-
-
Save dataslayermedia/714ec5a9601249d9ee754919dea49c7e to your computer and use it in GitHub Desktop.
| #!/bin/bash | |
| cd / | |
| sudo apt update | |
| echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list | |
| curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add - | |
| sudo apt-get update | |
| sudo apt-get install libedgetpu1-std | |
| sudo apt install -y devscripts debhelper | |
| sudo apt install dkms | |
| sudo apt-get install dh-dkms | |
| # Clone the Gasket driver repository | |
| sudo git clone https://github.com/google/gasket-driver.git | |
| # Change directory to the cloned repository | |
| cd gasket-driver | |
| # Build the Gasket driver package | |
| sudo debuild -us -uc -tc -b | |
| # Go back to the parent directory | |
| cd .. | |
| # Install the built Gasket driver package | |
| sudo dpkg -i gasket-dkms_1.0-18_all.deb | |
| sudo sh -c "echo 'SUBSYSTEM==\"apex\", MODE=\"0660\", GROUP=\"apex\"' >> /etc/udev/rules.d/65-apex.rules" | |
| sudo groupadd apex | |
| sudo adduser $USER apex | |
| # Update the boot configuration for Raspberry Pi | |
| echo "kernel=kernel8.img" | sudo tee -a /boot/firmware/config.txt | |
| # Back up the Device Tree Blob (DTB) | |
| sudo cp /boot/firmware/bcm2712-rpi-5-b.dtb /boot/firmware/bcm2712-rpi-5-b.dtb.bak | |
| # Decompile the DTB into a DTS file | |
| sudo dtc -I dtb -O dts /boot/firmware/bcm2712-rpi-5-b.dtb -o ~/test.dts | |
| # Modify the Device Tree Source (DTS) | |
| sudo sed -i '/pcie@110000 {/,/};/{/msi-parent = <[^>]*>;/{s/msi-parent = <[^>]*>;/msi-parent = <0x67>;/}}' ~/test.dts | |
| # Recompile the DTS back into a DTB | |
| sudo dtc -I dts -O dtb ~/test.dts -o ~/test.dtb | |
| # Replace the old DTB with the new one | |
| sudo mv ~/test.dtb /boot/firmware/bcm2712-rpi-5-b.dtb | |
| sudo reboot now |
Had to change line #29 to:
echo "kernel=kernel8.img" | sudo tee -a /boot/firmware/config.txt
Had to change line #29 to:
echo "kernel=kernel8.img" | sudo tee -a /boot/firmware/config.txt
Good catch for Raspbian OS latest on the RPI 5.
cat /boot/config.txt
DO NOT EDIT THIS FILE
The file you are looking for has moved to /boot/firmware/config.txt
Yeaha! I have the Dual TPU Running at the AI Hat with bigs
thank you @feranick feranick GH Page and thank you @dataslayermedia for this script
a few changes at RPi5:
Linux escapepod 6.6.20+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.20-1+rpt1 (2024-03-07) aarch64 GNU/Linux
echo "kernel=kernel8.img" | sudo tee -a /boot/firmware/config.txt
echo "dtparam=pciex1" | sudo tee -a /boot/firmware/config.txt
echo "pciex1_gen=3" | sudo tee -a /boot/firmware/config.txt
Maybe this is possible in an Onliner
change dts stuff, thanks at @ChetanKukreja
sudo sed -i '/pcie@110000 {/,/msi-parent = <0x2[fc]>;/{s/<0x2f>/<0x67>/; s/<0x2c>/<0x67>/}' ~/test.dts
Python Modules @ Python 3.11:
- numpy 1.26.4
- pillow 10.3.0
- pycoral 0.2.0 from Feranick/pycoral
python -c "import pycoral; print(pycoral.__version__)"
2.0.1
- FFLite-runtime from Feranick/TFlite-builds
python -c "import tflite_runtime; print(tflite_runtime.__version__)"
2.16.1
i ve init my project with rye my pyproject.toml is like this
[project]
name = "coral-t"
version = "0.1.0"
description = "Add your description here"
dependencies = [
"tflite-runtime @ https://github.com/feranick/TFlite-builds/releases/download/v.2.16.1/tflite_runtime-2.16.1-cp311-cp311-linux_aarch64.whl",
"pycoral @ https://github.com/feranick/pycoral/releases/download/v2.0.1TF2.16.1/pycoral-2.0.1-cp311-cp311-linux_aarch64.whl",
]
readme = "README.md"
requires-python = ">= 3.11"
[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"
[tool.rye]
managed = true
dev-dependencies = []
[tool.hatch.metadata]
allow-direct-references = true
[tool.hatch.build.targets.wheel]
packages = ["src/coral_t"]
Output:
(coral-t) ➜ coral_t git:(master) ✗ python3 examples/classify_image.py
--model test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite
--labels test_data/inat_bird_labels.txt
--input test_data/parrot.jpg
----Debug2----
test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite
----INFERENCE TIME----
Note: The first inference on Edge TPU is slow because it includes loading the model into Edge TPU memory.
