Open Technology

Continuing the topic of slave DMA functionality on SuperH, using the shdma.c dmaengine driver, patches have been developed and submitted, adding DMA functionality to the SuperH sh-sci.c serial driver. I started by implementing the transmit direction, and even if that proved to be not very simple, there at least you know in advance the amount of data you're going to send. In the receive case, however, it becomes even more interesting, because we don't know the amount of data that's arriving, and the DMA controller on SH will only trigger an interrupt, when it has filled the receive buffer. So, unless you use a one-byte large buffer, you will not be informed of smaller than your buffer size transactions. So, eventually, I had to use an interrupt on the serial controller to detect the beginning of incoming data, and then use a timer to see, if fewer than buffer size bytes have arrived. In the latter case those bytes have to be collected from the DMA driver forcibly, for which there is currently no support in the dmaengine API. Therefore I had to extend and use the .device_terminate_all() method to retrieve the number of transferred bytes in the incomplete transaction. Review in process...

A number of patches for 2.6.34 - nothing major, support for VIDIOC_S_PARM / VIDIOC_G_PARM added to soc-camera, a couple of other minor improvements:

01/08: sh_mobile_ceu_camera: pass .set_parm and .get_parm down to subdevices
02/08: soc-camera: add support for VIDIOC_S_PARM and VIDIOC_G_PARM ioctls
03/08: soc-camera: return -ENODEV if sensor is missing
04/08: pxa_camera: remove init() callback
05/08: soc-camera: adjust coding style to match V4L preferences
06/08: soc-camera: ov772x: Modify buswidth control
07/08: tw9910: use TABs for indentation
08/08: soc-camera: update mt9v022 to take into account board signal routing

Here comes the long awaited documentation update:

01/01: v4l: document new Bayer and monochrome pixel formats

although, it looks like it still has problems with mainlining - something gets broken with it.

This was my first pull request since V4L transition from Mercurial to git. I am still not quite sure how to handle updating, branching, rebasing of git-trees on http://linuxtv.org/, but we'll figure it out with time. My preference would be to have just one tree on the server with two branches - fixes for the kernel, currently under development, and development - for the next kernel version, and to rebase those branches on the top of respective v4l trees. Still not sure whether this will work reasonably with the current v4l git setup. Locally I can set up a tree to trace the two v4l trees (fixes and development) as remotes, create branches from those remote tops and rebase them as needed. Then, if I just do a "git push <repository> HEAD:devel" or "git push <repository> HEAD:fixes" - would that work as expected? (comments welcome) "As expected" also presumes, that I won't be pushing all mainline commits up my ADSL link. Anyway, my first v4l git pull request happened to be on the fixes side and contained one patch:

01/01: soc-camera: mt9t112: modify exiting conditions from standby mode

Three fixes for 2.6.33 have been pushed upstream:

01/03: V4L/DVB mx1_camera: don't check platform_get_irq's return value against zero
02/03: V4L/DVB sh_mobile_ceu: don't check platform_get_irq's return value against zero
03/03: rj54n1cb0c: remove compiler warning

Currently in its second version, the following patchseries has been developed and submitted to relevant mailing lists. It adds ASoC (ALSA SoC) drivers for the Sound Interface Unit (SIU), found on several sh CPUs. The series also adds a driver for the WM8978 stereo audio codec from Wolfson Microelectronics, as well as board support for the sh7722-based Migo-R board. Drivers have been tested with stereo playback and capture with various sampling rates. They are based on earlier work from another author, as can be seen from driver headers, but drivers have been practically completely rewritten. The original driver contained built-in firmware, was a pure ALSA driver, used direct register accesses to configure DMA, pin functions, used physical addresses to access the hardware, used hard-coded register addresses. The new version is an ASoC driver, it uses the dmaengine interface for DMA functionality (see a previous post for slave DMA support for sh), firmware is loaded from the user-space using the standard hotplug functionality, it uses runtime pm and generic clock APIs to manage CPU clocks. Although currently only tested on sh7722, the same SIU hardware is also present on other chips, e.g., sh7343, sh7354, sh7367.

The current sh dmaengine driver supports only public DMA operations, i.e., memcpy offloading to the DMA controller. A patchseries has been submitted to add a DMA slave capability to the driver. This functionality allows the use of the DMA engine on sh processors for data transfers to and from on-board peripheral devices.

