Re: [RFC PATCH 02/06] input/rmi4: Core files

[Christopher Heiny <cheiny@synaptics.com>]

On 10/23/2012 05:11 PM, Dmitry Torokhov wrote:
> On Tuesday, October 23, 2012 04:46:28 PM Christopher Heiny wrote:
>> On 10/11/2012 01:13 AM, Dmitry Torokhov wrote:
>>> On Thu, Oct 11, 2012 at 04:15:56AM +0000, Christopher Heiny wrote:
>>>> On Thursday, October 11, 2012 02:21:53 AM you wrote:
>>>>> On Sat, Oct 6, 2012 at 6:09 AM, Christopher Heiny <cheiny@synaptics.com>
> wrote:

[snip]

>>>>>> +static int process_interrupt_requests(struct rmi_device *rmi_dev)
>>>>>> +{
>>>>>> +       struct rmi_driver_data *data = dev_get_drvdata(&rmi_dev->dev);
>>>>>> +       struct device *dev = &rmi_dev->dev;
>>>>>> +       struct rmi_function_container *entry;
>>>>>> +       u8 irq_status[data->num_of_irq_regs];
>>>>>
>>>>> Looking at this...
>>>>>
>>>>> What does the data->num_of_irq_regs actually contain?
>>>>>
>>>>> I just fear that it is something constant like always 2 or always 4,
>>>>> so there is actually, in reality, a 16 or 32 bit register hiding in
>>>>> there.
>>>>>
>>>>> In that case what you should do is to represent it as a u16 or u32 here,
>>>>> just or the bits into a status word, and then walk over that status
>>>>> word with something like ffs(bitword); ...
>>>>
>>>> Nope, it's not constant.  In theory, and RMI4 based sensor can have up
>>>> to 128 functions (in practice, it's far fewer), and each function can
>>>> have as many as 7 interrupts.  So the number of IRQ registers can vary
>>>> from RMI4 sensor to RMI4 sensor, and needs to be computed during the
>>>> scan of the product descriptor table.
>>>
>>> Is it a good idea to have it on stack then? Should it be part of
>>> rmi_device instead?
>>
>> It's not coming off the stack.  We're allocating it via devm_kzalloc()
>> in rmi_driver_probe().
>
> No, look at the part of the code that was quoted. "u8 irq_status[data-
> num_of_irq_regs];" is on stack.

Sorry - I thought you were referring to data->num_of_irq_regs rather 
than irq_status.  We'll move that.