Re: [PATCH net-next v2] net: opt loopback into instance locking

From
Stanislav Fomichev <hidden>
Date
2026-09-22 19:21:49
Reply to
Jakub Kicinski, 2026-09-22
On 09/22, Jakub Kicinski wrote:
On Mon, 21 Sep 2026 14:14:19 -0700 Stanislav Fomichev wrote:
quoted
On 09/21, Jakub Kicinski wrote:
quoted
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
v2:
 - move setting the request flag from gen_lo_setup() to loopback_setup()
   to avoid splats on the blackhole dev which we don't care about  
Acked-by: Stanislav Fomichev <sdf@fomichev.me>

(mostly because dummy is also ops locked, don't think loopback should
bring any problems)
Stan, the issue Matt reported in the BPF CI is a generic sw device +
netdev lock issue. We are triggering it on lo because.. probability,
but it reproduces with dummy + netkit already today.

Do you recall why we have separate classes for each device type?

Classes make the lock_set_cmp_fn() ineffective across different device
types. Quick test removing the classes and sticking to just the cmp fn
seems make lockdep happy, I'm running the ksft suite now. But maybe you
can tell me if it's the right idea before it finishes..?
No, I don't think there is a reason we added both. It might have been
that we started with a class and then added cmp_fn on top. This whole
function is a 10+ years of whack-a-mole with lockdep :-( Agreed that
removing class (which prevents cmp_fn from working properly across
all instance locks) sounds sensible.
Thread (6 messages) 6 messages, 4 authors, last reply 2026-09-24
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help