Re: [PATCH 1/6] remoteproc: Introduce rproc_detach_device() wrapper

[Suman Anna]

On 5/27/21 11:17 PM, Bjorn Andersson wrote:
> On Fri 21 May 19:03 CDT 2021, Suman Anna wrote:
> 
>> The .attach() rproc ops is invoked through the helper
>> rproc_attach_device(), but the .detach() ops is invoked
>> directly at present. Introduce a similar wrapper function
>> rproc_detach_device() for .detach() ops so that the code
>> is symmetric.
>>
>> Signed-off-by: Suman Anna <s-anna@xxxxxx>
>> ---
>>  drivers/remoteproc/remoteproc_core.c     | 2 +-
>>  drivers/remoteproc/remoteproc_internal.h | 8 ++++++++
>>  2 files changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/remoteproc/remoteproc_core.c b/drivers/remoteproc/remoteproc_core.c
>> index 6348aaa42bbb..6019f46001c8 100644
>> --- a/drivers/remoteproc/remoteproc_core.c
>> +++ b/drivers/remoteproc/remoteproc_core.c
>> @@ -1869,7 +1869,7 @@ static int __rproc_detach(struct rproc *rproc)
>>  	}
>>  
>>  	/* Tell the remote processor the core isn't available anymore */
>> -	ret = rproc->ops->detach(rproc);
>> +	ret = rproc_detach_device(rproc);
>>  	if (ret) {
>>  		dev_err(dev, "can't detach from rproc: %d\n", ret);
>>  		return ret;
>> diff --git a/drivers/remoteproc/remoteproc_internal.h b/drivers/remoteproc/remoteproc_internal.h
>> index a328e634b1de..931d50b6a0d1 100644
>> --- a/drivers/remoteproc/remoteproc_internal.h
>> +++ b/drivers/remoteproc/remoteproc_internal.h
>> @@ -121,6 +121,14 @@ static inline int rproc_attach_device(struct rproc *rproc)
>>  	return 0;
>>  }
>>  
>> +static inline int rproc_detach_device(struct rproc *rproc)
>> +{
>> +	if (rproc->ops->detach)
>> +		return rproc->ops->detach(rproc);
>> +
>> +	return 0;
> 
> I was going to complain that this will silently succeed to detach a
> remoteproc when the driver doesn't implement detach, but then I realized
> that in the current code path we just failed if it wasn't set.
> 
> So this only becomes a problem if we're out of sync between the wish to
> detach and the implementation of detach, in the later patch.
> 
> But based on this, why do we allow rproc_attach_device() to succeed even
> though a driver doesn't implement attach? Could we achieve the symmetry
> by going the other way?

We don't, it does throw an error. See rproc_validate(). The error-checking is
somewhat asymmetric. Any remoteproc requiring attach behavior is supposed to be
setting the rproc state as RPROC_DETACHED. The remoteproc core state-machine is
dictated by that value between start and attach. rproc_validate() does check the
required ops between RPROC_OFFLINE and RPROC_DETACHED states.

Do you mean use return -EINVAL by default in both the wrappers? Atm, you will
never exercise this particular code paths in either of these wrapper functions,
because there are checks enforced even before these wrappers are invoked.

regards
Suman

> 
> Regards,
> Bjorn
> 
>> +}
>> +
>>  static inline
>>  int rproc_fw_sanity_check(struct rproc *rproc, const struct firmware *fw)
>>  {
>> -- 
>> 2.30.1
>>