2013-06-12 05:39:29

by Benjamin Herrenschmidt

[permalink] [raw]
Subject: [PATCH] of: Fix locking vs. interrupts

The OF code uses irqsafe locks everywhere except in a handful of functions
for no obvious reasons. Since the conversion from the old rwlocks, this
now triggers lockdep warnings when used at interrupt time. At least one
driver (ibmvscsi) seems to be doing that from softirq context.

This converts the few non-irqsafe locks into irqsafe ones, making them
consistent with the rest of the code.

Signed-off-by: Benjamin Herrenschmidt <[email protected]>
CC: <[email protected]> [v3.9+]
---

Note: It's silly to access the device-tree at interrupt time in most cases,
and we should probably fix ibmvscsi, but for the time being, let's fix the
obvious bug. Thomas, this can probably still go into 3.10... If not, I've
CCed stable.

diff --git a/arch/sparc/kernel/prom_common.c b/arch/sparc/kernel/prom_common.c
index 9f20566..79cc0d1 100644
--- a/arch/sparc/kernel/prom_common.c
+++ b/arch/sparc/kernel/prom_common.c
@@ -54,6 +54,7 @@ EXPORT_SYMBOL(of_set_property_mutex);
int of_set_property(struct device_node *dp, const char *name, void *val, int len)
{
struct property **prevp;
+ unsigned long flags;
void *new_val;
int err;

@@ -64,7 +65,7 @@ int of_set_property(struct device_node *dp, const char *name, void *val, int len
err = -ENODEV;

mutex_lock(&of_set_property_mutex);
- raw_spin_lock(&devtree_lock);
+ raw_spin_lock_irqsave(&devtree_lock, flags);
prevp = &dp->properties;
while (*prevp) {
struct property *prop = *prevp;
@@ -91,7 +92,7 @@ int of_set_property(struct device_node *dp, const char *name, void *val, int len
}
prevp = &(*prevp)->next;
}
- raw_spin_unlock(&devtree_lock);
+ raw_spin_unlock_irqrestore(&devtree_lock, flags);
mutex_unlock(&of_set_property_mutex);

/* XXX Upate procfs if necessary... */
diff --git a/drivers/of/base.c b/drivers/of/base.c
index f53b992..951452c 100644
--- a/drivers/of/base.c
+++ b/drivers/of/base.c
@@ -192,14 +192,15 @@ EXPORT_SYMBOL(of_find_property);
struct device_node *of_find_all_nodes(struct device_node *prev)
{
struct device_node *np;
+ unsigned long flags;

- raw_spin_lock(&devtree_lock);
+ raw_spin_lock_irqsave(&devtree_lock, flags);
np = prev ? prev->allnext : of_allnodes;
for (; np != NULL; np = np->allnext)
if (of_node_get(np))
break;
of_node_put(prev);
- raw_spin_unlock(&devtree_lock);
+ raw_spin_unlock_irqrestore(&devtree_lock, flags);
return np;
}
EXPORT_SYMBOL(of_find_all_nodes);
@@ -421,8 +422,9 @@ struct device_node *of_get_next_available_child(const struct device_node *node,
struct device_node *prev)
{
struct device_node *next;
+ unsigned long flags;

- raw_spin_lock(&devtree_lock);
+ raw_spin_lock_irqsave(&devtree_lock, flags);
next = prev ? prev->sibling : node->child;
for (; next; next = next->sibling) {
if (!__of_device_is_available(next))
@@ -431,7 +433,7 @@ struct device_node *of_get_next_available_child(const struct device_node *node,
break;
}
of_node_put(prev);
- raw_spin_unlock(&devtree_lock);
+ raw_spin_unlock_irqrestore(&devtree_lock, flags);
return next;
}
EXPORT_SYMBOL(of_get_next_available_child);
@@ -735,13 +737,14 @@ EXPORT_SYMBOL_GPL(of_modalias_node);
struct device_node *of_find_node_by_phandle(phandle handle)
{
struct device_node *np;
+ unsigned long flags;

- raw_spin_lock(&devtree_lock);
+ raw_spin_lock_irqsave(&devtree_lock, flags);
for (np = of_allnodes; np; np = np->allnext)
if (np->phandle == handle)
break;
of_node_get(np);
- raw_spin_unlock(&devtree_lock);
+ raw_spin_unlock_irqrestore(&devtree_lock, flags);
return np;
}
EXPORT_SYMBOL(of_find_node_by_phandle);


2013-06-12 08:26:07

by Thomas Gleixner

[permalink] [raw]
Subject: Re: [PATCH] of: Fix locking vs. interrupts

On Wed, 12 Jun 2013, Benjamin Herrenschmidt wrote:

> The OF code uses irqsafe locks everywhere except in a handful of functions
> for no obvious reasons. Since the conversion from the old rwlocks, this
> now triggers lockdep warnings when used at interrupt time. At least one
> driver (ibmvscsi) seems to be doing that from softirq context.
>
> This converts the few non-irqsafe locks into irqsafe ones, making them
> consistent with the rest of the code.

