[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20260116161607.00005588@huawei.com>
Date: Fri, 16 Jan 2026 16:16:07 +0000
From: Jonathan Cameron <jonathan.cameron@...wei.com>
To: <dan.j.williams@...el.com>
CC: Terry Bowman <terry.bowman@....com>, <dave@...olabs.net>,
<dave.jiang@...el.com>, <alison.schofield@...el.com>, <bhelgaas@...gle.com>,
<shiju.jose@...wei.com>, <ming.li@...omail.com>,
<Smita.KoralahalliChannabasappa@....com>, <rrichter@....com>,
<dan.carpenter@...aro.org>, <PradeepVineshReddy.Kodamati@....com>,
<lukas@...ner.de>, <Benjamin.Cheatham@....com>,
<sathyanarayanan.kuppuswamy@...ux.intel.com>, <linux-cxl@...r.kernel.org>,
<vishal.l.verma@...el.com>, <alucerop@....com>, <ira.weiny@...el.com>,
<linux-kernel@...r.kernel.org>, <linux-pci@...r.kernel.org>
Subject: Re: [PATCH v14 19/34] cxl/port: Fix devm resource leaks around with
dport management
> rc = check_no_existing_dport(port, dport_dev);
> if (rc)
> return ERR_PTR(rc);
>
> then can rename new_dport to dport for this patch.
> Or don't bother hiding it and just use dport variable for both.
>
>
> > int rc;
> >
> > @@ -1615,29 +1649,47 @@ static struct cxl_dport *cxl_port_add_dport(struct cxl_port *port,
> > return ERR_PTR(-EBUSY);
> > }
> >
> > - struct cxl_dport *new_dport __free(del_cxl_dport) =
> > - devm_cxl_add_dport_by_dev(port, dport_dev);
> > - if (IS_ERR(new_dport))
> > - return new_dport;
> > -
> > - cxl_switch_parse_cdat(new_dport);
> > + /*
> > + * With the first dport arrival it is now safe to start looking at
> > + * component registers. Be careful to not strand resources if dport
> > + * creation ultimately fails.
> > + */
> > + struct cxl_port *port_group __free(cxl_port_release_group) =
> > + cxl_port_open_group(port);
>
> So this is relies on everything being registered against port->dev, whereas
> for root ports this then gets handed off to port->uport_dev.
> That's a bit obscure - hence request for some patch description text
> on who owns what resources + when.
>
> > + if (IS_ERR(port_group))
> > + return ERR_CAST(port_group);
> >
> > - if (ida_is_empty(&port->decoder_ida)) {
> > + if (port->nr_dports == 0) {
> > rc = devm_cxl_switch_port_decoders_setup(port);
> > if (rc)
> > return ERR_PTR(rc);
I'm not totally sure I have an appropriate base for messing with this but
with just this patch on top of cxl/for-7.0/cxl-init I'm getting problems
with the reorder here as the devm_cxl_switch_port fails to map HDM decoders.
On a simple single switch couple of devices test.
I'm probably suffering Friday afternoon syndrome (and naughty testing just
a middle patch in a series)
My 'guess' is that it's because we no longer add the port first.
I couldn't spot any other patch messing with related logic earlier in Terry's series
(and wanted to keep what I was messing with for trying alternative fixes to a minimum)
Any thoughts on what I'm missing?
Jonathan
> > - dev_dbg(&port->dev, "first dport%d:%s added with decoders\n",
> > - new_dport->port_id, dev_name(dport_dev));
> > - return no_free_ptr(new_dport);
> > + /*
> > + * Note, when nr_dports returns to zero the port is unregistered
> > + * and triggers cleanup. I.e. no need for open-coded release
> > + * action on dport removal. See cxl_detach_ep() for that logic.
> > + */
> > }
> >
> > + new_dport = cxl_add_dport_by_dev(port, dport_dev);
> > + if (IS_ERR(new_dport))
> > + return new_dport;
> > +
> > + rc = cxl_dport_autoremove(new_dport);
> > + if (rc)
> > + return ERR_PTR(rc);
> > +
> > + cxl_switch_parse_cdat(new_dport);
> > +
> > + /* group tracking no longer needed, dport successfully added */
> > + cxl_port_remove_group(no_free_ptr(port_group));
Powered by blists - more mailing lists