Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1479520 > unrolled thread

Re: [RFC/PATCH] posix-timers: make them configurable

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [RFC/PATCH] posix-timers: make them configurable John Stultz <john.stultz@linaro.org> - 2016-09-08 23:20 +0200
    Re: [RFC/PATCH] posix-timers: make them configurable Josh Triplett <josh@joshtriplett.org> - 2016-09-08 23:50 +0200
    Re: [RFC/PATCH] posix-timers: make them configurable Nicolas Pitre <nicolas.pitre@linaro.org> - 2016-09-09 02:00 +0200
      Re: [RFC/PATCH] posix-timers: make them configurable Josh Triplett <josh@joshtriplett.org> - 2016-09-09 02:40 +0200
    Re: [RFC/PATCH] posix-timers: make them configurable Richard Cochran <richardcochran@gmail.com> - 2016-09-09 09:50 +0200
      Re: [RFC/PATCH] posix-timers: make them configurable Josh Triplett <josh@joshtriplett.org> - 2016-09-09 10:10 +0200
        Re: [RFC/PATCH] posix-timers: make them configurable Thomas Gleixner <tglx@linutronix.de> - 2016-09-09 16:00 +0200
      Re: [RFC/PATCH] posix-timers: make them configurable Thomas Gleixner <tglx@linutronix.de> - 2016-09-09 15:50 +0200

#1479520 — Re: [RFC/PATCH] posix-timers: make them configurable

On Thu, Sep 8, 2016 at 2:05 PM, Nicolas Pitre <nico@fluxnic.net> wrote:
>
> Small embedded systems typically don't need them.  This removes about
> 16KB from the kernel binary size on ARM when configured out.
> Corresponding syscalls are routed to a stub logging the attempt to
> use those syscalls which should be enough of a clue if they were
> disabled without proper consideration.
>
> Signed-off-by: Nicolas Pitre <nico@linaro.org>

Hrm... So being able to trim down the kernel is important.

Although my sense is that momentum has been moving to clock_gettime()
over gettimeofday(), such that gettimeofday() is mostly a shim over
clock_gettime() logic wise. So this is sort of going the other
direction.

Also given many other syscalls take clockids and the backing logic
isn't really getting removed (probably could cut the dynamic posix
clocks core with the same conditional), I wonder if you could get a
similar size win by taking a slightly more narrow cutting of the
subsystem. That way you could preserve the more useful clock_gettime()
functionality, but maybe stub out some of the less often used
functionality.

Josh (cc'ed) also was talking awhile back about cutting out the core
NTP logic. Having a single minimal-time option might be nice rather
then having a bunch of different conditionals that might be combined.

thanks
-john

[toc] | [next] | [standalone]

#1479539

On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> On Thu, Sep 8, 2016 at 2:05 PM, Nicolas Pitre <nico@fluxnic.net> wrote:
> > Small embedded systems typically don't need them.  This removes about
> > 16KB from the kernel binary size on ARM when configured out.
> > Corresponding syscalls are routed to a stub logging the attempt to
> > use those syscalls which should be enough of a clue if they were
> > disabled without proper consideration.
> >
> > Signed-off-by: Nicolas Pitre <nico@linaro.org>
> 
> Hrm... So being able to trim down the kernel is important.
> 
> Although my sense is that momentum has been moving to clock_gettime()
> over gettimeofday(), such that gettimeofday() is mostly a shim over
> clock_gettime() logic wise. So this is sort of going the other
> direction.
> 
> Also given many other syscalls take clockids and the backing logic
> isn't really getting removed (probably could cut the dynamic posix
> clocks core with the same conditional), I wonder if you could get a
> similar size win by taking a slightly more narrow cutting of the
> subsystem. That way you could preserve the more useful clock_gettime()
> functionality, but maybe stub out some of the less often used
> functionality.

