Re: [PATCH 1/1] driver core: Avoid using fwnode in __fwnode_link_del()

[Herve Codina <herve.codina@bootlin.com>]

Hi Saravan,

On Fri, 10 Nov 2023 12:09:02 -0800
Saravana Kannan <saravanak@google.com> wrote:

> On Fri, Nov 10, 2023 at 9:01 AM Herve Codina <herve.codina@bootlin.com> wrote:
> >
> > A refcount issue can appeared in __fwnode_link_del() due to the
> > pr_debug() call:
> >   WARNING: CPU: 0 PID: 901 at lib/refcount.c:25 refcount_warn_saturate+0xe5/0x110
> >   Call Trace:
> >   <TASK>
> >   ? refcount_warn_saturate+0xe5/0x110
> >   ? __warn+0x81/0x130
> >   ? refcount_warn_saturate+0xe5/0x110
> >   ? report_bug+0x191/0x1c0
> >   ? srso_alias_return_thunk+0x5/0x7f
> >   ? prb_read_valid+0x1b/0x30
> >   ? handle_bug+0x3c/0x80
> >   ? exc_invalid_op+0x17/0x70
> >   ? asm_exc_invalid_op+0x1a/0x20
> >   ? refcount_warn_saturate+0xe5/0x110
> >   kobject_get+0x68/0x70
> >   of_node_get+0x1e/0x30
> >   of_fwnode_get+0x28/0x40
> >   fwnode_full_name_string+0x34/0x90
> >   fwnode_string+0xdb/0x140
> >   vsnprintf+0x17b/0x630
> >   va_format.isra.0+0x71/0x130
> >   vsnprintf+0x17b/0x630
> >   vprintk_store+0x162/0x4d0
> >   ? srso_alias_return_thunk+0x5/0x7f
> >   ? srso_alias_return_thunk+0x5/0x7f
> >   ? srso_alias_return_thunk+0x5/0x7f
> >   ? try_to_wake_up+0x9c/0x620
> >   ? rwsem_mark_wake+0x1b2/0x310
> >   vprintk_emit+0xe4/0x2b0
> >   _printk+0x5c/0x80
> >   __dynamic_pr_debug+0x131/0x160
> >   ? srso_alias_return_thunk+0x5/0x7f
> >   __fwnode_link_del+0x25/0xa0
> >   fwnode_links_purge+0x39/0xb0
> >   of_node_release+0xd9/0x180
> >   kobject_put+0x7b/0x190
> >   ...
> >
> > Indeed, an of_node is destroyed and so, of_node_release() is called
> > because the of_node refcount reached 0.
> > of_node_release() calls fwnode_links_purge() to purge the links and
> > ended with __fwnode_link_del() calls.
> > __fwnode_link_del calls pr_debug() to print the fwnodes (of_nodes)
> > involved in the link and so this call is done while one of them is no
> > more available (ie the one related to the of_node_release() call)
> >
> > Remove the pr_debug() call to avoid the use of the links fwnode while
> > destroying the fwnode itself.
> >
> > Fixes: ebd6823af378 ("driver core: Add debug logs when fwnode links are added/deleted")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Herve Codina <herve.codina@bootlin.com>
> > ---
> >  drivers/base/core.c | 2 --
> >  1 file changed, 2 deletions(-)
> >
> > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > index f4b09691998e..62088c663014 100644
> > --- a/drivers/base/core.c
> > +++ b/drivers/base/core.c
> > @@ -109,8 +109,6 @@ int fwnode_link_add(struct fwnode_handle *con, struct fwnode_handle *sup)
> >   */
> >  static void __fwnode_link_del(struct fwnode_link *link)
> >  {
> > -       pr_debug("%pfwf Dropping the fwnode link to %pfwf\n",
> > -                link->consumer, link->supplier);  
> 
> Valid issue, but a NACK for the patch.
> 
> The pr_debug has been very handy, so I don't want to delete it. Also,
> the fwnode link can't get deleted before the supplier/consumer. If it
> is, I need to take a closer look as I'd expect the list_del() to cause
> corruption. My guess is that the %pfwf is traversing stuff that's
> causing an issue. But let me take a closer look next week when I'll be
> at LPC.
> 

The issue is really related to print the full name (%pfwf) of the node
been destroyed by of_node_release() due to refcount == 0.
The issue does not appear with %pfwP.

Looked at printk(). On %pfwf fwnode_handle_{get,put}() is called for
current node and its parents whereas %pfwP does not call
fwnode_handle_{get,put}() on the current node.

A fix can probably be done at printk() level to avoid the
fwnode_handle_{get,put}() calls for the current node in case of %pfwf.

I will do a patch in this way instead of removing the pr_debug() call
in __fwnode_link_del().

Best regards,
Hervé