Bug 13676

Not sure who to assign this one to, sorry if it's wrong.

CC: (none) => anssi.hannula, ennael1
Whiteboard: (none) => 5alpha1

Created attachment 5249 [details]
report.bug.xz

There are a number of these..

<6>[   26.720376] traps: monitor-get-edi[371] trap invalid opcode ip:7fabd8ebfb5f sp:7fffc37cd520 error:0 in libx86.so.1[7fabd8eb8000+20000]

I can reproduce the issue when running /usr/sbin/monitor-get-edid-using-vbe manually, but I have no clue about how to fix it. And as the error seems to happen inside libx86.so, it may as well comes from there.

Created attachment 5257 [details]
GDB trace with symbols

A gcc miscompilation?
Code hasn't changed for years...

Fixed in libx86-1.1-15.mga5

Status: NEW => RESOLVED
Resolution: (none) => FIXED

I am just seeing this message on the current boot-nonfree.iso installer while upgrading mga4 to mga5 RC.

CC: (none) => dirteat

At what stage do you see this error?  This should be fixed in stage2 as of a few weeks ago.

I don't know, but you may have a look to the full report there:

https://bugs.mageia.org/show_bug.cgi?id=15781

For instance, check report_1.bug at time 66.81. If I am not mistaken, this is at stage 1 then (during kernel boot).

Yep, it still segfaults on x86_64

Status: RESOLVED => REOPENED
Resolution: FIXED => (none)

There was a discussion on the dev list where Chris complained about a date in stage1 that was from February and Pascal said that stage1 hadn't been regenerated since then.  Does it just need to be regenerated with the fixed libx86?

He meat stage1 as just the stage1 binary.
stage1 is rebuild whenever there's a new kernel.
But libx86 is used in stage2 anyway...

Ahh, OK.  Unfortunately I don't have x86_64 mirrored locally so I can't test it there.  I'm guessing the x86emu backend isn't working there though, so I can try pushing a build with that changed.

Build failed:
http://pkgsubmit.mageia.org/uploads/failure/cauldron/core/release/20150428160911.ennael.valstar.1446/log/libx86-1.1-24.mga5/build.0.20150428161003.log

For some reasons, I now see that trap error appearing 3 times in the logs when I boot the machine:

May 09 08:47:05 brenva kernel: traps: monitor-get-edi[1026] trap invalid opcode ip:7f74b94c6b5f sp:7ffe6ef27030 error:0 in libx86.so.1[7f74b94bf000+1f000]
May 09 08:47:06 brenva kernel: traps: monitor-get-edi[1027] trap invalid opcode ip:7fa19d73cb5f sp:7ffd8cc1f060 error:0 in libx86.so.1[7fa19d735000+1f000]
May 09 08:47:06 brenva kernel: traps: monitor-get-edi[1028] trap invalid opcode ip:7f4f0cee1b5f sp:7ffffb27dd90 error:0 in libx86.so.1[7f4f0ceda000+1f000]

(In reply to Chris Denice from comment #14)
> For some reasons, I now see that trap error appearing 3 times in the logs
> when I boot the machine:

It might be because of mga-bg-res.service which calls /usr/sbin/monitor-probe.

ah,
that may explain also the rest (also posted on the mailing list), the log of mga-be-res not working:



May 09 08:47:07 brenva mga-bg-res[942]: Unrecognized option: -probeonly
May 09 08:47:07 brenva mga-bg-res[942]: use: X [:<display>] [option]
May 09 08:47:07 brenva mga-bg-res[942]: -a #                   default pointer acceleration (factor)
May 09 08:47:07 brenva mga-bg-res[942]: -ac                    disable access control restrictions
May 09 08:47:07 brenva mga-bg-res[942]: -audit int             set audit trail level

May 09 08:47:07 brenva mga-bg-res[942]: (EE)
May 09 08:47:07 brenva mga-bg-res[942]: Fatal server error:
May 09 08:47:07 brenva mga-bg-res[942]: (EE) Unrecognized option: -probeonly
May 09 08:47:07 brenva mga-bg-res[942]: (EE)
May 09 08:47:07 brenva mga-bg-res[942]: (EE)
May 09 08:47:07 brenva mga-bg-res[942]: Please consult the The X.Org Foundation support
May 09 08:47:07 brenva mga-bg-res[942]: at http://bugs.mageia.org
May 09 08:47:07 brenva mga-bg-res[942]: for help.
May 09 08:47:07 brenva mga-bg-res[942]: (EE)
May 09 08:47:07 brenva mga-bg-res[942]: X probe failed
May 09 08:47:07 brenva mga-bg-res[942]: /usr/libexec/mga-bg-res: line 28: 1000*/: syntax error: operand expected (error token is "/")
May 09 08:47:07 brenva mga-bg-res[942]: Could not find this file: /usr/share/mga/backgrounds/Mageia-Default-.jpg.
May 09 08:47:07 brenva mga-bg-res[942]: Please check that the mageia-theme-Default package is properly installed, or disable the mga-bg-res systemd service if you do not want to force using the Mageia theme.

cheers.

