Embedded ARM64 crash trying to zero-fill an 8GB ramdisk

[bugproxy]

---Problem Description---
Kernel crash when setting up ramdisk on embedded ARM

Contact Information = Chris Ward <email address hidden> Mohit Kapur <email address hidden>

---Additional Hardware Info---
Embedded ARM with FPGA

---uname output---
Linux cuttlefisharm1 5.4.0-xilinx-v2020.2 #1 SMP Thu Nov 18 18:44:45 UTC 2021 aarch64 aarch64 aarch64 GNU/Linux

Machine Type = IBM Research internal processor based on xilinx

---System Hang---
 Unresponsive. Power cycle reclaims

---Debugger---
A debugger is not configured

---Steps to Reproduce---
 Boot the system. Try dd if=/dev/zero of=/dev/ram0 bs=... count=...

A small ramdisk works and gives the expected error when trying to write 8G of data into an 8M ramdisk.
tjcw@cuttlefisharm1:~$ sudo dd if=/dev/zero of=/dev/ram0 bs=4096 count=1048576
[sudo] password for tjcw:
dd: error writing '/dev/ram0': No space left on device
2049+0 records in
2048+0 records out
8388608 bytes (8.4 MB, 8.0 MiB) copied, 0.0531788 s, 158 MB/s
tjcw@cuttlefisharm1:~$

A 2GB ramdisk works, a 4GB ramdisk causes a crash

[bugproxy (bugproxy)]

Default Comment by Bridge

[Chris Ward (tjcw)]

This is really on an ARM64 system, not IBM Power

[Chris Ward (tjcw)]

https://github.com/Xilinx/linux-xlnx is where we got the kernel in question from

[bugproxy (bugproxy)]

------- Comment From <email address hidden> 2021-12-06 09:43 EDT-------
It was filed with Canonical because we are running with a Ubuntu 20 userland; this is probably a mistake since it is clearly a kernel bug. How do I un-file a bug ? May be better to just wait until they reject it.

Do you think Xilinx technical support might help us ? I can ask my manager if he is paying them for that. But LTC can't mirror a bug to Xilinx the way they can to RedHat, SuSE and Canonical.

It's definitely a bug, and it is badly hurting the ability of IBM Research to deliver the next version of the IBM Cloud.

Should I try to fix it myself ? re-familiarising myself with the linux kernel and the meaning of each of those hundreds of kernel debug messages would take ages.

15 years ago is better than never, I would say. Can I persuade my manager to fund you for some of this ? And you can 'ssh' to the machine, and work its power-cycling mechanism without us having to let someone external in through the IBM VPN and all that that entails.

Eagerly awaiting reply.

[Chris Ward (tjcw)]

Please 'park' this bug report, i.e. take no action on it. It is a bug in a Xilinx kernel, not a Ubuntu one. Please leave this bug report here as it is IBM's way to communicate the problem to Xilinx engineers. Thanks !

[Chris Ward (tjcw)]

You can only log in to the affected system after you have accessed the IBM network, i.e. when you are inside the firewall. VPN access can be granted (I believe) to whoever is assignde to fix this bug.

[Chris Ward (tjcw)]

We are using the kernel level from
tjcw:linux-xlnx$ git checkout xilinx-v2020.2