I agree with this.  Cutting out the syscall alone helps, but cutting out
the corresponding infrastructure for those timers would help even more.
I wouldn't suggest turning down this patch in isolation, but I'd very
much like to see it become a patch series that also removes the
underlying timer infrastructure.

> Josh (cc'ed) also was talking awhile back about cutting out the core
> NTP logic. Having a single minimal-time option might be nice rather
> then having a bunch of different conditionals that might be combined.

I do think having individual configuration options for families of
syscalls helps, in that people can more easily figure out the set of
syscalls their code calls.  But those options should also cut out the
underlying infrastructure whenever possible; that avoids the need to
have separate options for the infrastructure, which a user would have to
enable to see the options for the interfaces to that infrastructure.  If
it can't get called, it doesn't need compiling in.

If that infrastructure supports multiple families of syscalls that make
sense to drop independently, then it might make sense to have that
underlying option, ideally automatically selected.  For instance, a
legacy-free userspace might only enable signalfd and not traditional
signal delivery, or only enable timerfd and not enable other forms of
timers.

(Hopefully the kconfig-sat folks can successfully develop a system that
allows you to turn on an option whose dependencies aren't yet enabled
and automatically enable its dependencies, to improve the functionality
of "select".)

For this patch, I'd recommend expanding the documentation for the
Kconfig symbol to explicitly state the families of syscalls this omits.

I'd also suggest dropping the stubs that print a message, and just using
sys_ni_syscall.  As a debug mechanism, that infrastructure seems more
generally useful than just POSIX timer syscalls.  Instead, I'd suggest a
separate patch adding (optional, CONFIG_SHOW_MISSING_SYSCALLS) support
in sys_ni.c to print a one-time message per syscall showing the process
name, PID, syscall number, name, and Kconfig symbol needed to enable it.
You could modify cond_syscall to accept a Kconfig symbol argument, and
generate a unique stub that calls printk_once.

This infrastructure would itself take up space, and wouldn't make any
sense without CONFIG_PRINTK, hence making it optional.  But as a
debugging mechanism when working on shrinking a kernel for a target
application, it seems quite useful.  (The only caveat: a message
wouldn't always indicate a problem, since an application may handle
ENOSYS.)

[toc] | [prev] | [next] | [standalone]

#1479581

On Thu, 8 Sep 2016, Josh Triplett wrote:

> On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> > On Thu, Sep 8, 2016 at 2:05 PM, Nicolas Pitre <nico@fluxnic.net> wrote:
> > > Small embedded systems typically don't need them.  This removes about
> > > 16KB from the kernel binary size on ARM when configured out.
> > > Corresponding syscalls are routed to a stub logging the attempt to
> > > use those syscalls which should be enough of a clue if they were
> > > disabled without proper consideration.
> > >
> > > Signed-off-by: Nicolas Pitre <nico@linaro.org>
> > 
> > Hrm... So being able to trim down the kernel is important.
> > 
> > Although my sense is that momentum has been moving to clock_gettime()
> > over gettimeofday(), such that gettimeofday() is mostly a shim over
> > clock_gettime() logic wise. So this is sort of going the other
> > direction.
> > 
> > Also given many other syscalls take clockids and the backing logic
> > isn't really getting removed (probably could cut the dynamic posix
> > clocks core with the same conditional), I wonder if you could get a
> > similar size win by taking a slightly more narrow cutting of the
> > subsystem. That way you could preserve the more useful clock_gettime()
> > functionality, but maybe stub out some of the less often used
> > functionality.
> 
> I agree with this.  Cutting out the syscall alone helps, but cutting out
> the corresponding infrastructure for those timers would help even more.
> I wouldn't suggest turning down this patch in isolation, but I'd very
> much like to see it become a patch series that also removes the
> underlying timer infrastructure.

The patch does remove more than only syscalls which are in 
posix-timers.c. The whole of posix-timers.c, posix-clock.c and 
posix-cpu-timers.c are removed. The later is particularly footprint 
heavy for one specialized clock type.

