Hi Laurent,
> Am 29.09.2016 um 10:54 schrieb Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>:
>
> Hi Nikolaus,
>
> On Thursday 08 Sep 2016 17:48:33 H. Nikolaus Schaller wrote:
>> of_node_put() has already been called inside of_graph_get_next_endpoint().
>>
>> Otherwise we may get warnings like
>>
>> [ 10.118286] omap3isp 480bc000.isp: parsing endpoint
>> /ocp/isp@480bc000/ports/port@0/endpoint, interface 0 [ 10.118499] ERROR:
>> Bad of_node_put() on /ocp/isp@480bc000/ports/port@0/endpoint [ 10.118499]
>> CPU: 0 PID: 968 Comm: udevd Not tainted 4.7.0-rc4-letux+ #376 [
>> 10.118530] Hardware name: Generic OMAP36xx (Flattened Device Tree) [
>> 10.118560] [<c010f0e0>] (unwind_backtrace) from [<c010b6d8>]
>> (show_stack+0x10/0x14) [ 10.118591] [<c010b6d8>] (show_stack) from
>> [<c03ecc50>] (dump_stack+0x98/0xd0) [ 10.118591] [<c03ecc50>]
>> (dump_stack) from [<c03eecac>] (kobject_release+0x60/0x74) [ 10.118621]
>> [<c03eecac>] (kobject_release) from [<c05ab128>]
>> (__of_get_next_child+0x40/0x48) [ 10.118652] [<c05ab128>]
>> (__of_get_next_child) from [<c05ab158>] (of_get_next_child+0x28/0x44) [
>> 10.118652] [<c05ab158>] (of_get_next_child) from [<c05ab350>]
>> (of_graph_get_next_endpoint+0xe4/0x124) [ 10.118804] [<c05ab350>]
>> (of_graph_get_next_endpoint) from [<bf1c88a4>] (isp_probe+0xdc/0xd80
>> [omap3_isp]) [ 10.118896] [<bf1c88a4>] (isp_probe [omap3_isp]) from
>> [<c0482008>] (platform_drv_probe+0x50/0xa0) [ 10.118927] [<c0482008>]
>> (platform_drv_probe) from [<c04800e8>] (driver_probe_device+0x134/0x29c) [
>> 10.118957] [<c04800e8>] (driver_probe_device) from [<c04802d8>]
>> (__driver_attach+0x88/0xac) [ 10.118957] [<c04802d8>] (__driver_attach)
>> from [<c047e7b8>] (bus_for_each_dev+0x6c/0x90) [ 10.118957] [<c047e7b8>]
>> (bus_for_each_dev) from [<c047f798>] (bus_add_driver+0xcc/0x1e8) [
>> 10.118988] [<c047f798>] (bus_add_driver) from [<c0481228>]
>> (driver_register+0xac/0xf4) [ 10.118988] [<c0481228>] (driver_register)
>> from [<c010192c>] (do_one_initcall+0xac/0x154) [ 10.119018] [<c010192c>]
>> (do_one_initcall) from [<c02015bc>] (do_init_module+0x58/0x39c) [
>> 10.119049] [<c02015bc>] (do_init_module) from [<c01bd314>]
>> (load_module+0xe5c/0x1004) [ 10.119049] [<c01bd314>] (load_module) from
>> [<c01bd68c>] (SyS_finit_module+0x88/0x90) [ 10.119079] [<c01bd68c>]
>> (SyS_finit_module) from [<c0107040>] (ret_fast_syscall+0x0/0x1c)
>>
>> Signed-off-by: H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>
>> ---
>> drivers/media/platform/omap3isp/isp.c | 3 +--
>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/drivers/media/platform/omap3isp/isp.c
>> b/drivers/media/platform/omap3isp/isp.c index 5d54e2c..6e2624e 100644
>> --- a/drivers/media/platform/omap3isp/isp.c
>> +++ b/drivers/media/platform/omap3isp/isp.c
>> @@ -2114,7 +2114,6 @@ static int isp_of_parse_nodes(struct device *dev,
>>
>> isd = devm_kzalloc(dev, sizeof(*isd), GFP_KERNEL);
>> if (!isd) {
>> - of_node_put(node);
>
> I don't think this one is correct. Looking at the context
>
> while (notifier->num_subdevs < ISP_MAX_SUBDEVS &&
> (node = of_graph_get_next_endpoint(dev->of_node, node))) {
> struct isp_async_subdev *isd;
>
> isd = devm_kzalloc(dev, sizeof(*isd), GFP_KERNEL);
> if (!isd) {
> of_node_put(node);
> return -ENOMEM;
> }
>
> notifier->subdevs[notifier->num_subdevs] = &isd->asd;
>
> if (isp_of_parse_node(dev, node, isd)) {
> of_node_put(node);
> return -EINVAL;
> }
>
> isd->asd.match.of.node =
> of_graph_get_remote_port_parent(node);
> of_node_put(node);
>
> if (!isd->asd.match.of.node) {
> dev_warn(dev, "bad remote port parent\n");
> return -EINVAL;
> }
>
> isd->asd.match_type = V4L2_ASYNC_MATCH_OF;
> notifier->num_subdevs++;
> }
>
> of_graph_get_next_endpoint() increases the reference count of the node it
> returns, which needs a corresponding of_node_put() in the error paths. It
> thus look like to me that the function isn't correct in that the
> devm_kzalloc() and !isd->asd.match.of.node error paths.
