2013-07-07 20:34:08

by Sebastian Hesselbarth

[permalink] [raw]
Subject: [PATCH] net: mv643xx_eth: fix DT port device name

Device tree support added to Marvell MV643xx ethernet driver registers
port devices from port device nodes found on the corresponding controller
node. The current port device name will cause the second controller to
fail on registration because of two identical device names. This fixes
the issue by taking the device node's name also as port device name.

Signed-off-by: Sebastian Hesselbarth <[email protected]>
Reported-by: Jonas Gorski <[email protected]>
---
Cc: Lennert Buytenhek <[email protected]>
Cc: Jonas Gorski <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
---
drivers/net/ethernet/marvell/mv643xx_eth.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
index 6495bea..1f3a03d 100644
--- a/drivers/net/ethernet/marvell/mv643xx_eth.c
+++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
@@ -2521,7 +2521,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
of_property_read_u32(pnp, "duplex", &ppd.duplex);
}

- ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
+ ppdev = platform_device_alloc(pnp->name, ppd.port_number);
if (!ppdev)
return -ENOMEM;
ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
--
1.7.10.4


2013-07-07 21:43:50

by Jonas Gorski

[permalink] [raw]
Subject: Re: [PATCH] net: mv643xx_eth: fix DT port device name

On Sun, 7 Jul 2013 22:33:51 +0200
Sebastian Hesselbarth <[email protected]> wrote:

> Device tree support added to Marvell MV643xx ethernet driver registers
> port devices from port device nodes found on the corresponding controller
> node. The current port device name will cause the second controller to
> fail on registration because of two identical device names. This fixes
> the issue by taking the device node's name also as port device name.
>
> Signed-off-by: Sebastian Hesselbarth <[email protected]>
> Reported-by: Jonas Gorski <[email protected]>
> ---
> Cc: Lennert Buytenhek <[email protected]>
> Cc: Jonas Gorski <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> ---
> drivers/net/ethernet/marvell/mv643xx_eth.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
> index 6495bea..1f3a03d 100644
> --- a/drivers/net/ethernet/marvell/mv643xx_eth.c
> +++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
> @@ -2521,7 +2521,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
> of_property_read_u32(pnp, "duplex", &ppd.duplex);
> }
>
> - ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
> + ppdev = platform_device_alloc(pnp->name, ppd.port_number);
> if (!ppdev)
> return -ENOMEM;
> ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);

This breaks ethernet completely, as there is no platform driver
registered for pnp->name ("ethernetX-port"), only for MV643XX_ETH_NAME.

Also since I didn't see a patch for it and no mentioning of it:

There's still one further issue from having two ethernet-ports with
port_number 0, it causes a device leak:

static struct platform_device *port_platdev[3];

mv643xx_eth_shared_of_add_port()
{
...
port_platdev[ppd.port_number] = ppdev;
...
}

The second port at 0 will overwrite the first and thus will never be
deleted in

mv643xx_eth_shared_of_remove()
{
...
for (n = 0; n < 3; n++) {
platform_device_del(port_platdev[n]);
port_platdev[n] = NULL;
}
}

I doubt a insmod-rmmod-insmod will go well in that case ;-)


Regards
Jonas

2013-07-07 21:58:57

by Sebastian Hesselbarth

[permalink] [raw]
Subject: Re: [PATCH] net: mv643xx_eth: fix DT port device name

On 07/07/2013 11:43 PM, Jonas Gorski wrote:
> On Sun, 7 Jul 2013 22:33:51 +0200
> Sebastian Hesselbarth<[email protected]> wrote:
>
>> Device tree support added to Marvell MV643xx ethernet driver registers
>> port devices from port device nodes found on the corresponding controller
>> node. The current port device name will cause the second controller to
>> fail on registration because of two identical device names. This fixes
>> the issue by taking the device node's name also as port device name.
>>
>> Signed-off-by: Sebastian Hesselbarth<[email protected]>
>> Reported-by: Jonas Gorski<[email protected]>
>> ---
>> Cc: Lennert Buytenhek<[email protected]>
>> Cc: Jonas Gorski<[email protected]>
>> Cc: [email protected]
>> Cc: [email protected]
>> Cc: [email protected]
>> ---
>> drivers/net/ethernet/marvell/mv643xx_eth.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
>> index 6495bea..1f3a03d 100644
>> --- a/drivers/net/ethernet/marvell/mv643xx_eth.c
>> +++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
>> @@ -2521,7 +2521,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
>> of_property_read_u32(pnp, "duplex",&ppd.duplex);
>> }
>>
>> - ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
>> + ppdev = platform_device_alloc(pnp->name, ppd.port_number);
>> if (!ppdev)
>> return -ENOMEM;
>> ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
>
> This breaks ethernet completely, as there is no platform driver
> registered for pnp->name ("ethernetX-port"), only for MV643XX_ETH_NAME.