12.9ms
2.7ms
2.7ms
2.7ms
2.7ms
-------RESULTS--------
Ara macao (Scarlet Macaw): 0.75781
Next question is. How can i add the second tpu. I have an Dual TPU M2 Card but can only find /dev/apex_0
@dasTholo The Pi 5 only exposes one PCIe lane so you will need a HAT to multiplex both devices over the single lane. I don't think there are any dual-lane E HATs yet, but Pineberry Pi does make dual-slot HATs. You might be able to install two single-TPU devices, one A+E and one B+M key.
@mjforan
That is sad. I didn't know that. I had specially ordered the DualTPU. But I don't quite understand (I haven't found any other docs on this yet) how you can address both TPUs on an x86, for example. As they are both on the same PCIe port.
I have the Pineberry with M-Key and E-Key
https://pineberrypi.com/products/hatdrive-ai-nvme-2230-2242-gen-2-coral-edge-tpu-for-raspberry-pi-5
If I understand you correctly, then I could put another TPU in the other slot? This one here?
https://coral.ai/products/m2-accelerator-bm/
@dasTholo the m.2 specification is not just a single PCIe "port", but has many different configurations. The "key" is the shape of the connector. The A+E variant of the coral can plug in to the E key port on a computer which is commonly used for WiFi cards while the B+M variant can use the M key port typically used for SSDs. On the dual TPU version there are two separate PCIe devices, each of which requires a separate x1 PCIe interface (x1 means one lane). Even on larger computers, many E-key ports only have one PCIe interface - the coral page you linked to notes:
Although the M.2 Specification (section 5.1.2) declares E-key sockets provide two instances of PCIe x1, most manufacturers provide only one. To use both Edge TPUs, be sure your socket connects both instances to the host.
If I understand you correctly, then I could put another TPU in the other slot? This one here? https://coral.ai/products/m2-accelerator-bm/
Exactly. You will need to cut down the TPU to get it to fit (it is designed to do that). I can't personally guarantee that this will work - you might want to ask Pineberry Pi before buying another TPU. Let us know if it does work out for you!
@mjforan That is sad. I didn't know that. I had specially ordered the DualTPU. But I don't quite understand (I haven't found any other docs on this yet) how you can address both TPUs on an x86, for example. As they are both on the same PCIe port.
I have the Pineberry with M-Key and E-Key https://pineberrypi.com/products/hatdrive-ai-nvme-2230-2242-gen-2-coral-edge-tpu-for-raspberry-pi-5
If I understand you correctly, then I could put another TPU in the other slot? This one here? https://coral.ai/products/m2-accelerator-bm/
The dual TPU requires special considerations you need a specific adapter as well as a motherboard that will allow it. Magic Blue Smoke makes available a range of such adapters here.
But getting it to work on a Raspberry Pi 5 introduces yet another obstacle you would need a PCIe expansion slot and even then it's likely the motherboard will only make one TPU available. But I will try that soon to determine if it works.
I fixed the script as of April 7th, it looks like the gasket package now needs some finessing to work on the RPI. But the latest revision worked for me.
@dataslayermedia Thanks for the script. I set up this for a friend. I tested this against a fresh install and in addition to your script, I had to do following two steps in advance.
- "apt install dkms"
- Switching to X11 instead of Wayland using the raspi-config tool. Otherwise there was no GUI anymore, after reboot. Didn't found the root cause. Might be related to the HDMI Display type.
I ran into trouble with this script today with a fresh install of Rasp Pi OS on Rasp Pi 5. I believe the HAT is correctly installed considering the LED on it is green. Here's what I believe to be the relevant logs:
Setting up gasket-dkms (1.0-18) ...
