On 1/30/20 11:50 AM, Jan Engelhardt wrote:

On Thursday 2020-01-30 11:40, Thomas Zimmermann wrote:

...

I tried to run Tumbleweed on i586 today and when starting Gnome on X11 or got a number of errors caused by invalid opcodes.

[ 630.370899] traps: evolution-sourc[2962] trap invalid opcode ip:ae285db9 sp:bfd6fa00 error:0 in libjavascriptcoregtk- 4.0.so.18.14.8[ad937000+ca8000]

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

Yeah, and overwriting CFLAGS/CXXFLAGS from the environment. Adrian

On Thu, 2020-01-30 at 12:11 +0100, Thomas Zimmermann wrote:

Hi

Am 30.01.20 um 11:50 schrieb Jan Engelhardt:

...

On Thursday 2020-01-30 11:40, Thomas Zimmermann wrote:

...

I tried to run Tumbleweed on i586 today and when starting Gnome on X11 or got a number of errors caused by invalid opcodes.

[ 630.370899] traps: evolution-sourc[2962] trap invalid opcode ip:ae285db9 sp:bfd6fa00 error:0 in libjavascriptcoregtk- 4.0.so.18.14.8[ad937000+ca8000]

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

webkit2gtk3 build log: [ 64s] -- Performing Test C_COMPILER_SUPPORTS_-msse2 [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-msse2 - Success [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-mfpmath=sse [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-mfpmath=sse - Success [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-msse2 [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-msse2 - Success [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-mfpmath=sse [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-mfpmath=sse - Success

Glad to see that the processor is not yet entirely unsupported. Shall I open a bug report for this problem?

That will be tough: webkit2gtk claims in the CMake files: include(DetectSSE2) if (NOT SSE2_SUPPORT_FOUND) message(FATAL_ERROR "SSE2 support is required to compile WebKit") endif () Cheers, Dominique -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Hi Am 30.01.20 um 12:39 schrieb Dominique Leuenberger / DimStar:

On Thu, 2020-01-30 at 12:11 +0100, Thomas Zimmermann wrote:

...

Hi

Am 30.01.20 um 11:50 schrieb Jan Engelhardt:

...

On Thursday 2020-01-30 11:40, Thomas Zimmermann wrote:

...

I tried to run Tumbleweed on i586 today and when starting Gnome on X11 or got a number of errors caused by invalid opcodes.

[ 630.370899] traps: evolution-sourc[2962] trap invalid opcode ip:ae285db9 sp:bfd6fa00 error:0 in libjavascriptcoregtk- 4.0.so.18.14.8[ad937000+ca8000]

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

webkit2gtk3 build log: [ 64s] -- Performing Test C_COMPILER_SUPPORTS_-msse2 [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-msse2 - Success [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-mfpmath=sse [ 65s] -- Performing Test C_COMPILER_SUPPORTS_-mfpmath=sse - Success [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-msse2 [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-msse2 - Success [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-mfpmath=sse [ 65s] -- Performing Test CXX_COMPILER_SUPPORTS_-mfpmath=sse - Success

Glad to see that the processor is not yet entirely unsupported. Shall I open a bug report for this problem?

That will be tough: webkit2gtk claims in the CMake files:

include(DetectSSE2) if (NOT SSE2_SUPPORT_FOUND) message(FATAL_ERROR "SSE2 support is required to compile WebKit") endif ()

Wonderful! It forces its requirements on every piece of the software stack that depends on it. However, I wonder what they do on arm. I don't strictly depend on Gnome. I use the machine for kernel development and Gnome/X11 is just one of my smoke tests. Twm works nicely BTW! :) Best regards Thomas

Cheers, Dominique

-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

On Thu, 30 Jan 2020 12:54:28 +0100 Thomas Zimmermann wrote:

...

That will be tough: webkit2gtk claims in the CMake files:

include(DetectSSE2) if (NOT SSE2_SUPPORT_FOUND) message(FATAL_ERROR "SSE2 support is required to compile WebKit") endif ()

Wonderful! It forces its requirements on every piece of the software stack that depends on it. However, I wonder what they do on arm.