After a lengthy discussion, mediabus API patches have finally been approved and pushed upstream for the 2.6.33 kernel, along with a number of other patches, including a new driver for the mt9t112 camera sensor, multiple fixes and improvements, with this pull request:

01/30: tw9910: The driver can also handle revision 1 of the chip
02/30: soc-camera: remove no longer needed struct members
03/30: v4l: add new v4l2-subdev sensor operations, use g_skip_top_lines in soc-camera
04/30: soc-camera: fix multi-line comment coding style
05/30: sh_mobile_ceu_camera: do not mark host occupied, when adding a client fails
06/30: v4l: Add a 10-bit monochrome and missing 8- and 10-bit Bayer fourcc codes
07/30: soc-camera: add a private field to struct soc_camera_link
08/30: Subject: kernel-sync for arch/sh/boards/mach-ap325rxa/setup.c
09/30: soc-camera: switch drivers and platforms to use .priv in struct soc_camera_link
10/30: sh_mobile_ceu_camera: document the scaling and cropping algorithm
11/30: tw9910: Add revision control
12/30: tw9910: simplify chip ID calculation
13/30: tw9910: Tri-state pins when idle
14/30: tw9910: Add power control
15/30: tw9910: tw9910_set_hsync clean up
16/30: tw9910: Add revision control to tw9910_set_hsync
17/30: v4l: add a media-bus API for configuring v4l2 subdev pixel and frame formats
18/30: soc-camera: convert to the new mediabus API
19/30: Subject: kernel-sync for arch/sh/boards/mach-kfr2r09/setup.c
20/30: rj54n1cb0c: Add cropping, auto white balance, restrict sizes, add platform data
21/30: mt9t031: make the use of the soc-camera client API optional
22/30: sh_mobile_ceu: Add V4L2_FIELD_INTERLACED_BT/TB support
23/30: tw9910: use V4L2_FIELD_INTERLACED_BT
24/30: sh_mobile_ceu_camera: Add support for sync polarity selection
25/30: tw9910: modify V/H outpit pin setting to use VALID
26/30: tw9910: modify output format
27/30: tw9910: remove cropping
28/30: tw9910: Add sync polarity support
29/30: soc-camera: Add mt9t112 camera driver
30/30: sh_mobile_ceu_camera: Remove frame size page alignment

To remind, the mediabus API provides one of the two missing functionalities for the V4L subsystem, necessary to make v4l2-subdev client drivers reusable between soc-camera and other v4l2-subdev compliant bridge driver implementations. It handles data format negotiation on the "media bus" between host (bridge) and client drivers. The API is similar to the VIDIOC_{ENUM,G,S,TRY}_FMT user-space V4L2 API. The second missing functionality is the (media) bus parameter negotiation API. But even without it soc-camera client drivers can now be made usable for other bridge drivers in a limited way.

Three more fixes for 2.6.32:

01/03: v4l: add more missing linux/sched.h includes
02/03: soc-camera: properly initialise the device object when reusing
03/03: soc-camera: sh_mobile_ceu_camera: call pm_runtime_disable

It was my first attendance of an Embedded Linux Conference (ELC), which I generally would summarise as quite interesting and informative. However, comparing to FOSDEM, the latter hosts a much larger number of simultaneously held sessions, as a result, its usefulness for individual attendants might well be higher. FOSDEM talks are also much shorter - 20 to 40 minutes in comparison to 50 minutes at ELC-E. It must, however, be noted, that CELF holds four conferences per year, of which, the one held in the USA is, supposedly, the largest. In any case, I would like to thank CELF for this opportunity and wish them success with future conferences and, of course, more and better talks:-)

The conference took place in Grenoble in the World Trade Center and was co-located (in space and time) with Embedded Systems Week (11-16 October) and Embedded Systems Exhibition (14-15 October).

On a personal side, it was also my first visit to Grenoble, and I definitely enjoyed the nice environment, the marvellous outlook on French Alps, the hospitality of the town, patiently and forgivingly reacting to my degraded over years of non-use to an almost non-existence degree French, well functioning transport, accommodation and catering infrastructures, which seamlessly extended
their services to hundreds of conference attendants.

General information about the conference is available at CELF website at http://www.embeddedlinuxconference.com/elc_europe09/ and a list of talks with presentation slides at http://tree.celinuxforum.org/CelfPubWiki/ELCEurope2009Presentations.

Attended talks:

15.10.2009: Day 1

Keynote: Jon Masters, Porting Linux

The main purpose of the talk was to give some advise on porting Linux to new platforms to relative and absolute beginners in this area. Jon pointed out at the recent work of Arnd Bergmann on generic asm, which makes porting much easier, at usefulness of ftrace, kprobe, ksplice and perf kernel interfaces. Apart from being a kernel- and user-space Linux developer, Jon writes books and publishes daily podcasts, summarising Linux Kernel Mailing List traffic, which is a substantial work in itself. The talk was held in a live manner, on a high professional level, and left a solid positive impression. Slides available.

