| Message ID | 1361974273-17087-2-git-send-email-mkl@pengutronix.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Feb 27, 2013 at 03:11:13PM +0100, Marc Kleine-Budde wrote: > In patch "5d3c28b usb: otg: add device tree support to otg library" > devm_usb_get_phy_by_phandle() was added. It uses try_module_get() to lock the > phy driver in memory. The corresponding module_put() is missing in that patch. > > This patch adds try_module_get() to usb_get_phy() and usb_get_phy_dev(). > Further the missing module_put() is added to usb_put_phy(). > > Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com> > Acked-by: Felipe Balbi <balbi@ti.com> > Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> > Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de> > --- > drivers/usb/otg/otg.c | 10 +++++++--- > 1 file changed, 7 insertions(+), 3 deletions(-) > > diff --git a/drivers/usb/otg/otg.c b/drivers/usb/otg/otg.c > index e181439..2bd03d2 100644 > --- a/drivers/usb/otg/otg.c > +++ b/drivers/usb/otg/otg.c > @@ -130,7 +130,7 @@ struct usb_phy *usb_get_phy(enum usb_phy_type type) > spin_lock_irqsave(&phy_lock, flags); > > phy = __usb_find_phy(&phy_list, type); > - if (IS_ERR(phy)) { > + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { Ugh, really? We really are trying this type of module locking mess? Why? What is it solving? What's wrong with having the module be able to be unloaded whenever it wants to? No one should be doing that and expecting that their hardware would still work properly, right? I really don't like this type of thing, sorry. greg k-h
On 02/27/2013 05:21 PM, Greg KH wrote: > On Wed, Feb 27, 2013 at 03:11:13PM +0100, Marc Kleine-Budde wrote: >> In patch "5d3c28b usb: otg: add device tree support to otg library" >> devm_usb_get_phy_by_phandle() was added. It uses try_module_get() to lock the >> phy driver in memory. The corresponding module_put() is missing in that patch. >> >> This patch adds try_module_get() to usb_get_phy() and usb_get_phy_dev(). >> Further the missing module_put() is added to usb_put_phy(). >> >> Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com> >> Acked-by: Felipe Balbi <balbi@ti.com> >> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> >> Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de> >> --- >> drivers/usb/otg/otg.c | 10 +++++++--- >> 1 file changed, 7 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/usb/otg/otg.c b/drivers/usb/otg/otg.c >> index e181439..2bd03d2 100644 >> --- a/drivers/usb/otg/otg.c >> +++ b/drivers/usb/otg/otg.c >> @@ -130,7 +130,7 @@ struct usb_phy *usb_get_phy(enum usb_phy_type type) >> spin_lock_irqsave(&phy_lock, flags); >> >> phy = __usb_find_phy(&phy_list, type); >> - if (IS_ERR(phy)) { >> + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { > > Ugh, really? We really are trying this type of module locking mess? > > Why? What is it solving? What's wrong with having the module be able Without this patch, you can unload the phy module, while it is in use. As the phy framework doesn't use any existing abstraction to handle phy <-> user of phy pairing/unpairing (it's all open coded), any subsequent use of the phy's callback will result in deref'ing bogus pointers. > to be unloaded whenever it wants to? No one should be doing that and > expecting that their hardware would still work properly, right? Yes, but it should not result in an kernel oops. > I really don't like this type of thing, sorry. Can you point Kishon and me to a better implementation? Marc
On Wed, Feb 27, 2013 at 05:33:11PM +0100, Marc Kleine-Budde wrote: > On 02/27/2013 05:21 PM, Greg KH wrote: > > On Wed, Feb 27, 2013 at 03:11:13PM +0100, Marc Kleine-Budde wrote: > >> In patch "5d3c28b usb: otg: add device tree support to otg library" > >> devm_usb_get_phy_by_phandle() was added. It uses try_module_get() to lock the > >> phy driver in memory. The corresponding module_put() is missing in that patch. > >> > >> This patch adds try_module_get() to usb_get_phy() and usb_get_phy_dev(). > >> Further the missing module_put() is added to usb_put_phy(). > >> > >> Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com> > >> Acked-by: Felipe Balbi <balbi@ti.com> > >> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> > >> Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de> > >> --- > >> drivers/usb/otg/otg.c | 10 +++++++--- > >> 1 file changed, 7 insertions(+), 3 deletions(-) > >> > >> diff --git a/drivers/usb/otg/otg.c b/drivers/usb/otg/otg.c > >> index e181439..2bd03d2 100644 > >> --- a/drivers/usb/otg/otg.c > >> +++ b/drivers/usb/otg/otg.c > >> @@ -130,7 +130,7 @@ struct usb_phy *usb_get_phy(enum usb_phy_type type) > >> spin_lock_irqsave(&phy_lock, flags); > >> > >> phy = __usb_find_phy(&phy_list, type); > >> - if (IS_ERR(phy)) { > >> + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { > > > > Ugh, really? We really are trying this type of module locking mess? > > > > Why? What is it solving? What's wrong with having the module be able > > Without this patch, you can unload the phy module, while it is in use. > As the phy framework doesn't use any existing abstraction to handle phy > <-> user of phy pairing/unpairing (it's all open coded), any subsequent > use of the phy's callback will result in deref'ing bogus pointers. Then perhaps we should fix the open-coded-ness of the phy layer? > > to be unloaded whenever it wants to? No one should be doing that and > > expecting that their hardware would still work properly, right? > > Yes, but it should not result in an kernel oops. Agreed. > > I really don't like this type of thing, sorry. > > Can you point Kishon and me to a better implementation? Wasn't someone working on a "generic" phy layer for the kernel? That should probably resolve this issue there. But in looking at this further, you are right, this is about the only way this can be solved now. It is a pretty minor problem though. I'll wait for this to come in from Felipe into my trees. thanks, greg k-h
Hi, On Thursday 28 February 2013 06:11 AM, Greg KH wrote: > On Wed, Feb 27, 2013 at 05:33:11PM +0100, Marc Kleine-Budde wrote: >> On 02/27/2013 05:21 PM, Greg KH wrote: >>> On Wed, Feb 27, 2013 at 03:11:13PM +0100, Marc Kleine-Budde wrote: >>>> In patch "5d3c28b usb: otg: add device tree support to otg library" >>>> devm_usb_get_phy_by_phandle() was added. It uses try_module_get() to lock the >>>> phy driver in memory. The corresponding module_put() is missing in that patch. >>>> >>>> This patch adds try_module_get() to usb_get_phy() and usb_get_phy_dev(). >>>> Further the missing module_put() is added to usb_put_phy(). >>>> >>>> Reviewed-by: Kishon Vijay Abraham I <kishon@ti.com> >>>> Acked-by: Felipe Balbi <balbi@ti.com> >>>> Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de> >>>> Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de> >>>> --- >>>> drivers/usb/otg/otg.c | 10 +++++++--- >>>> 1 file changed, 7 insertions(+), 3 deletions(-) >>>> >>>> diff --git a/drivers/usb/otg/otg.c b/drivers/usb/otg/otg.c >>>> index e181439..2bd03d2 100644 >>>> --- a/drivers/usb/otg/otg.c >>>> +++ b/drivers/usb/otg/otg.c >>>> @@ -130,7 +130,7 @@ struct usb_phy *usb_get_phy(enum usb_phy_type type) >>>> spin_lock_irqsave(&phy_lock, flags); >>>> >>>> phy = __usb_find_phy(&phy_list, type); >>>> - if (IS_ERR(phy)) { >>>> + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { >>> >>> Ugh, really? We really are trying this type of module locking mess? >>> >>> Why? What is it solving? What's wrong with having the module be able >> >> Without this patch, you can unload the phy module, while it is in use. >> As the phy framework doesn't use any existing abstraction to handle phy >> <-> user of phy pairing/unpairing (it's all open coded), any subsequent >> use of the phy's callback will result in deref'ing bogus pointers. > > Then perhaps we should fix the open-coded-ness of the phy layer? > >>> to be unloaded whenever it wants to? No one should be doing that and >>> expecting that their hardware would still work properly, right? >> >> Yes, but it should not result in an kernel oops. > > Agreed. > >>> I really don't like this type of thing, sorry. >> >> Can you point Kishon and me to a better implementation? > > Wasn't someone working on a "generic" phy layer for the kernel? That > should probably resolve this issue there. Even in the generic PHY layer, we use *try_module_get* to avoid dereferencing bogus pointers. Thanks Kishon
diff --git a/drivers/usb/otg/otg.c b/drivers/usb/otg/otg.c index e181439..2bd03d2 100644 --- a/drivers/usb/otg/otg.c +++ b/drivers/usb/otg/otg.c @@ -130,7 +130,7 @@ struct usb_phy *usb_get_phy(enum usb_phy_type type) spin_lock_irqsave(&phy_lock, flags); phy = __usb_find_phy(&phy_list, type); - if (IS_ERR(phy)) { + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { pr_err("unable to find transceiver of type %s\n", usb_phy_type_string(type)); goto err0; @@ -228,7 +228,7 @@ struct usb_phy *usb_get_phy_dev(struct device *dev, u8 index) spin_lock_irqsave(&phy_lock, flags); phy = __usb_find_phy_dev(dev, &phy_bind_list, index); - if (IS_ERR(phy)) { + if (IS_ERR(phy) || !try_module_get(phy->dev->driver->owner)) { pr_err("unable to find transceiver\n"); goto err0; } @@ -301,8 +301,12 @@ EXPORT_SYMBOL(devm_usb_put_phy); */ void usb_put_phy(struct usb_phy *x) { - if (x) + if (x) { + struct module *owner = x->dev->driver->owner; + put_device(x->dev); + module_put(owner); + } } EXPORT_SYMBOL(usb_put_phy);