This requirement looks like introduced by a CPU manufacturer who wants to obsolete former CPU generations. From software point of view I can't believe that the benefit of SSE2 is so miraculous that something that is not doable without SSE2 is suddenly easily possible. BTW: I am also using Tumbleweed on an Athlon-XP. I upgraded from openSUSE 13.2 to Tumbleweed in Sept. 2018 using the procedure described in this message: https://lists.opensuse.org/opensuse-factory/2018-09/msg00128.html and keep it up to date ever since. In the first months konsole was not working, it crashed with an illegal opcode, but in the meantime this was resolved. There exist x86_64 CPUs with significantly less performance than the Athlon-XP, so I hope 32-bit support (without SSE2 requirement) will still be around for some time. Kind regards, Dieter -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thursday 2020-01-30 18:13, Felix Miata wrote:

Thomas Zimmermann composed on 2020-01-30 06:11 (UTC-0500):

...

Glad to see that the processor is not yet entirely unsupported. Shall I open a bug report for this problem?

btw, https://bugzilla.opensuse.org/show_bug.cgi?id=1162283 is now open

I think if you do that you'll get a wontfix because SSE2 is a required CPU configuration.

Says where? -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thu, Jan 30, 2020 at 7:50 AM Jan Engelhardt wrote:

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

And it clearly shows nobody uses TW in old hardware and that there should be a way to make the compiler gobally ignore or reinterpret certain flags -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thu, Jan 30, 2020 at 8:30 AM Richard Biener wrote:

You get what you feed it. Don't feed it crap.

Agreed, that's kinda difficult with thousands and thousands of packages though. -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Hi Am 30.01.20 um 12:36 schrieb Jan Engelhardt:

On Thursday 2020-01-30 12:13, Cristian Rodríguez wrote:

...

On Thu, Jan 30, 2020 at 7:50 AM Jan Engelhardt wrote:

...

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

And it clearly shows nobody uses TW in old hardware

That is bullcrap. I do use TW on two items of i586-class hardware. Not very often, but still. It's just that... webkit is not used by anything I care about (xfce4). According to .spec files:

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well? Best regards Thomas

atril balsa bijiben bookworm capnet-assist devhelp eclipse emacs empathy eolie epiphany evolution-data-server evolution geary gitg glade gnome-books gnome-boxes gnome-builder gnome-documents gnome-online-accounts gnucash gthumb libgepub meteo midori nemo-extensions notes-up pix python-wxPython quilter remmina shotwell sushi vimb vocal webkit2gtk3 wxGTK3-3_2 xiphos xreader yelp zenity

So mostly a GNOME thing. zenity is sometimes pulled in. For example:

clamtk live-fat-stick marco metacity muffin mutter opengl-games-utils openrct2 patterns-gnome patterns-mate python-dialite shellementary wqy-zenhei-fonts xfce4-panel[subpkg]

So yeah, as a xfce4 user, you don't really get to deal with webkit2gtk.

-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

Hi Am 30.01.20 um 15:13 schrieb Dominique Leuenberger / DimStar:

On Thu, 2020-01-30 at 14:50 +0100, Thomas Zimmermann wrote:

...

...

That is bullcrap. I do use TW on two items of i586-class hardware. Not very often, but still. It's just that... webkit is not used by anything I care about (xfce4). According to .spec files:

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well?

rsvg -> rust... nuff said

Admittedly, I lol'ed. But then it doesn't really help me with my problem. And according to zypper se --requires librsvg almost every graphical environment depends on librsvg. So in practice it imposes a minimal requirement for all things graphics. Best regards Thomas -- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

Am 30.01.20 um 15:58 schrieb Thomas Zimmermann:

Am 30.01.20 um 15:52 schrieb Thomas Zimmermann:

...

Hi

Am 30.01.20 um 15:13 schrieb Dominique Leuenberger / DimStar:

...

On Thu, 2020-01-30 at 14:50 +0100, Thomas Zimmermann wrote:

...

...

That is bullcrap. I do use TW on two items of i586-class hardware. Not very often, but still. It's just that... webkit is not used by anything I care about (xfce4). According to .spec files:

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well?

rsvg -> rust... nuff said

Admittedly, I lol'ed.

But then it doesn't really help me with my problem. And according to

zypper se --requires librsvg

almost every graphical environment depends on librsvg. So in practice it imposes a minimal requirement for all things graphics.

Or wait a sec, it doesn't seem to list anything KDE. There's hope...

