Performances of open-source Radeon driver

I am the happy owner of a new netbook with an AMD Fusion E-450 APU, which includes a Radeon graphics card. I am using the open-source driver on it, that is a 3.2-rc7 kernel for KMS, and xserver-xorg-video-radeon package from sid. I have to say I am not really happy about the performances.

No I don't speak about the graphical performances that are pretty good (especially compared to my Intel Atom N450 based previous netbook) but about the power consumption. With this setup and with the original battery I get 2h30 of autonomy. Switching to UMS and adding some power management options in xorg.conf improves it to 2h40, but breaks suspend to ram/disk (a pity for a netbook) and switch between VT. I then tried the non-free fglrx driver, it also suffers from the suspend to ram/disk issue, in addition to crashing xorg when playing videos... On the other hand I get an impressive 3h30 of autonomy, and additionally a silent netbook (contrary to the open-source driver, the fan doesn't spin at idle).

I have tried plenty of options, ranging from adding some power management options to xorg.conf, to passing dynclks=1 to the radeon module, including setting /sys/class/drm/card0/device/power\_method to dynpm. Right now I have worked around the issue by buying a bigger battery which brings me 5h30 of autonomy, but I would really appreciate any software way to improve it with the open-source driver.

Debian s390x port (aka 31 bits is not enough)

During Debconf 11, I got access to a fast s390 machine, and I have started to work on a Debian s390x port, the 64-bit version of the s390 port. One of my goal was to help the SPARC64 port, as some of the issues are the same: both are 64-bit big-endian, don't support unaligned access and behave differently between -fpic and -fPIC.

Why such a port?

When talking about 64-bit ports, we usually hear: "4GB is enough, handling 64-bit takes more memory". This really sounds like "640K ought to be enough for anybody". The s390 port is actually 31-bit from the address point of view (one bit is reserved for address space extension from 24 to 31 bits), so each process is limited to 2GB only. Nowadays applications which need more than 2GB are not that uncommon, especially on mainframes. Actually the 2GB limit already causes some problem in Debian: in some cases it's not possible to build haskell applications or even C applications using GCC. On the other hand, we already require a 64-bit kernel on the s390 port (only the userland is 32-bit), and applications are handling more and more 64-bit or greater values (files offset, time counters, uid, etc.).

What is the status?

Bootstrapping the architecture was not really easy (as for any other new architectures), due to a huge amount of dependencies and build-dependencies loops, as explained by Wookey during Debconf11. Now that this part is mostly done, an autobuilder has been started and currently more than 65% of the packages are built. The s390x port is hosted on debian-ports.org. Unfortunately it is not yet deboostrapable, though that should happen in the next few days (only a few packages are missing).

The main issues are currently packages which fail to build from source due to linker, gcc-4.6 and curl changes, or due to the libjpeg and multiarch transitions, and thus are not directly related to s390x. If your package is in this case, it would be a good idea to fix it. Otherwise if it has a lot of reverse dependencies and the bug is opened for a while, just expect an NMU (as allowed by the 0-day NMU policy). Of course for a few packages s390x specific fixes are needed, some of them are already in the BTS.

How can you help?

The list of bugs blocking the s390x port is available through the s390x usertag, fixing these bugs (a lot of them are general FTBFS) would help a lot. Alternatively if you have access to an s390x machine you can take a look at the packages failing to build.

Update: Fixed the explanation about the 32th bit, thanks to Bastian Blank for the comment.

Debian and the ARM hype

Thanks to the versatility of the Linux kernel, Debian has always been known for supporting a large number of architectures. It has also often been criticized for that as it is said to slow down the development of Debian.

Among these architectures, the ARM one was considered dead a few years ago, and some people wanted to get rid of it. Today all major distributions now have an ARM port, one of those distributions being even based on Debian. It seems Debian was right.

Now that Android has been ported to MIPS, we may see more and more MIPS based devices. Will the same scenario happen again?

bindv6only=1 and GNU/kFreeBSD

A few months ago the netbase package started to install the /etc/sysctl.d/bindv6only.conf file to switch the default bindv6only value from 0 to 1.