> > Josh (cc'ed) also was talking awhile back about cutting out the core
> > NTP logic. Having a single minimal-time option might be nice rather
> > then having a bunch of different conditionals that might be combined.
> 
> I do think having individual configuration options for families of
> syscalls helps, in that people can more easily figure out the set of
> syscalls their code calls.  But those options should also cut out the
> underlying infrastructure whenever possible; that avoids the need to
> have separate options for the infrastructure, which a user would have to
> enable to see the options for the interfaces to that infrastructure.  If
> it can't get called, it doesn't need compiling in.

Absolutely.  However this is something that would scale better with 
automatic code pruning performed by LTO or ld with -gc-sections.

> If that infrastructure supports multiple families of syscalls that make
> sense to drop independently, then it might make sense to have that
> underlying option, ideally automatically selected.  For instance, a
> legacy-free userspace might only enable signalfd and not traditional
> signal delivery, or only enable timerfd and not enable other forms of
> timers.

One problem is that embedded libc implementations often go with the 
legacy interface as it is often simpler and smaller.

> (Hopefully the kconfig-sat folks can successfully develop a system that
> allows you to turn on an option whose dependencies aren't yet enabled
> and automatically enable its dependencies, to improve the functionality
> of "select".)
> 
> For this patch, I'd recommend expanding the documentation for the
> Kconfig symbol to explicitly state the families of syscalls this omits.

Good point.

> I'd also suggest dropping the stubs that print a message, and just using
> sys_ni_syscall.  As a debug mechanism, that infrastructure seems more
> generally useful than just POSIX timer syscalls.  Instead, I'd suggest a
> separate patch adding (optional, CONFIG_SHOW_MISSING_SYSCALLS) support
> in sys_ni.c to print a one-time message per syscall showing the process
> name, PID, syscall number, name, and Kconfig symbol needed to enable it.
> You could modify cond_syscall to accept a Kconfig symbol argument, and
> generate a unique stub that calls printk_once.

Yes, this would be a good thing to have.  The ptrace infrastructure 
might be leveraged here simply to get the actual syscall number, etc.  

However I wouldn't start on that unless there is actually some more 
configurable syscalls making the rationalization worth it.

Here I wanted to have a minimal printout in case someone decided to turn 
off POSIX timers and wondered why his big desktop distro doesn't boot 
anymore.  Those syscalls have never been optional before.


Nicolas

[toc] | [prev] | [next] | [standalone]

#1479592

On Thu, Sep 08, 2016 at 07:56:28PM -0400, Nicolas Pitre wrote:
> On Thu, 8 Sep 2016, Josh Triplett wrote:
> 
> > On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> > > On Thu, Sep 8, 2016 at 2:05 PM, Nicolas Pitre <nico@fluxnic.net> wrote:
> > > > Small embedded systems typically don't need them.  This removes about
> > > > 16KB from the kernel binary size on ARM when configured out.
> > > > Corresponding syscalls are routed to a stub logging the attempt to
> > > > use those syscalls which should be enough of a clue if they were
> > > > disabled without proper consideration.
> > > >
> > > > Signed-off-by: Nicolas Pitre <nico@linaro.org>
> > > 
> > > Hrm... So being able to trim down the kernel is important.
> > > 
> > > Although my sense is that momentum has been moving to clock_gettime()
> > > over gettimeofday(), such that gettimeofday() is mostly a shim over
> > > clock_gettime() logic wise. So this is sort of going the other
> > > direction.
> > > 
> > > Also given many other syscalls take clockids and the backing logic
> > > isn't really getting removed (probably could cut the dynamic posix
> > > clocks core with the same conditional), I wonder if you could get a
> > > similar size win by taking a slightly more narrow cutting of the
> > > subsystem. That way you could preserve the more useful clock_gettime()
> > > functionality, but maybe stub out some of the less often used
> > > functionality.
> > 
> > I agree with this.  Cutting out the syscall alone helps, but cutting out
> > the corresponding infrastructure for those timers would help even more.
> > I wouldn't suggest turning down this patch in isolation, but I'd very
> > much like to see it become a patch series that also removes the
> > underlying timer infrastructure.
> 
> The patch does remove more than only syscalls which are in 
> posix-timers.c. The whole of posix-timers.c, posix-clock.c and 
> posix-cpu-timers.c are removed. The later is particularly footprint 
> heavy for one specialized clock type.

