Am 05.03.22 um 12:34 schrieb Stefan Wahren:
> Am 04.03.22 um 21:50 schrieb Stefan Wahren:
>> Am 03.03.22 um 02:29 schrieb Chris Adams:
>>> Once upon a time, Stefan Wahren <stefan.wahren(a)i2se.com> said:
>>>> Hi Chris,
>>>>
>>>> Am 02.03.22 um 10:51 schrieb Peter Robinson:
>>>>>> I have an RPi4 running Fedora 35. It hadn't been updated in
a while, so
>>>>>> I applied updates today. When I boot to the kernel 5.16.11-200,
I lose
>>>>>> the PPS device from my GPS hat. Boot back to 5.14.16-301.fc35
and it
>>>>>> works.
>>>>>>
>>>>>> I'm booting with EFI firmware, and with a device tree
config.txt that
>>>>> All our firmware are EFI, that's the only way we support booting,
is
>>>>> it the default U-Boot based one or are you using the edk2 one?
>>>>>
>>>>>> has "dtoverlay=pps-gpio,gpiopin=4" in it. I removed
the /boot/dtb
>>>>>> symlink and set /etc/u-boot.conf to not re-add it.
>>>>>>
>>>>>> When I boot 5.16, I see /proc/device-tree, and it has
>>>>>> /proc/device-tree/pps@4 in it (and the contents look correct),
but
>>>>>> loading the pps-gpio kernel module just gives:
>>>>>>
>>>>>> pps-gpio: probe of pps@4 failed with error -22
>>>>> Any chance you can tell us which kernel it started with, 5.14.x to
>>>>> 5.16.x is a big window to debug and looking at kernel logs for
>>>>> drivers/pps there's been no changes in that space in the 5.15+
kernels
>>>>> at all.
>>>>>
>>>>> I wonder if it's a change/regression in the firmware overlay
<->
>>>>> kernel interface.
>>>> do you use the DTB file from 5.14.x or 5.16x?
>>>>
>>>> I guess this is related to this commit:
>>>>
>>>>
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?...
>>>>
>>>> Unfortunately this is a change which requires this DTS fix:
>>>>
>>>>
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?...
>>>>
>>>> So DTB and kernel must "match".
>>> Okay, so what's the best way to do that?
>>>
>>> I have EFI firmware from
https://github.com/pftf/RPi4 in /boot/efi, and
>>> the /boot/dtb symlink removed. The bcm2711-rpi-4-b.dtb file in
>>> /boot/efi is from the RPi4_UEFI_Firmware_v1.32.zip file. I have a
>>> config.txt in /boot/efi that references overlays in /boot/efi/overlays
>>> from RPM bcm283x-overlays (like pps-gpio)..
>>>
>>> I'm not quite sure how I need to fit all this together.
>> Let me recap the situation. mainline and rpi vendor tree are independent
>> in development. Synchronization (incl. device tree) happens in both
>> directions.
>>
>> U-Boot bootloader decided to keep an own copy of the mainline DT files
>> (which is currently based on an older 5.15 version which lacks the
>> gpio-ranges property). According to this statement [1] the EDK2 UEFI
>> decided to use the rpi vendor tree and don't care about the mainline DT
>> files.
>>
>> I'm sorry but as a spare-time kernel developer, i don't have to time to
>> fight against all this mess.
>>
>> I hope i will have some time for debugging in the near future ...
>>
>> [1] -
https://github.com/pftf/RPi4/issues/193
>>
> Today i tried Fedora 35 Minimal for my first time, here are my test results:
>
> Raspberry Pi 400 32 bit => bus width issue
> Raspberry Pi 400 64 bit => hangs while show graphics
> Raspberry Pi 4 B 4 GB 32 bit => kernel crash
> Raspberry Pi 4 B 4 GB 64 bit => boot into setup
> Raspberry Pi 4 B 8 GB 32 bit => black screen, fails to start kernel?
> Raspberry Pi 4 B 8 GB 64 bit => boot into setup
>
> At least the 32 bit issues on Raspberry Pi 4 are expected since the
> kernel config doesn't have ARM_LPAE enabled.
>
Okay, here is the explanation for the different behavior on Raspberry Pi
400 and Raspberry Pi 4 B. The Raspberry Pi 400 has a newer BCM2711 SoC
(Stepping C0), which have less DMA restrictions for the emmc2 interface
(responsible for SD card access). For the Raspberry Pi 4 B there are
older boards which have Stepping B0 and all the new boards should have
Stepping C0 [1].
Unfortunately there is no 100% reliable way to detect the stepping from
the kernel side. So currently the Raspberry Pi firmware patches the
dma-ranges in the firmware DT [2]. So in case U-Boot [3] or another
bootloader ignores this firmware DT and read a fresh DTB the right
dma-ranges get lost. Finally this results in unexpected behavior as soon
the emmc2 switches to DMA mode [4].
Okay, at least i found a fix [1] for the 64 bit boot issues (Original
Fedora 35, Linux 5.14) with RPi 400/RPi 4 Stepping C0. This requires the
DTB files to be updated and U-Boot to choose between the B0 and the C0
variant of the Rpi 4 DTB file.
Unfortunately this doesn't fix the SD card issues on 32 bit.
Best regards
[1] -