A lot of people are not happy with this change, but it is not my goal to give my opinion here. On the other hand, people have propagated the rumour that it has been done as the FreeBSD kernel, used in the kfreebsd-amd64 and kfreebsd-i386 ports of Debian, only supports the mode corresponding to 1.

Let's give the truth:

  • GNU/kFreeBSD people haven't been contacted about this decision;
  • The FreeBSD kernel can support both modes through the net.inet6.ip6.v6only sysctl. However contrary to the Linux kernel it defaults to 1;
  • This option is available in the FreeBSD kernel since 2001, and in the Linux kernel since 2003.

EGLIBC and PowerPCSPE port

This has been roughly one year since Debian switched from GLIBC to EGLIBC, so it's probably the time to make a small report about this change.

First of all, on the GLIBC upstream side, things has improved a bit since we now have regular stable release, thanks to Petr Baudis aka Pasky. The good point is that the stable releases are imported into the EGLIBC stable repositories.

On the EGLIBC side the switch has helped to reduce the number of patches in the Debian package (for example, resolv.conf is automatically reloaded if needed), and has brought some bug fixes and improvements, especially for the arm, mips and powerpc targets.

It should be noted that the newly created PowerPCSPE port for PowerPC e500 series CPU also benefits from EGLIBC, as it is not natively supported by GLIBC.

Debian SH4 QEMU image available

Thanks to the great work of Nobuhiro Iwamatsu and others, Debian has an unofficial SH4 port which is close to complete (> 90% of the packages built).

In order to help developers to fix bugs on this architecture, I have produced an SH4 QEMU image which is available at the same location as my other QEMU images.

You will need a recent GIT HEAD QEMU to use it. Previous versions suffer from bugs in the MMU, causing segfaults and gratuitous TLB flushing. The MMU emulation is now hopefully correct, but still a bit slow. Also the emulated board is limited to 64 MB of memory, and this value can't be changed as memory extension would overlap the addresses used for peripherals.

Update: I have backported the necessary SH4 patches into the QEMU Debian package version 0.12.3+dfsg-4.

Squeeze will be released with eglibc 2.11

Contrary to what lucas announced (I don't know where he got this info), we plan to release Squeeze with eglibc 2.11. It is already packaged in experimental and is ready on all architectures except hppa where there are a few major regressions in the testsuite to fix. This is what prevent us to upload it to unstable.

Working on the eglibc package

In the last weeks, I stopped being motivated to work on the eglibc package, it's not fun as it was before. Maintaining this package is taking a lot of my (free) time, and I am not anymore able to follow the bug rate, especially for RC bugs or bugs that I consider high priority. In turn it does not give me time to integrate eglibc 2.11 or other wishlist features I would have liked to see (rework of the locales* packages, using multiarch paths, etc.).

I hope it's only a bad moment and that things will change soon, so I can find time to work on GNU/kFreeBSD, QEMU or to do electronics. In any case you can help by handling bug reports or writing patches. Everything is in the BTS!

Thought of the day

New features usually come with new versions. Before reporting a bug for a new feature, it may be a good idea to make sure you are using the latest version. apt-get can be really useful for that.

Learning IA-64 assembly

While testing EGLIBC 2.10.1 on all Debian architectures, I have discovered that the testsuite on IA-64 fails for the new POSIX 2008 math tests. I have reported the problem both upstream and on debian-ia64@lists.debian.org, but without success.

IA-64 being one of the last architecture (with HPPA) on which EGLIBC 2.10.1 fails to build from sources, I have decided to spend my day fixing the problem and digging into the corresponding IA-64 assembly code. The mathematical functions on this architecture are based on the Highly Optimized Mathematical Functions for the Intel® Itanium™ Architecture. Using the Intel® Itanium™ Architecture Software Developer's Manual and after a lot of tries, I have been able to add the missing code paths needed for POSIX 2008 compliance, and also fixed a few bugs on the stack frame allocation (arguments of the alloc instruction).

The resulting patch is now in the pkg-glibc SVN and in the upstream bugzilla.

And more important, I have learned the basic about IA-64 assembly!