Re: [PATCH 2/7] coresight: Convert disclaim functions to take a struct cs_access

[James Clark]

On 17/03/2025 6:29 pm, Leo Yan wrote:
> On Mon, Mar 17, 2025 at 11:36:40AM +0000, James Clark wrote:
> > On 13/03/2025 2:54 pm, Leo Yan wrote:
> > > On Tue, Feb 11, 2025 at 10:39:38AM +0000, James Clark wrote:
> > >
> > > [...]
> > >
> > > >    static inline bool coresight_is_claimed_any(struct coresight_device *csdev)
> > > >    {
> > > > -       return coresight_read_claim_tags(csdev) != 0;
> > > > +       return coresight_read_claim_tags(&csdev->access) != 0;
> > > >    }
> > >
> > > Likewise other claim functions, can coresight_is_claimed_any() change its
> > > argument type from struct coresight_device to struct csdev_access?
> >
> > I only wanted to change the ones that I had to. I think we should prioritize
> > passing csdev as much as possible in the coresight framework to make
> > everything consistent. Otherwise it's extra churn for no benefit, and if we
> > need something from csdev here in the future we'll have to change this one
> > back again.
>
> The function coresight_is_claimed_any() has been deleted in a later
> patch.  So this is fine for me.
>
> In theory, claim tags are low level operations and don't need a
> CoreSight device context, I prefer we can keep them as simple as
> possible.
>
> With this series, we can see coresight_claim_device() and
> coresight_disclaim_device() are inconsistent for their parameters:
> one is using "struct coresight_device *" and another is
> "struct csdev_access *".  Maybe we just proceed to use csdev_access
> for all claim tag functions?

That's because coresight_claim_device() has logging which requires the 
device name to make it useful. If anything, coresight_disclaim_device() 
should actually take a csdev and then print a better log message. I 
don't think there's a way to make it consistent when there are different 
requirements for each level of operations. The ones with logging need a 
csdev and the lower level ones used on probe need to work with only csa.

> If later we need to use a CoreSight device context when operating
> claim tags, it means we might have different scenarios and we can
> handle that separately.
>
> Thanks,
> Leo