[libre-riscv-dev] Fwd: power-gem5

lkcl . luke.leighton at gmail.com
Thu May 28 13:02:00 BST 2020


hi folks, i just received this from the IBM team behind the power-gem5
port (which hugh kindly alerted me to) - as you can see below they
managed to track down the memory leak.  this means that the power-gem5
simulator, which is cycle-accurate, is capable of booting a linux
kernel with an initramfs containing busybox and a hello-world program.
yes they implemented a RADIX MMU including hypervisor mode i believe.
linux kernel device-driver interrupts are working however according to
the last presentation on the topic, at present these are done using
polling.

l.


---------- Forwarded message ---------
From: Basavaraj Talawar
Luke,

A few weeks back the team were able to root-cause and fix the memory
leaks that originated from the address translation code. The changes
have been pushed to the 'gem5-experimental' branch of our repository.
https://github.com/power-gem5/gem5

Here's a link to the commit that fixes this.
https://github.com/power-gem5/gem5/commit/cae53531103ebc5bccddf874db85f2659b64000a

Memory usage now stays under 500MB.

BT.

On Mon, Mar 16, 2020 at 2:41 PM lkcl . <luke.leighton at gmail.com> wrote:
>
> i can confirm that using gcc/g++ version 8 still has the same massive
> memory leaks.  despite this, it is a huge achievement.
> best,
> l.



More information about the libre-riscv-dev mailing list