Re: [PATCH] tpm/tpm_tis: Free IRQ if probing fails

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 4/14/20 6:45 PM, Jarkko Sakkinen wrote:
> On Tue, Apr 14, 2020 at 07:04:07PM +0300, Jarkko Sakkinen wrote:
>> On Tue, Apr 14, 2020 at 10:26:32AM +0200, Hans de Goede wrote:
>>> Hi,
>>>
>>> On 4/14/20 9:13 AM, Jarkko Sakkinen wrote:
>>>> On Mon, Apr 13, 2020 at 08:11:15PM +0200, Hans de Goede wrote:
>>>>> Hi,
>>>>>
>>>>> On 4/13/20 8:07 PM, Jarkko Sakkinen wrote:
>>>>>> On Mon, Apr 13, 2020 at 12:04:25PM +0200, Hans de Goede wrote:
>>>>>>> Hi Jarkko,
>>>>>>>
>>>>>>> On 4/12/20 7:04 PM, Jarkko Sakkinen wrote:
>>>>>>>> Call devm_free_irq() if we have to revert to polling in order not to
>>>>>>>> unnecessarily reserve the IRQ for the life-cycle of the driver.
>>>>>>>>
>>>>>>>> Cc: stable@vger.kernel.org # 4.5.x
>>>>>>>> Reported-by: Hans de Goede <hdegoede@redhat.com>
>>>>>>>> Fixes: e3837e74a06d ("tpm_tis: Refactor the interrupt setup")
>>>>>>>> Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
>>>>>>>> ---
>>>>>>>>      drivers/char/tpm/tpm_tis_core.c | 5 ++++-
>>>>>>>>      1 file changed, 4 insertions(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
>>>>>>>> index 27c6ca031e23..ae6868e7b696 100644
>>>>>>>> --- a/drivers/char/tpm/tpm_tis_core.c
>>>>>>>> +++ b/drivers/char/tpm/tpm_tis_core.c
>>>>>>>> @@ -1062,9 +1062,12 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>>>>>>>>      		if (irq) {
>>>>>>>>      			tpm_tis_probe_irq_single(chip, intmask, IRQF_SHARED,
>>>>>>>>      						 irq);
>>>>>>>> -			if (!(chip->flags & TPM_CHIP_FLAG_IRQ))
>>>>>>>> +			if (!(chip->flags & TPM_CHIP_FLAG_IRQ)) {
>>>>>>>>      				dev_err(&chip->dev, FW_BUG
>>>>>>>>      					"TPM interrupt not working, polling instead\n");
>>>>>>>> +				devm_free_irq(chip->dev.parent, priv->irq,
>>>>>>>> +					      chip);
>>>>>>>> +			}
>>>>>>>
>>>>>>> My initial plan was actually to do something similar, but if the probe code
>>>>>>> is actually ever fixed to work as intended again then this will lead to a
>>>>>>> double free as then the IRQ-test path of tpm_tis_send() will have called
>>>>>>> disable_interrupts() which already calls devm_free_irq().
>>>>>>>
>>>>>>> You could check for chip->irq != 0 here to avoid that.
>>>
>>> Erm in case you haven't figured it out yet this should be priv->irq != 0, sorry.
>>
>> Yup.
>>
>>>>>>>
>>>>>>> But it all is rather messy, which is why I went with the "#if 0" approach
>>>>>>> in my patch.
>>>>>>
>>>>>> I think it is right way to fix it. It is a bug independent of the issue
>>>>>> we are experiencing.
>>>>>>
>>>>>> However, what you are suggesting should be done in addition. Do you have
>>>>>> a patch in place or do you want me to refine mine?
>>>>>
>>>>> I do not have a patch ready for this, if you can refine yours that would
>>>>> be great.
>>>>
>>>> Thanks! Just wanted to confirm.
>>>
>>> And thank you for working on a (temporary?) fix for this.
>>
>> As far as I see it, it is orthogonal fix that needs to be backported
>> to stable kernels. This bug predates the issue we're seeing now.
> 
> Hey, I came to other thoughts on "how". Would probably make sense
> to always call disable_interrupts() aka no sense to add two separate
> code paths. What do you think?

Sounds good, I guess it would be best to combine that with a:

	if (priv->irq == 0)
		return;

At the top of disable_interrupts() and then unconditionally
call disable_interrupts() where your v1 of this patch
calls devm_free_irq(). That would be a reasonable clean
solution I think.

Regards,

Hans