace" in the Plan 9 sense? Is it implemented
yet, or was this file written in the "future optimistic" tense? :-)

BTW: the definition isn't listed in my man page for clone(2), or
my Debian woody header files, however, see below...

I heard that Al Viro was doing something in this area some time
back, but don't know what happened to his work. (Sure enough, it
does seem to do something in 2.4.24 - is this a "best kept secret"?
If it works, why aren't we all using it?)

Martin

-- 
Martin C. Atkins			martin@parvat.com
Parvat Infotech Private Limited		http://www.parvat.com{/,/martin}

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> But when I speak it, I realize when I'm full of shit.

you said it.

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, C H Forsyth wrote:
>
> my impression, and it's only that, is that relatively few user mode
> applications use clone-for-threads at all on Linux, as yet.

No, there's tons of them. But they almost always use the pthreads
interfaces. Few applications use "clone()" directly, since there is seldom
any point to going to that low level.

So libc does a mapping (pretty trivial these days thanks to the new signal
interfaces) of pthreads to clone() and other low-level kernel
functionality.

There's a lot more to threads than just creating them, btw. Linux has a
whole "futex" infrastructure for doing fast user-level mutual exclusion
(which in a pthreads world get mapped to pthread_mutex calls etc). Again,
the kernel interfaces are _not_ the pthreads interfaces (since it would be
crazy to call into the kernel just for a simple lock - 99.99% of the time
you can do it with a simple atomic sequence in user space). But they are
there to support _any_ kind of thread synchronization, whether pthreads-
based or anything else.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Fco. J. Ballesteros wrote:
>
> > Hint: you can't message-pass a hash table that describes 200 megabytes
> > worth of filesystem names.
> >
> > Welcome to the real world, Neo.
>
> You can. You can pass a pointer to it.

Absolutely. But don't call it "message passing" then, because it damn well
isn't.

Sure, "message passing" is a wonderful idea if you are allowed to redefine
it to say "pass pointers around". In fact, let's go one step further, and
just call a function call "message passing", since what it does is to make
a message out of the arguments, and then let the CPU do a "message pass"
operation to the target.

That proves that "message passing" is as efficient as traditional code,
while still having the obvious advantage of being about "messages", which
we all know are much better than those horrible "function calls".

I can speak the newspeak as well as anybody else.

But when I speak it, I realize when I'm full of shit.

		Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Fco.J.Ballesteros]

>> what?  only ~20.  surely you need some more?
>
> So far, no. It's been growing over the years, but not quickly.

I think boyd was kidding, wasn't him?

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, boyd, rounin wrote:
>
> > The bits you can control the context copy with are:
>
> what?  only ~20.  surely you need some more?

So far, no. It's been growing over the years, but not quickly.

Realize that you don't want to even control this with a very fine
granularity. Every single new thing that you allow being copied or shared
ends up being more state that you have to reference-count and keep track
of. You want to have as little as possible of that kind of state.

So keep them to big fundamental things. The VM. The file table. The signal
state. The namespace. (And those f*ing horrible SysV IPC things).

Splitting it up more would just hurt. Just out of interest, what else
could you possibly want, and why?

After all, it's not like the kernel keeps track of all that much else than
VM, files and signals.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Geoff Collyer wrote:
>
> I think we're talking past each other due to different terminologies.
> Linus seems to use `thread' to mean `a process sharing address space
> (other than normal text segment sharing)', whereas in Plan 9, that's
> just a process; some share address space, some don't.  A Plan 9 thread
> is entirely a user-mode creation of the Plan 9 thread library, which
> doesn't implement POSIX threads.

Well, what Linux has is really what I privately call a "context of
execution".

The "clone()" system call in Linux just creates a new such "context of
execution", and you can choose to arbitrarily share pretty much any OS
state, by just saying which state you want to share in a bitmap. In
addition to the bitmap there are a few pointers you pass around, the full
required state is actually

	clone_flags: bitmap of how to create the new context of execution
	newsp: stack pointer of new context
	parent tidptr: pointer (in the parent) to the thread ID information
	child tidptr: pointer (in the child) to the thread ID information
	tls pointer: pointer to TLS (thread-local-storage) for the context

but not all of them are necessarily used (ie if you don't want to set TLS
or TID information, those pointers are obviously unused).

