RE: Re: [v4,net-next 2/4] net: schedule: add action gate offloading

RE: Re: [v4,net-next 2/4] net: schedule: add action gate offloading

[Po Liu]

Hi Vlad,



> -----Original Message-----
> From: Vlad Buslov <vlad@buslov.dev>
> Sent: 2020年4月30日 1:41
> To: Po Liu <po.liu@nxp.com>
> Cc: davem@davemloft.net; linux-kernel@vger.kernel.org;
> netdev@vger.kernel.org; vinicius.gomes@intel.com; Claudiu Manoil
> <claudiu.manoil@nxp.com>; Vladimir Oltean <vladimir.oltean@nxp.com>;
> Alexandru Marginean <alexandru.marginean@nxp.com>;
> michael.chan@broadcom.com; vishal@chelsio.com;
> saeedm@mellanox.com; leon@kernel.org; jiri@mellanox.com;
> idosch@mellanox.com; alexandre.belloni@bootlin.com;
> UNGLinuxDriver@microchip.com; kuba@kernel.org; jhs@mojatatu.com;
> xiyou.wangcong@gmail.com; simon.horman@netronome.com;
> pablo@netfilter.org; moshe@mellanox.com; m-karicheri2@ti.com;
> andre.guedes@linux.intel.com; stephen@networkplumber.org
> Subject: Re: [v4,net-next 2/4] net: schedule: add action gate
> offloading
> 
> 
> On Tue 28 Apr 2020 at 06:34, Po Liu <Po.Liu@nxp.com> wrote:
> > Add the gate action to the flow action entry. Add the gate parameters
> > to the tc_setup_flow_action() queueing to the entries of
> > flow_action_entry array provide to the driver.
> >
> > Signed-off-by: Po Liu <Po.Liu@nxp.com>
> > ---
> >  include/net/flow_offload.h   |  10 ++++
> >  include/net/tc_act/tc_gate.h | 113
> +++++++++++++++++++++++++++++++++++
> >  net/sched/cls_api.c          |  33 ++++++++++
> >  3 files changed, 156 insertions(+)
> >
> > diff --git a/include/net/flow_offload.h b/include/net/flow_offload.h
> > index 3619c6acf60f..94a30fe02e6d 100644
> > --- a/include/net/flow_offload.h
> > +++ b/include/net/flow_offload.h
> > @@ -147,6 +147,7 @@ enum flow_action_id {
> >       FLOW_ACTION_MPLS_PUSH,
> >       FLOW_ACTION_MPLS_POP,
> >       FLOW_ACTION_MPLS_MANGLE,
> > +     FLOW_ACTION_GATE,
> >       NUM_FLOW_ACTIONS,
> >  };
> >
> > @@ -255,6 +256,15 @@ struct flow_action_entry {
> >                       u8              bos;
> >                       u8              ttl;
> >               } mpls_mangle;
> > +             struct {
> > +                     u32             index;
> > +                     s32             prio;
> > +                     u64             basetime;
> > +                     u64             cycletime;
> > +                     u64             cycletimeext;
> > +                     u32             num_entries;
> > +                     struct action_gate_entry *entries;
> > +             } gate;
> >       };
> >       struct flow_action_cookie *cookie; /* user defined action cookie
> > */  }; diff --git a/include/net/tc_act/tc_gate.h
> > b/include/net/tc_act/tc_gate.h index 330ad8b02495..9e698c7d64cd
> 100644
> > --- a/include/net/tc_act/tc_gate.h
> > +++ b/include/net/tc_act/tc_gate.h
> > @@ -7,6 +7,13 @@
> >  #include <net/act_api.h>
> >  #include <linux/tc_act/tc_gate.h>
> >
> > +struct action_gate_entry {
> > +     u8                      gate_state;
> > +     u32                     interval;
> > +     s32                     ipv;
> > +     s32                     maxoctets;
> > +};
> > +
> >  struct tcfg_gate_entry {
> >       int                     index;
> >       u8                      gate_state;
> > @@ -44,4 +51,110 @@ struct tcf_gate {
> >
> >  #define to_gate(a) ((struct tcf_gate *)a)
> >
> > +static inline bool is_tcf_gate(const struct tc_action *a) { #ifdef
> > +CONFIG_NET_CLS_ACT
> > +     if (a->ops && a->ops->id == TCA_ID_GATE)
> > +             return true;
> > +#endif
> > +     return false;
> > +}
> > +
> > +static inline u32 tcf_gate_index(const struct tc_action *a) {
> > +     return a->tcfa_index;
> > +}
> > +
> > +static inline s32 tcf_gate_prio(const struct tc_action *a) {
> > +     s32 tcfg_prio;
> > +
> > +     rcu_read_lock();
> 
> This action no longer uses rcu, so you don't need protect with
> rcu_read_lock() in all these helpers.

I would remove all the rcu_read_lock() here in this patch. 

