Warning on module unload: "BUG: using smp_processor_id() in preemptible"

[aldem]

Looks very similar to #193, but I believe that the proposed PR actually does not fix it:

Mar 12 22:07:57 proton kernel: BUG: using smp_processor_id() in preemptible [00000000] code: rmmod/188620
Mar 12 22:07:57 proton kernel: caller is netflow_scan_and_export+0x221/0x6c0 [ipt_NETFLOW]
Mar 12 22:07:57 proton kernel: CPU: 0 PID: 188620 Comm: rmmod Tainted: G           OE      6.0.0-0.deb11.6-cloud-amd64 #1  Debian 6.0.12-1~bpo11+1
Mar 12 22:07:57 proton kernel: Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
Mar 12 22:07:57 proton kernel: Call Trace:
Mar 12 22:07:57 proton kernel:  <TASK>
Mar 12 22:07:57 proton kernel:  dump_stack_lvl+0x45/0x5e
Mar 12 22:07:57 proton kernel:  check_preemption_disabled+0xe1/0xf0
Mar 12 22:07:57 proton kernel:  netflow_scan_and_export+0x221/0x6c0 [ipt_NETFLOW]
Mar 12 22:07:57 proton kernel:  ipt_netflow_fini+0x78/0xe63 [ipt_NETFLOW]

It seems that problem is a bit different, I encountered this recently and it happens only on module unload (any version, including current master), when netflow_scan_and_export() finally called to flush the remaining flows:

ipt-netflow/ipt_NETFLOW.c

Lines 5801 to 5802 in b049e91

	_unschedule_scan_worker();
	netflow_scan_and_export(AND_FLUSH);

As long as the kernel message is nothing more than a warning, and module is not reloaded very often, this is not a big issue, though anything in kernel log starting with a "BUG: " is a bit scary.

However, it would be nice to get rid of this if possible - I fixed this by surrounding the final call with preempt_disable() and preempt_enable() (works for me) but not sure if it could have some negative consequences, especially if it takes some time to finish.

[kalamlacki]

I build kernel with preempting disabled just for this module purpose. Last time when I mentioned about this BUG someone made stupid change to hide "BUG: using smp_processor_id() in preemptible" and the change was accepted, so for me this module will slowly die on missing support.