Removing old gasket-1.0 DKMS files...
Deleting module gasket-1.0 completely from the DKMS tree.
Loading new gasket-1.0 DKMS files...
Deprecated feature: REMAKE_INITRD (/usr/src/gasket-1.0/dkms.conf)
Building for 6.6.20+rpt-rpi-2712 6.6.20+rpt-rpi-v8
Building initial module for 6.6.20+rpt-rpi-2712
Deprecated feature: REMAKE_INITRD (/var/lib/dkms/gasket/1.0/source/dkms.conf)
Error! Bad return status for module build on kernel: 6.6.20+rpt-rpi-2712 (aarch64)
Consult /var/lib/dkms/gasket/1.0/build/make.log for more information.
dpkg: error processing package gasket-dkms (--configure):
installed gasket-dkms package post-installation script subprocess returned error exit status 10
Errors were encountered while processing:
gasket-dkms
E: Sub-process /usr/bin/dpkg returned an error code (1)
DKMS make.log for gasket-1.0 for kernel 6.6.20+rpt-rpi-v8 (aarch64)
Tue 16 Apr 22:56:32 EDT 2024
make: Entering directory '/usr/src/linux-headers-6.6.20+rpt-rpi-v8'
/var/lib/dkms/gasket/1.0/build/gasket core.o
/var/lib/dkms/gasket/1.0/build/gasket_ioctl.o
/var/lib/dkms/gasket/1.0/build/gasket_interrupt.o
/var/lib/dkms/gasket/1.0/build/gasket_page_table.o
/var/lib/dkms/gasket/1.0/build/gasket_sysfs.o
/var/lib/dkms/gasket/1.0/build/apex_driver.o
/var/lib/dkms/gasket/1.0/build/apex_driver.c: 1143:12: warning: 'apex_pci_resume' defined but not used [-Wunused-function]
1143 | static int apex pci_resume (struct pci_dev *pci_dev)
/var/lib/dkms/gasket/1.0/build/apex_driver.c: 1128:12: warning: 'apex_pci_suspend' defined but not used [-Wunused-function]
1128 | static int apex pci_suspend (struct pci_dev *pci_dev, pm_message_t state) {
/var/lib/dkms/gasket/1.0/build/apex.o
/var/lib/dkms/gasket/1.0/build/gasket.o
MODPOST /var/lib/dkms/gasket/1.0/build/Module. symvers
/var/lib/dkms/gasket/1.0/build/gasket.mod.o
/var/lib/dkms/gasket/1.0/build/apex.mod. o
/var/lib/dkms/gasket/1.0/build/apex. ko
/var/lib/dkms/gasket/1.0/build/gasket. ko
make: Leaving directory '/usr/src/linux-headers-6.6.20+rpt-rpi-v8'
EDIT: I read the whole thread and used these commands before re-running the script and it seems to have worked!
sudo apt update
sudo apt upgrade
sudo apt install devscripts debhelper -y
Hi
Thankyou for creating this. I have been using it to install onto the Raspberry Pi OS Lite (64bit). I don't know if there are more things missing on the Lite version that the script relies on, but the build of gasket fails. When I repeat the build I got this:
sudo debuild -us -uc -tc -b
dpkg-buildpackage -us -uc -ui -tc -b
dpkg-buildpackage: info: source package gasket-dkms
dpkg-buildpackage: info: source version 1.0-18
dpkg-buildpackage: info: source distribution unstable
dpkg-buildpackage: info: source changed by Coral <coral-support@google.com>
dpkg-source --before-build .
dpkg-buildpackage: info: host architecture arm64
dpkg-checkbuilddeps: error: Unmet build dependencies: dkms
dpkg-buildpackage: warning: build dependencies/conflicts unsatisfied; aborting
dpkg-buildpackage: warning: (Use -d flag to override.)
debuild: fatal error at line 1182:
dpkg-buildpackage -us -uc -ui -tc -b failed
After a few attempts, I found that running before running your script meant it worked:
sudo apt-get install raspberrypi-kernel-headers
@pierslawson, are you seeing /dev/apex0 when you're done? I get successful builds but the Coral is not showing up as far as I can tell. Just looking for someone to confirm OS-Lite is working for them. ty!
