On 10/20/25 14:53, Song Liu wrote:
On Mon, Oct 20, 2025 at 9:45 AM Andrey Grodzovsky
<[email protected]> wrote:
On 10/20/25 12:03, Song Liu wrote:
On Mon, Oct 20, 2025 at 7:56 AM Andrey Grodzovsky
<[email protected]> wrote:
[...]
If you build the kernel from source code, there are some samples in
samples/livepatch that you can use for testing. PS: You need to enable
CONFIG_SAMPLE_LIVEPATCH=m
I hope this helps.
Thanks Song, working on repro, kernel rebuilt, test module is loading
but, bpftrace is refusing to attach now to fentries/fexits claiming the
costum kernel is not supporting it. It did
attach in the case of stock AWS kernel i copied the .config from. So
just trying to figure out now if some Kcofnig flags are missing or
different . Let me know in case you manage to confirm yourself in the
meanwhile the fix works for
you.
Yes, it worked in my tests.
[root@(none) /]# kpatch load linux/samples/livepatch/livepatch-sample.ko
loading patch module: linux/samples/livepatch/livepatch-sample.ko
[root@(none) /]# bpftrace.real -e 'fexit:cmdline_proc_show
{printf("fexit\n");}' &
[1] 388
[root@(none) /]# Attached 1 probe
[root@(none) /]# bpftrace.real -e 'fentry:cmdline_proc_show
{printf("fentry\n");}' &
[2] 397
[root@(none) /]# Attached 1 probe
[root@(none) /]# cat /proc/cmdline
this has been live patched
fentry
fexit
Thanks,
Song
Verified the failures I observe when trying to attach with BPF trace are
only in presence of patch you provided.
Please see attached dmesg for failures. Initial warning on boot.
Subsequebt warnings and errors at the point i try to run
sudo bpftrace -e "fexit:cmdline_proc_show { printf(\"fexit hit\\n\");
exit(); }"
sudo: unable to resolve host ip-10-10-115-238: Temporary failure in name
resolution
stdin:1:1-25: ERROR: kfunc/kretfunc not available for your kernel version.
ubuntu@ip-10-10-115-238:~/linux-6.8.1$ sudo cat
/sys/kernel/debug/tracing/available_filter_functions | grep
cmdline_proc_show
sudo: unable to resolve host ip-10-10-115-238: Temporary failure in name
resolution
cat: /sys/kernel/debug/tracing/available_filter_functions: No such device
After reboot and before trying to attacg with bpftrace,
/sys/kernel/debug/tracing/available_filter_functions is available and
shows all function.
Using stable kernel from
https://urldefense.com/v3/__https://cdn.kernel.org/pub/linux/kernel/v6.x/linux-6.8.1.tar.gz__;!!BmdzS3_lV9HdKG8!1ZJe4jY49_xIzp4h4i4AbqpkLKoAqrXLFX2wDxhoSUDg2kSeTjy3COy9MngNDRlZhJ1oUKgf1yPqmnTY9-Y50TkA$
for build.
FTRACE related KCONFIGs bellow
I can see the similar issue with the upstream kernel. I was testing on
stable 6.17 before just know because of another issue with upstream
kernel, and somehow 6.17 kernel doesn't seem to have the issue.
To fix this, I think we should land a fix similar to the earlier diff:
diff --git i/kernel/trace/ftrace.c w/kernel/trace/ftrace.c
index 42bd2ba68a82..8f320df0ac52 100644
--- i/kernel/trace/ftrace.c
+++ w/kernel/trace/ftrace.c
@@ -6049,6 +6049,9 @@ int register_ftrace_direct(struct ftrace_ops
*ops, unsigned long addr)
err = register_ftrace_function_nolock(ops);
+ if (err)
+ remove_direct_functions_hash(hash, addr);
+
out_unlock:
mutex_unlock(&direct_mutex);
Steven,
Does this change look good to you?
Seems reasonable to me, we are simply cleaning the entry on failure so
we don't encounter it late anymore.
So I will apply this patch ONLY and retest - correct ?
Another question - it seems you found where it broke ? I saw 'Cc:
[email protected] # v6.6+' in your prev. patch.'
If so , can you please point me to the offending patch so I add this to
my records of my discovery work of bpf coexistence
livepatching ?
Thanks,
Andrey
Thanks,
Song
