On Mon, Dec 04, 2023 at 10:13:34AM +0100, Peter Zijlstra wrote:

quoted

Just running test_progs it splats right away:

[   74.047757] kmemleak: Found object by alias at 0xffffffffa0001d80
[   74.048272] CPU: 14 PID: 104 Comm: kworker/14:0 Tainted: G        W
 O       6.7.0-rc3-00702-g41c30fec304d-dirty #5241
[   74.049118] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014
[   74.050042] Workqueue: events bpf_prog_free_deferred
[   74.050448] Call Trace:
[   74.050663]  <TASK>
[   74.050841]  dump_stack_lvl+0x55/0x80
[   74.051141]  __find_and_remove_object+0xdb/0x110
[   74.051521]  kmemleak_free+0x41/0x70
[   74.051828]  vfree+0x36/0x130

Durr, I'll see if I can get that stuff running locally, and otherwise
play with the robot as you suggested. Thanks!

I think it is bpf_jit_binary_pack_hdr(), which is using prog->bpf_func
as a start address for the image, instead of jit_data->image.

This used to be true, but now it's offset.

Let me see what to do about that...