I don't know if it's the same bug or a new.
I'm on :
[    0.000000] Linux version 3.19.8-desktop-1.mga5 (iurt@ecosse.mageia.org) (gcc version 4.9.2 (GCC) ) #1 SMP Mon May 11 16:35:09 UTC 2015
[    0.000000] Command line: BOOT_IMAGE=linux root=UUID=ad21247c-3d73-453e-b9d3-541bd8e6be74 splash quiet noiswmd resume=UUID=f5e4982e-2b05-4d5c-bbd3-63de6110a2ac vga=788
with a radeon 4670 and a Samsung 22' monitor in 16/10°
On Mageia 4.1 and M5rc When I boot, I don't have passage by Bios and Grub. I've black page and open directly a page with (obtain by dmesg) :

[drm:drm_edid_block_valid [drm]] *ERROR* EDID checksum is invalid, remainder is 248
[    2.692616] Raw EDID:
[    2.692618]          00 ff ff ff ff ff ff 00 4c 2d e5 03 32 32 57 54
[    2.692620]          26 12 01 03 80 2f 1e 78 2a ee 91 a3 54 4c 99 26
[    2.692622]          0f 50 54 bf ef 80 b3 00 81 80 81 40 71 4f 01 01
[    2.692623]          01 01 01 01 01 01 74 2e 90 a0 60 1a 1e 40 30 20
[    2.692625]          36 00 da 28 11 00 00 1a 00 00 00 fd 00 38 4b 1e
[    2.692626]          51 0e 00 0a 20 20 20 20 20 20 00 00 00 fc 00 53
[    2.692627]          79 6e 63 4d 61 73 74 65 72 0a 20 20 00 00 00 ff
[    2.692629]          00 48 39 58 51 39 32 30 35 32 37 0a 20 20 00 25
[    2.719874] [drm:drm_edid_block_valid [drm]] *ERROR* EDID checksum is invalid, remainder is 248
[    2.719876] Raw EDID:
[    2.719877]          00 ff ff ff ff ff ff 00 4c 2d e5 03 32 32 57 54
[    2.719879]          26 12 01 03 80 2f 1e 78 2a ee 91 a3 54 4c 99 26
[    2.719880]          0f 50 54 bf ef 80 b3 00 81 80 81 40 71 4f 01 01
[    2.719882]          01 01 01 01 01 01 74 2e 90 a0 60 1a 1e 40 30 20
[    2.719883]          36 00 da 28 11 00 00 1a 00 00 00 fd 00 38 4b 1e
[    2.719884]          51 0e 00 0a 20 20 20 20 20 20 00 00 00 fc 00 53
[    2.719886]          79 6e 63 4d 61 73 74 65 72 0a 20 20 00 00 00 ff
[    2.719888]          00 48 39 58 51 39 32 30 35 32 37 0a 20 20 00 25
[    2.747141] [drm:drm_edid_block_valid [drm]] *ERROR* EDID checksum is invalid, remainder is 248
[    2.747145] Raw EDID:
[    2.747147]          00 ff ff ff ff ff ff 00 4c 2d e5 03 32 32 57 54
[    2.747149]          26 12 01 03 80 2f 1e 78 2a ee 91 a3 54 4c 99 26
[    2.747150]          0f 50 54 bf ef 80 b3 00 81 80 81 40 71 4f 01 01
[    2.747152]          01 01 01 01 01 01 74 2e 90 a0 60 1a 1e 40 30 20
[    2.747154]          36 00 da 28 11 00 00 1a 00 00 00 fd 00 38 4b 1e
[    2.747155]          51 0e 00 0a 20 20 20 20 20 20 00 00 00 fc 00 53
[    2.747157]          79 6e 63 4d 61 73 74 65 72 0a 20 20 00 00 00 ff
[    2.747159]          00 48 39 58 51 39 32 30 35 32 37 0a 20 20 00 25
[    2.774403] [drm:drm_edid_block_valid [drm]] *ERROR* EDID checksum is invalid, remainder is 248
[    2.774405] Raw EDID:
[    2.774407]          00 ff ff ff ff ff ff 00 4c 2d e5 03 32 32 57 54
[    2.774408]          26 12 01 03 80 2f 1e 78 2a ee 91 a3 54 4c 99 26
[    2.774410]          0f 50 54 bf ef 80 b3 00 81 80 81 40 71 4f 01 01
[    2.774412]          01 01 01 01 01 01 74 2e 90 a0 60 1a 1e 40 30 20
[    2.774414]          36 00 da 28 11 00 00 1a 00 00 00 fd 00 38 4b 1e
[    2.774415]          51 0e 00 0a 20 20 20 20 20 20 00 00 00 fc 00 53
[    2.774417]          79 6e 63 4d 61 73 74 65 72 0a 20 20 00 00 00 ff
[    2.774419]          00 48 39 58 51 39 32 30 35 32 37 0a 20 20 00 25
[    2.774422] radeon 0000:01:00.0: DVI-I-1: EDID block 0 invalid.
[    2.774477] [drm:radeon_dvi_detect [radeon]] *ERROR* DVI-I-1: probed a monitor but no|invalid EDID

