Messages

136

General OpenPOWER Discussion / critical mass of developers

« on: November 07, 2020, 03:58:56 am »

Looking at my recent experiences with some packages such as Thunderbird sqlite, Gimp (linker), OBS Studio (can't remember what I tweaked but it builds a package now) and even making my own kernel and installer ISO with 4K page size, none of these things were too hard to resolve or work around but on the other hand, it becomes tedious if a developer encounters stuff like that every day.

With Thunderbird specifically, unless it is fixed upstream, I have to spend some time adapting my patch to each new version of the package.  It is not hard work, it may only take 30 minutes but it is tedious and takes energy away from new development.

Hopefully when Vikings make their products available in Europe, some more developers will decide to try the platform for their next workstation and this will distribute the burden more widely.

It is important to have some reach into other ecosystems, for example, if there were two or three developers in each big upstream community like Mozilla, GNOME and KDE/Qt using this platform, they would notice any issues much more quickly and issues would be fixed centrally, before anybody tries to release or package new versions of those products.  Does anybody already know about interest in any of those communities?

For OS level development it can be really useful for developers to have multiple workstations, for example, one as a stable workstation and a second machine that we can reboot multiple times per day to test kernels and hardware.  Only having one machine right now, I tend to avoid testing anything that might require a reboot because I don't want to close and re-open all the different applications that I'm using.

If anybody has any old hardware that they wanted to donate to developers or if anybody knows of any funding to subsidize hardware for developers in different projects that could help the platform gain critical mass.  As with Metcalfe's law in telecoms, each extra developer adds more value than their personal contribution.  Any other ideas would be welcome too.

137

Applications and Porting / Re: Thunderbird not working Debian buster (1:68.12.0-1~deb10u1 & 1:78.4.0-1~deb10u1)

« on: November 07, 2020, 03:33:42 am »

Here are some details about my investigation into this issue, I finally got it working by manually patching and rebuilding the package version 68.12 on Debian buster.

It may impact Fedora and other distributions too.

I started with an empty profile and press CTRL-SHIFT-J to bring up the error console.

There were a range of errors, here are some of them:

Code: [Select]

Component returned failure code: 0x80520001 (NS_ERROR_FILE_UNRECOGNIZED_PATH)

There are many search results for that.

Code: [Select]

thunderbird Uncaught (in promise) TypeError "this.asyncConnection" is nullcreateNounTable

Searching for that one immediately found a ppc64el-specific bug report in Ubuntu (bug 1845929)

The bug report includes a patch against Thunderbird 68.1.1, they configure Thunderbird to use the system sqlite3 library instead of the bundled version in the Mozilla source tree.

Looking at the debian/changelog (buster) for the Debian buster build, I found that the patch had been reverted at some point after that:

Code: [Select]

thunderbird (1:68.2.2-1~deb10u1) stable-security; urgency=medium

  * Rebuild for buster-security
  * [2c1bd00] d/mozconfig.default: use internal version of
    nspr, nss, sqlite and icu
  * [94d6ae4] d/control: remove lib{nspr4,nss3,sqlite3}-dev from B-D

Further on, around Thunderbird v76 packaging, I found they removed more of the system-sqlite stuff, there is a changelog comment suggesting that the Thunderbird internal sqlite codebase is now a fork.  The patch can no longer be used because the configure option --enable-system-sqlite no longer exists.  Base on my experience with Thunderbird v78, that fork is not working at all on ppc64el.  From the debian/changelog (sid / unstable):

Code: [Select]

thunderbird (1:76.0~b2-1) experimental; urgency=medium

  * [8386db0] d/control: Remove B-D on libjson-dev and libsqlite3-dev
    The built uses internal copies for libjson and libsqlite as there are
    made modifications to them. For now we can decrease the list of build
    dependencies by removing this two packages.

Going forward, I believe somebody will need to create unit tests for exercising the Mozilla fork of sqlite during builds on ppc64el hosts.  Debian and other distributions automatically run the unit tests on every architecture when creating the packages.

I took the Debian buster packaging branch, forked it at v68.12 and re-enabled the system-sqlite option.  This immediately fixed the problem I was experiencing.  For now I have a usable Thunderbird package on buster.

Some additional references I looked at suggested deleting some or all of the following may help people having problems with things not appearing in the folder pane after upgrading or migrating the profile to a different host/CPU architecture.  I personally deleted the cache but didn't need to delete the others:

Code: [Select]

~/.cache/thunderbird
~/.thunderbird/..../folderTree.json
~/.thunderbird/..../session.json


138

Applications and Porting / gimp fails load error: libgomp.so.1: cannot allocate memory in static TLS block

« on: November 06, 2020, 02:04:35 pm »

On Debian buster, when I try to run gimp, it fails like this:

Code: [Select]

GEGL-Message: 20:50:18.383: Module '/usr/lib/powerpc64le-linux-gnu/gegl-0.4/ff-save.so' load error: /lib/powerpc64le-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block
GEGL-Message: 20:50:18.388: Module '/usr/lib/powerpc64le-linux-gnu/gegl-0.4/matting-levin.so' load error: /lib/powerpc64le-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block
GEGL-Message: 20:50:18.390: Module '/usr/lib/powerpc64le-linux-gnu/gegl-0.4/raw-load.so' load error: /lib/powerpc64le-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block
GEGL-Warning: Module '/usr/lib/powerpc64le-linux-gnu/gegl-0.4/ff-load.so' load error: /lib/powerpc64le-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block


(gimp:39890): GLib-GObject-WARNING **: 20:50:18.483: Fatal error - Could not reload previously loaded plugin '(unknown)'

As a workaround, it works if I run it with this command line:

Code: [Select]

$ LD_PRELOAD=libgomp.so.1 gimp

There is some discussion about this particular issue here in the context of Anaconda, it may need to be adapted into a bug report against Gimp.

