On 1/16/26 11:11, Stefano Garzarella wrote:

On Fri, Jan 16, 2026 at 09:52:36AM +0100, Michal Luczaj wrote:

quoted

Blamed commit implemented logic to discover available vsock transports by
grepping /proc/kallsyms for known symbols. It incorrectly filtered entries
by type 'd'.

For some kernel configs having

   CONFIG_VIRTIO_VSOCKETS=m
   CONFIG_VSOCKETS_LOOPBACK=y

kallsyms reports

   0000000000000000 d virtio_transport	[vmw_vsock_virtio_transport]
   0000000000000000 t loopback_transport

Overzealous filtering might have affected vsock test suit, resulting in
insufficient/misleading testing.

Do not filter symbols by type. It never helped much.

Fixes: 3070c05b7afd ("vsock/test: Introduce get_transports()")
Signed-off-by: Michal Luczaj <redacted>
---
man nm says: 't' stands for symbol is in the text (code) section. Is this
correct for `static struct virtio_transport loopback_transport`?

I'm not an expert, but yeah I was expecting "d" too, but maybe since
it's static and built-in will be in the text section?

But it does not end up in the text section. Address points at RW NX
(read-writeable non-executable) pages. Please see below.

BTW I just checked and for example on my 6.18.4-100.fc42.x86_64 I have:

0000000000000000 t sock_fs_type
0000000000000000 t proto_net_ops
0000000000000000 t net_inuse_ops

And they are all static structs of built-in modules.
So it seems it is common.

$ mv .config .config.orig
$ vng -k --configitem RANDOMIZE_BASE=n --configitem CONFIG_PTDUMP_DEBUGFS=y
$ vng -b
$ vng --user root "grep sock_fs_type /proc/kallsyms"
ffffffff82c1f000 t sock_fs_type
$ vng --user root "grep 0xffffffff82 /sys/kernel/debug/page_tables/kernel"
0xffffffff82a00000-0xffffffff83000000     6M    RW    PSE   GLB NX pmd

And if I try a 2-year-old v6.13, kallsyms shows 'd' as we'd expect. I've
bisected it down to commit cb33ff9e063c ("x86/kexec: Move relocate_kernel
to kernel .data section"), but I still don't know if it's a bug or a feature.