Is anyone already running the testing kernel 5.15 on their WRT AC series? Judging by the topic linked below, at least 5.14 seemed to be working on the WRT3200 and WRT1900ACS. Space allotted to the kernel on all these devices is 6 MiB on both firmware partitions.
All I get is a blinking power LED when I switch master to 5.15 and run a build. @anomeome Seems 5.15 is running okay for you? I see you got self built master images from August 22nd with 5.15?
Did you check fw_printenv for the exact (kernel-) size read by u-boot (that is iirc smaller -the problem- than the alotted partition size on the nbg6617/ ipq4018)?
I'm not sure what I should be looking at here, so this is the full output from fw_printenv. Seems kernel size is set to 6 MiB? Should be sufficient no?
Even that guesstimate is appreciated . I'm going to open it up and hook up serial. Even my own 22.03 build won't boot. Gotta be something fishy that's goin on.
Well hooked up serial, and I'm seeing all my builds hang. I've tested official 22.03.1 and master images, they both boot. My own 22.03.1 build (kernel 5.10.147) and master build (with 5.15) both just hang at starting the kernel, e.g.::
Hit any key to stop autoboot: 0
NAND read: device 0 offset 0x3200000, size 0x600000
6291456 bytes read: OK
## Booting kernel from Legacy Image at 02000000 ...
Image Name: ARM OpenWrt Linux-5.15.72
Created: 2022-10-13 20:04:47 UTC
Image Type: ARM Linux Kernel Image (uncompressed)
Data Size: 3402962 Bytes = 3.2 MiB
Load Address: 00008000
Entry Point: 00008000
Verifying Checksum ... OK
Loading Kernel Image ... OK
OK
Starting kernel ...
I'm going to diff my config against the official one, my master build has the early printk stuff enabled, so if it was booting I should be seeing something. Will report back once I've had the time to inspect my build config.
@anomeome I've tested your build from December 2nd with 5.15 and that boots flawlessly (my own attempts still hang). What's the reason for changing the CPU subtype from vfpv3-d16 to vfpv3 if I may ask?
Purely performance by getting all available registers in-play, nothing to do with issues you are experiencing; there is an old PR somewhere with a number of us arguing against the change made in support of some one-off dev board which failed with > 16 registers churned out by GCC.
So I'm not an inch closer to pinning down the issue, but I did manage to mess up the single bootable firmware I had on the device somehow .
Error during boot of the installed firmware:
Marvell>> run altnandboot
NAND read: device 0 offset 0x3200000, size 0x600000
6291456 bytes read: OK
Wrong Image Format for bootm command
ERROR: can't get kernel image!
The first partition contains a broken OpenWrt build (kernel will just hang), so switching firmware is no option anymore at this point.
I can TFTP over the ramdisk just fine, and it seems like the memory address is okay as well, but it just hangs (just like with my own images, FWIW). This ramdisk is from the same images @anomeome built - and I successfully sysupgraded to that r21380 build; until now, it booted just fine.
Thanks, but what hit mvebu is a switch issue that only hits the vanilla 5.10 builds, and that happens way later. Here, it doesn't even load the kernel... And this is a custom master build with 5.15.
Seems I was approaching it wrong, also tried bootelf but didn't manage to get the ramdisk to boot either. Tried a 21.02 and 22.03 image as well.
Then I stumbled upon these instructions and it turns out you just to configure u-boot the right way so it will pull the factory image from the TFTP server with run update_both_images and write it to both firmware partitions again. So nothing like TFTPing it over yourself. No trace of that on the wiki somehow.
If anyone knows how to get TFTP boot to work the usual way, I'm all ears. By now I'd rather not overwrite the flash ten times a day .
Alrighty. I was carrying a patch that enabled CONFIG_KERNEL_EARLY_PRINTK by default on a few targets - seems that's what makes boot hang. Revert that for mvebu, and my builds boot happily again.