Yes, I see /dev/apex_0 One thing to check is that you can see the device before you do anything else (i.e. on a fresh install of the OS. At first I wasn't seeing it and assumed that was normal if the driver had not yet been installed. However, after disconnecting and reconnecting the hat (whilst reformatting the SD card for about the 10th time) I happened to notice it was there before the script had been run. Then after updating the headers first, it all worked.
Note this is where I got the idea for updating the headers: https://gist.github.com/Reddimus/c6948d08a4f4b54ee9d075270bd79c3b
However use dataslayermedia's script as it is been modified since reddimus branched it.
I've been having trouble getting this working. I get nothing in lspci:
00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21)
01:00.0 Ethernet controller: Device 1de4:0001
Firstly, I think the I think the msi-parent value has changed again, so just to clarify, I want msi-parent to match the phandle, right?
pcie@110000 {
compatible = "brcm,bcm2712-pcie";
reg = <0x10 0x110000 0x00 0x9310>;
device_type = "pci";
max-link-speed = <0x02>;
#address-cells = <0x03>;
#interrupt-cells = <0x01>;
#size-cells = <0x02>;
interrupt-parent = <0x01>;
interrupts = <0x00 0xdf 0x04 0x00 0xe0 0x04>;
interrupt-names = "pcie\0msi";
interrupt-map-mask = <0x00 0x00 0x00 0x07>;
interrupt-map = <0x00 0x00 0x00 0x01 0x01 0x00 0xdb 0x04 0x00 0x00 0x00 0x02 0x01 0x00 0xdc 0x04 0x00 0x00 0x00 0x03 0x01 0x00 0xdd 0x04 0x00 0x00 0x00 0x04 0x01 0x00 0xde 0x04>;
resets = <0x2c 0x07 0x2c 0x2b 0x2d>;
reset-names = "swinit\0bridge\0rescal";
msi-controller;
msi-parent = <0x6d>;
ranges = <0x2000000 0x00 0x00 0x1b 0x00 0x00 0xfffffffc 0x43000000 0x04 0x00 0x18 0x00 0x03 0x00>;
dma-ranges = <0x3000000 0x10 0x00 0x00 0x00 0x10 0x00>;
brcm,enable-l1ss;
status = "disabled";
phandle = <0x6d>;
};
But it's still not working - in dmesg I get nothing about apex, but I get this:
brcm-pcie 1000110000.pcie: link down
Just to clarify - should lspci show the device before starting? And if it doesn't show it's likely a hardware fault or physical installation issue - is that correct?
I've been having trouble getting this working. I get nothing in lspci:
00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21) 01:00.0 Ethernet controller: Device 1de4:0001Firstly, I think the I think the msi-parent value has changed again, so just to clarify, I want msi-parent to match the phandle, right?
pcie@110000 { compatible = "brcm,bcm2712-pcie"; reg = <0x10 0x110000 0x00 0x9310>; device_type = "pci"; max-link-speed = <0x02>; #address-cells = <0x03>; #interrupt-cells = <0x01>; #size-cells = <0x02>; interrupt-parent = <0x01>; interrupts = <0x00 0xdf 0x04 0x00 0xe0 0x04>; interrupt-names = "pcie\0msi"; interrupt-map-mask = <0x00 0x00 0x00 0x07>; interrupt-map = <0x00 0x00 0x00 0x01 0x01 0x00 0xdb 0x04 0x00 0x00 0x00 0x02 0x01 0x00 0xdc 0x04 0x00 0x00 0x00 0x03 0x01 0x00 0xdd 0x04 0x00 0x00 0x00 0x04 0x01 0x00 0xde 0x04>; resets = <0x2c 0x07 0x2c 0x2b 0x2d>; reset-names = "swinit\0bridge\0rescal"; msi-controller; msi-parent = <0x6d>; ranges = <0x2000000 0x00 0x00 0x1b 0x00 0x00 0xfffffffc 0x43000000 0x04 0x00 0x18 0x00 0x03 0x00>; dma-ranges = <0x3000000 0x10 0x00 0x00 0x00 0x10 0x00>; brcm,enable-l1ss; status = "disabled"; phandle = <0x6d>; };But it's still not working - in dmesg I get nothing about apex, but I get this:
brcm-pcie 1000110000.pcie: link downJust to clarify - should lspci show the device before starting? And if it doesn't show it's likely a hardware fault or physical installation issue - is that correct?
