[PATCH net] net: phy: Decrement phy_fixed_addr during unregister

If we have a system which uses fixed PHY devices and calls
fixed_phy_register() then fixed_phy_unregister() we can exhaust the
number of fixed PHYs available after a while, since we keep incrementing
the variable phy_fixed_addr, but we never decrement it.

This patch fixes that by decrementing phy_fixed_addr during
fixed_phy_del(), and in order to do that, we need to move the
phy_fixed_addr integer and its spinlock above that function.

Fixes: a75951217472 ("net: phy: extend fixed driver with fixed_phy_register()")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
---
 drivers/net/phy/fixed_phy.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/net/phy/fixed_phy.c b/drivers/net/phy/fixed_phy.c
index 2d2e4339f0df..050bc5657b9d 100644
--- a/drivers/net/phy/fixed_phy.c
+++ b/drivers/net/phy/fixed_phy.c

@@ -286,6 +286,9 @@ err_regs:
 }
 EXPORT_SYMBOL_GPL(fixed_phy_add);
 
+static int phy_fixed_addr;
+static DEFINE_SPINLOCK(phy_fixed_addr_lock);
+
 static void fixed_phy_del(int phy_addr)
 {
 	struct fixed_mdio_bus *fmb = &platform_fmb;

@@ -297,14 +300,14 @@ static void fixed_phy_del(int phy_addr)
 			if (gpio_is_valid(fp->link_gpio))
 				gpio_free(fp->link_gpio);
 			kfree(fp);
+			spin_lock(&phy_fixed_addr_lock);
+			phy_fixed_addr--;
+			spin_unlock(&phy_fixed_addr_lock);
 			return;
 		}
 	}
 }
 
-static int phy_fixed_addr;
-static DEFINE_SPINLOCK(phy_fixed_addr_lock);
-
 struct phy_device *fixed_phy_register(unsigned int irq,
 				      struct fixed_phy_status *status,
 				      int link_gpio,

-- 
2.7.4

On Fri, Jun 24, 2016 at 03:44:11PM -0700, Florian Fainelli wrote:

If we have a system which uses fixed PHY devices and calls
fixed_phy_register() then fixed_phy_unregister() we can exhaust the
number of fixed PHYs available after a while, since we keep incrementing
the variable phy_fixed_addr, but we never decrement it.

This patch fixes that by decrementing phy_fixed_addr during
fixed_phy_del(), and in order to do that, we need to move the
phy_fixed_addr integer and its spinlock above that function.

Is this really a good idea?

What if we have two fixed phys register, and the first one is
unregistered and a new one subsequently registered?

First phy registered, gets address 0, phy_fixed_addr becomes 1.
Second phy registered, gets address 1, phy_fixed_addr becomes 2.
First phy is unregistered, phy_fixed_addr becomes 1.
Third phy registered, gets address 1, conflicts with the second phy.

Obviously not a good outcome.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

On 06/24/2016 03:55 PM, Russell King - ARM Linux wrote:

On Fri, Jun 24, 2016 at 03:44:11PM -0700, Florian Fainelli wrote:

quoted

If we have a system which uses fixed PHY devices and calls
fixed_phy_register() then fixed_phy_unregister() we can exhaust the
number of fixed PHYs available after a while, since we keep incrementing
the variable phy_fixed_addr, but we never decrement it.

This patch fixes that by decrementing phy_fixed_addr during
fixed_phy_del(), and in order to do that, we need to move the
phy_fixed_addr integer and its spinlock above that function.

Is this really a good idea?

In the sense that it is symetrical to the register code, probably.

What if we have two fixed phys register, and the first one is
unregistered and a new one subsequently registered?

First phy registered, gets address 0, phy_fixed_addr becomes 1.
Second phy registered, gets address 1, phy_fixed_addr becomes 2.
First phy is unregistered, phy_fixed_addr becomes 1.
Third phy registered, gets address 1, conflicts with the second phy.