Ah ok, the of_node_put() is not always wrong.
>
>> return -ENOMEM;
>> }
>>
>> @@ -2126,7 +2125,7 @@ static int isp_of_parse_nodes(struct device *dev,
>> }
>>
>> isd->asd.match.of.node = of_graph_get_remote_port_parent(node);
>> - of_node_put(node);
>> +
>
> This change is correct not because of_graph_get_next_endpoint() has called
> of_node_put() but because it *will* call it on the next iteration.
>
> How about the following patch instead ?
>
> commit 4ed9893bf52c90181c8d5b1ae29a37556b89f1da
> Author: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
> Date: Thu Sep 29 11:41:24 2016 +0300
>
> omap3isp: Fix OF node double put when parsing OF graph
>
> When parsing the graph the driver loops over all endpoints using
> of_graph_get_next_endpoint(). The function handles reference counting of
> the passed and returned nodes, so the returned node's reference count
> must not be decreased manually in the normal path.
>
> Move the offending of_node_put() call to the error path that requires
> manual reference count handling.
>
> Reported-by: H. Nikolaus Schaller <hns@xxxxxxxxxxxxx>
> Signed-off-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
>
> diff --git a/drivers/media/platform/omap3isp/isp.c b/drivers/media/platform/omap3isp/isp.c
> index 5e212668f726..f8b437cc8943 100644
> --- a/drivers/media/platform/omap3isp/isp.c
> +++ b/drivers/media/platform/omap3isp/isp.c
> @@ -2131,23 +2131,18 @@ static int isp_of_parse_nodes(struct device *dev,
> struct isp_async_subdev *isd;
>
> isd = devm_kzalloc(dev, sizeof(*isd), GFP_KERNEL);
> - if (!isd) {
> - of_node_put(node);
> - return -ENOMEM;
> - }
> + if (!isd)
> + goto error;
>
> notifier->subdevs[notifier->num_subdevs] = &isd->asd;
>
> - if (isp_of_parse_node(dev, node, isd)) {
> - of_node_put(node);
> - return -EINVAL;
> - }
> + if (isp_of_parse_node(dev, node, isd))
> + goto error;
>
> isd->asd.match.of.node = of_graph_get_remote_port_parent(node);
> - of_node_put(node);
> if (!isd->asd.match.of.node) {
> dev_warn(dev, "bad remote port parent\n");
> - return -EINVAL;
> + goto error;
> }
>
> isd->asd.match_type = V4L2_ASYNC_MATCH_OF;
> @@ -2155,6 +2150,10 @@ static int isp_of_parse_nodes(struct device *dev,
> }
>
> return notifier->num_subdevs;
> +
> +error:
> + of_node_put(node);
> + return -EINVAL;
> }
>
> static int isp_subdev_notifier_bound(struct v4l2_async_notifier *async,
>
>> if (!isd->asd.match.of.node) {
>> dev_warn(dev, "bad remote port parent\n");
>> return -EINVAL;
Seems to fix the reported warning, but obviously I can't test the error paths.
So let's go this way (until someone shows a bug in the error path).
BR and thanks,
Nikolaus
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
