Raspberry Pi 4

Can anybody tell me if there is or will be a version of Risc OS for the new Pi 4?

Thanks

In time…

So much for the “no more Broadcom CPU updates after RPi 3B+, the BCM is end-of-the-line” stories…a pleasant surprise I guess, although I have not yet looked into the details of Cortex-A72 and 32bit compatibility…

Cortex A72 runs 32bit code just fine.

Yes, the core is fine, but the rest of the chip is all different – see https://gitlab.riscosopen.org/RiscOS/Sources/HAL/HAL_BCM2835/merge_requests/2

The rest of the board is different too – USB-C for power, different HDMI sockets and change in overall layout of the various connection ports.

I don’t mind all the different connectors, as long as the software works and talks to my keyboard, mouse, network, monitor & hard drive/SSD.

Would I be correct in thinking that a working USB 2.0 would give a basic usability (mouse and keyboard)

Followed, perhaps by USB external network adapter and SSD.

Or, is there a lot more to it than that?

Does the SDFS still work the same way? (not that I like SDcards!)

So would a working USB3…? I think the issue is not that it’s 2 or 3, but that the controller is actually different. The older Pi models were all extensions of the same basic idea. This one is….different.

Might be that the ethernet is largely supported? It depends on how similar it is to the OMAP way of doing things (if I recall, the OMAP3 had hardware Ethernet separate from USB, didn’t it?) Again, lack of documentation is (always) a problem.
Edit: totally different beast.

Video is likely to need a rethink, with dual 4K capable outputs, it’s certainly not going to be the same GPU. How is it handled? Two GPUs? One with two framebuffers?

Also, since this is a lot of questions awaiting answers – does the Pi 4 still follow the slightly backwards principle of the GPU being the one in control with the ARM being almost a co-processor? Have they fixed IIC yet? :-p
Edit: no, looks like it’s ARM in charge with the GPU acting as a GPU. And yes, clock stretching should work now. Finally.

(if I recall, the OMAP3 had hardware Ethernet separate from USB, didn’t it?)

The manual for the BeagleBoard -xM has ‘Ethernet 10/100 from USB HUB’.

Thanks Fred. My mistake. USB on the OMAP3 is implemented as a PHY (OSI layer terminology) rather than “here’s a USB port”, and I got mixed up as I’m more used to seeing that attached to Ethernet ports.
So, yeah, it’s a shared USB/Ethernet jobbie…

At least the Pi4 isn’t. :-)

First experience of the Raspberry Pi 4 is not good. I ordered a Pi4 and up to date NOOBS card (version 3.1.1) together from Pi Hut but on power up I just get a multi-coloured square on the screen. So not even Linux is working yet!

I presume there will be some help for users on the Pi foundation web site but it seems it has been released prematurely…

So I tried the other HDMI port instead and that does seem to work on Linux. The NOOBS card was 16GB but has only Raspbian on it (4GB) – not very impressive considering RISC OS only uses 2GB. Why has it been taken off again?

There are traps for the unwary:

If you wish to obtain RISC OS 5.26 for the Raspberry Pi 3A+, use NOOBS Lite to download the software – Be sure to use NOOBS Lite (Network install only). The installation will fail if you use the full NOOBS.

Please note that RISC OS 5.27 cannot use the Raspberry Pi firmware files (bootloader.bin, fixup.dat and start.elf) supplied with earlier releases of RISC OS. Instructions on how to obtain updated firmware can be found at Software information: Raspberry Pi: Firmware.

A code change in late January 2019 (the only change in January 2019 was marked OMAP5 so that should keep people guessing) means that RISC OS 5.27 cannot use the firmware supplied with earlier distributions, but there is a risk in using the absolute latest firmware as recent changes in the firmware could have had a detrimental effect on RISC OS operation. The current recommendation is to use January 2019 firmware obtained from the download site

RISC OS 5.26 is not available for download, nor is the RISC OS 5.26 SD card image, which has been available indirectly via NOOBS for some months.

I cannot be sure whether these traps have been set deliberately or not. It would seem sensible to have an SD card image download available for 5.26 surely? And perhaps a supplement for anyone trying the model 4?

If the intention is to set the hurdles sufficiently high that no new users are attracted unless they are very experienced programmers with a considerable determination to prevail in adverse circumstances, then it is likely to be successful.

I shall try the 5.26 SD card image on the Pi 4 (which I obtained via a difficult process using NOOBS Lite) and report further…

But doesn’t the link to the Pi4 news article on the ROOL homepage say that RISC OS isn’t fully ready to use yet…

The traps are not specific to the Pi 4.

The 2GB Raspberry Pi 4 has been here over a week now and I am very pleased with it. It came from the Pi Hut with their NOOBS card. The latest Debian duster based Raspbian is fine here.

At this point RISC OS is ‘work in progress’, which as a start requires a self-build of an RPi4 git development branch.

The RPi4 is noticeably faster than the RPi3B+, but just mind you don’t burn yourself on it! Mine, as yet uncased, is currently at 73C.