Looking at this again, it looks like this *does* already cut out the
dynamic POSIX clocks core that John Stultz mentioned, unless I'm missing
something.  John, what did you have in mind here?

And which other syscalls did you have in mind that take a clockid_t?

> > > Josh (cc'ed) also was talking awhile back about cutting out the core
> > > NTP logic. Having a single minimal-time option might be nice rather
> > > then having a bunch of different conditionals that might be combined.
> > 
> > I do think having individual configuration options for families of
> > syscalls helps, in that people can more easily figure out the set of
> > syscalls their code calls.  But those options should also cut out the
> > underlying infrastructure whenever possible; that avoids the need to
> > have separate options for the infrastructure, which a user would have to
> > enable to see the options for the interfaces to that infrastructure.  If
> > it can't get called, it doesn't need compiling in.
> 
> Absolutely.  However this is something that would scale better with 
> automatic code pruning performed by LTO or ld with -gc-sections.

Agreed, that may help eliminate some internal-only selected Kconfig
options eventually.

> > If that infrastructure supports multiple families of syscalls that make
> > sense to drop independently, then it might make sense to have that
> > underlying option, ideally automatically selected.  For instance, a
> > legacy-free userspace might only enable signalfd and not traditional
> > signal delivery, or only enable timerfd and not enable other forms of
> > timers.
> 
> One problem is that embedded libc implementations often go with the 
> legacy interface as it is often simpler and smaller.

In some cases, yes.  But, for instance, signalfd needs significantly
less infrastructure than traditional signals/sigreturn/etc.

> > (Hopefully the kconfig-sat folks can successfully develop a system that
> > allows you to turn on an option whose dependencies aren't yet enabled
> > and automatically enable its dependencies, to improve the functionality
> > of "select".)
> > 
> > For this patch, I'd recommend expanding the documentation for the
> > Kconfig symbol to explicitly state the families of syscalls this omits.
> 
> Good point.
> 
> > I'd also suggest dropping the stubs that print a message, and just using
> > sys_ni_syscall.  As a debug mechanism, that infrastructure seems more
> > generally useful than just POSIX timer syscalls.  Instead, I'd suggest a
> > separate patch adding (optional, CONFIG_SHOW_MISSING_SYSCALLS) support
> > in sys_ni.c to print a one-time message per syscall showing the process
> > name, PID, syscall number, name, and Kconfig symbol needed to enable it.
> > You could modify cond_syscall to accept a Kconfig symbol argument, and
> > generate a unique stub that calls printk_once.
> 
> Yes, this would be a good thing to have.  The ptrace infrastructure 
> might be leveraged here simply to get the actual syscall number, etc.  

Assuming you have the ptrace infrastructure. :)  But yes, looking up the
syscall number in a table might work better, though it'd add complexity
versus just generating multiple stubs.

> However I wouldn't start on that unless there is actually some more 
> configurable syscalls making the rationalization worth it.

Many other syscalls have Kconfig options; this would help with those
too.

> Here I wanted to have a minimal printout in case someone decided to turn 
> off POSIX timers and wondered why his big desktop distro doesn't boot 
> anymore.  Those syscalls have never been optional before.

I agree; I'd just prefer to see the general version rather than a
syscall-specific solution.

