On Thursday, March 12th, 2026 at 6:37 AM, Yuao Ma <[email protected]> wrote:
> Himadri Chhaya-Shailesh <[email protected]> 于 2026年3月12日周四 05:58写道： > >> On Saturday, March 7th, 2026 at 9:33 AM, Yuao Ma [email protected] wrote: >> >>> Hi all, >>> >>> Please find attached the writeup for the Experiment-0(1-2 done, 3-4 >>> >>> WIP). Feel free to take a look whenever you get a chance. >>> >>> Thanks, >>> >>> Yuao >> >> Hi Yuao, >> >> Thanks for the update. Your experiment set-up looks good. >> >> Himadri > > Thanks for the review! Unfortunately, I'm running into an issue with my > `trace-cmd` environment setup. When I configure the agent on the guest and > record on the host, I consistently get the following error: > > $ sudo trace-cmd record --verbose=all -e sched -A 3:823 --name vm1 -e sched > sleep 1 > libtraceevent: No such file or directory > [kvm:kvm_inj_exception] bad op token } > libtraceevent: No such file or directory > Error: expected type 5 (DELIM) but read 0 (ERROR) > Negotiated ptp time sync protocol with guest vm1 > libtracecmd: No such file or directory > reading client -110 (Unknown error -110) > trace-cmd: No such file or directory > cannot create output handle > > For reference, both my host and guest are running Fedora 43 (to minimize > kernel differences). Note that my host server is itself a VM with nested > virtualization enabled. I'm not sure whether my setup is misconfigured or if > something else is going on - could you help take a look? For example, sharing > your kernel version or trace-cmd version would be a great start. > > Thanks in advance, > Yuao Sure. Maybe it is worth noting that my host runs Debian and I use virtme-ng for launching the VMs. The environment that I was using for running the experiments mentioned in the thesis had: trace-cmd-v3.2 with libtraceevent-1.7.3 and libtracefs-1.7.0 (all built from source), and host-guest kernel (v6.14) builds from a virtme-ng minimum .config + the enabled config symbols for the host-guest tracing. The same environment also works fine with the latest versions of trace-cmd. (Unless I am mistaken,) your setup uses Fedora distribution kernels for both the host and the guest, along with the distro package for trace-cmd. Can you please confirm that all the kernel config symbols required for host-guest tracing are enabled in your (virtualized) host kernel and in both of the (nested) guest kernels? I also suggest testing that local tracing on the host and inside the guests work fine, and confirming that the problem only occurs upon attempting the host-guest tracing. Next thing to check would be your vsock set-up, i.e. is your trace-cmd agent in the guest listening to the the same cid that you are passing as the argument in your trace-cmd record command on the host? I would also advice against using nested virtualization for these experiments. If you don't mind compiling the kernel, then I suggest trying out virtme-ng. We have @Andrea as a co-mentor, so questions related to virtme-ng's usage for our experiment set-up are also welcome. Himadri >>>
