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

[Horia Geanta <horia.geanta@nxp.com>]

On 7/4/2019 2:45 AM, Leonard Crestez wrote:
> On 7/3/2019 8:14 PM, Andrey Smirnov wrote:
>> On Wed, Jul 3, 2019 at 6:51 AM Leonard Crestez <leonard.crestez@nxp.com> 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.
>>>
Actually there is a previous report (in i.MX BSP) of CAAM job ring generating
an interrupt at probe time, between request_irq() and reset:
https://source.codeaurora.org/external/imx/linux-imx/commit/drivers/crypto/caam?h=imx_4.14.98_2.0.0_ga&id=aa7b3f51971ec1f60f41fe8ea71870b215376b8a

So yes, there might be cases when interrupts are pending.

>>
>> 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.
> 
There's some history here... (which is in contradiction with above-mentioned
report).

Commit 9620fd959fb1 ("crypto: caam - handle interrupt lines shared across rings")
moved request_irq() before JR reset:
"
    - resetting a job ring triggers an interrupt, so move request_irq
    prior to jr_reset to avoid 'got IRQ but nobody cared' messages.
"

but IIUC that ("resetting a job ring triggers an interrupt") was actually
due to disabling the IRQ line using disable_irq() instead of masking
the interrupt in HW using JRCFGR_LS[IMSK].

The way JR reset sequence is implemented now - in caam_reset_hw_jr() - should not
trigger any interrupt.

Thus, it should be safe to move request_irq() after everything is set up,
at the end of probing.

My suggestion is to move both tasklet_init() and request_irq() at the bottom
of the probe callback.
However, I'd say this is a fix that should be marked accordingly and
should be posted either separately, or at the top of the patch set.

Thanks,
Horia