Hi Takashi-san

...

commit da215354eb55c ("ASoC: simple-card: merge simple-scu-card") merged simple-card and simple-scu-card. Then it had refcount underflow bug. This patch fixup it. We will get below error without this patch.

OF: ERROR: Bad of_node_put() on /sound CPU: 3 PID: 237 Comm: kworker/3:1 Not tainted 5.0.0-rc6+ #1514 Hardware name: Renesas H3ULCB Kingfisher board based on r8a7795 ES2.0+ (DT) Workqueue: events deferred_probe_work_func Call trace: dump_backtrace+0x0/0x150 show_stack+0x24/0x30 dump_stack+0xb0/0xec of_node_release+0xd0/0xd8 kobject_put+0x74/0xe8 of_node_put+0x24/0x30 __of_get_next_child+0x50/0x70 of_get_next_child+0x40/0x68 asoc_simple_card_probe+0x604/0x730 platform_drv_probe+0x58/0xa8 ... Reported-by: Vicente Bergas vicencb@gmail.com Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com

Please don't forget to put Fixes tag for a regression fix.

Oops, indeed. Thanks. will add it on v2 patch.

And, looking at the code in 5.0-rc, it seems still leaving the reference in some condition. The 5.0-rc code corrected your patch looks like:

node = of_get_child_by_name(top, PREFIX "dai-link"); if (!node) { node = of_node_get(top); loop = 0; }

do { .... node = of_get_next_child(top, node); } while (loop && node);

So when loop=0, it aborts the loop at the first iteration. If node is non-NULL at that point, it still keeps the refcount taken in of_get_next_child().

That is, you'd need to call of_node_put(node) after the loop.

Ah.. yes..

And, this leads me checking the whole code, and I'm afraid that we have many similar unblanced refcounts. For example, a tricky one is for_each_child_of_node() loop. When you abort at the middle of the loop, you'd need to unreference the remaining node. But we return or abort immediately without unreferencing in many places.

Hrm...

Yes, I had noticed this refcounts issue. The code can be very confusable for this refcounts...

Best regards --- Kuninori Morimoto