Hell Labs

“What is nash-hotplug and why is it consuming 100% CPU?“, someone asked in a Xen-related message I found after experiencing similar problems booting Mandriva 2008.1 with finit in the Asus Eeepc. This situation seems to be especially common in different Linux distributions running as Xen guests and the usual advice is to just kill nash-plugin after the system boots. Instead of doing that ugly workaround, we decided to investigate and find out what’s happening. Whatever nash-hotplug is supposed to do, I can’t say it keeps running consuming 100% of your cpu because it’s a well-written program (notice the bizarre IPC protocol, how it tries to read from a descriptor after closing it and how easily it can get trapped in infinite loops). Use the following quick fix to avoid the problem.

• mkinitrd-6.0.34-nash-hotplug.patch

Die, nash, die.

Last week I came across a no-brand, very cheap-looking USB webcam with id 0ac8:0323 and recognized by Linux as a “Z-Star Microelectronics Corp. Luxya WC-1200 USB 2.0 Webcam”, which seems to be misleading in this case.

From the USB ID it is supposed to contain a Vimicro VC0323 controller, and it was recognized as so by the GSPCA driver. This camera, however, seems to have a number of quirks that cause the driver to decode the image incorrectly: the image is fed as YUV instead of Jpeg (like the VC0321), image start offset is different from what the driver expects for VC0323 or VC0321, and finally the sensor appears to be mounted upside-down inside the camera. Read on for fix details and GSPCA patches.

(more…)

On New Year’s Eve a loose heatsink fried my old Asus motherboard northbridge, so I replaced the entire system with a nice and inexpensive Dell Inspiron 530. It’s a G33/ICH9-based system, which runs almost out-of-the box with Linux 2.6.23, except for the ethernet controller (you’ll need an upgraded e1000 driver from Intel) and HDA sound (multichannel and headphone detection not working). Use the following ALSA patch to fix both sound problems:

• Hg patch for alsa-kernel: fixes 5.1 sound and headphone detection in the Dell Inspiron 530
• Patch for Linux 2.6.23.12: same patch as above, backported to Linux 2.6.23.12

It adds the ALC888 6stack-dell mode for the Inspiron 530 and hopefully for other Dell systems based on the ALC888. The patch has already been submited upstream and should show up in the next kernel version. The patch has also been added to the current Mandriva Cooker kernel.

The following patch allows the Syntek 174f:5212 camera found in some HP laptops (including the Compaq Presario C700 series) to be used in Linux with the linux-uvc driver. The image quality is fairly good, with decent colors, white balance and frame rate in luvcview. The patch is a bit intrusive since it requires quirk testing at the start of the isochronous payload frame decoding to completely bypass synchronization, but it seems to work very well. This patch is against revision 158 of the linux-uvc subversion repository.

• Add quirk for Syntek 174f:5212 UVC camera Wrong! Use Herton’s Laurent’s patch below.
• Add quirk for Syntek 174f:5212 UVC camera (fixed version)
• Laurent Pinchart’s final patch

Syntek camera framegrabs: 1, 2.

Update: Posted Herton Krzesinski’s fixed version. Posted Laurent Pinchart’s final version.

Following the development of the investigations on the recently reported Mesa problem triggered by Neverball and Metisse, Ademar reports that “the workaround is not effective with at least ATI 9250 video cards, where we now have a crash at a different place”. We set up a system with an ATI Radeon 9250, and, indeed, it still crashes:

Mesa: Mesa 7.0.1 DEBUG build Oct  1 2007 18:52:02
Mesa warning: couldn't open libtxc_dxtn.so, software DXTn compression/
decompression unavailable
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread -1218062640 (LWP 17221)]
_generic_read_RGBA_span_RGB565_MMX () at x86/read_rgba_span_x86.S:590
590             pushl   MASK_565_H

A segfault in a push instruction sounds very odd. Read on to see what the differential diagnosis session with Boto and Salem lead us into. And unlike the previous patch, this one resulted in a real fix for a real problem:

• read_RGBA_span_RGB565_MMX patch for Mesa 7.0.1

Don’t read the patch if you want to find the bug yourself based on the scenario description below.