It's a mixed bag. KDE still starts up, but: [ 134.757646] traps: opensuse-welcom[1807] trap invalid opcode ip:af0a7d89 sp:bf9ca0dc error:0 in libQt5WebEngineCore.so.5.14.0[af047000+61b8000] [ 172.001956] traps: gst-plugin-scan[2058] trap invalid opcode ip:b5c7a412 sp:bfdabca4 error:0 in libgstopengl.so[b5c54000+2a000] [ 172.205344] traps: gst-plugin-scan[2057] trap invalid opcode ip:b5bd2412 sp:bfc6f8a4 error:0 in libgstopengl.so[b5bac000+2a000] [ 172.315414] traps: gst-plugin-scan[2064] trap invalid opcode ip:b73e2412 sp:bf964bd4 error:0 in libgstopengl.so[b73bc000+2a000] [ 172.442156] traps: gst-plugin-scan[2068] trap invalid opcode ip:b7441412 sp:bfb27f34 error:0 in libgstopengl.so[b741b000+2a000] So it's better for what I what, but not good in general.

...

Best regards Thomas

-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 30.01.2020 18:46, Dominique Leuenberger / DimStar пишет:

On Thu, 2020-01-30 at 15:52 +0100, Thomas Zimmermann wrote:

...

Hi

Am 30.01.20 um 15:13 schrieb Dominique Leuenberger / DimStar:

...

On Thu, 2020-01-30 at 14:50 +0100, Thomas Zimmermann wrote:

...

...

That is bullcrap. I do use TW on two items of i586-class hardware. Not very often, but still. It's just that... webkit is not used by anything I care about (xfce4). According to .spec files:

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well?

rsvg -> rust... nuff said

Admittedly, I lol'ed.

But then it doesn't really help me with my problem. And according to

zypper se --requires librsvg

almost every graphical environment depends on librsvg. So in practice it imposes a minimal requirement for all things graphics.

Just came across an interesting blog post[0] it contains:

""" The 32-bit x86 Rust compiler also targets the "i686" CPU architecture. Sortof. Rust doesn't actually work on these processors, despite claims, because it actually targets the Pentium 4 processor, by using SSE2 instructions that were first introduced on that chip. If you try to run it on a real i686 (a Pentium Pro, Pentium II, or Pentium III), it will try to execute SSE2 instructions which that chip knows nothing about, and crash. """

This post is five years old. I hope the things changed since then. Anyway, I think that if rust generates SSE for i586 target, this should be reported upstream. It is a bug. Or there should be an option to disable. Go compiler may be tuned not to use SSE for i386 go target.

If there is sufficient demand and our rust maintainer(s) are willing to bite the bullet, we could offer rust as i586 and i686 package. But please, not to many of those i686 packages - or we have to sit together with the OBS team to get the product builder 'smarter' about this; I have to manually list all i686 binaries for now - and so far this was only glibc, where we have .i586 AND .i686 and Firefox (which are .i686 already)

Cheers, Dominique

[0] https://fraserblog.codewise.org/rust-on-i586/

-----BEGIN PGP SIGNATURE----- iQEzBAEBCAAdFiEEFnRguwljCXV9IjzOAv0PX1iySn8FAl4y/3kACgkQAv0PX1iy Sn9c+wf+Kgf/soi10jaLb9wvHLb+5k8CZFRTi9hxludBXZiA9Jzz3oVHxAIzJ2Ge LmPzJVoBDDcpzsXBe6FI48ZUJi5iRCoRlg9NHl/7wdxIVGipL8HcNfomdzPVIckY mea+lm8fXoOsZV+gLEaZU+7tTS5qcAtvtggGqyWF/F5pc9mR85YuMd8y2dfyQ5H8 qjERCRxZs6gmgUuXzA2kyx9cX4RmW+eAap276ioWFxMO+SykKl+nCpJ5O2JF+Hrl Ye8pmfQtjLy0xyi0/wWBV21hgeA1CGGqWXkqNf4rPpIUpDOQAWfCebuKr7n3S2B+ 0nuvqfrDLZSBPkL+t0UZOFFFOf3+/w== =BpMp -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

On Thursday 2020-01-30 17:14, Dominique Leuenberger / DimStar wrote:

Do I need to say that OBS has quite new workers and not a single, real, i586 CPU? :)

OBS also has no native riscv64 workers as far as I can tell and yet it produces RPMs of that kind. ;-) -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 30.01.2020 19:39, Dominique Leuenberger / DimStar пишет:

On Thu, 2020-01-30 at 17:19 +0100, Jan Engelhardt wrote:

...

