netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Mahesh Bandewar <maheshb@google.com>
To: Jay Vosburgh <jay.vosburgh@canonical.com>
Cc: Veaceslav Falico <vfalico@redhat.com>,
	Andy Gospodarek <andy@greyhouse.net>,
	David Miller <davem@davemloft.net>,
	netdev <netdev@vger.kernel.org>,
	Eric Dumazet <edumazet@google.com>,
	Maciej Zenczykowski <maze@google.com>
Subject: Re: [PATCH] bonding: Do not try to send packets over dead link in TLB mode.
Date: Thu, 10 Jul 2014 07:40:01 -0700	[thread overview]
Message-ID: <CAF2d9jgQ=0e11hPvyBuh3FsPt1AzOk+UV0yZhtYNPWi6f_pGJQ@mail.gmail.com> (raw)
In-Reply-To: <7156.1404999570@localhost.localdomain>

On Thu, Jul 10, 2014 at 6:39 AM, Jay Vosburgh
<jay.vosburgh@canonical.com> wrote:
> Mahesh Bandewar <maheshb@google.com> wrote:
>
>>In TLB mode if tlb_dynamic_lb is NOT set, slaves from the bond
>>group are selected based on the hash distribution. This does not
>>exclude dead links which are part of the bond. Also if there is a
>>temporary link event which brings down the interface, packets
>>hashed on that interface would be dropped too.
>>
>>This patch fixes these issues and distributes flows across the
>>UP links only. Also the array construction of links which are
>>capable of sending packets happen in the control path leaving
>>only link-selection duing the data-path.
>>
>>One possible side effect of this is - at a link event; all
>>flows will be shuffled to get good distribution. But impact of
>>this should be minimum with the assumption that a member or
>>members of the bond group are not available is a very temporary
>>situation.
>
>         Why limit this to just TLB mode?  Other similar situations,
> e.g., the TX slave selection in 802.3ad or balance-xor via
> bond_xmit_slave_id, would presumably see a performance gain from using
> an array lookup instead of traversing a linked list (even though those
> modes do already skip slaves that are down).
>
>         The 802.3ad case is a bit more complex, as an array would have
> to include only slaves that are members of the active aggregator, but
> the bond_xmit_slave_id case shouldn't be much more difficult than what
> you've done below.
>
That's a good point. My original intent was to fix the issue with the
TLB mode. Eric pointed out the gains with array approach and I started
thinking about gains in other modes too. So I'll fix the current patch
(remove spinlock if possible and few other minor nits) but will try to
cook another patch to address all modes later. So the TLB issue gets
addressed first.
>         -J
>
>>Signed-off-by: Mahesh Bandewar <maheshb@google.com>
>>---
>> drivers/net/bonding/bond_alb.c | 52 +++++++++++++++++++++++++++++++++++++-----
>> drivers/net/bonding/bond_alb.h | 11 +++++++++
>> drivers/net/bonding/bonding.h  |  6 +++++
>> 3 files changed, 63 insertions(+), 6 deletions(-)
>>
>>diff --git a/drivers/net/bonding/bond_alb.c b/drivers/net/bonding/bond_alb.c
>>index 76c0dade233f..1f39d41fde4b 100644
>>--- a/drivers/net/bonding/bond_alb.c
>>+++ b/drivers/net/bonding/bond_alb.c
>>@@ -195,6 +195,9 @@ static int tlb_initialize(struct bonding *bond)
>>
>>       _unlock_tx_hashtbl_bh(bond);
>>
>>+      /* Initialize the TLB array spin-lock */
>>+      spin_lock_init(&bond_info->slave_arr_lock);
>>+
>>       return 0;
>> }
>>
>>@@ -209,6 +212,9 @@ static void tlb_deinitialize(struct bonding *bond)
>>       bond_info->tx_hashtbl = NULL;
>>
>>       _unlock_tx_hashtbl_bh(bond);
>>+
>>+      if (bond_is_nondyn_tlb(bond) && bond_info->slave_arr)
>>+              kfree_rcu(bond_info->slave_arr, rcu);
>> }
>>
>> static long long compute_gap(struct slave *slave)
>>@@ -1406,9 +1412,37 @@ out:
>>       return NETDEV_TX_OK;
>> }
>>
>>+static int bond_tlb_update_slave_arr(struct bonding *bond)
>>+{
>>+      struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
>>+      struct slave *tx_slave;
>>+      struct list_head *iter;
>>+      struct tlb_up_slave *new_arr, *old_arr;
>>+
>>+      new_arr = kzalloc(offsetof(struct tlb_up_slave, arr[bond->slave_cnt]),
>>+                        GFP_KERNEL);
>>+      if (!new_arr)
>>+              return -ENOMEM;
>>+
>>+      bond_for_each_slave(bond, tx_slave, iter) {
>>+              if (bond_slave_can_tx(tx_slave))
>>+                      new_arr->arr[new_arr->count++] = tx_slave;
>>+      }
>>+
>>+      spin_lock(&bond_info->slave_arr_lock);
>>+      old_arr = bond_info->slave_arr;
>>+      rcu_assign_pointer(bond_info->slave_arr, new_arr);
>>+      spin_unlock(&bond_info->slave_arr_lock);
>>+      if (old_arr)
>>+              kfree_rcu(old_arr, rcu);
>>+
>>+      return 0;
>>+}
>>+
>> int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
>> {
>>       struct bonding *bond = netdev_priv(bond_dev);
>>+      struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
>>       struct ethhdr *eth_data;
