4:19 a.m.
Hi all.
Does Clear Linux support running binaries for the x32 ABI?
If I look at the following line
# CONFIG_X86_X32 is not set
in https://github.com/clearlinux-pkgs/linux-lts/blob/master/config ,
it seems to be disabled.
For many applications, x32 clearly outperforms i386 and x86-64, e.g.
(for functional programming):
https://lists.gnu.org/archive/html/chicken-users/2015-09/msg00000.html
https://sympa.inria.fr/sympa/arc/bigloo/2016-12/msg00015.html
In the long run, also compiling x32 binaries could be an important
performance option for Clear Linux.
Ciao
Sven
10:21 a.m.
On 12/25/2016 1:19 AM, Sven Hartrumpf wrote:
Hi all.
Does Clear Linux support running binaries for the x32 ABI?
currently we do not support this; we can certainly look into it,
the most complicated factor will be build infrastructure rather
than packaging.
I'm surprised at some of the conclusions, there is an implication
that this helps garbage collection.... while the golang folks pretty
much see the opposite effect, where 64 pointers help their garbage collector
out a lot versus 32 bit pointers
it's a bit hard to read online ;-)
also wonder how much is a distro effect; if it's easy to run such
benchmarks (still poking at it)... one can run a clear linux
docker instance on ubuntu easily
something like
docker run -it clearlinux
and then inside the docker install a pile of dev tools
swupd update
swupd bundle-add os-clr-on-clr
and you should have a pretty darn complete system with our full toolchain/etc.
Depending on the benchmark/workload, there's typically [*] a 0 to 10% perf delta
between something like ubuntu and clear linux.
[*] based on various phoronix benchmarks, such as
http://www.phoronix.com/scan.php?page=article&item=eoy16-ws-server&…
11:58 a.m.
Hello Arjan.
Thanks for your answers.
Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
On 12/25/2016 1:19 AM, SH wrote:
That's true if you want to allow compiling in x32 mode.
It's easier for running x32 binaries (maybe produced on another system):
this requires only that the kernel
contains the given feature (CONFIG_X86_X32).
For example, I regularly build my x32 binaries on Ubuntu (which has
a x32 tool chain and libraries) and can run the (statically
linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
no x32 libraries).
Does Clear Linux support running binaries for the
x32 ABI?
currently we do not support this; we can certainly look into it,
the most complicated factor will be build infrastructure rather
than packaging. I'm surprised at some of the conclusions, there is
an implication
that this helps garbage collection.... while the golang folks pretty
much see the opposite effect, where 64 pointers help their garbage
collector out a lot versus 32 bit pointers
Interesting.
In the Scheme implementation I looked at, the amount of garbage (and
most stored data structures)
is doubled if stepping from i386/x32 to x86-64. Hence, the garbage
collector and other components need much more time.
it's a bit hard to read online ;-)
Yes, the ML archive software is not ideal for this kind of text :-(
Here are the relevant percentage lines (for a monospaced font):
i386 257016.0 430 760 1750 200 1090 1220 430 1690 870 950
x86-64 -17.7 -13.1 -26.7 -12.5 0.1 -18.7 -24.5 -4.4 -46.9 -18.7 4.3
x32 -26.2 -16.8 -31.5 -16.5 -5.0 -25.4 -38.5 -4.4 -47.7 -44.2 -15.3
i386 5750 930 920 950 1250 10 830 1180 8850 1820 AVG
x86-64 -18.8 -37.4 -15.8 -4.8 -16.0 0.1 -24.0 -1.4 -19.6 -32.6 -16.5
x32 -18.0 -41.1 -28.8 -8.6 -34.6 0.1 -32.5 -37.8 -27.1 -32.6 -25.6
also wonder how much is a distro effect; if it's
easy to run such
benchmarks (still poking at it)... one can run a clear linux
docker instance on ubuntu easily
something like
docker run -it clearlinux
and then inside the docker install a pile of dev tools
swupd update
swupd bundle-add os-clr-on-clr
and you should have a pretty darn complete system with our full
toolchain/etc.
Thanks for the suggestions.
Ciao
Sven
1:19 p.m.
On 12/25/2016 8:58 AM, Sven Hartrumpf wrote:
Hello Arjan.
Thanks for your answers.
Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
well the flipside argument is one of security attack surface; adding a whole
new surface for something we don't ship/use ourselves will give the security
folks some shivers..... it's a bit of a balance as a result.
anyway will play a bit with some of this...
On 12/25/2016 1:19 AM, SH wrote:
That's true if you want to allow compiling in x32 mode.
It's easier for running x32 binaries (maybe produced on another system):
this requires only that the kernel
contains the given feature (CONFIG_X86_X32).
For example, I regularly build my x32 binaries on Ubuntu (which has
a x32 tool chain and libraries) and can run the (statically
linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
no x32 libraries).
Does Clear Linux support running binaries for the
x32 ABI?
currently we do not support this; we can certainly look into it,
the most complicated factor will be build infrastructure rather
than packaging.
4:24 p.m.
Arjan wrote, 2016-12-25:
On 12/25/2016 8:58 AM, Sven wrote:
In the past 2 years, there has been no security problem that was X32-specific
AFAIK.
Hence, I would like to ask you to reconsider the ability to run X32 binaries.
Sven
Hello Arjan.
Thanks for your answers.
Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
well the flipside argument is one of security attack surface; adding a
whole
new surface for something we don't ship/use ourselves will give the
security
folks some shivers..... it's a bit of a balance as a result. On 12/25/2016 1:19 AM, SH wrote:
That's true if you want to allow compiling in x32 mode.