Obviously not a good outcome.

What would you suggest we do instead? Would switching to IDA/IDR give us
better results for instance (I have not looked too closely yet)?
-- 
Florian

On Fri, Jun 24, 2016 at 03:58:39PM -0700, Florian Fainelli wrote:

On 06/24/2016 03:55 PM, Russell King - ARM Linux wrote:

quoted

On Fri, Jun 24, 2016 at 03:44:11PM -0700, Florian Fainelli wrote:

quoted

If we have a system which uses fixed PHY devices and calls
fixed_phy_register() then fixed_phy_unregister() we can exhaust the
number of fixed PHYs available after a while, since we keep incrementing
the variable phy_fixed_addr, but we never decrement it.

This patch fixes that by decrementing phy_fixed_addr during
fixed_phy_del(), and in order to do that, we need to move the
phy_fixed_addr integer and its spinlock above that function.

Is this really a good idea?

In the sense that it is symetrical to the register code, probably.

quoted

What if we have two fixed phys register, and the first one is
unregistered and a new one subsequently registered?

First phy registered, gets address 0, phy_fixed_addr becomes 1.
Second phy registered, gets address 1, phy_fixed_addr becomes 2.
First phy is unregistered, phy_fixed_addr becomes 1.
Third phy registered, gets address 1, conflicts with the second phy.

Obviously not a good outcome.

What would you suggest we do instead? Would switching to IDA/IDR give us
better results for instance (I have not looked too closely yet)?

I would expect an IDA to be suitable, because the IDA would track which
indexes (==addresses) are currently in-use.

If you want to go further, using an IDR would allow fixed_mdio_read() to
find the right fixed_phy struct without needing to loop over fmb->phys.
Whether that's worth it or not depends if you have a large number of
fixed phys.  I suspect we're talking about small quantities here though.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

On 06/24/2016 04:06 PM, Russell King - ARM Linux wrote:

On Fri, Jun 24, 2016 at 03:58:39PM -0700, Florian Fainelli wrote:

quoted

On 06/24/2016 03:55 PM, Russell King - ARM Linux wrote:

quoted

On Fri, Jun 24, 2016 at 03:44:11PM -0700, Florian Fainelli wrote:

quoted

If we have a system which uses fixed PHY devices and calls
fixed_phy_register() then fixed_phy_unregister() we can exhaust the
number of fixed PHYs available after a while, since we keep incrementing
the variable phy_fixed_addr, but we never decrement it.

This patch fixes that by decrementing phy_fixed_addr during
fixed_phy_del(), and in order to do that, we need to move the
phy_fixed_addr integer and its spinlock above that function.

Is this really a good idea?

In the sense that it is symetrical to the register code, probably.

quoted

What if we have two fixed phys register, and the first one is
unregistered and a new one subsequently registered?

First phy registered, gets address 0, phy_fixed_addr becomes 1.
Second phy registered, gets address 1, phy_fixed_addr becomes 2.
First phy is unregistered, phy_fixed_addr becomes 1.
Third phy registered, gets address 1, conflicts with the second phy.

Obviously not a good outcome.

What would you suggest we do instead? Would switching to IDA/IDR give us
better results for instance (I have not looked too closely yet)?

I would expect an IDA to be suitable, because the IDA would track which
indexes (==addresses) are currently in-use.

OK, thanks!

If you want to go further, using an IDR would allow fixed_mdio_read() to
find the right fixed_phy struct without needing to loop over fmb->phys.

Since I am targetting this as a bugfix, the switch to IDA seems more
appropriate to be backported, but yes, that's a good idea though.

Whether that's worth it or not depends if you have a large number of
fixed phys.  I suspect we're talking about small quantities here though.

Yes, at the moment we are limited to 32 PHYs maximum, just like a real
MDIO bus, which in some systems could actually be not enough, but then
you run into other problems, like the need to register more than a
single fixed MDIO bus driver to get a larger address space...
-- 
Florian