On Tue, Sep 02, 2025 at 11:38:18AM +0900, Mikko Perttunen wrote:
> > @@ -282,20 +411,27 @@ static int tegra20_vi_enable(struct tegra_vi *vi,
> > bool on)
> > static int tegra20_channel_host1x_syncpt_init(struct tegra_vi_channel
> > *chan)
> > {
> > struct tegra_vi *vi = chan->vi;
> > - struct host1x_syncpt *out_sp;
> > + struct host1x_syncpt *out_sp, *fs_sp;
> >
> > out_sp = host1x_syncpt_request(&vi->client,
> > HOST1X_SYNCPT_CLIENT_MANAGED);
> > if (!out_sp)
> > - return dev_err_probe(vi->dev, -ENOMEM, "failed to request
> > syncpoint\n");
> > + return dev_err_probe(vi->dev, -ENOMEM, "failed to request mw
> > ack syncpoint\n");
>
> Existing issue, but dev_err_probe doesn't print anything when the error is
> -ENOMEM, since "there is already enough output". But that's not necessarily
> the case with failing syncpoint allocation. Maybe we should be using a
> different error code like EBUSY?
>
I'm not sure I love the rule that -ENOMEM doesn't print a message.
Deleting error messages is fine because it makes the code simpler and
saves a little memory. But with dev_err_probe() the message is still
there in the memory, we just don't print it. Printing the error message
doesn't hurt anything.
But if we go down that road, we should make it make it a checkpatch
warning to pass a hard coded -ENOMEM to dev_err_probe().
regards,
dan carpenter
