+Grant Hi, On Tuesday 17 November 2015 07:08 AM, Brian Norris wrote: > On Mon, Nov 16, 2015 at 12:33:14PM +0100, Julia Lawall wrote: >> for_each_available_child_of_node performs an of_node_get on each iteration, >> so a return from the middle of the loop requires an of_node_put. >> >> A simplified version of the semantic patch that finds this problem is as >> follows (http://coccinelle.lip6.fr): >> >> // <smpl> >> @@ >> expression root,e; >> local idexpression child; >> @@ >> >> for_each_available_child_of_node(root, child) { >> ... when != of_node_put(child) >> when != e = child >> ( >> return child; >> | >> * return ...; >> ) >> ... >> } >> // </smpl> >> >> Signed-off-by: Julia Lawall <Julia.Lawall@xxxxxxx> >> >> --- > > For this patch: > > Acked-by: Brian Norris <computersforpeace@xxxxxxxxx> > >> drivers/phy/phy-brcmstb-sata.c | 17 ++++++++++++----- >> 1 file changed, 12 insertions(+), 5 deletions(-) > > [snip patch, which fixes of_node_put() handling for > for_each_available_child_of_node() loop, which creates PHY devices with > devm_phy_create()] > > This reminds me of a potential problem I'm looking at in other > subsystems: from code reading (I haven't seen any issues in practice, > probably because I don't use OF_DYNAMIC) it looks like device-creating > infrastructure like the PHY subsystem should be acquiring a reference to > the device_node when they stash it away. But drivers/phy/phy-core.c does > not do this, AFAICT. > > See phy_create(), which does > > phy->dev.of_node = node ?: dev->of_node; > > and later might reuse this of_node pointer, even though it never called > of_node_get() on this node. > > Potential patch to fix this (not tested). > > Signed-off-by: Brian Norris <computersforpeace@xxxxxxxxx> > > diff --git a/drivers/phy/phy-core.c b/drivers/phy/phy-core.c > index fc48fac003a6..8df29caeeef9 100644 > --- a/drivers/phy/phy-core.c > +++ b/drivers/phy/phy-core.c > @@ -697,6 +697,7 @@ struct phy *phy_create(struct device *dev, struct device_node *node, > phy->dev.class = phy_class; > phy->dev.parent = dev; > phy->dev.of_node = node ?: dev->of_node; > + of_node_get(phy->dev.of_node); > phy->id = id; > phy->ops = ops; > > @@ -726,6 +727,7 @@ struct phy *phy_create(struct device *dev, struct device_node *node, > return phy; > > put_dev: > + of_node_put(phy->dev.of_node); > put_device(&phy->dev); /* calls phy_release() which frees resources */ > return ERR_PTR(ret); > > @@ -775,6 +777,7 @@ EXPORT_SYMBOL_GPL(devm_phy_create); > */ > void phy_destroy(struct phy *phy) > { > + of_node_put(phy->dev.of_node); I think it's better to have this patch in phy-core though OF_DYNAMIC is not enabled? Grant, Is it safe to assume of_node_get() will prevent "anyone else" from deleting the node? Here phy core uses the node pointer (passed to it by phy providers) and we would like to avoid "anyone" from removing this node pointer resulting in phy core having an invalid node pointer. Using of_node_get() in phy core should be sufficient for this? We are also interested in this todo tasklist for Devicetree.. "Document node lifecycle for CONFIG_OF_DYNAMIC" Please find the complete thread of this mail chain here [1] [1] -> http://www.gossamer-threads.com/lists/linux/kernel/2304857 Thanks Kishon > pm_runtime_disable(&phy->dev); > device_unregister(&phy->dev); > } > -- To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
