| Message ID | 20250225110020.59221-3-urezki@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v3,1/3] rcutorture: Allow a negative value for nfakewriters | expand |
On Tue, Feb 25, 2025 at 12:00:20PM +0100, Uladzislau Rezki (Sony) wrote: > Switch for using of get_state_synchronize_rcu_full() and > poll_state_synchronize_rcu_full() pair for debug a normal > synchronize_rcu() call. > > Just using "not" full APIs to identify if a grace period > is passed or not might lead to a false kernel splat. > > Link: https://lore.kernel.org/lkml/Z5ikQeVmVdsWQrdD@pc636/T/ > Fixes: 988f569ae041 ("rcu: Reduce synchronize_rcu() latency") > Reported-by: cheung wall <zzqq0103.hey@gmail.com> > Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com> This one passes light testing, so: Reviewed-by: Paul E. McKenney <paulmck@kernel.org> > --- > include/linux/rcupdate_wait.h | 3 +++ > kernel/rcu/tree.c | 8 +++----- > 2 files changed, 6 insertions(+), 5 deletions(-) > > diff --git a/include/linux/rcupdate_wait.h b/include/linux/rcupdate_wait.h > index f9bed3d3f78d..4c92d4291cce 100644 > --- a/include/linux/rcupdate_wait.h > +++ b/include/linux/rcupdate_wait.h > @@ -16,6 +16,9 @@ > struct rcu_synchronize { > struct rcu_head head; > struct completion completion; > + > + /* This is for debugging. */ > + struct rcu_gp_oldstate oldstate; > }; > void wakeme_after_rcu(struct rcu_head *head); > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c > index 8625f616c65a..48384fa2eaeb 100644 > --- a/kernel/rcu/tree.c > +++ b/kernel/rcu/tree.c > @@ -1632,12 +1632,10 @@ static void rcu_sr_normal_complete(struct llist_node *node) > { > struct rcu_synchronize *rs = container_of( > (struct rcu_head *) node, struct rcu_synchronize, head); > - unsigned long oldstate = (unsigned long) rs->head.func; > > WARN_ONCE(IS_ENABLED(CONFIG_PROVE_RCU) && > - !poll_state_synchronize_rcu(oldstate), > - "A full grace period is not passed yet: %lu", > - rcu_seq_diff(get_state_synchronize_rcu(), oldstate)); > + !poll_state_synchronize_rcu_full(&rs->oldstate), > + "A full grace period is not passed yet!\n"); > > /* Finally. */ > complete(&rs->completion); > @@ -3247,7 +3245,7 @@ static void synchronize_rcu_normal(void) > * snapshot before adding a request. > */ > if (IS_ENABLED(CONFIG_PROVE_RCU)) > - rs.head.func = (void *) get_state_synchronize_rcu(); > + get_state_synchronize_rcu_full(&rs.oldstate); > > rcu_sr_normal_add_req(&rs); > > -- > 2.39.5 >
On Tue, Feb 25, 2025 at 07:48:55AM -0800, Paul E. McKenney wrote: > On Tue, Feb 25, 2025 at 12:00:20PM +0100, Uladzislau Rezki (Sony) wrote: > > Switch for using of get_state_synchronize_rcu_full() and > > poll_state_synchronize_rcu_full() pair for debug a normal > > synchronize_rcu() call. > > > > Just using "not" full APIs to identify if a grace period > > is passed or not might lead to a false kernel splat. > > > > Link: https://lore.kernel.org/lkml/Z5ikQeVmVdsWQrdD@pc636/T/ > > Fixes: 988f569ae041 ("rcu: Reduce synchronize_rcu() latency") > > Reported-by: cheung wall <zzqq0103.hey@gmail.com> > > Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com> > > This one passes light testing, so: > > Reviewed-by: Paul E. McKenney <paulmck@kernel.org> > Thank you! -- Uladzislau Rezki
Hi Ulad, On Tue, Feb 25, 2025 at 12:00:20PM +0100, Uladzislau Rezki (Sony) wrote: > Switch for using of get_state_synchronize_rcu_full() and > poll_state_synchronize_rcu_full() pair for debug a normal > synchronize_rcu() call. > > Just using "not" full APIs to identify if a grace period > is passed or not might lead to a false kernel splat. > Could you provide detailed explanation on this? I.e. why is _full() is needed? I find the current commit message is a bit vague. Regards, Boqun > Link: https://lore.kernel.org/lkml/Z5ikQeVmVdsWQrdD@pc636/T/ > Fixes: 988f569ae041 ("rcu: Reduce synchronize_rcu() latency") > Reported-by: cheung wall <zzqq0103.hey@gmail.com> > Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com> > --- > include/linux/rcupdate_wait.h | 3 +++ > kernel/rcu/tree.c | 8 +++----- > 2 files changed, 6 insertions(+), 5 deletions(-) > > diff --git a/include/linux/rcupdate_wait.h b/include/linux/rcupdate_wait.h > index f9bed3d3f78d..4c92d4291cce 100644 > --- a/include/linux/rcupdate_wait.h > +++ b/include/linux/rcupdate_wait.h > @@ -16,6 +16,9 @@ > struct rcu_synchronize { > struct rcu_head head; > struct completion completion; > + > + /* This is for debugging. */ > + struct rcu_gp_oldstate oldstate; > }; > void wakeme_after_rcu(struct rcu_head *head); > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c > index 8625f616c65a..48384fa2eaeb 100644 > --- a/kernel/rcu/tree.c > +++ b/kernel/rcu/tree.c > @@ -1632,12 +1632,10 @@ static void rcu_sr_normal_complete(struct llist_node *node) > { > struct rcu_synchronize *rs = container_of( > (struct rcu_head *) node, struct rcu_synchronize, head); > - unsigned long oldstate = (unsigned long) rs->head.func; > > WARN_ONCE(IS_ENABLED(CONFIG_PROVE_RCU) && > - !poll_state_synchronize_rcu(oldstate), > - "A full grace period is not passed yet: %lu", > - rcu_seq_diff(get_state_synchronize_rcu(), oldstate)); > + !poll_state_synchronize_rcu_full(&rs->oldstate), > + "A full grace period is not passed yet!\n"); > > /* Finally. */ > complete(&rs->completion); > @@ -3247,7 +3245,7 @@ static void synchronize_rcu_normal(void) > * snapshot before adding a request. > */ > if (IS_ENABLED(CONFIG_PROVE_RCU)) > - rs.head.func = (void *) get_state_synchronize_rcu(); > + get_state_synchronize_rcu_full(&rs.oldstate); > > rcu_sr_normal_add_req(&rs); > > -- > 2.39.5 >
Hello, Boqun. > Hi Ulad, > > On Tue, Feb 25, 2025 at 12:00:20PM +0100, Uladzislau Rezki (Sony) wrote: > > Switch for using of get_state_synchronize_rcu_full() and > > poll_state_synchronize_rcu_full() pair for debug a normal > > synchronize_rcu() call. > > > > Just using "not" full APIs to identify if a grace period > > is passed or not might lead to a false kernel splat. > > > > Could you provide detailed explanation on this? I.e. why is _full() is > needed? I find the current commit message is a bit vague. > <snip> rcu: Use _full() API to debug synchronize_rcu() Switch for using of get_state_synchronize_rcu_full() and poll_state_synchronize_rcu_full() pair to debug a normal synchronize_rcu() call. Just using "not" full APIs to identify if a grace period is passed or not might lead to a false-positive kernel splat. It can happen, because get_state_synchronize_rcu() compresses both normal and expedited states into one single unsigned long value, so a poll_state_synchronize_rcu() can miss GP-completion when synchronize_rcu()/synchronize_rcu_expedited() concurrently run. To address this, switch to poll_state_synchronize_rcu_full() and get_state_synchronize_rcu_full() APIs, which use separate variables for expedited and normal states. <snip> Does it look better? -- Uladzislau Rezki
diff --git a/include/linux/rcupdate_wait.h b/include/linux/rcupdate_wait.h index f9bed3d3f78d..4c92d4291cce 100644 --- a/include/linux/rcupdate_wait.h +++ b/include/linux/rcupdate_wait.h @@ -16,6 +16,9 @@ struct rcu_synchronize { struct rcu_head head; struct completion completion; + + /* This is for debugging. */ + struct rcu_gp_oldstate oldstate; }; void wakeme_after_rcu(struct rcu_head *head); diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c index 8625f616c65a..48384fa2eaeb 100644 --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -1632,12 +1632,10 @@ static void rcu_sr_normal_complete(struct llist_node *node) { struct rcu_synchronize *rs = container_of( (struct rcu_head *) node, struct rcu_synchronize, head); - unsigned long oldstate = (unsigned long) rs->head.func; WARN_ONCE(IS_ENABLED(CONFIG_PROVE_RCU) && - !poll_state_synchronize_rcu(oldstate), - "A full grace period is not passed yet: %lu", - rcu_seq_diff(get_state_synchronize_rcu(), oldstate)); + !poll_state_synchronize_rcu_full(&rs->oldstate), + "A full grace period is not passed yet!\n"); /* Finally. */ complete(&rs->completion); @@ -3247,7 +3245,7 @@ static void synchronize_rcu_normal(void) * snapshot before adding a request. */ if (IS_ENABLED(CONFIG_PROVE_RCU)) - rs.head.func = (void *) get_state_synchronize_rcu(); + get_state_synchronize_rcu_full(&rs.oldstate); rcu_sr_normal_add_req(&rs);
Switch for using of get_state_synchronize_rcu_full() and poll_state_synchronize_rcu_full() pair for debug a normal synchronize_rcu() call. Just using "not" full APIs to identify if a grace period is passed or not might lead to a false kernel splat. Link: https://lore.kernel.org/lkml/Z5ikQeVmVdsWQrdD@pc636/T/ Fixes: 988f569ae041 ("rcu: Reduce synchronize_rcu() latency") Reported-by: cheung wall <zzqq0103.hey@gmail.com> Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com> --- include/linux/rcupdate_wait.h | 3 +++ kernel/rcu/tree.c | 8 +++----- 2 files changed, 6 insertions(+), 5 deletions(-)