On 09/21/2010 08:24 PM, Ben Dooks wrote:
> On Sat, Sep 04, 2010 at 12:07:48PM -0400, Michael Williamson wrote:
>> There is a bug in the i2c-davinci device init routine that attempts
>> to use default platform data if none is supplied (e.g., is NULL).
>> This patch fixes the bug.
>>
>> Signed-off-by: Michael Williamson <michael.williamson@xxxxxxxxxxxxxxxx>
>> ---
>> drivers/i2c/busses/i2c-davinci.c | 8 +++++---
>> 1 files changed, 5 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/i2c/busses/i2c-davinci.c b/drivers/i2c/busses/i2c-davinci.c
>> index 2222c87..6d4eeeb 100644
>> --- a/drivers/i2c/busses/i2c-davinci.c
>> +++ b/drivers/i2c/busses/i2c-davinci.c
>> @@ -235,10 +235,12 @@ static void i2c_davinci_calc_clk_dividers(struct davinci_i2c_dev *dev)
>> */
>> static int i2c_davinci_init(struct davinci_i2c_dev *dev)
>> {
>> - struct davinci_i2c_platform_data *pdata = dev->dev->platform_data;
>> + struct davinci_i2c_platform_data *pdata;
>>
>> - if (!pdata)
>> - pdata = &davinci_i2c_platform_data_default;
>> + if (!dev->dev->platform_data)
>> + dev->dev->platform_data = &davinci_i2c_platform_data_default;
>> +
>> + pdata = dev->dev->platform_data;
>
> At first glance this looks like a code shift, however is the platform
> data used later in the driver by referencing the device.platform_data
> field?
>
Yes. The problem call is i2c_davinci_calc_clk_dividers(), which is called
pretty early in the i2c_davinci_init() routine. There is a reference to
pdata->bus_freq without a check to see if pdata is valid. I think my proposed
fix is not the right approach as you have pointed out.
Perhaps a better patch would be (?):
---
diff --git a/drivers/i2c/busses/i2c-davinci.c b/drivers/i2c/busses/i2c-davinci.c
index 2222c87..15d0cea 100644
--- a/drivers/i2c/busses/i2c-davinci.c
+++ b/drivers/i2c/busses/i2c-davinci.c
@@ -195,6 +195,9 @@ static void i2c_davinci_calc_clk_dividers(struct davinci_i2c_dev *dev)
u32 clkl;
u32 input_clock = clk_get_rate(dev->clk);
+ if (!pdata)
+ pdata = &davinci_i2c_platform_data_default;
+
/* NOTE: I2C Clock divider programming info
* As per I2C specs the following formulas provide prescaler
* and low/high divider values
---
> Could you rewrite the header to show this, and if possible note any
> actual oops report that has been seen with this.
>
I found this because I was told (while proposing a new platform) to use NULL
as platform data as the parameters I was using matched the default data. I
didn't get an oops, it just hung (it's pretty early in the board init).
Our boot strapper/load resets the DDR where the kernel buffer is, so getting
any oops info is a bit time consuming. The above three lines (and the original
patch) resolved the problem.
I can update the header to indicate a default parameter set if you like. New patch OK?
-Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
