| Message ID | 20230817121631.1878897-2-ruanjinjie@huawei.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: Fix return value check for fixed_phy_register() | expand |
On Thu, Aug 17, 2023 at 08:16:28PM +0800, Ruan Jinjie wrote: > Since fixed_phy_get_gpiod() return NULL instead of ERR_PTR(), > if it fails, the IS_ERR() can never return the error. So check NULL > and return ERR_PTR(-EINVAL) if fails. > > Fixes: 71bd106d2567 ("net: fixed-phy: Add fixed_phy_register_with_gpiod() API") > Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com> Reviewed-by: Andrew Lunn <andrew@lunn.ch> Andrew
On Thu, Aug 17, 2023 at 08:16:28PM +0800, Ruan Jinjie wrote: > Since fixed_phy_get_gpiod() return NULL instead of ERR_PTR(), > if it fails, the IS_ERR() can never return the error. So check NULL > and return ERR_PTR(-EINVAL) if fails. No, this is totally and utterly wrong, and this patch introduces a new bug. The original code is _correct_. > Fixes: 71bd106d2567 ("net: fixed-phy: Add fixed_phy_register_with_gpiod() API") > Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com> > --- > drivers/net/phy/fixed_phy.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/phy/fixed_phy.c b/drivers/net/phy/fixed_phy.c > index aef739c20ac4..4e7406455b6e 100644 > --- a/drivers/net/phy/fixed_phy.c > +++ b/drivers/net/phy/fixed_phy.c > @@ -239,8 +239,8 @@ static struct phy_device *__fixed_phy_register(unsigned int irq, > /* Check if we have a GPIO associated with this fixed phy */ > if (!gpiod) { > gpiod = fixed_phy_get_gpiod(np); > - if (IS_ERR(gpiod)) > - return ERR_CAST(gpiod); > + if (!gpiod) > + return ERR_PTR(-EINVAL); Let's look at fixed_phy_get_gpiod(): gpiod = fwnode_gpiod_get_index(of_fwnode_handle(fixed_link_node), "link", 0, GPIOD_IN, "mdio"); if (IS_ERR(gpiod) && PTR_ERR(gpiod) != -EPROBE_DEFER) { ... gpiod = NULL; } ... return gpiod; If fwnode_gpiod_get_index() returns -EPROBE_DEFER, _then_ we return an error pointer. So it _does_ return an error pointer. It also returns NULL when there is no device node passed to it, or if there is no fixed-link specifier, or there is some other error from fwnode_gpiod_get_index(). Otherwise, it returns a valid pointer to a gpio descriptor. The gpio is optional. The device node is optional. When fixed_phy_get_gpiod() returns NULL, it is _not_ an error, it means that we don't have a GPIO. Just because something returns NULL does _not_ mean it's an error - please get out of that thinking, because if you don't your patches will introduce lots of new bugs. Only when fwnode_gpiod_get_index() wants to defer probe do we return an error. So, sorry but NAK to this patch, it is incorrect.
On 2023/8/17 21:10, Russell King (Oracle) wrote: > On Thu, Aug 17, 2023 at 08:16:28PM +0800, Ruan Jinjie wrote: >> Since fixed_phy_get_gpiod() return NULL instead of ERR_PTR(), >> if it fails, the IS_ERR() can never return the error. So check NULL >> and return ERR_PTR(-EINVAL) if fails. > > No, this is totally and utterly wrong, and this patch introduces a new > bug. The original code is _correct_. > >> Fixes: 71bd106d2567 ("net: fixed-phy: Add fixed_phy_register_with_gpiod() API") >> Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com> >> --- >> drivers/net/phy/fixed_phy.c | 4 ++-- >> 1 file changed, 2 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/net/phy/fixed_phy.c b/drivers/net/phy/fixed_phy.c >> index aef739c20ac4..4e7406455b6e 100644 >> --- a/drivers/net/phy/fixed_phy.c >> +++ b/drivers/net/phy/fixed_phy.c >> @@ -239,8 +239,8 @@ static struct phy_device *__fixed_phy_register(unsigned int irq, >> /* Check if we have a GPIO associated with this fixed phy */ >> if (!gpiod) { >> gpiod = fixed_phy_get_gpiod(np); >> - if (IS_ERR(gpiod)) >> - return ERR_CAST(gpiod); >> + if (!gpiod) >> + return ERR_PTR(-EINVAL); > > Let's look at fixed_phy_get_gpiod(): > > gpiod = fwnode_gpiod_get_index(of_fwnode_handle(fixed_link_node), > "link", 0, GPIOD_IN, "mdio"); > if (IS_ERR(gpiod) && PTR_ERR(gpiod) != -EPROBE_DEFER) { > ... > gpiod = NULL; > } > ... > return gpiod; > > If fwnode_gpiod_get_index() returns -EPROBE_DEFER, _then_ we return an > error pointer. So it _does_ return an error pointer. > > It also returns NULL when there is no device node passed to it, or > if there is no fixed-link specifier, or there is some other error > from fwnode_gpiod_get_index(). > > Otherwise, it returns a valid pointer to a gpio descriptor. > > The gpio is optional. The device node is optional. When > fixed_phy_get_gpiod() returns NULL, it is _not_ an error, it means > that we don't have a GPIO. Just because something returns NULL does > _not_ mean it's an error - please get out of that thinking, because > if you don't your patches will introduce lots of new bugs. Thank you, I understand what you mean, NULL is not an error here, so it is not handled. > > Only when fwnode_gpiod_get_index() wants to defer probe do we return > an error. > > So, sorry but NAK to this patch, it is incorrect. >
diff --git a/drivers/net/phy/fixed_phy.c b/drivers/net/phy/fixed_phy.c index aef739c20ac4..4e7406455b6e 100644 --- a/drivers/net/phy/fixed_phy.c +++ b/drivers/net/phy/fixed_phy.c @@ -239,8 +239,8 @@ static struct phy_device *__fixed_phy_register(unsigned int irq, /* Check if we have a GPIO associated with this fixed phy */ if (!gpiod) { gpiod = fixed_phy_get_gpiod(np); - if (IS_ERR(gpiod)) - return ERR_CAST(gpiod); + if (!gpiod) + return ERR_PTR(-EINVAL); } /* Get the next available PHY address, up to PHY_MAX_ADDR */
Since fixed_phy_get_gpiod() return NULL instead of ERR_PTR(), if it fails, the IS_ERR() can never return the error. So check NULL and return ERR_PTR(-EINVAL) if fails. Fixes: 71bd106d2567 ("net: fixed-phy: Add fixed_phy_register_with_gpiod() API") Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com> --- drivers/net/phy/fixed_phy.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)