Fun. https://lkml.org/lkml/2013/2/4/416 seems to have got lost

> Signed-off-by: Benjamin Herrenschmidt <[email protected]>
> CC: <[email protected]> [v3.9+]

Acked-by: Thomas Gleixner <[email protected]>

> ---
>
> Note: It's silly to access the device-tree at interrupt time in most cases,
> and we should probably fix ibmvscsi, but for the time being, let's fix the

Right.

> obvious bug. Thomas, this can probably still go into 3.10... If not, I've
> CCed stable.

Should go through Grant I think.

Thanks,

tglx

2013-06-12 08:56:56

by Benjamin Herrenschmidt

[permalink] [raw]
Subject: Re: [PATCH] of: Fix locking vs. interrupts

On Wed, 2013-06-12 at 10:25 +0200, Thomas Gleixner wrote:
> On Wed, 12 Jun 2013, Benjamin Herrenschmidt wrote:
>
> > The OF code uses irqsafe locks everywhere except in a handful of functions
> > for no obvious reasons. Since the conversion from the old rwlocks, this
> > now triggers lockdep warnings when used at interrupt time. At least one
> > driver (ibmvscsi) seems to be doing that from softirq context.
> >
> > This converts the few non-irqsafe locks into irqsafe ones, making them
> > consistent with the rest of the code.
>
> Fun. https://lkml.org/lkml/2013/2/4/416 seems to have got lost
>
> > Signed-off-by: Benjamin Herrenschmidt <[email protected]>
> > CC: <[email protected]> [v3.9+]
>
> Acked-by: Thomas Gleixner <[email protected]>
>
> > ---
> >
> > Note: It's silly to access the device-tree at interrupt time in most cases,
> > and we should probably fix ibmvscsi, but for the time being, let's fix the
>
> Right.
>
> > obvious bug. Thomas, this can probably still go into 3.10... If not, I've
> > CCed stable.
>
> Should go through Grant I think.

Right, thinko. Sent to you due to the bug being exposed by your
conversion to spinlocks. Anyway, Grant got it now.

Cheers,
Ben.


2013-06-12 09:06:23

by Grant Likely

[permalink] [raw]
Subject: Re: [PATCH] of: Fix locking vs. interrupts

On Wed, 12 Jun 2013 10:25:56 +0200 (CEST), Thomas Gleixner <[email protected]> wrote:
> On Wed, 12 Jun 2013, Benjamin Herrenschmidt wrote:
>
> > The OF code uses irqsafe locks everywhere except in a handful of functions
> > for no obvious reasons. Since the conversion from the old rwlocks, this
> > now triggers lockdep warnings when used at interrupt time. At least one
> > driver (ibmvscsi) seems to be doing that from softirq context.
> >
> > This converts the few non-irqsafe locks into irqsafe ones, making them
> > consistent with the rest of the code.
>
> Fun. https://lkml.org/lkml/2013/2/4/416 seems to have got lost
>
> > Signed-off-by: Benjamin Herrenschmidt <[email protected]>
> > CC: <[email protected]> [v3.9+]
>
> Acked-by: Thomas Gleixner <[email protected]>
>
> > ---
> >
> > Note: It's silly to access the device-tree at interrupt time in most cases,
> > and we should probably fix ibmvscsi, but for the time being, let's fix the
>
> Right.
>
> > obvious bug. Thomas, this can probably still go into 3.10... If not, I've
> > CCed stable.
>
> Should go through Grant I think.

Applied, thanks.

g.

2013-06-12 09:15:44

by David Miller

[permalink] [raw]
Subject: Re: [PATCH] of: Fix locking vs. interrupts

From: Grant Likely <[email protected]>
Date: Wed, 12 Jun 2013 10:06:12 +0100

> On Wed, 12 Jun 2013 10:25:56 +0200 (CEST), Thomas Gleixner <[email protected]> wrote:
>> On Wed, 12 Jun 2013, Benjamin Herrenschmidt wrote:
>>
>> > The OF code uses irqsafe locks everywhere except in a handful of functions
>> > for no obvious reasons. Since the conversion from the old rwlocks, this
>> > now triggers lockdep warnings when used at interrupt time. At least one
>> > driver (ibmvscsi) seems to be doing that from softirq context.
>> >
>> > This converts the few non-irqsafe locks into irqsafe ones, making them
>> > consistent with the rest of the code.
>>
>> Fun. https://lkml.org/lkml/2013/2/4/416 seems to have got lost
>>
>> > Signed-off-by: Benjamin Herrenschmidt <[email protected]>
>> > CC: <[email protected]> [v3.9+]
>>
>> Acked-by: Thomas Gleixner <[email protected]>
>>
>> > ---
>> >
>> > Note: It's silly to access the device-tree at interrupt time in most cases,
>> > and we should probably fix ibmvscsi, but for the time being, let's fix the
>>
>> Right.
>>
>> > obvious bug. Thomas, this can probably still go into 3.10... If not, I've
>> > CCed stable.
>>
>> Should go through Grant I think.
>
> Applied, thanks.

FWIW:

Acked-by: David S. Miller <[email protected]>