On Sun, May 20, 2018 at 09:42:19PM -0700, Joel Fernandes wrote:
> The funnel locking loop in rcu_start_this_gp uses rcu_root as a
> temporary variable while walking the combining tree. This causes a
> tiresome exercise of a code reader reminding themselves that rcu_root
> may not be root. Lets just call it rnp, and rename other variables as
> well to be more appropriate.
>
> Original patch: https://patchwork.kernel.org/patch/10396577/
>
> Signed-off-by: Joel Fernandes <j...@joelfernandes.org>
Nice!
Please see feedback interspersed below.
Thanx, Paul
> ---
> kernel/rcu/tree.c | 48 ++++++++++++++++++++++++-----------------------
> 1 file changed, 25 insertions(+), 23 deletions(-)
>
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 0ffd41ba304f..879c67a31116 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -1526,7 +1526,7 @@ static void trace_rcu_this_gp(struct rcu_node *rnp,
> struct rcu_data *rdp,
>
> /*
> * rcu_start_this_gp - Request the start of a particular grace period
> - * @rnp: The leaf node of the CPU from which to start.
> + * @rnp_start: The leaf node of the CPU from which to start.
> * @rdp: The rcu_data corresponding to the CPU from which to start.
> * @gp_seq_req: The gp_seq of the grace period to start.
> *
> @@ -1540,12 +1540,12 @@ static void trace_rcu_this_gp(struct rcu_node *rnp,
> struct rcu_data *rdp,
> *
> * Returns true if the GP thread needs to be awakened else false.
> */
> -static bool rcu_start_this_gp(struct rcu_node *rnp, struct rcu_data *rdp,
> +static bool rcu_start_this_gp(struct rcu_node *rnp_start, struct rcu_data
> *rdp,
> unsigned long gp_seq_req)
> {
> bool ret = false;
> struct rcu_state *rsp = rdp->rsp;
> - struct rcu_node *rnp_root;
> + struct rcu_node *rnp, *rnp_root = NULL;
Unless I am going blind, this patch really isn't using rnp_root. It
could be removed.
>
> /*
> * Use funnel locking to either acquire the root rcu_node
> @@ -1556,34 +1556,36 @@ static bool rcu_start_this_gp(struct rcu_node *rnp,
> struct rcu_data *rdp,
> * scan the leaf rcu_node structures. Note that rnp->lock must
> * not be released.
> */
> - raw_lockdep_assert_held_rcu_node(rnp);
> - trace_rcu_this_gp(rnp, rdp, gp_seq_req, TPS("Startleaf"));
> - for (rnp_root = rnp; 1; rnp_root = rnp_root->parent) {
> - if (rnp_root != rnp)
> - raw_spin_lock_rcu_node(rnp_root);
> - if (ULONG_CMP_GE(rnp_root->gp_seq_needed, gp_seq_req) ||
> - rcu_seq_started(&rnp_root->gp_seq, gp_seq_req) ||
> - (rnp != rnp_root &&
> - rcu_seq_state(rcu_seq_current(&rnp_root->gp_seq)))) {
> - trace_rcu_this_gp(rnp_root, rdp, gp_seq_req,
> + raw_lockdep_assert_held_rcu_node(rnp_start);
> + trace_rcu_this_gp(rnp_start, rdp, gp_seq_req, TPS("Startleaf"));
> + for (rnp = rnp_start; 1; rnp = rnp->parent) {
> + if (rnp != rnp_start)
> + raw_spin_lock_rcu_node(rnp);
> + if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req) ||
> + rcu_seq_started(&rnp->gp_seq, gp_seq_req) ||
> + (rnp != rnp_start &&
> + rcu_seq_state(rcu_seq_current(&rnp->gp_seq)))) {
> + trace_rcu_this_gp(rnp, rdp, gp_seq_req,
> TPS("Prestarted"));
> goto unlock_out;
> }
> - rnp_root->gp_seq_needed = gp_seq_req;
> - if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) {
> + rnp->gp_seq_needed = gp_seq_req;
> + if (rcu_seq_state(rcu_seq_current(&rnp_start->gp_seq))) {
The original had a performance bug, which is quite a bit more obvious
given the new names, so thank you for that! The above statement should
instead be as follows:
if (rcu_seq_state(rcu_seq_current(&rnp->gp_seq))) {
It does not make sense to keep checking the starting rcu_node because
changes to ->gp_seq happen first at the top of the tree. So we might
take an earlier exit by checking the current rnp instead of rechecking
rnp_start over and over.
Please feel free to make this change, which is probably best as a separate
patch. That way this rename patch can remain a straightforward rename patch.
> /*
> * We just marked the leaf, and a grace period
> * is in progress, which means that rcu_gp_cleanup()
> * will see the marking. Bail to reduce contention.
> */
> - trace_rcu_this_gp(rnp, rdp, gp_seq_req,
> + trace_rcu_this_gp(rnp_start, rdp, gp_seq_req,
> TPS("Startedleaf"));
> goto unlock_out;
> }
> - if (rnp_root != rnp && rnp_root->parent != NULL)
> - raw_spin_unlock_rcu_node(rnp_root);
> - if (!rnp_root->parent)
> + if (rnp != rnp_start && rnp->parent != NULL)
> + raw_spin_unlock_rcu_node(rnp);
> + if (!rnp->parent) {
> + rnp_root = rnp;
Since rnp_root is otherwise unused in the new version, the above statement
can be dropped along with the "if" statement's braces and the declaration.
> break; /* At root, and perhaps also leaf. */
> + }
> }
>
> /* If GP already in progress, just leave, otherwise start one. */
> @@ -1601,11 +1603,11 @@ static bool rcu_start_this_gp(struct rcu_node *rnp,
> struct rcu_data *rdp,
> trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gp_seq),
> TPS("newreq"));
> ret = true; /* Caller must wake GP kthread. */
> unlock_out:
> - if (rnp != rnp_root)
> - raw_spin_unlock_rcu_node(rnp_root);
> + if (rnp != rnp_start)
> + raw_spin_unlock_rcu_node(rnp);
> /* Push furthest requested GP to leaf node and rcu_data structure. */
> - if (ULONG_CMP_GE(rnp_root->gp_seq_needed, gp_seq_req)) {
> - rnp->gp_seq_needed = gp_seq_req;
> + if (ULONG_CMP_GE(rnp->gp_seq_needed, gp_seq_req)) {
> + rnp_start->gp_seq_needed = gp_seq_req;
> rdp->gp_seq_needed = gp_seq_req;
> }
> return ret;
> --
> 2.17.0.441.gb46fe60e1d-goog
>