139

Applications and Porting / Thunderbird not working Debian buster (1:68.12.0-1~deb10u1 & 1:78.4.0-1~deb10u1)

« on: November 06, 2020, 01:43:15 pm »

When I started using my Talos II system, I moved my $HOME into Debian buster with Thunderbird 1:68.12.0-1~deb10u1

Launching Thunderbird, I found that all the panes (folder list, message list, message preview) were empty and toolbar appears mildly corrupt

I thought I would just wait for Debian to package a newer version.

Thunderbird 1:78.4.0-1~deb10u1 is now available through the Debian buster security updates so I tried that today.

The same problem occurred, all panes empty

I looked for information online and found a few suggestions but these things didn't make any difference:

- running Thunderbird in safe mode

- running Thunderbird in safe mode with all add-ons disabled and selecting the option to reset the toolbar

- running Thunderbird in safe mode to create a new profile

When I try to start in a new profile, I notice that the wizard to create a new mail account never appears.  Trying to access the accounts window from the menu, the window doesn't appear either.

Has anybody else had problems like this on Debian or any other distribution?

Are there any tweaks in the Fedora build of Thunderbird that may be necessary for the Debian build?

140

General OpenPOWER Discussion / Re: European logistics / delivery, warranty returns, etc

« on: November 05, 2020, 06:45:14 am »

This is definitely interesting news.  Will you provide servers or only workstations?

Can people order the coolers now?

Will you provide any installation media to help people get around the 64k page size problem?  I built my own Debian installer image with a 4k kernel but not everybody will be able to do that.  I can share my ISO image if that is helpful.  I didn't build one for Fedora 33 yet.

There are some other outstanding issues that will be irritating for workstation users, I've recently seen glitches in Gimp, Blender, Thunderbird and gstreamer.  The complete absence of Qt WebEngine is a hassle because many things need it but on a positive front, I suspect most or all of those things will immediately start working after Qt WebEngine is fixed.

My overall impression is that none of these issues require months of work, most of them only require a few hours or days of developer time but apart from a few IBM bounties on BountySource (e.g. ffmpeg), nobody has stepped forward and offered funding for this work.

I don't write all this to complain, I write this because I know you put a lot of effort into the European distribution and it is really important that the people who buy in have a good first impression.  Even if there is a roadmap with tentative dates for fixing the issues in this list it will be very reassuring to people.

I'm personally quite busy until the end of the year but I may have a couple of weeks in January or February that I could dedicate to full time development on this platform if there is any funding available.

141

Blackbird / Re: [amdgpu] [Fiji] Fedora 32 Linux kernel 5.7.x crashes

« on: October 24, 2020, 03:13:22 pm »

Does anybody have any idea what this means for Big Navi?

As they already merged patches into the kernel, should it just work out of the box?

The Big Navi launch is supposed to be on Wednesday, 28 October and we can potentially buy the cards in November.

142

Talos II / Re: Using IPMI Tool

« on: October 23, 2020, 10:19:37 am »

I regularly use it, for example, to check fan speeds

Code: [Select]

sudo ipmitool sensor | grep RPM


ipmitool version 1.8.18-6 on Debian buster


There is an existing Perl script for Ganglia integration too:

https://www.gnu.org/software/freeipmi/ganglia_ipmi_sensors.pl


I was thinking about extending ganglia-modules-linux to report data from IPMI, it could be written more efficiently as a C plugin.

143

General Discussion / Re: Fractal Design cases

« on: October 08, 2020, 04:47:31 pm »

As discussed elsewhere, the PWM fan described here is now running at 1800 RPM.  The system is actually idle so this feels a bit odd.

It was 300 RPM when the system booted and it increased gradually over time.

From the perspective of the Fractal case, the default Fractal fans are rated for 1000 RPM.  Given a PWM fan, the system seems to prefer a lot more than 1000 RPM and that implies the default Fractal fans are not sufficient for a multi-CPU build like this.  They may still be suitable for somebody with single CPU, e.g. Talos II Lite with 4 or 8 core.

144

Mod Zone / is Talos II overclocking the fan?

« on: October 08, 2020, 04:42:59 pm »

ipmitool tells me that the PWM fan I put in the rear of my case is running at 1800 RPM

Noctua rates it for 1500 RPM +/- 10%, that suggests that 1650 RPM is the maximum

PWM, by definition, can't overclock or overvoltage anything, it simply increases the pulse width up to the point where pulses are 100%.  You can't PWM over 100%

Is the fan simply running outside the spec or could this be a reporting error?