Pascal Pellet, Linux Embedded applications in Machine Vision

The presenter has described a specialised Linux-based machine vision application, in which a camera sensor is connected to an FPGA, which is then connected to an ARM9 SoC from Atmel. The system communicates with hosts over ethernet, the whole system implements the GigEVision protocol. The Control part of the protocol is handled by the ARM CPU, whereas the Streaming data is pushed by the FPGA from the camera directly into the ethernet controller, completely bypassing the OS and the CPU.

Guennadi Liakhovetski, Embedded video capture under Linux: the soc-camera framework (given)

I talked about the soc-camera Linux kernel framework, its current functionality and supported hardware, and gave an overview of future work. Slides available.

Matt Porter, Mythbusters: Android

Quite an interesting and informative talk from an Android developer. What would otherwise take me some time and effort to find out, has been readily presented and summarised. What, supposedly, all closely working with Android would know, and what was news to me, is the fact, that although Android uses the Linux kernel, and can therefore be called "Linux," it is not a GNU/Linux, because it doesn't the GNU glibc, nor it is compatible with it, like Linux variants, based on uClibc or eglibc. Android uses an own, developed by Google version of libc, called Bionic, which has been originally derived from a BSD libc, and is very incompatible with all other libc versions, used by popular Linux distributions. This incompatibility makes porting programs between traditional Linux and Android a complicated and time-consuming task. This talk has won the first prize. Slides available.

Stefan Schwarzer, Disko v1.6 - an application framework for embedded devices

This topic is pretty far from my area of interest, so, I will not be able to competently comment on it. A general impression was, that Disko is a "yet another" graphics library for Linux, developed by a commercial company, and enjoying a limited but active customer base. It is being actively developed, however, it seems its development is driven by customer requirements, which might lead to a development time-line, lacking a well thought-through concept and strategy. However, it seems to be growing well, adding new features, and trying to compete with other popular solutions like Qt, GTK+, making the embedded market its only focus. Watch out for slides.

Tim Bird, Analyzing Kernel Function Execution with ftrace

Tim Bird has extended the ftrace functionality with a new tracer, measuring function duration. The patch set is pretty elaborate, has been submitted to LKML, its new version is expected to eventually make it into the mainline. The initial implementation has been done for ARM, where it has to modify the kernel stack layout - exciting stuff:-) He has also written user-space tools to assist in the analysis and interpretation of collected traces. Slides available.

Marcin Juszkiewicz, Hacking with OpenEmbedded

Marcin gave an overview of the OpenEmbedded project and its main use cases. He described, how OpenEmbedded can be used for cross-building complete package-based embedded distributions, separate packages for OE, Maemo-4, SharpROM and OpenWRT systems, as well as native packages for the host. Slides available.

Pierre Pronchery, Hackable Devices (BoF)

Pierre described ideas behind the Hackable Devices project, presented a new soon-to-come project web-portal, and further asked participants to express their opinions on the subject of open hardware. One of the examples of such hardware designs, discussed during this BoF, were the famous Neo 1973 and Neo FreeRunner OpenMoko mobile phones.

Micheal Opdenacker, Small Business (BoF)

This BoF collected owners and employers of small companies, specialising in embedded Linux software development and services. Following a short introduction from Michael, he suggested, that participants shortly introduce themselves, and then answer some of his questions, concerning typical problems, that small businesses in this area often face. A live conversation followed. It has been further decided to open a mailing list, on which such problems can be discussed. It is expected, that such meetings will become regular at CELF conferences. On the whole it seems, there is indeed a lot, that small open source businesses can learn from each other and with what they can otherwise benefit from each other.

Dinner / social event

Day one of the conference ended with a dinner in La Bastille de Grenoble. It is located on the top of a mountain on the other shore of the river IsÚre. To get there we took Téléphérique de Grenoble Bastille. Just Google for any of those to get an idea what that is:-) As for personal impressions - wonderful view (even at night) on the city, very pleasant location, good food and wine, and, most importantly, of course, a great chance for informal communication!

16.10.2009: Day 2

Keynote: Philippe Gerum, State of real-time Linux: Don't stop until history follows

Philippe is the author and maintainer of Xenomai - a "Real-Time Framework for Linux." In this keynote he described the current state of his project, its main features and advantages over alternative soft- and hard-real-time extensions to the Linux kernel.

Nicolas Palix, Julia Lawall, Gilles Muller, Coccinelle: A Program Matching and Transformation Tool for Systems Code

Coccinelle is a tool to automatically analyse and rewrite C code. Regular readers of the Linux kernel mailing list certainly got used to patches from Julia Lawall (maybe also from others, not sure), like this one, most of us have also learned to appreciate these patches;) It was indeed quite interesting to learn about the technology behind them. Slides available.