[drm:radeon_dvi_detect [radeon]] *ERROR* DVI-I-1: probed a monitor but no|invalid EDID
[   38.746591] [drm:radeon_dvi_detect [radeon]] *ERROR* DVI-I-1: probed a monitor but no|invalid EDID
[   42.331722] [drm:radeon_dvi_detect [radeon]] *ERROR* DVI-I-1: probed a monitor but no|invalid EDID
and sustem blocked
I open Mageia by DVD and copy the xorg.conf onto /etc/X11/xorg.conf and so ontain to open Mageia, but allways with darkpage and this EDID ERROR page before obtaining Mageia in 1024*768.
Sometines, I've 1650x1080, but more and more rare.
Is it a new ug or the same, I don't know. Say me what and excuse me to disturb.
Cheers

CC: (none) => dantou

To David (comment 13); I suceeded to compile it on x86_64 but only by adding explicitely
BACKEND=x86emu

I modified the spec accordingly; but I am not sure this is fine. Do you mind to have a look and resubmit if ok?

cheers.

On svn I did one more thing:
Adding BACKEND=x86emu allows compilation under x86_64; but libx86 was still segfaulting on x86_64.

David, in fact the segfault on x86_64 is due to disabling patch2! So I reverted your last change and restored patch2; but it was conflicting with patch4 (fixing a HLT hexadecimal code); so I *tried* to fix patch4 to be compatible with patch2 by doing this:

-       *mmap_addr = 0x4f; /* Make sure that we end up jumping back to a
+       *mmap_addr = 0xf4; /* Make sure that we end up jumping back to a
                              halt instruction */

Of course, I am completely out of my league here, I guess Thierry you should really have a look.

Anyway, on x86_64, that compiles and monitor-get-edid-using-vbe does not segfaut anymore (actually that seems to work).

Before asking for freeze push, please review my changes and try to build it locally and check if monitor-get-edid-using-vbe works (on i586 in particular).

cheers.

I guess we need to make it conditional patching for now, so P2 is only added on x86_64 until we can fix it properly.

David, do you confirm, can you test it in its current form on i586?

What's the status for this bug report? Has it been fixed? Is it still valid? Does it require an Errata?

Keywords: (none) => NEEDINFO

On x86_64, it is fixed; I am waiting for someone to test it on i586 as I think it should be too.

Why waiting?

I think I remember someone saying it wasn't fixed in i586 because the patch was applied only for x86_64, but I can't say more and might be wrong.

I hadn't been CC'd on this bug so I hadn't seen the recent discussion, but I tested many combinations before and committed the only one that worked.  If you reverted any of my changes (which it appears that you have), then you have likely broken it again on i586.  I'll test to confirm that later today.  It may need to do something conditional depending on the arch.

CC: (none) => luigiwalser

Yes, please check. For i586, the only thing that has changed is restoring patch 2. If this breaks i586, then, as tmb suggested, an easy fix is to conditionally apply the patch to x86_64 only.

I compiled the current version on core/update_testing; that should be easier to test.

You have indeed broken it, as I suspected.

Please test libx86_1-1.1-25.mga5.  It conditionally applies patch 2.

Great, thanks! Works for me on x86_64!

Pushed by tmb yesterday (rebuilt as libx86-1.1-2.mga5).  Marking as FIXED.

Status: REOPENED => RESOLVED
Resolution: (none) => FIXED

(In reply to David Walser from comment #30)
> Pushed by tmb yesterday (rebuilt as libx86-1.1-26.mga5).  Marking as FIXED.

Missed a key there.  Release tag is 26.