Re: [PATCH v3] drivers: char: ipmi: ipmi_msghandler: Pass lockdep expression to RCU lists

From: Corey Minyard
Date: Thu Feb 13 2020 - 08:04:13 EST


On Wed, Feb 12, 2020 at 01:49:56PM -0800, Paul E. McKenney wrote:
> On Wed, Feb 12, 2020 at 07:45:52AM -0600, Corey Minyard wrote:
> > On Fri, Jan 17, 2020 at 06:55:22PM +0530, Amol Grover wrote:
> > > intf->cmd_rcvrs is traversed with list_for_each_entry_rcu
> > > outside an RCU read-side critical section but under the
> > > protection of intf->cmd_rcvrs_mutex.
> > >
> > > ipmi_interfaces is traversed using list_for_each_entry_rcu
> > > outside an RCU read-side critical section but under the protection
> > > of ipmi_interfaces_mutex.
> > >
> > > Hence, add the corresponding lockdep expression to the list traversal
> > > primitive to silence false-positive lockdep warnings, and
> > > harden RCU lists.
> > >
> > > Add macro for the corresponding lockdep expression to make the code
> > > clean and concise.
> > >
> > > Signed-off-by: Amol Grover <frextrite@xxxxxxxxx>
> >
> > After reading everything, I think this is correct, but I would like
> > Paul's stamp of approval on this.
>
> Acked-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
>
> But note that I did not trace the locking in the case of ipmi_add_smi().
> I did the others, so lockdep can do the last one. ;-)

Thanks, it's in my queue for 5.7.

-corey

>
> Thanx, Paul
>
> > Thanks,
> >
> > -corey
> >
> > > ---
> > > v3:
> > > - Remove rcu_read_lock_held() from lockdep expression since it is
> > > implicitly checked.
> > > - Remove unintended macro usage.
> > >
> > > v2:
> > > - Fix sparse error
> > > CHECK: Alignment should match open parenthesis
> > >
> > > drivers/char/ipmi/ipmi_msghandler.c | 14 ++++++++++----
> > > 1 file changed, 10 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/drivers/char/ipmi/ipmi_msghandler.c b/drivers/char/ipmi/ipmi_msghandler.c
> > > index cad9563f8f48..64ba16dcb681 100644
> > > --- a/drivers/char/ipmi/ipmi_msghandler.c
> > > +++ b/drivers/char/ipmi/ipmi_msghandler.c
> > > @@ -618,6 +618,8 @@ static DEFINE_MUTEX(ipmidriver_mutex);
> > >
> > > static LIST_HEAD(ipmi_interfaces);
> > > static DEFINE_MUTEX(ipmi_interfaces_mutex);
> > > +#define ipmi_interfaces_mutex_held() \
> > > + lockdep_is_held(&ipmi_interfaces_mutex)
> > > static struct srcu_struct ipmi_interfaces_srcu;
> > >
> > > /*
> > > @@ -1321,7 +1323,8 @@ static void _ipmi_destroy_user(struct ipmi_user *user)
> > > * synchronize_srcu()) then free everything in that list.
> > > */
> > > mutex_lock(&intf->cmd_rcvrs_mutex);
> > > - list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link) {
> > > + list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link,
> > > + lockdep_is_held(&intf->cmd_rcvrs_mutex)) {
> > > if (rcvr->user == user) {
> > > list_del_rcu(&rcvr->link);
> > > rcvr->next = rcvrs;
> > > @@ -1599,7 +1602,8 @@ static struct cmd_rcvr *find_cmd_rcvr(struct ipmi_smi *intf,
> > > {
> > > struct cmd_rcvr *rcvr;
> > >
> > > - list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link) {
> > > + list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link,
> > > + lockdep_is_held(&intf->cmd_rcvrs_mutex)) {
> > > if ((rcvr->netfn == netfn) && (rcvr->cmd == cmd)
> > > && (rcvr->chans & (1 << chan)))
> > > return rcvr;
> > > @@ -1614,7 +1618,8 @@ static int is_cmd_rcvr_exclusive(struct ipmi_smi *intf,
> > > {
> > > struct cmd_rcvr *rcvr;
> > >
> > > - list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link) {
> > > + list_for_each_entry_rcu(rcvr, &intf->cmd_rcvrs, link,
> > > + lockdep_is_held(&intf->cmd_rcvrs_mutex)) {
> > > if ((rcvr->netfn == netfn) && (rcvr->cmd == cmd)
> > > && (rcvr->chans & chans))
> > > return 0;
> > > @@ -3450,7 +3455,8 @@ int ipmi_add_smi(struct module *owner,
> > > /* Look for a hole in the numbers. */
> > > i = 0;
> > > link = &ipmi_interfaces;
> > > - list_for_each_entry_rcu(tintf, &ipmi_interfaces, link) {
> > > + list_for_each_entry_rcu(tintf, &ipmi_interfaces, link,
> > > + ipmi_interfaces_mutex_held()) {
> > > if (tintf->intf_num != i) {
> > > link = &tintf->link;
> > > break;
> > > --
> > > 2.24.1
> > >