Hello, what version of the kernel are you using?
uname -ra ?
I'm using 6.6.29-v8-16k+ #1760 SMP PREEMPT Mon Apr 29 14:44:20 BST 2024 aarch64 GNU/Linux and PCIe devices also stopped working
I can confirm that with the current kernel - 6.6.28+rpt-rpi-v8
[ 1.473214] brcm-pcie 1000110000.pcie: host bridge /axi/pcie@110000 ranges:
[ 1.480216] brcm-pcie 1000110000.pcie: No bus range found for /axi/pcie@110000, using [bus 00-ff]
[ 1.489758] brcm-pcie 1000110000.pcie: MEM 0x1b00000000..0x1bfffffffb -> 0x0000000000
[ 1.498061] brcm-pcie 1000110000.pcie: MEM 0x1800000000..0x1affffffff -> 0x0400000000
[ 1.506361] brcm-pcie 1000110000.pcie: IB MEM 0x0000000000..0x0fffffffff -> 0x1000000000
[ 1.515832] brcm-pcie 1000110000.pcie: Forcing gen 3
[ 1.520819] brcm-pcie 1000110000.pcie: Unable to find MSI PCI address
[ 1.527284] brcm-pcie: probe of 1000110000.pcie failed with error -22
[ 1.533802] brcm-pcie 1000120000.pcie: host bridge /axi/pcie@120000 ranges:
[ 1.540792] brcm-pcie 1000120000.pcie: No bus range found for /axi/pcie@120000, using [bus 00-ff]
[ 1.549879] brcm-pcie 1000120000.pcie: MEM 0x1f00000000..0x1ffffffffb -> 0x0000000000
[ 1.558176] brcm-pcie 1000120000.pcie: MEM 0x1c00000000..0x1effffffff -> 0x0400000000
[ 1.566477] brcm-pcie 1000120000.pcie: IB MEM 0x1f00000000..0x1f003fffff -> 0x0000000000
[ 1.574775] brcm-pcie 1000120000.pcie: IB MEM 0x0000000000..0x0fffffffff -> 0x1000000000
[ 1.584227] brcm-pcie 1000120000.pcie: Forcing gen 2
[ 1.589405] brcm-pcie 1000120000.pcie: PCI host bridge to bus 0000:00
Update #1: wget https://github.com/raspberrypi/firmware/raw/master/boot/bcm2712-rpi-5-b.dtb and moving the file mv bcm2712-rpi-5-b.dtb /boot/firmware/bcm2712-rpi-5-b.dtb then rebooting did the trick - /dev/apex_0 is an actual device now.
@lukaszraczylo @ezaul I think this is because the msi-parent now needs to be 0x6d in the latest updates
@scaredyfish Thank you so much! Updated the msi-parent, now getting:
script side: RuntimeError: Error in device opening (/dev/apex_0)!
dmesg: apex 0000:01:00.0: Couldn't reinit interrupts: -28
ls -lA /dev/apex_0
crw-rw-rw- 1 root apex 120, 0 May 6 22:30 /dev/apex_0
Still battling :)
Hi,
Every time I change msi-parent to
msi-parent = <0x66> or msi-parent = <0x67>
I lost my TPU card.