> 
> > +     tcfg_prio = to_gate(a)->param.tcfg_priority;
> > +     rcu_read_unlock();
> > +
> > +     return tcfg_prio;
> > +}
> > +
> > +static inline u64 tcf_gate_basetime(const struct tc_action *a) {
> > +     u64 tcfg_basetime;
> > +
> > +     rcu_read_lock();
> > +     tcfg_basetime = to_gate(a)->param.tcfg_basetime;
> > +     rcu_read_unlock();
> > +
> > +     return tcfg_basetime;
> > +}
> > +
> > +static inline u64 tcf_gate_cycletime(const struct tc_action *a) {
> > +     u64 tcfg_cycletime;
> > +
> > +     rcu_read_lock();
> > +     tcfg_cycletime = to_gate(a)->param.tcfg_cycletime;
> > +     rcu_read_unlock();
> > +
> > +     return tcfg_cycletime;
> > +}
> > +
> > +static inline u64 tcf_gate_cycletimeext(const struct tc_action *a) {
> > +     u64 tcfg_cycletimeext;
> > +
> > +     rcu_read_lock();
> > +     tcfg_cycletimeext = to_gate(a)->param.tcfg_cycletime_ext;
> > +     rcu_read_unlock();
> > +
> > +     return tcfg_cycletimeext;
> > +}
> > +
> > +static inline u32 tcf_gate_num_entries(const struct tc_action *a) {
> > +     u32 num_entries;
> > +
> > +     rcu_read_lock();
> > +     num_entries = to_gate(a)->param.num_entries;
> > +     rcu_read_unlock();
> > +
> > +     return num_entries;
> > +}
> > +
> > +static inline struct action_gate_entry
> > +                     *tcf_gate_get_list(const struct tc_action *a) {
> > +     struct action_gate_entry *oe;
> > +     struct tcf_gate_params *p;
> > +     struct tcfg_gate_entry *entry;
> > +     u32 num_entries;
> > +     int i = 0;
> > +
> > +     rcu_read_lock();
> > +
> > +     p = &to_gate(a)->param;
> > +     num_entries = p->num_entries;
> > +
> > +     list_for_each_entry(entry, &p->entries, list)
> > +             i++;
> > +
> > +     if (i != num_entries)
> > +             return NULL;
> > +
> > +     oe = kzalloc(sizeof(*oe) * num_entries, GFP_KERNEL);
> 
> Can't allocate with GFP_KERNEL flag in rcu read blocks, but you don't need
> the rcu read lock here anyway. However, tc_setup_flow_action() calls this
> function while holding tcfa_lock spinlock, which also precludes allocating
> memory with that flag. You can test for such problems by enabling
> CONFIG_DEBUG_ATOMIC_SLEEP. To help uncover such errors all new act

Thanks a lot. I added this config for debug. I would use GFP_ATOMIC flag avoid sleeping alloc and using kcalloc for the array.

> APIs and action implementations are usually accompanied by tdc tests. If
> you chose to implement such tests you can look at 6e52fca36c67 ("tc-tests:
> Add tc action ct tests") for recent example.

I would look into the test. Thanks!

> 
> > +     if (!oe)
> > +             return NULL;
> 
> This returns without releasing rcu read lock, but as I said before you
> probably don't need rcu protection here anyway.

Thanks for remind, that is helpful.

> 
> > +
> > +     i = 0;
> > +     list_for_each_entry(entry, &p->entries, list) {
> > +             oe[i].gate_state = entry->gate_state;
> > +             oe[i].interval = entry->interval;
> > +             oe[i].ipv = entry->ipv;
> > +             oe[i].maxoctets = entry->maxoctets;
> > +             i++;
> > +     }
> > +
> > +     rcu_read_unlock();
> > +
> > +     return oe;
> > +}
> >  #endif
> > diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c index
> > 11b683c45c28..7e85c91d0752 100644
> > --- a/net/sched/cls_api.c
> > +++ b/net/sched/cls_api.c
> > @@ -39,6 +39,7 @@
> >  #include <net/tc_act/tc_skbedit.h>
> >  #include <net/tc_act/tc_ct.h>
> >  #include <net/tc_act/tc_mpls.h>
> > +#include <net/tc_act/tc_gate.h>
> >  #include <net/flow_offload.h>
> >
> >  extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1]; @@
> > -3526,6 +3527,27 @@ static void tcf_sample_get_group(struct
> > flow_action_entry *entry,  #endif  }
> >
> > +static void tcf_gate_entry_destructor(void *priv) {
> > +     struct action_gate_entry *oe = priv;
> > +
> > +     kfree(oe);
> > +}
> > +
> > +static int tcf_gate_get_entries(struct flow_action_entry *entry,
> > +                             const struct tc_action *act) {
> > +     entry->gate.entries = tcf_gate_get_list(act);
> > +
> > +     if (!entry->gate.entries)
> > +             return -EINVAL;
> > +
> > +     entry->destructor = tcf_gate_entry_destructor;
> > +     entry->destructor_priv = entry->gate.entries;
> > +
> > +     return 0;
> > +}
> > +
> >  int tc_setup_flow_action(struct flow_action *flow_action,
> >                        const struct tcf_exts *exts)  { @@ -3672,6
> > +3694,17 @@ int tc_setup_flow_action(struct flow_action *flow_action,
> >               } else if (is_tcf_skbedit_priority(act)) {
> >                       entry->id = FLOW_ACTION_PRIORITY;
> >                       entry->priority = tcf_skbedit_priority(act);
> > +             } else if (is_tcf_gate(act)) {
> > +                     entry->id = FLOW_ACTION_GATE;
> > +                     entry->gate.index = tcf_gate_index(act);
> > +                     entry->gate.prio = tcf_gate_prio(act);
> > +                     entry->gate.basetime = tcf_gate_basetime(act);
> > +                     entry->gate.cycletime = tcf_gate_cycletime(act);
> > +                     entry->gate.cycletimeext = tcf_gate_cycletimeext(act);
> > +                     entry->gate.num_entries = tcf_gate_num_entries(act);
> > +                     err = tcf_gate_get_entries(entry, act);
> > +                     if (err)
> > +                             goto err_out;
> >               } else {
> >                       err = -EOPNOTSUPP;
> >                       goto err_out_locked;

Thanks a lot.

Br,
Po Liu