(more…)

Ademar complained about certain programs such as Neverball and Metisse causing a crash in the X.org GLX code in systems with some Intel and ATI cards, specifically in a call to quadfunc_unfilled_rgba() when glPopMatrix() flushes vertices. After a quick inspection it was clear that the cause was a null pointer dereference in Mesa 7.0.1, so If you’re experiencing a similar problem you may want to try this trivial workaround:

• Add sanity check to quadfunc_unfilled_rgba

Further information and possibly a better fix available at the original bug report.

Less than two months ago xmp was dug out of its grave, and today I’m pleased to announce the availability of version 2.2.0 — the first stable release in more than six years (RPM packages for Mandriva 2008.0 are already available). This version fixes a truckload of bugs and adds support for many module formats, including some from Atari ST/Falcon, Acorn Archimedes and Apple IIgs. Other highlights are the MacOS X driver and the Audacious plugin.

There’s a lot of ugly things in the code that are still to be fixed, but this version is certainly much, much better than the previous stable. Let’s hope the next version doesn’t take nearly as much time to be released as did this one.

Update: Ok, so we have a patch already. This is needed to build the package in non-Linux systems, especially OSX.

Update 2: You may want to get xmp 2.2.1 instead.

These are mostly small quirk table additions to allow certain systems to be recognized by HDA driver without specifiying a parameter model when loading the module. The first three patches are for HP systems (Samba, Nettle, Lucknow, dx2200 and dx2250) and the fourth is for the Asus P5LD2 with the Realtek ALC882 codec:

• Add quirk for ALC888 3stack HP: for HP Samba (ECS Alhena5-GLB8)
• Use generic model names for ALC888-based HPs: for Nettle (ECS MCP61-PM) and Lucknow
• Add quirks for ALC862-based HPs: for dx2200 (MSI MS-7254) and dx2250 (MSI MS-7297)
• Add quirk for ASUS P5LD2

For more ALSA patches for HP systems including the Nettle and Lucknow, see my previous ALSA HDA patches. 5.1 sound output in Realtek-based HDA systems is quite easy to fix, so feel free to contact me if you have such a system that’s not working properly (usually no 6ch option available, channels missing or no sound from the rear connectors).

Owners of the LG LW20 laptop (like me) have had poor sound support in Linux for over a year and possibly never noticed that. I had never used its 5.1 surround support, recorded from the line-in input or used the si3054-based modem before this week — had I tried it earlier, I would have noticed that these features didn’t work. So to complement the patch to fix 5.1 output I published a few days ago, here’s a couple of small fixes for the rest of the aforementioned issues:

• Add LG LW20 si3054 modem id
• Add support for LG LW20 line-in sound recording

Hopefully these patches can also fix issues in the LG LW25, which seems similar in many aspects to the LW20. If you have issues with sound on the LW20 or LW25, feel free to contact me so we can fix any remaining problem.

Let’s hypothesize for a moment that you must make an old, binary-only Conexant HSF softmodem driver (that you, for some reason, can’t upgrade) peacefully coexist with a newer 2.6.22 kernel ALSA HDA driver. “Oh, what an embarrassing situation!” you might think — but I assure you, this can actually happen to some unfortunate souls among us. A workaround for this problem is twofold: first, you must graft the HSF appendages to your existing HDA driver (the HSF driver comes with its own HDA driver that potentially won’t have some features you need), and then resolve a recursion in module loading since 2.6.22 merged snd-hda-codec and snd-hda-intel in a single module and the binary-only culprit used to sit between the two.

So without further ado, here is what it takes to have this setup working:

1. Revert the hda-codec and hda-intel merge, to work around the HSF initialization recursion.
2. Apply Herton’s HSF patch against 2.6.22 so you can use the binary-only portion with the stock HDA driver from your kernel.

Generally speaking, binary-only modules can be a pain in the ass and this situation illustrates well how. Always verify driver availability before purchasing hardware, and that can save you some headaches.