This is result when I use DTB default
$ uname -a
Linux smarthome 6.6.28+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.28-1+rpt1 (2024-04-22) aarch64 GNU/Linux
List PCI devices
$ lspci
0000:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21)
0000:01:00.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:02:03.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:02:07.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:03:00.0 System peripheral: Global Unichip Corp. Coral Edge TPU
0000:04:00.0 Non-Volatile memory controller: Phison Electronics Corporation E8 PCIe3 NVMe Controller (rev 01)
0001:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21)
0001:01:00.0 Ethernet controller: Device 1de4:0001
Detect module
$ lspci -nn | grep 089a
0000:03:00.0 System peripheral [0880]: Global Unichip Corp. Coral Edge TPU [1ac1:089a]
$ ls /dev/apex_0
/dev/apex_0
The dmesg log
dmesg | grep brcm-pcie
[ 0.405470] brcm-pcie 1000110000.pcie: host bridge /axi/pcie@110000 ranges:
[ 0.405482] brcm-pcie 1000110000.pcie: No bus range found for /axi/pcie@110000, using [bus 00-ff]
[ 0.405493] brcm-pcie 1000110000.pcie: MEM 0x1b00000000..0x1bfffffffb -> 0x0000000000
[ 0.405500] brcm-pcie 1000110000.pcie: MEM 0x1800000000..0x1affffffff -> 0x0400000000
[ 0.405506] brcm-pcie 1000110000.pcie: IB MEM 0x0000000000..0x0fffffffff -> 0x1000000000
[ 0.407017] brcm-pcie 1000110000.pcie: Forcing gen 3
[ 0.407253] brcm-pcie 1000110000.pcie: PCI host bridge to bus 0000:00
[ 0.516181] brcm-pcie 1000110000.pcie: link up, 5.0 GT/s PCIe x1 (!SSC)
[ 0.546021] brcm-pcie 1000120000.pcie: host bridge /axi/pcie@120000 ranges:
[ 0.546029] brcm-pcie 1000120000.pcie: No bus range found for /axi/pcie@120000, using [bus 00-ff]
[ 0.546040] brcm-pcie 1000120000.pcie: MEM 0x1f00000000..0x1ffffffffb -> 0x0000000000
[ 0.546046] brcm-pcie 1000120000.pcie: MEM 0x1c00000000..0x1effffffff -> 0x0400000000
[ 0.546054] brcm-pcie 1000120000.pcie: IB MEM 0x1f00000000..0x1f003fffff -> 0x0000000000
[ 0.546059] brcm-pcie 1000120000.pcie: IB MEM 0x0000000000..0x0fffffffff -> 0x1000000000
[ 0.547144] brcm-pcie 1000120000.pcie: Forcing gen 2
[ 0.547196] brcm-pcie 1000120000.pcie: PCI host bridge to bus 0001:00
[ 0.656178] brcm-pcie 1000120000.pcie: link up, 5.0 GT/s PCIe x4 (!SSC)
I want to ask where the numbers for msi-parent like 0x66, 0x67 come from?
How do I know which number is correct?
@EnziinSystem I'm no expert, but in trying to diagnose my own problem, here's what I learned:
The device tree is compiled from the source files here - the dtsi files are included in the dts. The relevant portion is in bcm2712.dtsi
https://github.com/raspberrypi/linux/blob/rpi-6.6.y/arch/arm/boot/dts/broadcom/bcm2712-rpi-5-b.dts
https://github.com/raspberrypi/linux/blob/rpi-6.6.y/arch/arm/boot/dts/broadcom/bcm2712-rpi.dtsi
https://github.com/raspberrypi/linux/blob/rpi-6.6.y/arch/arm/boot/dts/broadcom/bcm2712.dtsi
The original source uses names, and when this is compiled, the names are replaced with phandles. These phandles are not fixed - any time Raspberry Pi makes an upstream change to the device tree, the phandle may change.
pcie1: pcie@110000 {
...
msi-parent = <&mip1>;
...
}
mip1: msi-controller@131000 {
...
}
becomes
pcie@110000 {
...
msi-parent = <0x2f>;
...
phandle = <0x6d>;
};
msi-controller@131000 {
...
phandle = <0x2f>;
}
We want to change msi-parent to pcie1 - i.e. its msi-parent is itself - in this case, 0x6d (I don't know what this means, or why it works, but someone more expert than me came up with it: https://forums.raspberrypi.com/viewtopic.php?p=2157674#p2157674)
You can now use dtoverlay=pineboards-hat-ai instead of having to change the msi-parent. It requires the most current kernel raspberrypi/linux@269a721
You can now use dtoverlay=pineboards-hat-ai instead of having to change the msi-parent. It requires the most current kernel raspberrypi/linux@269a721
My kernel:
$ uname -a
Linux smarthome 6.6.28+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.28-1+rpt1 (2024-04-22) aarch64 GNU/Linux
Can I use dtoverlay=pineboards-hat-ai ?