It's easier for running x32 binaries (maybe produced on another system):
this requires only that the kernel
contains the given feature (CONFIG_X86_X32).
For example, I regularly build my x32 binaries on Ubuntu (which has
a x32 tool chain and libraries) and can run the (statically
linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
no x32 libraries).
Does Clear Linux support running binaries for the
x32 ABI?
currently we do not support this; we can certainly look into it,
the most complicated factor will be build infrastructure rather
than packaging.
4:26 p.m.
On 12/29/2018 1:24 PM, Sven Hartrumpf wrote:
Arjan wrote, 2016-12-25:
https://lkml.org/lkml/2018/12/10/1151
just when Linus and co are discussing dropping it entirely ...
On 12/25/2016 8:58 AM, Sven wrote:
In the past 2 years, there has been no security problem that was X32-specific
AFAIK.
Hence, I would like to ask you to reconsider the ability to run X32 binaries. Hello Arjan.
Thanks for your answers.
Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
well the flipside argument is one of security attack surface; adding a
whole
new surface for something we don't ship/use ourselves will give the
security
folks some shivers..... it's a bit of a balance as a result. On 12/25/2016 1:19 AM, SH wrote:
> Does Clear Linux support running binaries for the x32 ABI?
currently we do not support this; we can certainly look into it,
the most complicated factor will be build infrastructure rather
than packaging.
That's true if you want to allow compiling in x32 mode.
It's easier for running x32 binaries (maybe produced on another system):
this requires only that the kernel
contains the given feature (CONFIG_X86_X32).
For example, I regularly build my x32 binaries on Ubuntu (which has
a x32 tool chain and libraries) and can run the (statically
linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
no x32 libraries).
6:10 p.m.
Arjan wrote, 2016-12-25:
>> On 12/25/2016 8:58 AM, Sven wrote:
>>> Hello Arjan.
>>>
>>> Thanks for your answers.
>>>
>>> Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
>>>> On 12/25/2016 1:19 AM, SH wrote:
>>>>> Does Clear Linux support running binaries for the x32 ABI?
>>>>
>>>> currently we do not support this; we can certainly look into it,
>>>> the most complicated factor will be build infrastructure rather
>>>> than packaging.
>>>
>>> That's true if you want to allow compiling in x32 mode.
>>> It's easier for running x32 binaries (maybe produced on another
>>> system):
>>> this requires only that the kernel
>>> contains the given feature (CONFIG_X86_X32).
>>> For example, I regularly build my x32 binaries on Ubuntu (which has
>>> a x32 tool chain and libraries) and can run the (statically
>>> linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
>>> no x32 libraries).
>>>
>>
>> well the flipside argument is one of security attack surface; adding a
>> whole
>> new surface for something we don't ship/use ourselves will give the
>> security
>> folks some shivers..... it's a bit of a balance as a result.
Is security concerns the reason why no one has built WINE and put it in
the repository?
Jeff
5:29 a.m.
Happy new Year!
Arjan wrote, 2018-12-29:
On 12/29/2018 1:24 PM, Sven Hartrumpf wrote:
I followed this discussion in December.
The drop proposal came from Andy Lutomirski and Linus asked for
"complaints" - and he received some (IMHO relevant) complaints from quite
diverse areas of users and use cases. (I can collect the URLs of
these complaints if anyone is interested.)
Unfortunately, the main discussion led to technical details of possible
improvements/mistakes and the original question
("Who is using x32 and would complain if x32 were dropped?")
was somehow lost in the thread.
It was pointed out by H.J. Lu that building a CONFIG_X86_X32-enabled kernel
with X86-64 binutils and GCC is straigtforward, see
https://www.spinics.net/lists/kernel/msg2986535.html
Anyway, the discussion seems to have died without any decision.
I can understand that this discussion leads to irritations and doubts.
If so, we could wait a little longer till the irritation disappears
(hopefully :-) ) and the dust settles.
Sven
Arjan wrote, 2016-12-25:
https://lkml.org/lkml/2018/12/10/1151
just when Linus and co are discussing dropping it entirely ... On 12/25/2016 8:58 AM, Sven wrote:
In the past 2
years, there has been no security problem that was
X32-specific
AFAIK.
Hence, I would like to ask you to reconsider the ability to run X32
binaries. Hello Arjan.
Thanks for your answers.
Arjan van de Ven <arjan(a)linux.intel.com> wrote, 2016-12-25 07:21:
> On 12/25/2016 1:19 AM, SH wrote:
>> Does Clear Linux support running binaries for the x32 ABI?
>
> currently we do not support this; we can certainly look into it,
> the most complicated factor will be build infrastructure rather
> than packaging.
That's true if you want to allow compiling in x32 mode.
It's easier for running x32 binaries (maybe produced on another system):
this requires only that the kernel
contains the given feature (CONFIG_X86_X32).
For example, I regularly build my x32 binaries on Ubuntu (which has
a x32 tool chain and libraries) and can run the (statically
linked) x32 binary on openSUSE boxes (which have no x32 tool chain and
no x32 libraries).
well the flipside argument is one of security attack surface; adding a
whole
new surface for something we don't ship/use ourselves will give the
security
folks some shivers..... it's a bit of a balance as a result.
1548
days inactive
2285
days old
8 comments
3 participants
participants (3)
-
Arjan van de Ven -
Jeffrey Hurwit -
Sven Hartrumpf