On Wed, Feb 26, 2020 at 01:37:17PM +0000, Mike Leach wrote:
> Hi Mathieu,
>
> On Fri, 14 Feb 2020 at 22:58, Mathieu Poirier
> <mathieu.poirier@xxxxxxxxxx> wrote:
> >
> > On Tue, Feb 11, 2020 at 10:58:07AM +0000, Mike Leach wrote:
> > > Adds in sysfs links for connections where the connected device is another
> > > coresight device. This allows examination of the coresight topology.
> > >
> > > Non-coresight connections remain just as a reference name.
> > >
> > > Signed-off-by: Mike Leach <mike.leach@xxxxxxxxxx>
> > > ---
> > > drivers/hwtracing/coresight/coresight-cti.c | 41 ++++++++++++++++++++-
> > > 1 file changed, 40 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/hwtracing/coresight/coresight-cti.c b/drivers/hwtracing/coresight/coresight-cti.c
> > > index 9e18e176831c..f620e9460e7d 100644
> > > --- a/drivers/hwtracing/coresight/coresight-cti.c
> > > +++ b/drivers/hwtracing/coresight/coresight-cti.c
> > > @@ -441,6 +441,37 @@ int cti_channel_setop(struct device *dev, enum cti_chan_set_op op,
> > > return err;
> > > }
> > >
> > > +static void cti_add_sysfs_link(struct cti_drvdata *drvdata,
> > > + struct cti_trig_con *tc)
> > > +{
> > > + struct coresight_sysfs_link link_info;
> > > +
> > > + link_info.orig = drvdata->csdev;
> > > + link_info.orig_name = tc->con_dev_name;
> > > + link_info.target = tc->con_dev;
> > > + link_info.target_name = dev_name(&drvdata->csdev->dev);
> > > + coresight_add_sysfs_link(&link_info);
> >
> > I understand there isn't much to do if a problem occurs so just catch the error
> > and add a comment to assert you're doing this on purpose.
> >
>
> When I revisited this code I saw an imbalance between the case where
> the CTI is created first and the associated CSdev is created first.
> The result could be shutdown path where the CTI removes sysfs links
> after the csdev has been removed - which really shouldn't happen.
> Also we could try to remove a sysfs link that we failed to set in the
> first place - again not ideal
>
> I've reworked this code to fix this, and now if the sysfs link fails
> to be set, then we do not set the association between CTI and csdev
> components.
Ok
> This is not sufficient to fail either component from registering, as
> we may have successfully registered previous associations with the
> same CTI.
>
That is also my opinion.
> It seems unlikely that the sysfs link could fail, but if it does, is
> it worth using a dev_warn() to at least log the failure?
>
Yes, that would surely be helpful.
> Regards
>
> Mike
>
>
> > > +}
> > > +
> > > +static void cti_remove_all_sysfs_links(struct cti_drvdata *drvdata)
> > > +{
> > > + struct cti_trig_con *tc;
> > > + struct cti_device *ctidev = &drvdata->ctidev;
> > > + struct coresight_sysfs_link link_info;
> > > +
> > > + /* origin device and target link name constant for this cti */
> > > + link_info.orig = drvdata->csdev;
> > > + link_info.target_name = dev_name(&drvdata->csdev->dev);
> > > +
> > > + list_for_each_entry(tc, &ctidev->trig_cons, node) {
> > > + if (tc->con_dev) {
> > > + link_info.target = tc->con_dev;
> > > + link_info.orig_name = tc->con_dev_name;
> > > + coresight_remove_sysfs_link(&link_info);
> > > + }
> > > + }
> > > +}
> > > +
> > > /*
> > > * Look for a matching connection device name in the list of connections.
> > > * If found then swap in the csdev name, set trig con association pointer
> > > @@ -452,6 +483,8 @@ cti_match_fixup_csdev(struct cti_device *ctidev, const char *node_name,
> > > {
> > > struct cti_trig_con *tc;
> > > const char *csdev_name;
> > > + struct cti_drvdata *drvdata = container_of(ctidev, struct cti_drvdata,
> > > + ctidev);
> > >
> > > list_for_each_entry(tc, &ctidev->trig_cons, node) {
> > > if (tc->con_dev_name) {
> > > @@ -462,6 +495,7 @@ cti_match_fixup_csdev(struct cti_device *ctidev, const char *node_name,
> > > devm_kstrdup(&csdev->dev, csdev_name,
> > > GFP_KERNEL);
> > > tc->con_dev = csdev;
> > > + cti_add_sysfs_link(drvdata, tc);
> > > return true;
> > > }
> > > }
> > > @@ -546,10 +580,12 @@ static void cti_update_conn_xrefs(struct cti_drvdata *drvdata)
> > > struct cti_device *ctidev = &drvdata->ctidev;
> > >
> > > list_for_each_entry(tc, &ctidev->trig_cons, node) {
> > > - if (tc->con_dev)
> > > + if (tc->con_dev) {
> > > /* set tc->con_dev->ect_dev */
> > > coresight_set_assoc_ectdev_mutex(tc->con_dev,
> > > drvdata->csdev);
> > > + cti_add_sysfs_link(drvdata, tc);
> > > + }
> > > }
> > > }
> > >
> > > @@ -602,6 +638,9 @@ static void cti_device_release(struct device *dev)
> > > mutex_lock(&ect_mutex);
> > > cti_remove_conn_xrefs(drvdata);
> > >
> > > + /* clear the dynamic sysfs associate with connections */
> >
> > s/associate/associated
> >
> > > + cti_remove_all_sysfs_links(drvdata);
> > > +
> > > /* remove from the list */
> > > list_for_each_entry_safe(ect_item, ect_tmp, &ect_net, node) {
> > > if (ect_item == drvdata) {
> >
> > With the above:
> >
> > Reviewed-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
> >
> > > --
> > > 2.17.1
> > >
>
>
>
> --
> Mike Leach
> Principal Engineer, ARM Ltd.
> Manchester Design Centre. UK
