2012-02-20 23:38:55

by Hauke Mehrtens

[permalink] [raw]
Subject: [RFC] b43: fix memory leak on bcm5354

When using the bcm5354 with a recent firmware >= 478.104 it runs into a
memory very shortly after doing an active scan or any thing else where
packages are send. This was cased by a gpio misconfiguration, the
firmware triggered the GPIO pins used for buttons on some devices and
that caused an other driver (OpenWrt diag) listening for these buttons
to send many user space messages.
This patch fixes the bug for my devices (Asus WL-520GU) and makes it
work with firmware 666.2, but I do not know if this patch is correct.
The spec for this part is out of date:
http://bcm-specs.sipsolutions.net/GPIO

GPIO pin layout:
pin# name type
0 power led
1 wlan led
2 reset button
3 ses buttom

related nvram configuration:
wl0gpio2=11
wl0gpio3=11
wl0gpio0=11
wl0gpio1=0x02
reset_gpio=2

Signed-off-by: Hauke Mehrtens <[email protected]>
---
drivers/net/wireless/b43/main.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/net/wireless/b43/main.c b/drivers/net/wireless/b43/main.c
index 23ffb1b..f782723 100644
--- a/drivers/net/wireless/b43/main.c
+++ b/drivers/net/wireless/b43/main.c
@@ -2720,6 +2720,8 @@ static int b43_gpio_init(struct b43_wldev *dev)
mask |= 0x0060;
set |= 0x0060;
}
+ if (dev->dev->chip_id == 0x5354)
+ set &= 0xff02;
if (0 /* FIXME: conditional unknown */ ) {
b43_write16(dev, B43_MMIO_GPIO_MASK,
b43_read16(dev, B43_MMIO_GPIO_MASK)
--
1.7.5.4



2012-02-21 19:13:01

by Hauke Mehrtens

[permalink] [raw]
Subject: Re: [RFC] b43: fix memory leak on bcm5354

On 02/21/2012 01:57 AM, Larry Finger wrote:
> On 02/20/2012 05:38 PM, Hauke Mehrtens wrote:
>> When using the bcm5354 with a recent firmware>= 478.104 it runs into a
>> memory very shortly after doing an active scan or any thing else where
>
> You are missing a word after memory.
It leaked to somewhere else. ;-)

>> packages are send. This was cased by a gpio misconfiguration, the
>> firmware triggered the GPIO pins used for buttons on some devices and
>> that caused an other driver (OpenWrt diag) listening for these buttons
>> to send many user space messages.
>> This patch fixes the bug for my devices (Asus WL-520GU) and makes it
>> work with firmware 666.2, but I do not know if this patch is correct.
>> The spec for this part is out of date:
>> http://bcm-specs.sipsolutions.net/GPIO
>
> I just updated the specs based on additional info I found. Everything I
> see say that there are 4 bits for the LEDs. Do you have a source for
> what you posted below?
Thanks for updating the spec, but this is nothing special for the
bcm5354. My device just has one WLAN LED, like many SoCs.

>> GPIO pin layout:
>> pin# name type
>> 0 power led
>> 1 wlan led
>> 2 reset button
>> 3 ses buttom

This is from the Asus WL-520GU section of the the OpenWrt diag driver
[0] and I verified it myself on this device.

>> related nvram configuration:
>> wl0gpio2=11
>> wl0gpio3=11
>> wl0gpio0=11
>> wl0gpio1=0x02
>> reset_gpio=2

This was extracted from the nvram of the Asus W-520GU with "nvram show
|grep gpio". This was a used device, so I do not know if someone else
already did some modifications to the nvram.

Hauke

[0]: https://dev.openwrt.org/browser/trunk/package/broadcom-diag/src/diag.c

2012-02-21 00:57:36

by Larry Finger

[permalink] [raw]
Subject: Re: [RFC] b43: fix memory leak on bcm5354

On 02/20/2012 05:38 PM, Hauke Mehrtens wrote:
> When using the bcm5354 with a recent firmware>= 478.104 it runs into a
> memory very shortly after doing an active scan or any thing else where

You are missing a word after memory.

> packages are send. This was cased by a gpio misconfiguration, the
> firmware triggered the GPIO pins used for buttons on some devices and
> that caused an other driver (OpenWrt diag) listening for these buttons
> to send many user space messages.
> This patch fixes the bug for my devices (Asus WL-520GU) and makes it
> work with firmware 666.2, but I do not know if this patch is correct.
> The spec for this part is out of date:
> http://bcm-specs.sipsolutions.net/GPIO

I just updated the specs based on additional info I found. Everything I see say
that there are 4 bits for the LEDs. Do you have a source for what you posted below?

>
> GPIO pin layout:
> pin# name type
> 0 power led
> 1 wlan led
> 2 reset button
> 3 ses buttom
>
> related nvram configuration:
> wl0gpio2=11
> wl0gpio3=11
> wl0gpio0=11
> wl0gpio1=0x02
> reset_gpio=2

Larry

2012-02-21 09:49:14

by Florian Fainelli

[permalink] [raw]
Subject: Re: [RFC] b43: fix memory leak on bcm5354

Hi Hauke,

Le 02/21/12 00:38, Hauke Mehrtens a écrit :
> When using the bcm5354 with a recent firmware>= 478.104 it runs into a
> memory very shortly after doing an active scan or any thing else where
> packages are send. This was cased by a gpio misconfiguration, the
> firmware triggered the GPIO pins used for buttons on some devices and
> that caused an other driver (OpenWrt diag) listening for these buttons
> to send many user space messages.

and this is what really exhausts memory in fact (the send of many
messages). The real bug is actually the B43 driver taking over GPIOs
used by the system for other purposes: buttons and LEDs, this is what
your subject should reflect.

> This patch fixes the bug for my devices (Asus WL-520GU) and makes it
> work with firmware 666.2, but I do not know if this patch is correct.
> The spec for this part is out of date:
> http://bcm-specs.sipsolutions.net/GPIO
>
> GPIO pin layout:
> pin# name type
> 0 power led
> 1 wlan led
> 2 reset button
> 3 ses buttom
>
> related nvram configuration:
> wl0gpio2=11
> wl0gpio3=11
> wl0gpio0=11
> wl0gpio1=0x02
> reset_gpio=2
>
> Signed-off-by: Hauke Mehrtens<[email protected]>
> ---
> drivers/net/wireless/b43/main.c | 2 ++
> 1 files changed, 2 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/net/wireless/b43/main.c b/drivers/net/wireless/b43/main.c
> index 23ffb1b..f782723 100644
> --- a/drivers/net/wireless/b43/main.c
> +++ b/drivers/net/wireless/b43/main.c
> @@ -2720,6 +2720,8 @@ static int b43_gpio_init(struct b43_wldev *dev)
> mask |= 0x0060;
> set |= 0x0060;
> }
> + if (dev->dev->chip_id == 0x5354)
> + set&= 0xff02;
> if (0 /* FIXME: conditional unknown */ ) {
> b43_write16(dev, B43_MMIO_GPIO_MASK,
> b43_read16(dev, B43_MMIO_GPIO_MASK)