On 10/18/2020 3:58 PM, Vladimir Oltean wrote:
On Sun, Oct 18, 2020 at 04:13:28PM +0200, Heiner Kallweit wrote:
On a side note, because I'm also just dealing with it: this_cpu_ptr()
is safe only if preemption is disabled. Is this the case here? Else
there's get_cpu_ptr/put_cpu_ptr.
lockdep would shout about using smp_processor_id() in preemptible
context? Because it doesn't do that, and never did, but I don't really
understand why, coming to think of it.
Also, if irq's aren't disabled there might be a need to use
u64_stats_update_begin_irqsave() et al. See: 2695578b896a ("net:
usbnet: fix potential deadlock on 32bit hosts") But I don't know dsa
good enough to say whether this is applicable here.
DSA xmit and receive path do not run from hardirq context, just softirq,
so I believe the issue reported to Eric there does not apply here.
How about when used as a netconsole? We do support netconsole over DSA
interfaces.
--
Florian