Jonas,

I asked you to test the above earlier today. You said this would fix
the issue. I have no Kirkwood board with two ethernet's available, so
I really needed on your help here.

Please prepare a patch yourself, that fixes the issue.

Sebastian

> Also since I didn't see a patch for it and no mentioning of it:
>
> There's still one further issue from having two ethernet-ports with
> port_number 0, it causes a device leak:
>
> static struct platform_device *port_platdev[3];
>
> mv643xx_eth_shared_of_add_port()
> {
> ...
> port_platdev[ppd.port_number] = ppdev;
> ...
> }
>
> The second port at 0 will overwrite the first and thus will never be
> deleted in
>
> mv643xx_eth_shared_of_remove()
> {
> ...
> for (n = 0; n< 3; n++) {
> platform_device_del(port_platdev[n]);
> port_platdev[n] = NULL;
> }
> }
>
> I doubt a insmod-rmmod-insmod will go well in that case ;-)
>
>
> Regards
> Jonas

2013-07-07 22:05:43

by Jonas Gorski

[permalink] [raw]
Subject: Re: [PATCH] net: mv643xx_eth: fix DT port device name

On Sun, 7 Jul 2013 23:43:41 +0200
Jonas Gorski <[email protected]> wrote:

> On Sun, 7 Jul 2013 22:33:51 +0200
> Sebastian Hesselbarth <[email protected]> wrote:
>
> > Device tree support added to Marvell MV643xx ethernet driver registers
> > port devices from port device nodes found on the corresponding controller
> > node. The current port device name will cause the second controller to
> > fail on registration because of two identical device names. This fixes
> > the issue by taking the device node's name also as port device name.
> >
> > Signed-off-by: Sebastian Hesselbarth <[email protected]>
> > Reported-by: Jonas Gorski <[email protected]>
> > ---
> > Cc: Lennert Buytenhek <[email protected]>
> > Cc: Jonas Gorski <[email protected]>
> > Cc: [email protected]
> > Cc: [email protected]
> > Cc: [email protected]
> > ---
> > drivers/net/ethernet/marvell/mv643xx_eth.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
> > index 6495bea..1f3a03d 100644
> > --- a/drivers/net/ethernet/marvell/mv643xx_eth.c
> > +++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
> > @@ -2521,7 +2521,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
> > of_property_read_u32(pnp, "duplex", &ppd.duplex);
> > }
> >
> > - ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
> > + ppdev = platform_device_alloc(pnp->name, ppd.port_number);
> > if (!ppdev)
> > return -ENOMEM;
> > ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
>
> This breaks ethernet completely, as there is no platform driver
> registered for pnp->name ("ethernetX-port"), only for MV643XX_ETH_NAME.

Looking back at our conversation, this is my fault.
I actually did not change this part as you asked, but I saw the
alloc/del issue with port 0, then added the counter and also only
replaced the ppd.port_number in the alloc with it. I had completely
forgotten at that time to replace the device name; else I would
have caught it back then.

I only caught it now because I tried your patch and wondered why there
wasn't anything registered, not because I saw the problem by review.

Sorry for that.

Regards
Jonas

2013-07-07 22:45:32

by Jonas Gorski

[permalink] [raw]
Subject: [PATCH] net: mv643xx_eth: do not use port number as platform device id

The port number is only local to the ethernet block, not global, so
there can be two ethernet blocks both using the same port, like
kirkwood with both using port 0.

Fix this by using the array index offset for the allocated platform
devices as the id.

Signed-off-by: Jonas Gorski <[email protected]>
---
drivers/net/ethernet/marvell/mv643xx_eth.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c b/drivers/net/ethernet/marvell/mv643xx_eth.c
index 6495bea..c35db73 100644
--- a/drivers/net/ethernet/marvell/mv643xx_eth.c
+++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
@@ -2483,6 +2483,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
struct resource res;
const char *mac_addr;
int ret;
+ int dev_num = 0;

memset(&ppd, 0, sizeof(ppd));
ppd.shared = pdev;
@@ -2503,6 +2504,14 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
return -EINVAL;
}

+ while (dev_num < 3 && port_platdev[dev_num])
+ dev_num++;
+
+ if (dev_num == 3) {
+ dev_err(&pdev->dev, "too many ports registered\n");
+ return -EINVAL;
+ }
+
mac_addr = of_get_mac_address(pnp);
if (mac_addr)
memcpy(ppd.mac_addr, mac_addr, 6);
@@ -2521,7 +2530,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
of_property_read_u32(pnp, "duplex", &ppd.duplex);
}

- ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
+ ppdev = platform_device_alloc(MV643XX_ETH_NAME, dev_num);
if (!ppdev)
return -ENOMEM;
ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
@@ -2538,7 +2547,7 @@ static int mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
if (ret)
goto port_err;

- port_platdev[ppd.port_number] = ppdev;
+ port_platdev[dev_num] = ppdev;

return 0;

--
1.8.3.2

2013-07-10 22:59:48

by Sebastian Hesselbarth

[permalink] [raw]
Subject: Re: [PATCH] net: mv643xx_eth: fix DT port device name

On 07/10/2013 12:29 PM, Gerlando Falauto wrote:
> Hi Sebastian, Jonas,
>
> first of all thank you for your HUGE efforts in this area.
>
> On 07/08/2013 12:05 AM, Jonas Gorski wrote:
>> On Sun, 7 Jul 2013 23:43:41 +0200
>> Jonas Gorski <[email protected]> wrote:
>>
>>> On Sun, 7 Jul 2013 22:33:51 +0200
>>> Sebastian Hesselbarth <[email protected]> wrote:
>>>
>>>> Device tree support added to Marvell MV643xx ethernet driver registers
>>>> port devices from port device nodes found on the corresponding
>>>> controller
>>>> node. The current port device name will cause the second controller to
>>>> fail on registration because of two identical device names. This fixes
>>>> the issue by taking the device node's name also as port device name.
>>>>
>>>> Signed-off-by: Sebastian Hesselbarth <[email protected]>
>>>> Reported-by: Jonas Gorski <[email protected]>
>>>> ---
>>>> Cc: Lennert Buytenhek <[email protected]>
>>>> Cc: Jonas Gorski <[email protected]>
>>>> Cc: [email protected]
>>>> Cc: [email protected]
>>>> Cc: [email protected]
>>>> ---
>>>> drivers/net/ethernet/marvell/mv643xx_eth.c | 2 +-
>>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/net/ethernet/marvell/mv643xx_eth.c
>>>> b/drivers/net/ethernet/marvell/mv643xx_eth.c
>>>> index 6495bea..1f3a03d 100644
>>>> --- a/drivers/net/ethernet/marvell/mv643xx_eth.c
>>>> +++ b/drivers/net/ethernet/marvell/mv643xx_eth.c
>>>> @@ -2521,7 +2521,7 @@ static int
>>>> mv643xx_eth_shared_of_add_port(struct platform_device *pdev,
>>>> of_property_read_u32(pnp, "duplex", &ppd.duplex);
>>>> }
>>>>
>>>> - ppdev = platform_device_alloc(MV643XX_ETH_NAME, ppd.port_number);
>>>> + ppdev = platform_device_alloc(pnp->name, ppd.port_number);
>>>> if (!ppdev)
>>>> return -ENOMEM;
>>>> ppdev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
>>>
>>> This breaks ethernet completely, as there is no platform driver
>>> registered for pnp->name ("ethernetX-port"), only for MV643XX_ETH_NAME.
>>
>> Looking back at our conversation, this is my fault.
>> I actually did not change this part as you asked, but I saw the
>> alloc/del issue with port 0, then added the counter and also only
>> replaced the ppd.port_number in the alloc with it. I had completely
>> forgotten at that time to replace the device name; else I would
>> have caught it back then.
>>
>> I only caught it now because I tried your patch and wondered why there
>> wasn't anything registered, not because I saw the problem by review.
>>
>
> The way I understand it, you are "manually" creating platform devices
> off the port subnodes (which, incidentally, have a compatible =
> "marvell,kirkwood-eth-port" property), but matching with the driver is
> indeed performed by device name.
> Hence Sebastian's first patch breaks everything and Jonas' latest one
> fixes it. Is that correct? Is there any more work ongoing in this area?
>
> One more question: if my understanding above is correct, what's the
> reason for NOT having an of_device_id[] table, which looks like was
> present in Florian Fainelli's first proposal [1]?

Gerlando,

for the long run, I plan to break with controller/port device
separation. I haven't looked into netdev internals, but I guess
you can have more than one netdev per platform_device.

Using one device per controller will remove some temporal dependencies
of when which driver gets loaded (there will be only one) and also the
root cause of the issue fixed by Jonas.

But it is not on my top priority list, but after fixing MAC address
obliviousness, suspend/resume and initial port setup.

> I'm also willing to help testing your present/future kirkwood patches if
> you like. Please just put me on Cc:

I once put too many people on Cc and that caused ML rejects, but if I
remember it I will keep you posted.

Sebastian