Thanks.
you may need to use rpi-update to get the latest version https://www.raspberrypi.com/documentation/computers/os.html#rpi-update
It needs to be at 6.6.30 or higher
just tried dtoverlay=pineboards-hat-ai on Linux raspberrypi 6.6.30-v8+ #1761 SMP PREEMPT Thu May 2 16:54:52 BST 2024 aarch64 GNU/Linux
no, this doesn't work, I still have
ls -l /dev/apex_0
ls: cannot access '/dev/apex_0': No such file or directory
You need the gasket-dkms, not sure you have it installed. You can compile it following this info:
https://github.com/google/gasket-driver
or use prebuilt debs.
https://github.com/feranick/gasket-driver/releases/download/1.0-18.2/gasket-dkms_1.0-18.2_all.deb
It needs to be at 6.6.30 or higher
The config dtoverlay=pineboards-hat-ai doesn't work.
Here are the steps:
$ rpi-eeprom-update -a
$ sudo rpi-update
$ uname -a
Linux smarthome 6.6.30-v8+ #1761 SMP PREEMPT Thu May 2 16:54:52 BST 2024 aarch64 GNU/Linux
Configure
sudo nano /boot/firmware/config.txt
Add lines
[all]
# Enable the PCIe External connector.
dtparam=pciex1
dtparam=pciex1_gen=3
kernel=kernel8.img
# Enable Pineboard
dtoverlay=pineboards-hat-ai
Turn off PCI ASPM
sudo nano /boot/firmware/cmdline.txt
Add **pcie_aspm=off** before **rootwait**
sudo reboot
Install the PCIe driver and Edge TPU runtime
sudo apt update
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update
sudo apt-get install cmake
sudo apt-get install -y libedgetpu1-std
sudo apt install -y devscripts debhelper
sudo apt install -y dkms
sudo apt-get install -y dh-dkms
# Clone the Gasket driver repository
sudo git clone https://github.com/google/gasket-driver.git
cd gasket-driver
sudo debuild -us -uc -tc -b
# Go back to the parent directory
cd ..
# Install the built Gasket driver package
sudo dpkg -i gasket-dkms_1.0-18_all.deb
Selecting previously unselected package gasket-dkms.
(Reading database ... 175909 files and directories currently installed.)
Preparing to unpack gasket-dkms_1.0-18_all.deb ...
Unpacking gasket-dkms (1.0-18) ...
Setting up gasket-dkms (1.0-18) ...
Loading new gasket-1.0 DKMS files...
Deprecated feature: REMAKE_INITRD (/usr/src/gasket-1.0/dkms.conf)
Building for 6.6.30-v8+
Module build for kernel 6.6.30-v8+ was skipped since the
kernel headers for this kernel do not seem to be installed.
Note: Module build for kernel 6.6.30-v8+ was skipped since the kernel headers for this kernel do not seem to be installed
Set udev rule
sudo sh -c "echo 'SUBSYSTEM==\"apex\", MODE=\"0660\", GROUP=\"apex\"' >> /etc/udev/rules.d/65-apex.rules"
sudo groupadd apex
sudo adduser $USER apex
sudo reboot now
Result
$ lspci
0000:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21)
0000:01:00.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:02:03.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:02:07.0 PCI bridge: ASMedia Technology Inc. ASM1182e 2-Port PCIe x1 Gen2 Packet Switch
0000:03:00.0 System peripheral: Global Unichip Corp. Coral Edge TPU
0000:04:00.0 Non-Volatile memory controller: Phison Electronics Corporation E8 PCIe3 NVMe Controller (rev 01)
0001:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries Device 2712 (rev 21)
0001:01:00.0 Ethernet controller: Device 1de4:0001
List TPU
$ lspci -nn | grep 089a
0000:03:00.0 System peripheral [0880]: Global Unichip Corp. Coral Edge TPU [1ac1:089a]
$ ls /dev/apex_0
ls: cannot access '/dev/apex_0': No such file or directory
$ sudo modprobe gasket
modprobe: FATAL: Module gasket not found in directory /lib/modules/6.6.30-v8+
$ sudo modprobe apex
modprobe: FATAL: Module apex not found in directory /lib/modules/6.6.30-v8+
Note: I did not make changes to MSI-parent and did not recompile the DTB
For any others experiencing the same issues I had, this helped:
google-coral/edgetpu#808 (comment)
For reference:
I followed Reddimus' guide.
I am on kernel 6.6.20 using the PineberryPi HatDrive! AI.
My issue was that my accelerator wasn't found (solved this using the aforementioned msi-parent = <0x67>).
My next issue was that gasket-dkms package would not correctly install.
What finally worked for me was the following from google-coral/edgetpu issue #808 as follows:
Hope this also helps others in the same boat!