Frank Rowand, A Survey of Linux Measurement and Diagnostics Tools

Frank made a full use of the large screen, made available to him in the largest of three rooms, used by the conference. He listed a few tools, useful in diagnosing Linux performance - new ones as well as UNIX classical ones. He showed their typical usage scenarios, their strong and weak sides, example output dumps and briefly explained their meaning. Needless to explain the importance of performance measurement tools for any Operating System, even though any web search engine readily delivers dozens of hits, it is very helpful to have a concise comparison of most important such tools, composed by a professional with a good understanding of the topic. Slides available.

Samuel Ortiz, Linux wifi solutions for mobile platforms

Samuel is the current maintainer of the IrDA and MFD subsystems in the Linux kernel and a significant contributor to various other areas of the Linux kernel, including the mac80211 subsystem and wireless network drivers. In his talk he concentrated on embedded aspects of wireless networking, specifically on power-saving techniques, currently available to Linux wifi drivers. Watch out for slides.

Vitaly Wool, Using device trees on ARM platforms

Flat Device Trees (FDT) have relatively recently became a standard way to pass platform-specific device configuration from the bootloader to the kernel on embedded PowerPC devices. It is based on the data representation specified in the IEEE 1275-1994 Open Firmware standard. Open Firmware has been used since a long time on Apple PowerPC Macintosh, IBM POWER, Sun Sparc and some other systems. FDT also brought this to the embedded PPC world. More recently further advances has been made in this area, unifying PPC and Sparc FDT implementations. Since the introduction of FDT the idea has been brought forward multiple times to also use it on other architectures, notably, on ARM. The last attempt to do this has been undertaken this year and it has been accompanied by a series of patches and a lengthy but this time mostly technical discussion on Linux ARM mailing lists. The outcome has been negative for FDT - it didn't make it in for ARM this time too. This talk looked at requirements of a typical ARM boot process, i.e., what information a successful FDT implementation for ARM would have to provide, and why and how the current mainline FDT interfaces are not suitable for this. Vitaly also considered whether using FDT can at all be feasible on ARM Linux, and what would have to be done to achieve this. Slides available.

Francesco Virlinzi, A Generic Clock Framework implementation (first 5 minutes)

Currently support for clock devices is platform-specific in Linux, which, for example, makes dynamically exporting clock devices from platform-independent drivers impossible. This talk presented yet another attempt to introduce a platform-neutral clock API, but after the first 5 minutes I decided, that the solution is way too raw to be seriously considered. The problem is quite complex, even though a clock doesn't seem to be a very sophisticated device at a first glance. What makes this topic so complicated is, that clocks form tree-like hierarchies, and have to be linked to devices, that use them. All this also makes power management for clock instances non-trivial. To be accepted into the kernel any such patch-set first has to go through its paces on the LKML and arch-specific mailing lists, for which this specific attempt didn't seem quite prepared yet. Slides available.

Michael Opdenacker, Update on boot time reduction techniques

The presentation walked the audience through a number of standard and not so well known boot time reduction methodologies on a real life example. After that the reporter invited listeners to talk about their ways to speed up Linux boot process, which brought up a couple more interesting possibilities.

Closing session: Tim Bird & Klaas van Gend, The Butterfly effect of CELF

Closing the conference, the two reporters talked about CELF and its activities. Parallel to this the audience has been invited to take part in a highly intellectual (:-)) game with winners awarded with nice prizes, ranging from chocolate to a home multimedia server.

ELC-E 2009 is now over. My talk was quite well received, there have been a few questions too. I've uploaded my slides to the CELF site, feel free to have a look and comment either directly to me per mail, or to the linux-media list. Apart from internal soc-camera issues I also briefly went through a couple of recent RFCs, that will be or, eventually, might become relevant for soc-camera.

The following 6 patches

01/06: sh_mobile_ceu: add soft reset function
02/06: sh_mobile_ceu_camera: add VBP error support
03/06: sh_mobile_ceu_camera: fix cropping for scaling clients
04/06: soc-camera: add a new driver for the RJ54N1CB0C camera sensor from Sharp
05/06: pxa_camera: fix camera pixel format configuration
06/06: fix use-after-free Oops, resulting from a driver-core API change

have been pushed for a late 2.6.32 inclusion.

The following two patches

01/02: V4L2: Add a v4l2-subdev (soc-camera) driver for OmniVision OV9640 sensor
02/02: sh_mobile_ceu_camera: fix compile breakage, caused by a bad merge

pushed upstream.

A patch

01/01: SH: fix ap325rxa compilation breakage

fixing a typo in an earlier commit, has been pushed to v4l-dvb.

A patch

01/01: soc-camera: remove now unneeded subdevice group ID assignments

which also fixes a compiler warning, has been pushed to v4l-dvb.