On Monday, 11 June 2018 14:41:33 MSK Thierry Reding wrote:
> On Mon, Jun 11, 2018 at 01:35:03PM +0200, Thierry Reding wrote:
> > On Wed, Jun 06, 2018 at 04:42:01PM +0300, Dmitry Osipenko wrote:
> > > On 06.06.2018 14:02, Thierry Reding wrote:
> > > > On Mon, Jun 04, 2018 at 01:36:54AM +0300, Dmitry Osipenko wrote:
> [...]
>
> > > >> + if (!child_count) {
> > > >> + dev_err(emc->dev, "no memory timings in DT node\n");
> > > >> + return -ENOENT;
> > > >> + }
> > > >> +
> > > >> + emc->timings = devm_kcalloc(emc->dev, child_count,
sizeof(*timing),
> > > >> + GFP_KERNEL);
> > > >> + if (!emc->timings)
> > > >> + return -ENOMEM;
> > > >> +
> > > >> + emc->num_timings = child_count;
> > > >> + timing = emc->timings;
> > > >> +
> > > >> + for_each_child_of_node(node, child) {
> > > >> + err = load_one_timing_from_dt(emc, timing++, child);
> > > >> + if (err) {
> > > >> + of_node_put(child);
> > > >> + return err;
> > > >> + }
> > > >> + }
> > > >> +
> > > >> + sort(emc->timings, emc->num_timings, sizeof(*timing),
cmp_timings,
> > > >> + NULL);
> > > >> +
> > > >> + return 0;
> > > >> +}
> > > >> +
> > > >> +static struct device_node *
> > > >> +tegra_emc_find_node_by_ram_code(struct tegra_emc *emc, u32 ram_code)
> > > >> +{
> > > >> + struct device_node *np;
> > > >> + int err;
> > > >> +
> > > >> + for_each_child_of_node(emc->dev->of_node, np) {
> > > >> + u32 value;
> > > >> +
> > > >> + err = of_property_read_u32(np, "nvidia,ram-code",
> > > >> &value);
> > > >> + if (err || value != ram_code)
> > > >> + continue;
> > > >> +
> > > >> + return np;
> > > >> + }
> > > >> +
> > > >> + dev_info(emc->dev, "no memory timings for RAM code %u found in
> > > >> DT\n",
> > > >> + ram_code);
> > > >
> > > > This seems like it should be dev_warn() or perhaps even dev_err()
> > > > given
> > > > that the result of it is the driver failing to probe. dev_info() may
> > > > go
> > > > unnoticed.
> > >
> > > Absence of memory timings is a valid case, hence dev_info() suit well
> > > here.
> > >
> > > I can't see anything wrong with returning a errno if driver has nothing
> > > to do and prefer to keep it because in that case managed resources
> > > would be free'd by the driver core, though returning '0' also would
> > > work.
> >
> > I disagree. A driver failing to probe will show up as a kernel log entry
> > and is something that people will have to whitelist if they're filtering
> > for error messages in the boot log.
> >
> > I think it's more user-friendly to just let the driver succeed the probe
> > in an expected case, even if that means there's really nothing to do. If
> > you're really concerned about the managed resources staying around, I
> > think you could probably get rid of them explicitly. By the looks of it
> > devres_release_all() isn't an exported symbol, so it can't be called
> > from driver code, but perhaps that's something that we can change.
>
> Maybe an easier way to avoid keeping the managed resources around would
> be to move the check a little further up. That way, we can abort earlier
> if no EMC timings are available, before any resources are even
> allocated.
>
> The tegra_emc_find_node_by_ram_code() function would need to take a
> struct device * instead of struct tegra_emc *, but otherwise it should
> work fine.
Good point, thank you!