145

General CPU Discussion / Re: POWER9 water cooling kit

« on: October 08, 2020, 03:44:06 pm »

At risk of going off-topic, I'd like to emphasize that water-cooling is only 50% of the problem when house-training the POWER9

The thread about powering-down unused cores is equally important

The CPU cores, especially in a dual-CPU configuration, using over 100W when idle, will waste electricity and generate unwanted heat.

Electric heating is particularly inefficient, 60-70% of electricity is lost in the energy network.  The only exception to this is the use of a local generator (renewable, cogeneration, etc), as it is generated locally, very little is lost in transmission lines.

In my own case, I'm using solar PV and the excess heat is therefore useful and efficient for 7-8 months of the year, it simply offsets the underfloor heating.

But for most people this could be expensive, wasteful and uncomfortable.

Solving both of these issues together (the water cooling and making some cores sleep when idle) will eliminate most concerns from buyers.

146

GPU Compute / Accelerators / Sapphire Pulse Radeon RX 550 4GB

« on: October 08, 2020, 03:20:49 pm »

The RX 580 was running fine but I don't use a lot of GPU so I decided to swap it for the Sapphire Pulse Radeon RX 550 4GB, it is half the price.

This card appears to be running fine too.

Monitoring it with the radeontop utility, I'm using 2.5GB of VRAM while driving two screens, one is 4K and the other is 1920x1200

As it is really cheap, I think this card is useful as a spare or for people who don't use GPU-intensive workloads.

The card is less than 2 PCIe slots width so it is an interesting choice for people who are using both slots on a Blackbird.

The sensors command tells me the fan is running at 1231 RPM idle, it is a bit faster than the fans on the RX 580.  I feel it may make just a little bit more noise compared to an idle RX 580 which has two fans that appear to be bigger and quieter.

I may decide to return this one as well and swap it for the Big Navi in November or maybe I will get a Big Navi and keep this as a spare.

My only real concern then is the fan noise, some people may want to pay that little bit more to get a more silent alternative.

147

General Discussion / Re: Fractal Design cases

« on: October 08, 2020, 03:02:58 pm »

I ran ipmitool and I didn't see any fan speed from the built-in fan hub of the Fractal case.

I decided to try replacing one of the default 3-pin case fans with a 4-pin PWM Noctua fan

If the fan is at the rear of the case, the 30cm cable from Noctua doesn't quite reach the fan connectors on the front of the motherboard.

The fan hub has some 4-pin connectors and 3-pin connectors.  It appears that it takes the speed sensor from the first 4-pin connector and passes that back to the motherboard.  Connecting the fan to the first 4-pin connector, I was able to see the speed with ipmitool.  I tried moving it to each of the 4-pin connectors and they all report 0, it only reports the speed if it is on the first connector.

Using the fan hub in this way also solves the 30cm problem because the fan hub is at the back of the case too.

At idle, I don't believe the system is significantly quieter using the Noctua fan.  ipmitool reports that the fan is running at its slowest speed, 300 RPM.  I suspect that any noise I can hear is coming from other parts of the system and not the Noctua fan.

Overall, it is convenient being able to see the fan speed like this but maybe I could see the fan speed by using a more advanced fan hub that can monitor the 3-pin case fans.

148

General CPU Discussion / Re: POWER9 water cooling kit

« on: October 07, 2020, 08:47:43 am »

I found the IBM HSFs quiet enough when idle as well, for everything else, like office or living room situations, those HSFs are unusable IMO.

I agree a lot of people will have trouble with the HSFs

Another thing that comes to my mind is the performance of the case fans: if somebody has some very large and quiet case fans, e.g. 140mm and they move a lot of air across the HSF will the motherboard fan logic be able to reduce the HSF speeds?

Looking at the Noctua fans, for example, they are offering fans with different maximum speeds, for example, 800, 1200, 1500, I even saw one with 3000.  They will be noisy as well but will the overall noise be lower?

149

Applications and Porting / Re: Debian packaging and backport requests for OpenPOWER, Talos II, Blackbird?

« on: October 07, 2020, 06:58:11 am »

For qtwebengine, it is very easy to enable ppc64el in the Debian control file but it won't work until it is enabled in QtWebEngine itself

As commented elsewhere, I was able to build OBS, I'll test it and make the packages available shortly.

150

Applications and Porting / QtWebEngine status (depends on Chrome / Chromium patches)

« on: October 07, 2020, 06:56:35 am »

A lot of interesting things depend on QtWebEngine

QtWebEngine depends on Chromium, they periodically take a snapshot of the Chromium code from Google

Chromium patching is already in progress, is it complete?

I opened a bug for it within QtWebEngine

When somebody confirms the Chrome patches are ready, the QtWebEngine developers could enable it in their list of supported platforms.  They will probably need to know the minimum supported version of the Chromium code, they probably won't have time to look through the Chromium history, if anybody could add comments to the Qt bug report that would be really helpful.

If they can cherry-pick patches that haven't been accepted by Google, please also comment on that Qt bug report