The bits you can control the context copy with are:

  CSIGNAL         	/* signal mask to be sent at exit */

  CLONE_VM        	/* set if VM shared between processes */
  CLONE_FS        	/* set if fs info shared between processes */
  CLONE_FILES     	/* set if open files shared between processes */
  CLONE_SIGHAND   	/* set if signal handlers and blocked signals shared */
  CLONE_IDLETASK  	/* set if new pid should be 0 (kernel only)*/
  CLONE_PTRACE    	/* set if we want to let tracing continue on the child too */
  CLONE_VFORK     	/* set if the parent wants the child to wake it up on mm_release */
  CLONE_PARENT    	/* set if we want to have the same parent as the cloner */
  CLONE_THREAD    	/* Same thread group? */
  CLONE_NEWNS     	/* New namespace group? */
  CLONE_SYSVSEM   	/* share system V SEM_UNDO semantics */
  CLONE_SETTLS    	/* create a new TLS for the child */
  CLONE_PARENT_SETTID   /* set the TID in the parent */
  CLONE_CHILD_CLEARTID  /* clear the TID in the child */
  CLONE_DETACHED        /* Unused, ignored */
  CLONE_UNTRACED        /* set if the tracing process can't force CLONE_PTRACE on this clone */
  CLONE_CHILD_SETTID    /* set the TID in the child */
  CLONE_STOPPED         /* Start in stopped state */

(CSIGNAL isn't a bit - it's the low 8 bits, and it specifies the signal
you want to send to your parent when you die).

So a "fork()" is literally really just a "clone(SIGCHLD)". We're saying
that we don't want to share anything, and that we want to send a SIGCHLD
at exit.

Setting the CLONE_VM bit says that the VM gets shared. That means that
instead of copying the page tables, we just copy the pointer to the
"struct mm_struct", which describes everything in the VM, and we increment
its reference count.

There is no "partial copy". If you say that you want to share the VM, you
get the WHOLE VM. Or you will get a totally private VM. Similarly, i fyou
say that you want to share the file descriptors (CLONE_FILES), they will
all be shared: one context doing an "open()" will have that fd be valid in
all other contexts that share it.

(The difference between CLONE_FILES and a regular fork() is that a
CLONE_FILES will increment just _one_ reference count: the reference count
for the whole array of pointers to files. In contrast, a fork-like
non-shared case will create a whole new array of pointers to files, and
then for each file increment the pointer for that file).

What most "unix people" call threads is somethign that is created with
pretty much all flags set - we share pretty much everything except for the
register state and the kernel stack between the contexts. And when I say
"share", I really mean share: most of the bits end up being copying a
kernel pointer and incrementing the reference count for that object.

Some of the bits are "administrative": the VFORK bit isn't about sharing,
it's about the parent waiting until the child releases the VM back to it
(btw, that uses a "completion" structure on the parents stack). Similarly,
the SETTID/CLEARTID bits are about writing the TID ("thread ID" as opposed
to "process ID") to the VM space atomically with the creation (or in the
case of CLEARTID, teardown) of the thread. That ends up helping the thread
management (from user space) a _lot_.

(Tangential to this discussion is the TLS or "thread local storage" bit -
some architecture-specific way of indicating a small thread-specific
storage area. It's not the stack, it's just a regular allocation, and
different architectures have different ways of pointing to it. Usually
there's some architected register set aside for it).

And you can mix and match things. You can literally create a new context
that shares the file descriptors (so that one process doing an "open()"
will open files in the other one), but doesn't share the VM space.

Although some of them are interdependent - CLONE_THREAD (which is really
just "all signal state" despite the name - it has nothing to do with VM
per se) depends on CLONE_SIGHAND (which is just the set of signal handlers
associated with the context), which in turn depends on CLONE_VM (because
it doesn't make sense to be able to take a signal in different contexts
unless they share the same VM).