It’s worth remembering that the Pi 4 is, internally, a very different creature so there will be some issues to work out along the way.
Though I’m not entirely convinced that a firmware update shaving off a couple of degrees is really going to cut it. Shave off twenty, then we’re talking. ;-)

Please note that RISC OS 5.27 cannot use the Raspberry Pi firmware files (bootloader.bin, fixup.dat and start.elf) supplied with earlier releases of RISC OS.

I briefly used 5.27 a few weeks ago with the firmware from 5.24, and everything appeared to be OK. Is there something specific that doesn’t work?

Video is likely to need a rethink, with dual 4K capable outputs, it’s certainly not going to be the same GPU. How is it handled? Two GPUs? One with two framebuffers?

There is a lot of confusion about the term GPU seemingly with SoCs in general, but in particular with VideoCore/RaspberryPi. The GPU (3D rendering via OpenGL-ES and general computation via OpenCL) is a self-contained IP block that operates from memory-to-memory and does not output directly to the display pipe/HDMI output. GPUs are proprietary and complex blocks on SoCs, and the one in the Broadcom chips prior to RPi4 was known as the V3D block. This block has been extended and uprated.

The screen output is done via a separate IP block, a display controller which reads from the memory, and the HDMI encoder block. I don’t know whether the former has been duplicated, but the latter definitely will have been.

Also, since this is a lot of questions awaiting answers – does the Pi 4 still follow the slightly backwards principle of the GPU being the one in control with the ARM being almost a co-processor?
It’s not really backwards, just a matter of how you think about it; VideoCore contains two full 2D vector processors with their own scalar instruction set, running compiled C code and every bit as capable as a Cortex-A9 as far as the instruction set goes.

These are independent of the V3D (actual ‘GPU’) block and have an entire associated software stack. In the BCM2835, an ARM11 CPU was added into the design but not used in many/most deployments; VideoCore IV was designed as a multimedia co-processor to a separate (typically ARM) IC for mobile phone applications. The phone manufacturer already had its own SoC and operating system stack performing this role.

The ARM11 was obviously replaced by a more capable CPU cluster for subsequent RPi SoCs.

Hope some of that helps.

So not even Linux is working yet!
bq. So I tried the other HDMI port instead and that does seem to work on Linux.

I’ve had a few boot failures, but the firmware and Linux will start up on either HDMI output; you’ve probably seen an intermittent failure and attributed it to the port. I encountered the same but went back to try again with the first part. Since I’ve not yet run with two screens, I can’t say whether it prefers one initially or whether it duplicates the screen output on both until Linux starts.

The SoC does run very hot; you may want to try attaching a heatsink and/or pointing a fan at it, particularly at the moment. That may improve booting or it could be the declared nascent start of the firmware, which will surely be fixed in time because it’s now possible to update it (I’m looking forward to network booting with PXE; much easier!).

Re temperature, I’d expect that the power and speed control will have been the last thing implemented, or not yet implemented, and it can make a big difference if the design has been properly clock-gated to reduce power consumption as we did for earlier devices. I’m not privy to the new work done for the RPi4 SoC so cannot answer that.

I briefly used 5.27 a few weeks ago with the firmware from 5.24, and everything appeared to be OK. Is there something specific that doesn’t work?

On 6 February, Jan de Boer reported that using November 2018 Pi firmware with recent 5.27 builds resulted in an all-black screen in 256-colour modes – upgrading to the January “start.elf” fixed the problem.

A code change in late January 2019 (the only change in January 2019 was marked OMAP5 so that should keep people guessing)

Try this code change on 19 January
And this one, also on 19 January
And finally this one on 30 January

How odd – I was looking on the git rather than the cvs – there is only one change in January 2019 on git.

Anyway I’m off to download the source tarball for the Pi to get my Pi 4 working (with a serial terminal) with the various tweaks to the flags that are required.

Call it another “quirk” (if that’s the right word) of the Events page. In this case it’s not a long-term concern, as the page only shows 50 changes (so the older CVS changes will soon be pushed off the end).

The changes are all in Git, of course. For example, here’s the BCMVideo change made on 19 January.

Anyway I’m off to download the source tarball for the Pi to get my Pi 4 working (with a serial terminal) with the various tweaks to the flags that are required.

Do you have a Pi Hat to break out the UART signals as RS-232? I believe something like https://thepihut.com/products/modmypi-serial-hat-rs232 should work but does anyone know for sure (I’ve never done any electrical interfacing to a Pi) Since I happen to live near Cambridge I shall probably visit the store again.

Do you have a Pi Hat to break out the UART signals as RS-232?

Yes. I currently have a Pi model B1 256MB with a 9-way DIN socket which connects to pins 1, 6, 8 and 10 of the header and uses a level shifter.

I also have a Windows XP laptop with terminal software and the right RS232 9 pin connector.

Do we have a progress report on how the Pi 4 port is going? Over seven weeks on I would hope it is at least mostly complete?