Re: [PATCH v4 03/16] crypto: caam - move tasklet_init() call down

From: Leonard Crestez
Date: Wed Jul 03 2019 - 19:45:44 EST


On 7/3/2019 8:14 PM, Andrey Smirnov wrote:
> On Wed, Jul 3, 2019 at 6:51 AM Leonard Crestez <leonard.crestez@xxxxxxx> wrote:
>> On 7/3/2019 11:14 AM, Andrey Smirnov wrote:
>>> Move tasklet_init() call further down in order to simplify error path
>>> cleanup. No functional change intended.
>>>
>>> diff --git a/drivers/crypto/caam/jr.c b/drivers/crypto/caam/jr.c
>>> index 4b25b2fa3d02..a7ca2bbe243f 100644
>>> --- a/drivers/crypto/caam/jr.c
>>> +++ b/drivers/crypto/caam/jr.c
>>> @@ -441,15 +441,13 @@ static int caam_jr_init(struct device *dev)
>>>
>>> jrp = dev_get_drvdata(dev);
>>>
>>> - tasklet_init(&jrp->irqtask, caam_jr_dequeue, (unsigned long)dev);
>>> -
>>> /* Connect job ring interrupt handler. */
>>> error = request_irq(jrp->irq, caam_jr_interrupt, IRQF_SHARED,
>>> dev_name(dev), dev);
>>> if (error) {
>>> dev_err(dev, "can't connect JobR %d interrupt (%d)\n",
>>> jrp->ridx, jrp->irq);
>>> - goto out_kill_deq;
>>> + return error;
>>> }
>>
>> The caam_jr_interrupt handler can schedule the tasklet so it makes sense
>> to have it be initialized ahead of request_irq. In theory it's possible
>> for an interrupt to be triggered immediately when request_irq is called.
>>
>> I'm not very familiar with the CAAM ip, can you ensure no interrupts are
>> pending in HW at probe time? The "no functional change" part is not obvious.
>>
>
> Said tasklet will use both jrp->outring and jrp->entinfo array
> initialized after IRQ request call in both versions of the code
> (before/after this patch). AFAICT, the only case where this patch
> would change initialization safety of the original code is if JR was
> scheduled somehow while ORSFx is 0 (no jobs done), which I don't think
> is possible.

I took a second look at caam_jr_init and there is apparently a whole
bunch of other reset/init stuff done after request_irq. For example
caam_reset_hw_jr is done after request_irq and masks interrupts?

What I'd expect is that request_irq is done last after all other
initialization is performed. But I'm not familiar with how CAAM JRs work
so feel free to ignore this.

--
Regards,
Leonard