On Thursday 2020-01-30 17:14, Dominique Leuenberger / DimStar wrote:

...

Do I need to say that OBS has quite new workers and not a single, real, i586 CPU? :)

OBS also has no native riscv64 workers as far as I can tell and yet it produces RPMs of that kind. ;-)

Yes, in both cases we use qemu to build the packages. And our config for i586 builds reports:

[ 40s] 2nd stage started in virtual machine [ 40s] machine type: i686

And as we already know: i686 is only half the truth - as there is i686 with different feature sets; some with SSE2, some without. Surely, there is a way to tell qemu to not put sse2 features into the VM. but that seems not to be done so far.

Keep in mind that disabling SSE2 on i586 has also an impact on 64bit users that install things like wine, steam and so on - basically every 64bit user that has -32bit packages installed.

It depends on our understanding of what i586 is. In the ancient times of openSUSE 11.2 I have fully positive experience of running it on i586 CPU without SSE. Nothing ever crashed with SIGILL.

Cheers, Dominique

-----BEGIN PGP SIGNATURE----- iQEzBAEBCAAdFiEEFnRguwljCXV9IjzOAv0PX1iySn8FAl4zGbQACgkQAv0PX1iy Sn/fEAf/RlsDE5gBxMW2NieS2KpkM8JFUKhOcymPU5jQ04TQRbVYG4QYMI5Pf6Xx PUuewde58dgS0TAa7Yhfww9B/PtkkJdfZ89+dEdpzsJB2WdOalO7TSKrMQ//im8o bPozdW/0UC5QTa2ezOq6ntPtz29JsM9obPhcgCfyWDL3FA2Q8hIDMqqBD2dyP3Lz uub3Svgno1XcKn5SHLBnOwKxkzv7Gbv8cplhRcey/QZTJfE5RCxeX6gD9rwBtWL/ j76V+8nakl14NNKKzc0QoIPL5gfNYumQkCYioAjeDnArY7ePl54z3ZsGpywWqSt3 wxqqdBliK3IEuNX8gbJhNMl28Wwy2A== =M8eG -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Am 30.01.20 um 16:31 schrieb Matěj Cepl:

On 2020-01-30, 14:13 GMT, Dominique Leuenberger / DimStar wrote:

...

...

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well?

rsvg -> rust... nuff said

Is it? Could you elaborate, please?

Or stay on topic. Your choice.

Matěj

-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

On 2020-01-30 15:13, Dominique Leuenberger / DimStar wrote:

On Thu, 2020-01-30 at 14:50 +0100, Thomas Zimmermann wrote:

...

Well, I just tried xfce4 and found that it uses librsvg. Hence it triggers the same bug. Do you see this on your system as well?

rsvg -> rust... nuff said

Rust is available for i686 (i586), and LLVM will do the right thing. Where do you see the problem? -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org

Am 30.01.20 um 12:13 schrieb Cristian Rodríguez:

On Thu, Jan 30, 2020 at 7:50 AM Jan Engelhardt wrote:

...

It's the usual crap that programs make use of sse2 without checking cpuid AT DAMN RUNTIME.

And it clearly shows nobody uses TW in old hardware and that there

No it doesn't. It shows that people (me, at least) use TW on old hardware.

should be a way to make the compiler gobally ignore or reinterpret certain flags

-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer

30.01.2020 14:40, Jan Engelhardt пишет:

On Thursday 2020-01-30 12:35, Matwey V. Kornilov wrote:

...

In the past, I remember that i586 binaries must not use SSE extensions. I am not sure about i686 thought. If nothing changed since that ancient times, then it should be considered as a packaging bug.

Instruction-wise, i686 is mostly about CMOV (and maybe FXSAVE). SSE is not mandatory. SSE2 isn't either - as exemplified by the Athlon XP and the Pentium3.

I had even more exotic CPUs: https://github.com/SUSE/machinery/issues/2243 -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-factory+owner@opensuse.org