This has gotten fairly far off the notion of stacks and VM.. But I hope
it's clear to everybody that I heartily agree with rfork-like
functionality. It's just segmented/private stacks I can't understand.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[Linus Torvalds]

On Fri, 27 Feb 2004, Rob Pike wrote:
>
> > So in a C/UNIX-like environment, private stacks are wrong. You could
> > imagine _other_ environments where they might be valid, but even those
> > other environments would not invalidate my points about efficiency and
> > simplicity.
>
> as i said before, the stacks are not private.  you're right, that's a
> bad thing.
> that's why they're not private.
>
> the segment called 'stack' is private, but that's a different thing.
> i stress: stack != stack segment.  stack is where your sp is; stack
> segment is a figment of the VM system.

Well, in another email I already said that "private stack" and "segments"
are really exactly the same thing - some people think of segments as
paging things, others think of them in the x86 sense, but in the end it
all comes down to the fact that a "stack address" ends up making sense
only within a specific context (and that context can sometimes be
partially visible to other threads by using explicit segment registers or
other magic, like special instructions that can take another address
space).

And private/segmented stacks are bad.

They are bad exactly because they magically make automatic variables
fundamentally different from other variables.  And they really have no
reason they should be different.

There is absolutely nothing wrong with having a thread take the address of
some automatic variable, and then just pass that address off to another
routine. And if that other routine decides that it is going to create a
hundred threads to solve the problem that the variable described in
parallell, then that should JUST WORK. Anything else would be EVIL.

Having a pointer that sometimes works, and sometimes doesn't, based on who
uses it - that's just crazy talk.

> i ask again: how does linux create per-thread storage?

The same way it creates any other storage: with mmap() and brk(). You just
malloc the thing, and you pass in the new stack as an argument to the
thread creation mechanism (which linux calls "clone()", just to be
different).

And because that storage is just storage, things like the one I described
above "just work". If you pass another thread a pointer to your stack, the
other thread can happily manipulate it, and never even needs to know that
it's an automatic variable somewhere else.

And sure, you can shoot yourself in the foot that way. You can pass off a
pointer to another thread, and then return from the function without
synchronizing with that other thread properly, and now the other thread
will scribble all over your stack. But that's really nothing different
than using "alloca()" and passing off that to something that remembers the
address.

> the way the plan 9 thread library works is so different from linux's
> that they're hard to compare.  program design in the two worlds is
> radically different.  so your claim of 'better' is curious to me.  by
> 'better' you seem to mean 'faster' and 'cleaner'.  faster at least can
> be measured.

To me, the final decision on "better" tends to be a fairly wide issue.
Performance is part of it - especially infrastructure that everybody
depends on should always strive to at least _allow_ good performance, even
if not everybody ends up caring.

But the concept, to me, is more important. Basically, I do not see how you
can really have a portable and even _remotely_ efficient partial VM
sharing. And if you can't have it, then you shouldn't design the
interfaces around it.

> you speak with certainty.  have you seen performance comparisons? i
> haven't, although it wouldn't surprise me to learn that there are useful
> programs for which linux outperforms plan 9, and vice versa of course.

When it comes to threads, I only see three interesting performance
metrics: how fast can you create them, how fast can you synchronize them
(both "join" and locking), and how well do you switch between them.

The locking is pretty much OS-independent, since fast locking has to be
done in user space anyway (with just the contention case falling back to
the OS, and if your app cares about performance it hopefully won't have
much contention).

So we're left with create, tear-down and switch. All of which are
_fundamentally_ faster if you just have a "share everything" model.
Create and tear-down are just increment/decrement a reference counter
(there's a spinlock involved too). Task switch is a no-op from a VM
standpoint (except we have a per-thread lazy TLB invalidate that will
trigger).

In contrast, partial sharing is a major pain. You definitely don't just do
a reference count increment for your VM.

			Linus

Re: [9fans] Re: Threads: Sewing badges of honor onto a Kernel

[boyd, rounin]

> The bits you can control the context copy with are:

what?  only ~20.  surely you need some more?