- Josh Triplett

[toc] | [prev] | [next] | [standalone]

#1479709

On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> Also given many other syscalls take clockids and the backing logic
> isn't really getting removed (probably could cut the dynamic posix
> clocks core with the same conditional), I wonder if you could get a
> similar size win by taking a slightly more narrow cutting of the
> subsystem. That way you could preserve the more useful clock_gettime()
> functionality, but maybe stub out some of the less often used
> functionality.

I want to support tinification, but I also doubt the utility of
removing clock_gettime() and clock_nanosleep().  I can't imagine ever
building a user space without those.  In fact, thinking about IoT,
having good time is critical, and so these are the *last* functions I
would remove when downsizing.

Thanks,
Richard

[toc] | [prev] | [next] | [standalone]

#1479726

On Fri, Sep 09, 2016 at 09:48:57AM +0200, Richard Cochran wrote:
> On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> > Also given many other syscalls take clockids and the backing logic
> > isn't really getting removed (probably could cut the dynamic posix
> > clocks core with the same conditional), I wonder if you could get a
> > similar size win by taking a slightly more narrow cutting of the
> > subsystem. That way you could preserve the more useful clock_gettime()
> > functionality, but maybe stub out some of the less often used
> > functionality.
> 
> I want to support tinification, but I also doubt the utility of
> removing clock_gettime() and clock_nanosleep().  I can't imagine ever
> building a user space without those.  In fact, thinking about IoT,
> having good time is critical, and so these are the *last* functions I
> would remove when downsizing.

1) If you already have another function providing time and don't need two.
2) If you run an entirely event-driven loop and don't sleep.

[toc] | [prev] | [next] | [standalone]

#1480014

On Fri, 9 Sep 2016, Josh Triplett wrote:

> On Fri, Sep 09, 2016 at 09:48:57AM +0200, Richard Cochran wrote:
> > On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> > > Also given many other syscalls take clockids and the backing logic
> > > isn't really getting removed (probably could cut the dynamic posix
> > > clocks core with the same conditional), I wonder if you could get a
> > > similar size win by taking a slightly more narrow cutting of the
> > > subsystem. That way you could preserve the more useful clock_gettime()
> > > functionality, but maybe stub out some of the less often used
> > > functionality.
> > 
> > I want to support tinification, but I also doubt the utility of
> > removing clock_gettime() and clock_nanosleep().  I can't imagine ever
> > building a user space without those.  In fact, thinking about IoT,
> > having good time is critical, and so these are the *last* functions I
> > would remove when downsizing.
> 
> 1) If you already have another function providing time and don't need two.

Agreed.

> 2) If you run an entirely event-driven loop and don't sleep.

I hope you wanted to say: and don't use *nanosleep() :) Otherwise you'd
have a full busy polling event loop which I doubt is desirable ...

Thanks,

	tglx


 

[toc] | [prev] | [next] | [standalone]

#1480008

On Fri, 9 Sep 2016, Richard Cochran wrote:

> On Thu, Sep 08, 2016 at 02:19:24PM -0700, John Stultz wrote:
> > Also given many other syscalls take clockids and the backing logic
> > isn't really getting removed (probably could cut the dynamic posix
> > clocks core with the same conditional), I wonder if you could get a
> > similar size win by taking a slightly more narrow cutting of the
> > subsystem. That way you could preserve the more useful clock_gettime()
> > functionality, but maybe stub out some of the less often used
> > functionality.
> 
> I want to support tinification, but I also doubt the utility of
> removing clock_gettime() and clock_nanosleep().  I can't imagine ever
> building a user space without those.  In fact, thinking about IoT,
> having good time is critical, and so these are the *last* functions I
> would remove when downsizing.

gettimeofday() gives you microsecond resolution which is good enough for
99.9% of all problems.

The only real downside is removing clock_nanosleep() because nanosleep()
does not give you the ability to schedule timers on absolute time.

Thanks,

	tglx

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web