>>       struct slave *tx_slave = NULL;
>>       u32 hash_index;
>>@@ -1429,12 +1463,13 @@ int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
>>                                                             hash_index & 0xFF,
>>                                                             skb->len);
>>                       } else {
>>-                              struct list_head *iter;
>>-                              int idx = hash_index % bond->slave_cnt;
>>-
>>-                              bond_for_each_slave_rcu(bond, tx_slave, iter)
>>-                                      if (--idx < 0)
>>-                                              break;
>>+                              struct tlb_up_slave *slaves;
>>+                              rcu_read_lock();
>>+                              slaves = rcu_dereference(bond_info->slave_arr);
>>+                              if (slaves && slaves->count)
>>+                                      tx_slave = slaves->arr[hash_index %
>>+                                                             slaves->count];
>>+                              rcu_read_unlock();
>>                       }
>>                       break;
>>               }
>>@@ -1721,6 +1756,11 @@ void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char
>>                        */
>>               }
>>       }
>>+
>>+      if (bond_is_nondyn_tlb(bond)) {
>>+              if (bond_tlb_update_slave_arr(bond))
>>+                      pr_err("Failed to build slave-array for TLB mode.\n");
>>+      }
>> }
>>
>> /**
>>diff --git a/drivers/net/bonding/bond_alb.h b/drivers/net/bonding/bond_alb.h
>>index 5fc76c01636c..731a8e830639 100644
>>--- a/drivers/net/bonding/bond_alb.h
>>+++ b/drivers/net/bonding/bond_alb.h
>>@@ -139,12 +139,23 @@ struct tlb_slave_info {
>>                        */
>> };
>>
>>+struct tlb_up_slave {
>>+      unsigned int    count;
>>+      struct rcu_head rcu;
>>+      struct slave    *arr[0];
>>+};
>>+
>> struct alb_bond_info {
>>       struct tlb_client_info  *tx_hashtbl; /* Dynamically allocated */
>>       spinlock_t              tx_hashtbl_lock;
>>       u32                     unbalanced_load;
>>       int                     tx_rebalance_counter;
>>       int                     lp_counter;
>>+      /* -------- non-dynamic tlb mode only ---------*/
>>+      struct tlb_up_slave __rcu *slave_arr;     /* Up slaves */
>>+      spinlock_t                slave_arr_lock; /* Lock to manage concurrent
>>+                                                 * writers
>>+                                                 */
>>       /* -------- rlb parameters -------- */
>>       int rlb_enabled;
>>       struct rlb_client_info  *rx_hashtbl;    /* Receive hash table */
>>diff --git a/drivers/net/bonding/bonding.h b/drivers/net/bonding/bonding.h
>>index 0b4d9cde0b05..ad9b3dce07d8 100644
>>--- a/drivers/net/bonding/bonding.h
>>+++ b/drivers/net/bonding/bonding.h
>>@@ -265,6 +265,12 @@ static inline bool bond_is_lb(const struct bonding *bond)
>>              BOND_MODE(bond) == BOND_MODE_ALB;
>> }
>>
>>+static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
>>+{
>>+      return (BOND_MODE(bond) == BOND_MODE_TLB)  &&
>>+             (bond->params.tlb_dynamic_lb == 0);
>>+}
>>+
>> static inline bool bond_mode_uses_arp(int mode)
>> {
>>       return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
>>--
>>2.0.0.526.g5318336
>
> ---
>         -Jay Vosburgh, jay.vosburgh@canonical.com

      reply	other threads:[~2014-07-10 14:40 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-07-09  1:09 [PATCH] bonding: Do not try to send packets over dead link in TLB mode Mahesh Bandewar
2014-07-09  7:10 ` Eric Dumazet
     [not found]   ` <CAF2d9jh1jDL7NMZapGn_Ohy8Y2JzHrWaKA5gFgR0tU=_KydtPg@mail.gmail.com>
2014-07-09 17:09     ` Eric Dumazet
2014-07-09 17:21       ` Mahesh Bandewar
2014-07-09 10:14 ` Nikolay Aleksandrov
2014-07-09 10:24 ` Veaceslav Falico
2014-07-09 10:25   ` Nikolay Aleksandrov
2014-07-09 12:04     ` Veaceslav Falico
2014-07-09 13:17       ` Eric Dumazet
2014-07-09 13:27         ` Veaceslav Falico
2014-07-09 17:15           ` Mahesh Bandewar
2014-07-09 17:24             ` Mahesh Bandewar
2014-07-09 21:07               ` Nikolay Aleksandrov
2014-07-10  8:25                 ` Maciej Żenczykowski
2014-07-09 16:52     ` Mahesh Bandewar
2014-07-10 13:39 ` Jay Vosburgh
2014-07-10 14:40   ` Mahesh Bandewar [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAF2d9jgQ=0e11hPvyBuh3FsPt1AzOk+UV0yZhtYNPWi6f_pGJQ@mail.gmail.com' \
    --to=maheshb@google.com \
    --cc=andy@greyhouse.net \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=jay.vosburgh@canonical.com \
    --cc=maze@google.com \
    --cc=netdev@vger.kernel.org \
    --cc=vfalico@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).